Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

A population-based cohort study evaluating the association between inhaled corticosteroid use and statin… Raymakers, Adam John Nelson 2017

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2017_september_raymakers_adam.pdf [ 2.85MB ]
Metadata
JSON: 24-1.0354577.json
JSON-LD: 24-1.0354577-ld.json
RDF/XML (Pretty): 24-1.0354577-rdf.xml
RDF/JSON: 24-1.0354577-rdf.json
Turtle: 24-1.0354577-turtle.txt
N-Triples: 24-1.0354577-rdf-ntriples.txt
Original Record: 24-1.0354577-source.json
Full Text
24-1.0354577-fulltext.txt
Citation
24-1.0354577.ris

Full Text

A	POPULATION-BASED	COHORT	STUDY	EVALUATING	THE	ASSOCIATION	BETWEEN	INHALED	CORTICOSTEROID	USE	AND	STATIN	USE	WITH	LUNG	CANCER	RISK	IN	CHRONIC	OBSTRUCTIVE	PULMONARY	DISEASE	PATIENTS			by	ADAM	JOHN	NELSON	RAYMAKERS	BSc	(Honours),	Dalhousie	University,	2004	MSc	Health	Economics,	Policy,	and	Law,	Erasmus	University,	2011		A	THESIS	SUBMITTED	IN	PARTIAL	FULFILLMENT	OF	THE	REQUIREMENTS	FOR	THE	DEGREE	OF		DOCTOR	OF	PHILOSOPHY	in	THE	FACULTY	OF	GRADUATE	AND	POSTDOCTORAL	STUDIES	(Pharmaceutical	Sciences)	THE	UNIVERSITY	OF	BRITISH	COLUMBIA	(Vancouver)	August	2017		©	Adam	John	Nelson	Raymakers,	2017	ii  Abstract		Background:	Lung	cancer	incidence	is	elevated	in	patients	with	chronic	obstructive	pulmonary	disease	(COPD),	often	due	to	smoking,	but	potentially	also	resulting	from	inflammation.	COPD	patients	are	also	often	diagnosed	with	comorbidities,	the	most	prominent	being	cardiovascular	disease	(CVD).	Statins,	due	to	the	increased	prevalence	of	CVD,	and	inhaled	corticosteroids	(ICS),	are	two	commonly	prescribed	medications	for	COPD	patients	that	may	reduce	lung	cancer	risk.			Objective:	To	evaluate	the	association	between	lung	cancer	risk	with	ICS	and	statin	use	in	COPD	patients.	A	priori,	the	hypothesis	of	this	study	was	that	use	of	these	medications	would	be	associated	with	a	reduction	in	lung	cancer	risk.		Methods:	This	study	used	population-based	data	for	the	province	of	British	Columbia	to	identify	a	cohort	COPD	patients.	To	be	included,	patients	were	to	have	filled	three	prescriptions	for	COPD-related	medications	within	a	twelve-month	period.	To	evaluate	the	association	between	statin	and	ICS	use	with	lung	cancer	risk,	an	array	of	methods	of	defining	medication	exposure	were	used,	including		a	novel	recency-weighed	approach.		Results:	In	the	analysis	evaluating	the	association	between	ICS	use	and	lung	cancer	diagnosis,	time-dependent	ICS	exposure	was	associated	with	a	30%	reduction	in	lung	cancer	risk.	The	recency-weighted	duration	of	use	exposure	metric	also	demonstrated	a	iii  protective	effect	from	ICS	exposure	(HR:	0.74	(95%	CI:	0.66-0.82).	This	protective	effect	was	consistent	over	all	exposure	metrics.	Evaluation	of	the	association	between	statin	use	and	lung	cancer	risk	produced	less	consistent	results.	However,	the	best-fitting	model	which	incorporated	the	recency-weighted	duration	of	use	exposure	metric	indicated	a	protective	effect	from	statin	exposure	(HR:	0.85	(95%	CI:	0.77-0.93).	Statin	exposure	in	patients	65	or	over	was	protective	against	lung	cancer	diagnosis	consistently	for	all	exposure	metrics.	An	interaction	term	between	ICS	and	statin	use	was	also	explored,	but	was	not	found	to	be	statistically	significant.			Conclusions:	These	results	suggest	that	the	benefits	of	ICS	and	statin	use	might	extend	beyond	their	primary	indication.	The	results	also	underscore	the	importance	of	using	appropriate	methods	for	measuring	medication	exposure	in	observational	studies,	particularly	those	using	administrative	data.	Finally,	this	work	highlights	the	importance	of	‘real-world’	evidence.		iv  Lay	Summary		 Chronic	obstructive	pulmonary	disease	(COPD)	is	a	debilitating	disease	that	is	associated	with	increased	patient	morbidity	and	mortality.	The	disease	is	also	associated	with	several	comorbidities,	with	one	of	the	most	common	being	cardiovascular	disease	(CVD).	Patients	with	COPD	face	a	higher	risk	of	lung	cancer,	partially	due	to	a	history	of	smoking.	However,	the	evidence	also	suggests	that	the	increased	risk	of	lung	cancer	extends	beyond	what	can	be	attributed	to	smoking.		Patients	with	COPD	are	often	prescribed	inhaled	corticosteroids	(ICS).	Many	COPD	patients	also	receive	statins,	due	to	the	presence,	or	risk	of,	CVD.		Evidence	suggests	that	ICS	and	statins	might	be	associated	with	reduced	lung	cancer	risk,	but	this	evidence	is	limited	in	its	generalizability.	Therefore,	this	study	aims	to	evaluate	whether	ICS	and	statin	use	in	COPD	patients	is	associated	with	a	reduced	risk	of	lung	cancer,	using	data	for	the	province	of	British	Columbia.	 v  Preface		This	dissertation	comprises	my	research	in	the	evaluation	of	the	association	between	inhaled	corticosteroid	and	statin	use	with	lung	cancer	risk	in	a	population-based	cohort	of	chronic	obstructive	pulmonary	disease	patients.	I	was	chiefly	responsible	for	all	data	cleaning	and	setup,	data	analysis,	compiling,	interpretation,	and	presentation	of	results.	Advice	and	suggestions	on	data	cleaning	and	the	analytic	approach	was	provided	from	Drs	Larry	Lynd,	Huiqing	(Kathy)	Li,	Mohsen	Sadatsafavi,	and	Ms	Maja	Grubisic.	Ethical	approval	was	granted	for	this	study	(certificate	number	H08-00241).		Chapter	1:	Introduction.	This	chapter	was	completed	by	Adam	Raymakers	with	comments	and	revisions	from	Drs	Larry	Lynd,	and	Mark	FitzGerald.		Chapter	2:	Adam	Raymakers	was	the	primary	author	of	this	chapter,	and	was	responsible	for	the	literature	search,	collation	of	research,	critical	review,	and	writing	of	the	chapter.	Revisions	and	comments	were	provided	by	Drs	Larry	Lynd,	Mohsen	Sadatsafavi,	and	Mark	FitzGerald.		.		Chapter	3:	The	primary	literature	search	was	conducted	by	Adam	Raymakers	with	Natalie	MacCormick	acting	as	second	reviewer.	Drs	Larry	Lynd,	Carlo	Marra,	Mark	FitzGerald,	and	Don	Sin,	all	provided	helpful	feedback	for	the	development	of	this	chapter.	This	chapter	vi  was	also	improved	through	the	peer-review	process	at	the	journal	Respirology,	where	it	has	been	published.		Chapter	4:	Adam	Raymakers	was	responsible	for	the	design,	analysis,	and	writing	of	this	study.	Dr	Mohsen	Sadatsafavi	provided	an	important	contribution	to	the	development	of	this	chapter.	In	addition,	Dr	Mary	De	Vera	and	Dr	Larry	Lynd	provided	valuable	insight	and	contributions	to	develop	this	chapter.	Dr	Don	Sin	provided	important	clinical	insight.	Two	referees	from	the	journal	Chest	also	provided	critical	peer-review	to	this	chapter.		Chapter	5	and	Chapter	6:	Adam	Raymakers	was	responsible	for	the	design,	statistical	analysis,	interpretation,	and	writing	of	these	two	chapters.	Drs	Larry	Lynd	and	Mohsen	Sadatsafavi	offered	critical	insight	and	feedback	to	develop	this	chapter.	Drs	Don	Sin	and	Mark	FitzGerald	were	a	valuable	source	of	clinical	input.	Dr	Carlo	Marra	also	provided	comments	and	feedback	during	the	development	of	these	two	chapters.	Drs	FitzGerald,	Lynd,	Marra,	Sadatsafavi,	and	Sin	all	reviewed	these	two	chapters	are	reviewed	these	two	chapters	to	ensure	their	quality.		Chapter	7:	Adam	Raymakers	was	responsible	for	the	writing	of	compiling	the	results	of	the	previous	chapters	and	integrating	them	into	this	final	chapter,	complete	with	limitations,	and	future	avenues	of	research.	Drs	Larry	Lynd,	Mark	FitzGerald,	and	Mohsen	Sadatsafavi	offered	helpful	feedback	in	the	development	and	completion	of	this	chapter.		vii  Table	of	Contents		Abstract	....................................................................................................................................................	ii	Lay	Summary	.........................................................................................................................................	iv	Preface	......................................................................................................................................................	v	Table	of	Contents	................................................................................................................................	vii	List	of	Tables	.........................................................................................................................................	xi	List	of	Figures	.....................................................................................................................................	xiii	List	of	Abbreviations	........................................................................................................................	xiv	Acknowledgements	............................................................................................................................	xv	Dedication	..........................................................................................................................................	xvii	Chapter	1:	Introduction	......................................................................................................................	1	1.1	 Background	..............................................................................................................................................	1	1.2	 The	role	of	local	and	systemic	inflammation	............................................................................	4	1.3	 Comorbidities	of	COPD	.......................................................................................................................	7	1.4	 Lung	cancer	in	COPD	patients	..........................................................................................................	9	1.5	 Inhaled	corticosteroids	....................................................................................................................	11	1.6	 Statins	......................................................................................................................................................	15	1.7	 Knowledge	gaps	..................................................................................................................................	19	1.8	 Specific	objectives	and	overview	of	thesis	..............................................................................	20	1.9	 Closing	remarks	..................................................................................................................................	22	Chapter	2:	A	review	of	methods	for	defining	medication	exposure	in	observational	studies	....................................................................................................................................................	24	2.1	 Introduction	..........................................................................................................................................	24	2.2	 Measures	of	medication	exposure	..............................................................................................	28	2.2.1	 Ever/never	use	..........................................................................................................................	28	2.2.2	 'Current'	medication	use	.......................................................................................................	31	2.2.3	 Cumulative	dose	........................................................................................................................	35	2.2.4	 Medication	adherence	and	discontinuation	.................................................................	37	2.2.5	 Recency-weighted	exposure	measures	...........................................................................	39	2.3	 Discussion	..............................................................................................................................................	43	2.4	 Concluding	remarks	..........................................................................................................................	45	Chapter	3:	Do	inhaled	corticosteroids	(ICS)	protect	against	lung	cancer	in	patients	with	chronic	obstructive	pulmonary	disease?	A	systematic	review	.................................	47	3.1	 Background	...........................................................................................................................................	47	3.2	 Methods	..................................................................................................................................................	52	3.2.1	 Search	strategy	..........................................................................................................................	52	3.2.2	 Study	selection	criteria	..........................................................................................................	54	3.2.3	 Data	extraction	..........................................................................................................................	54	viii  3.3	 Results	.....................................................................................................................................................	58	3.3.1	 Randomized	controlled	trials	..............................................................................................	59	3.3.2	 Observational	studies	.............................................................................................................	63	3.4	 Discussion	..............................................................................................................................................	69	3.4.1	 Limitations	..................................................................................................................................	73	3.5	 Closing	remarks	..................................................................................................................................	75	Chapter	4:	Mortality	in	COPD	patients	that	use	statins:	a	population-based	cohort	study	........................................................................................................................................................	76	4.1	 Introduction	..........................................................................................................................................	77	4.2	 Methods	..................................................................................................................................................	80	4.2.1	 Identification	of	the	COPD	cohort	.....................................................................................	81	4.2.2	 Identification	of	statin	users	................................................................................................	81	4.2.3	 Outcome	ascertainment	.........................................................................................................	84	4.2.4	 Adjustment	for	potential	confounders	............................................................................	84	4.2.5	 Statistical	analyses	...................................................................................................................	85	4.2.6	 Sensitivity	analyses	.................................................................................................................	85	4.3	 Results	.....................................................................................................................................................	86	4.3.1	 Cohort	of	COPD	patients	........................................................................................................	86	4.3.2	 Statin	use	in	the	cohort	of	COPD	patients	......................................................................	88	4.3.3	 All-cause	and	pulmonary-related	mortality	.................................................................	90	4.3.4	 Sensitivity	analyses	.................................................................................................................	91	4.3.4.1	 Exposure	ascertainment	window	...................................................................................	91	4.3.4.2	 Statin	adherence	...................................................................................................................	93	4.3.4.3	 Statin	cumulative	dose	.......................................................................................................	95	4.4	 Discussion	..............................................................................................................................................	95	4.4.1	 Strengths	and	limitations	......................................................................................................	98	4.5	 Concluding	remarks	.......................................................................................................................	100	Chapter	5:	An	evaluation	of	inhaled	corticosteroid	use	and	lung	cancer	risk	in	chronic	obstructive	pulmonary	disease	patients:	a	population-based	cohort	study101	5.1	 Introduction	.......................................................................................................................................	102	5.2	 Methods	...............................................................................................................................................	106	5.2.1	 Database	description	...........................................................................................................	106	5.2.2	 Cohort	identification	............................................................................................................	107	5.2.3	 Inhaled	corticosteroid	use	.................................................................................................	108	5.2.4	 Latency	period	........................................................................................................................	109	5.2.5	 Exposure	measurement	......................................................................................................	110	5.2.6	 Adjustment	for	potential	confounders	.........................................................................	113	5.2.7	 Statistical	analysis	.................................................................................................................	114	5.2.8	 Secondary	analysis:	medication	possession	ratio	...................................................	115	ix  5.2.9	 Sub-group	analysis:	lung	cancer	histology	.................................................................	115	5.2.10	 Sensitivity	analyses	..............................................................................................................	115	5.3	 Results	..................................................................................................................................................	116	5.3.1	 ICS	use	in	the	COPD	cohort	................................................................................................	119	5.3.2	 Lung	cancer	among	COPD	patients	................................................................................	120	5.3.3	 Bivariate	results:	potential	confounders	.....................................................................	120	5.3.4	 Bivariate	results:	exposure	definitions	........................................................................	122	5.3.5	 Multivariable	analysis:	main	results	.............................................................................	123	5.3.6	 Secondary	analysis:	medication	possession	ratio	...................................................	126	5.3.7	 Lung	cancer	histology	..........................................................................................................	127	5.3.8	 Sensitivity	analyses	..............................................................................................................	128	5.4	 Discussion	...........................................................................................................................................	129	5.4.1	 Strengths	and	limitations	...................................................................................................	134	5.5	 Concluding	remarks	.......................................................................................................................	136	Chapter	6:	An	evaluation	of	the	association	between	statin	use	and	lung	cancer	risk	in	chronic	obstructive	pulmonary	disease	patients:	a	population-based	cohort	study138	6.1	 Introduction	.......................................................................................................................................	139	6.2	 Methods	...............................................................................................................................................	143	6.2.1	 Latency	period	........................................................................................................................	143	6.2.2	 Exposure	measurement	......................................................................................................	144	6.2.3	 Adjustment	for	potential	confounders	.........................................................................	147	6.2.4	 Statistical	analysis	.................................................................................................................	148	6.2.5	 Secondary	analysis:	medication	possession	ratio	...................................................	148	6.2.6	 Sub-group	analysis:	lung	cancer	histology	.................................................................	149	6.2.7	 Sensitivity	analyses	..............................................................................................................	149	6.2.8	 Negative	control	exposure	.................................................................................................	150	6.3	 Results	..................................................................................................................................................	151	6.3.1	 Statin	use	in	the	COPD	cohort	..........................................................................................	151	6.3.2	 Bivariate	and	age/sex	adjusted	results:	statin	exposure	definitions	..............	154	6.3.3	 Multivariable	analysis	.........................................................................................................	157	6.3.4	 Lung	cancer	histology	..........................................................................................................	160	6.3.5	 Sensitivity	analyses	..............................................................................................................	161	6.4	 Discussion	...........................................................................................................................................	162	6.4.1	 Strengths	and	limitations	...................................................................................................	166	6.5	 Conclusions	........................................................................................................................................	169	Chapter	7:	Discussion	and	Conclusions	....................................................................................	170	7.1	 Research	findings	and	implications	........................................................................................	170	7.2	 Research	contributions	.................................................................................................................	177	7.3	 Limitations	of	this	research	........................................................................................................	183	x  7.4	 Future	direction	of	research	.......................................................................................................	185	7.5	 Knowledge	translation	..................................................................................................................	187	7.6	 Conclusions	........................................................................................................................................	188	Bibliography	......................................................................................................................................	189	Appendices	........................................................................................................................................	208	Appendix	A	Primary	search	strategy	.....................................................................................................	208	Appendix	B	Risk	of	bias	assessment	......................................................................................................	217	Sub-appendix:	B1	Summary	table	.....................................................................................................	217	Sub-appendix:	B2	Detail	table	of	bias	assessment	(randomized	controlled	trial	studies	and	observational	studies)	(110)	......................................................................................................	218	Appendix	C	Preferred	Reporting	Items	for	Systematic	Reviews	and	Meta-Analyses	(PRISMA)	Checklist	.......................................................................................................................................	221		xi  List	of	Tables	Table	1.1.	GOLD	stages	of	disease.	.....................................................................................................................	2	Table	1.2.	Lung	cancer	risk	and	associated	levels	of	systemic	inflammation.	.............................	13	Table	1.3.	Studies	reporting	associations	between	inhaled	corticosteroid	use	and	systemic	inflammation.	..........................................................................................................................................................	14	Table	1.4.	Studies	reporting	associations	between	statin	use	and	systemic	inflammation.	.	18	Table	3.1.	Characteristics	of	included	studies.	..........................................................................................	56	Table	3.2.	Characteristics	of	ICS	use	among	patients	with	COPD	in	individual	identified	studies.	........................................................................................................................................................................	62	Table	3.3.	Results	from	identified	studies	for	the	relative	risk	of	lung	cancer	diagnosis	for	ICS	exposed/treated	and	unexposed/untreated	patients	with	COPD.	...........................................	68	Table	3.4.	Results	from	identified	studies	for	the	relative	risk	of	lung	cancer	specific	death	in	for	ICS	treated	and	untreated	patients	with	COPD.	...........................................................................	74	Table	4.1.	Patient	characteristics,	stratified	by	exposure	status	to	statins.	.................................	87	Table	4.2.	Hazard	ratios	from	bivariate	Cox	regression	analysis	to	assess	potential	confounders,	with	time	to	all-cause	mortality	as	the	outcome	variable.	......................................	89	Table	4.3.	Hazard	ratios	obtained	from	multivariable	Cox	regression	analysis	for	association	between	statin	exposure	and	all-cause	mortality	...........................................................	90	Table	4.4.	Multivariable	Cox	regression	analysis	for	the	association	between	statin	exposure	and	pulmonary-related	mortality).	............................................................................................	92	Table	4.5.	Multivariable	Cox	regression	models	with	time	to	all-cause	mortality	as	the	outcome	using	alternative	specifications	for	the	exposure	variable.	.............................................	93	Table	5.1.	Demographics	of	the	COPD	cohort	.	......................................................................................	118	Table	5.2.	Bivariate	regression	results	for	covariates	considered	for	inclusion	in	the	multivariable	model,	with	time	to	lung	cancer	diagnosis	as	the	outcome	variable.	..............	121	Table	5.3.	Bivariate,	and	age	and	sex	adjusted,	regression	results	(hazard	ratio	and	95%	CI)	for	each	ICS	exposure	definition	with	time	to	lung	cancer	diagnosis	as	the	outcome	variable......................................................................................................................................................................................	123	Table	5.4.	Multivariable	results	for	each	ICS	exposure	metric	and	associated	Akaike	Information	Criterion	values,	with	time	to	lung	cancer	diagnosis	as	the	outcome	variable......................................................................................................................................................................................	125	Table	5.5.	Multivariable	regression	analysis	using	the	medication	possession	ratio	to	capture	exposure	to	ICS.	..................................................................................................................................	126	Table	5.6.	Sub-group	analyses	based	on	lung	cancer	histology.	Multivariable	regression	analysis	with	time	to	NSCLC	or	SCLC	diagnosis	as	the	outcome	variables.	...............................	128	Table	5.7.	Sensitivity	analyses	of	different	lengths	of	the	latency	period	using	the	time-dependent	ever	metric	of	ICS	medication	exposure.	...........................................................................	130	Table	6.1.	Demographics	of	the	COPD	cohort.	........................................................................................	152	xii  Table	6.2.	Bivariate,	and	age/sex	adjusted	regression	results		for	each	exposure	definition	with	time	to	lung	cancer	diagnosis	as	the	outcome.	............................................................................	155	Table	6.3.	Bivariate	regression	model	results,	with	time	to	lung	cancer	diagnosis	as	the	outcome,	for	covariates	to	be	considered	for	inclusion	in	the	multivariable	model.	............	156	Table	6.4.	Multivariable	regression	results	for	each	statin	exposure	metric	with	time	to	lung	cancer	diagnosis	as	the	outcome	variable.	..............................................................................................	158	Table	6.5.	Evaluation	of	the	medication	possession	ratio	to	capture	exposure	to	statins	over	the	study	follow-up	period,	and	lung	cancer	risk.	................................................................................	159	Table	6.6.	Evaluation	of	association	between	statin	exposure	and	lung	cancer	histology.	160	Table	6.7.	Sensitivity	analyses:	evaluation	of	different	lengths	of	the	latency	period	and	a	cohort	age	restriction,	using	time-dependent	exposure	and	the	recency-weighted	duration	of	exposure	metrics,	with	time	to	lung	cancer	as	the	outcome.	.....................................................	163	Table	7.1.	Key	findings	for	each	specific	chapter	of	this		dissertation	.	.......................................	176		xiii  List	of	Figures	Figure	1.1.	Systemic	inflammation	is	associated	with	several	conditions	in	COPD	patients,	as	well	as	a	decreased	quality	of	life.	................................................................................................................	8	Figure	1.2.	COPD	is	associated	with	systemic	inflammation	and	systemic	inflammation	is	associated	with	increased	lung	cancer	risk.	Inhaled	corticosteroids	and	statins	have	anti-inflammatory	properties	and,	thus,	have	the	potential	to	reduce	lung	cancer	risk.	................	19	Figure	2.1.	Induction	and	latency	periods	associated	with	the	onset	of	a	particular	disease.	In	this	example,	the	‘exposure’	is	that	which	begins	the	disease	process.	....................................	27	Figure	2.2.	Using	a	fixed,	‘ever/never’	exposure	definition,	a	patient	can	be	considered	exposed	if	ever	having	been	dispensed	the	medication	in	the	follow-up	period.	.....................	29	Figure	2.3.	Exposure	and	outcome	ascertainment	windows.	.............................................................	31	Figure	2.4.	'Current'	medication	use.	............................................................................................................	32	Figure	2.5.	Addition	of	a	‘lag’	time	as	a	variation	on	‘current’	use.	..................................................	34	Figure	2.6.	Calculation	of	medication	adherence	using	the	medication	possession	ratio	(MPR).	.........................................................................................................................................................................	38	Figure	3.1.	Study	selection	process.	...............................................................................................................	53	Figure	4.1.	Exposure	and	outcome	ascertainment	periods	in	the	primary	analysis.	...............	83	Figure	4.2.	Distribution	of	statins	prescribed	within	the	one-year	exposure	ascertainment	window	in	the	cohort	of	COPD	patients.	......................................................................................................	88	Figure	4.3.	Number	of	statin	exposed	patients,	by	MPR	category,	in	the	1-year	exposure	ascertainment	window.	.......................................................................................................................................	94	Figure	5.1.	The	basic	conceptual	framework	for	the	study	presented	in	Chapter	5.	............	106	Figure	5.2.	The	latency	period	associated	with	medication	exposure	and	lung	cancer	diagnosis.	................................................................................................................................................................	109	Figure	5.3.	Distribution	of	all	ICS	prescriptions	dispensed	to	the	COPD	cohort	within	the	follow-up	period.	.................................................................................................................................................	117	Figure	5.4.	Medication	Possession	Ratio	(MPR)	for	ICS	use	in	the	COPD	cohort	over	the	study	follow-up	period,	by	category.	..........................................................................................................	119	Figure	5.5.	Distribution	of	lung	cancer	cases,	according	to	histology.	.........................................	127	Figure	6.1.	Conceptual	framework	for	the	analysis	presented	in	Chapter	6.	...........................	142	Figure	6.2.	A	graphical	representation	of	the	latency	period,	and	how	medication	exposure	is	considered	with	respect	to	this	latency	period.	................................................................................	144	Figure	6.3.	Distribution	of	all	statins	dispensed	among	statin	users	in	the	COPD	cohort.	.	153	Figure	6.4.	Medication	possession	ratio	(MPR)	for	statin	users	in	the	COPD	cohort.	...........	153	Figure	7.1.	The	conceptual	framework	for	this	thesis..	......................................................................	172		xiv  List	of	Abbreviations	 AECOPD:	Acute	Exacerbation	of	Chronic	Obstructive	Pulmonary	Disease	AIC:	Akaike	Information	Criterion	CI:	Confidence	Interval	CCB:	Calcium	Channel	Blocker	CCI:	Charlson	Comorbidity	Index		COPD:	Chronic	Obstructive	Pulmonary	Disease	CVD:	Cardiovascular	Disease	EMBASE:	Excerpta	Medica	Database	FEV1:	Forced	Expiratory	Volume	in	1	Second	HR:	Hazard	Ratio	ICD	(9	and	10):	International	Statistical	Classification	of	Diseases	and	Related	Health	Problems	ICS:	Inhaled	Corticosteroid	IQR:	Interquartile	Range	MPR:	Medication	Possession	Ratio	MSP:	Medical	Services	Plan	HR:	Hazard	Ratio	MEDLINE:	Medical	Literature	Analysis	and	Retrieval	System	Online	MI:	Myocardial	Infarction	OR:	Odds	Ratio	PRISMA:	Preferred	Reporting	Standards	for	Systematic	Reviews	and	Meta-Analyses	RCT:	Randomized	Control	Trial	RR:	Relative	Risk	SABA:	Short-Acting	Beta	Agonist	SD:	Standard	Deviation	STATCOPE:	Simvastatin	for	the	Prevention	of	Exacerbations	in	Moderate	to	Severe	COPD	TORCH:	Towards	a	Revolution	in	COPD	Health	xv  Acknowledgements	 I	want	to	thank	Dr.	Larry	Lynd	for	his	supervision	over	the	last	number	of	years.	I	am	truly	thankful	for	his	support	and	encouragement	along	this	path;	his	guidance,	feedback	and	direction	have	helped	to	achieve	my	goal.	I	am	indebted	to	him	for	his	assistance	and	any	of	my	future	successes	can,	in	part,	be	attributed	to	his	supervision.		I	would	like	to	thank	Drs	Carlo	Marra	and	Mark	FitzGerald	for	initially	hiring	me	for	a	staff	position	at	the	Collaboration	for	Outcomes	Research	and	Evaluation	(CORE).	They	gave	me	a	great	opportunity	to	gain	experience	in	fields	of	health	economics	and	outcomes	research,	and	were	both	instrumental	in	helping	make	the	decision	to	move	on	to	embarking	on	a	PhD.	I	also	want	to	thank	both	of	them	for	their	continued	interest	in	my	development	by	agreeing	to	serve	as	members	of	my	PhD	Supervisory	Committee.	I	would	also	like	to	thank	Drs	Mohsen	Sadatsafavi	and	Don	Sin	for	their	support	and	guidance	as	part	of	my	PhD	Committee,	your	insights	into	the	methodological	and	clinical	issues	associated	with	the	work	that	comprised	my	dissertation	were	invaluable	to	completing	this	work.	I	would	also	like	to	thank	Dr	Huiqing	(Kathy)	Li	for	her	support	in	completing	the	statistical	analysis	conducted	in	this	dissertation,	particularly	in	the	latter	stages.			I	would	also	like	to	extend	my	thanks	to	my	colleagues	at	CORE	for	their	support	and	for	creating	an	environment	that	is	conducive	to	success.	The	CORE	staff	made	being	in	the	office	an	enjoyable	and	pleasant	experience.		xvi  Finally,	I	would	like	to	acknowledge	the	support	my	friends	and	family.	Their	support	in	helping	me	to	complete	this	work,	but	also	to	remind	of	the	importance	of	a	balanced	life,	is	truly	invaluable.	Thank	you.	xvii  Dedication		I	would	like	to	dedicate	this	dissertation	to	my	parents,	Arno	and	Kim	Raymakers,	my	aunt,	Gail	Rudderham,	and	my	partner,	Angelina	Woof;	you	have	all	been	instrumental	to	the	completion	of	this	work.	Your	support,	patience,	criticisms,	inspiration,	and	patience,	is	all	very	much	appreciated.	This	endeavor,	and	any	future	endeavors,	are	possible	because	of	you	all.	Thank	you.			 1	Chapter	1: Introduction	 1.1 Background	Chronic	obstructive	pulmonary	disease	(COPD)	is	a	progressive	and	mostly	irreversible	disease	comprising	emphysema	and	chronic	bronchitis,	and	is	associated	with	considerable	morbidity	and	mortality	(1).	The	prevalence	of	COPD	is	approximately	10.1%	worldwide,	with	males	having	a	higher	prevalence	than	females	(11.7%	versus	8.5%)	(2).	In	Canada,	the	estimates	of	prevalence	are	similar,	with	studies	reporting	a	prevalence	of	9.3%	for	males,	and	7.3%	for	females	(2).	These	numbers	have	recently	been	rising,	with	a	shift	in	the	prevalence	toward	females	from	males	(3).	Increasing	numbers	of	COPD	patients	will	pose	increasingly	difficult	problems	to	health	systems	worldwide	with	projections	estimating	that	the	number	of	cases	of	COPD	will	increase	to	approximately	5.8	million	in	the	next	20	years	(4).	The	societal	economic	burden	of	this	disease,	which	is	already	substantial,	has	been	commensurately	projected	to	increase	from	$4.5	to	$7.3	billion	($CAD)	in	Canada	(4).				COPD	is	diagnosed	according	to	a	post-bronchodilator	ratio	of	forced	expiratory	volume	in	1	second	(FEV1)	to	forced	vital	capacity	(FVC)	of	less	than	0.70.	Forced	expiratory	volume	is	further	utilized	to	provide	a	stage	of	the	disease	according	to	the	degree	of	airflow	limitation.	The	Global	Initiative	for	Chronic	Obstructive	Lung	Disease	(GOLD)	guidelines	(5,6)	state	that	Stage	1	(mild)	disease	patients	have	post-bronchodilator	FEV1	of	≥	80%	of		 2	predicted.	Stage	2	(moderate)	disease	patients	have	FEV1	less	than	80%	and	above	50%.	Stage	3	(severe)	disease	patients	have	FEV1	greater	than	30%	and	less	than	50%	of	predicted.	Stage	4	(very	severe)	patients	have	FEV1	of	less	than	30%	of	predicted	(5,6).	These	stages	are	summarized	in	Table	1.1	below.			Table	1.1.	GOLD	stages	of	disease.	GOLD	Stage	 Description	 FEV1*		 	 	1	 Mild	 FEV1	≥	80%	predicted	2	 Moderate	 FEV1	<	80%	and	≥	50%	predicted	3	 Severe	 FEV1	<	50%	and	≥	30%	predicted	4	 Very	Severe	 FEV1	<	30%	predicted	*	Post-bronchodilator.	GOLD:	Global	Initiative	for	Chronic	Obstructive	Lung	Disease;	FEV1:	Forced	Expiratory	Volume	in	1	second.			The	treatment	of	COPD	is	largely	based	on	the	aforementioned	disease	staging,	and	often	also	considers	the	risk	and	history	of	acute	exacerbations	associated	with	the	disease.	Pharmacologic	treatment	for	patients	with	Stage	1	disease	typically	comprises	short-acting	anticholinergics	and	short-acting	beta-agonists	(SABAs).	Patients	with	Stage	2	disease	can	be	prescribed	long-acting	anticholinergics	or	long-acting	beta-agonists	(LABA)	(6)	while		 3	Stage	3	and	4	disease	call	for	inhaled	corticosteroids	(ICS)	in	combination	with	a	long-acting	beta-agonist	or	a	long-acting	anticholinergic	(6).			While	these	criteria	help	to	classify	COPD	patients,	it	should	be	noted	that	COPD	patients	are	heterogeneous	and	the	disease	may	have	several	likely	phenotypes	(7).	Han	et	al.	(7)	reported	that	FEV1,	as	a	measure	of	lung	function,	does	not	comprehensively	describe	this	heterogeneity.	The	authors	suggest	that	(frequent)	acute	exacerbations	can	be	considered	as	an	outcome	of	COPD	but	also	that	exacerbations	may	be	a	manifestation	of	a	specific	phenotype	of	COPD	patients.	This	phenotype	may	also	be	linked	to	systemic	inflammation,	discussed	in	Section	1.2	below.			Chronic	obstructive	pulmonary	disease,	similar	to	many	chronic	diseases,	typically	progresses	in	severity	with	the	duration	of	disease	(8).	As	a	result,	it	is	expected	that	lung	function	will	decline	and	patients'	outcomes	will	worsen	as	they	become	older.	Mannino	et	al.	(9)	showed	that	declining	lung	function	is	a	component	of	the	natural	history	of	COPD	and	is	associated	with	increased	morbidity	and	mortality.	As	noted	above,	severe	acute	exacerbations	are	a	major	concern	for	patients	with	COPD.	These	acute	exacerbations	of	COPD	(AECOPD)	also	become	increasingly	likely	as	lung	function	declines	and	previous	AECOPD	tend	to	be	the	best	predictor	of	future	exacerbations	(10).	Patients	who	experience	acute	exacerbations,	particularly	repeat/frequent	acute	exacerbations,	are	often	hospitalized	and	have	an	increased	risk	of	mortality.	As	such,	Suissa	et	al.	(10)	state	that	repeated	acute	exacerbations	are	associated	with	a	rapid	decline	in	health	status	and	that	considerable	effort	should	be	directed	toward	alleviating	the	potential	for	exacerbations	to		 4	improve	survival.	Similarly,	Najafzadeh	et	al.	(4)	used	a	dynamic	simulation	model,	populated	with	data	from	Canada,	to	estimate	the	impact	of	several	(hypothetical)	interventions	on	the	future	economic	burden	of	COPD	in	Canada	over	a	25-year	time	horizon	and	found	that	strategies	to	reduce	acute	exacerbations	of	COPD	would	have	the	greatest	effect	in	reducing	this	burden	over	this	period	of	time.	Recently,	model-based	projections	have	estimated	that	inpatient	hospitalizations	among	COPD	patients,	typically	resulting	from	AECOPD,	will	increase	from	150-185%	by	2030	(11),	thereby	highlighting	the	impact	of	AECOPD,	not	only	in	terms	of	patients’	outcomes,	but	also	to	health	systems	and/or	payers.		1.2 The	role	of	local	and	systemic	inflammation	A	COPD	diagnosis	is	characterized	by	reduced	lung	function	due	to	localized	inflammation	which	is	often	a	result	of	a	patients’	current	smoking	status	or	previous	smoking	history.	The	negative	aspects	of	smoking	behaviour,	however,	extend	beyond	this	localized	inflammation	as	levels	of	systemic	inflammation	are	also	increased	in	subjects	with	a	history	of	smoking	(12).	However,	smoking	history	does	not	account	entirely	for	the	increased	levels	of	systemic	inflammation	in	COPD	patients,	as	evidence	suggests	that	levels	of	systemic	inflammation	observed	in	COPD	patients	are	greater	than	that	attributable	to	patients'	smoking	history	or	current	status	(13).			While	airflow	limitation	might	be	intuitively	thought	of	as	affected	by	localized	inflammation	in	the	airway,	moderate	and	severe	airflow	limitation	has	further	been		 5	associated	with	increased	levels	of	systemic	inflammation.	For	example,	Sin	et	al.	(14)	found	that	those	with	severe	airflow	limitation	(defined	as	FEV1	≤	50%)	were	two	times	more	likely	to	have	elevated	levels	of	systemic	inflammation,	measured	by	levels	of	circulating	C-reactive	protein	(CRP)	levels	(OR:	2.18	(95%	CI:		1.46-3.27))	than	subjects	with	no	airflow	obstruction.	However,	the	heterogeneity	of	COPD	(as	noted	previously)	means	that	systemic	inflammation	is	variable	in	COPD	patients	(depending	on	the	marker	that	is	chosen)	and	there	is	no	clear	relationship	between	any	particular	aspect	of	COPD	and	systemic	inflammation	(7).	However,	in	those	COPD	patients	with	increased	levels	of	systemic	inflammation,	this	is	an	important	issue	due	to	an	observed	association	with	reduced	survival	(15).		The	presence	of	increased	levels	systemic	inflammation	in	COPD	patients	has,	in	recent	years,	brought	forth	a	new	conceptualization	of	COPD	whereby	a	specific	phenotype	of	COPD	may	exist	that	might	be	characterized	by	systemic	inflammation	(16).	Whether	this	is	a	specific	phenotype	or	simply	a	more	severe	disease	state	remains	uncertain	but	markers	for	systemic	inflammation	appear	to	be	more	present	in	patients	that	experience	acute	exacerbations	and	with	later	stage	(more	severe)	disease	(13).	Fabbri	et	al.	(17)	suggest	that	adding	the	term	'chronic	systemic	inflammatory	syndrome'	to	COPD	a	diagnosis	may	help	convey	the	strong	association	between	inflammation	and	the	relationship	to	comorbidities	in	patients	with	COPD.	Agusti	et	al.	(18)	report	that	in	a	subgroup	of	COPD	patients	with	persistent	systemic	inflammation,	there	was	an	association	with	poor	clinical	outcomes	compared	to	those	with	no	evidence	of	systemic	inflammation,	despite	similar		 6	lung	function1.	Further,	Wouters	et	al.	(19)	suggest	that	understanding	the	role	of	systemic	inflammation,	and	not	simply	focusing	on	respiratory	symptoms,	is	important	in	understanding	the	aetiology	of	acute	exacerbations	in	COPD	patients.			There	are	differing	views	of	where	systemic	inflammation	fits	in	the	disease	course	in	COPD.	The	question	has	been	asked,	does	COPD	result	in	systemic	inflammation	or	is	COPD	a	manifestation	of	systemic	inflammation	(13)?	This	may	have	implications	for	how	the	disease	is	managed	and	treated.			While	the	relationship	between	elevated	levels	of	inflammation	and	lung	cancer	has	been	recognized,	the	exact	mechanism	by	which	inflammation	increases	lung	cancer	risk	is	less	well	understood.	There	are	several	possible	links	that	have	been	put	forward	which	might	explain	the	association	between	inflammation	in	the	lungs	of	COPD	patients	and	increased	lung	cancer	risk.	Two	potential	causes	of	increased	lung	cancer	risk	resulting	from	inflammation	associated	with	COPD	are	increased	oxidative	stress	and	surfactant	dysregulation	in	the	lungs,	which	are	associated	with	markers	of	elevated	systemic	inflammation	(12,20–22).	In	particular,	increased	oxidative	stress	has	been	associated	with	a	reduction	in	apoptosis	(20)	while	surfactant	dysregulation	reduces	the	ability	of	pulmonary	surfactant	protein	to	attenuate	inflammation	in	the	lungs.	Additional	hypotheses	have	been	put	forth	that	inflammation	in	the	airway	and	in	the	lungs	might	                                                1	In	this	study,	systemic	inflammation	was	quantified	using	six	different	biomarkers:	white	blood	cell,	C-reactive	protein,	interleukin-6,	interleukin-8,	fibrinogen,	and	TNF-alpha	levels	measured	in	peripheral	blood.		 7	stimulate	recurring	injury	and	repair	of	cells	in	the	lungs,	leading	to	uncontrolled	cell	growth	(18),	or	that	inflammation	in	the	lungs	may	induce	the	activation	of	proteins	that	promote	oncogenesis	(23).	In	terms	of	genetic	factors,	research	suggests	that	the	regulation	of	anti-oxidant	genes	in	patients	with	COPD	is	impaired,	thereby	increasing	oxidative	stress,	and	increasing	the	risk	of	lung	cancer	(13).	Therefore,	there	are	several	potential	explanations	for	how	local	and	systemic	inflammation	might	make	the	lungs	susceptible	to	processes	that	increase	lung	cancer	risk,	but	the	exact	mechanism	remains	an	area	for	future	study	(18,23).			 1.3 Comorbidities	of	COPD	There	are	several	comorbidities	common	among	COPD	patients,	including:	ischemic	heart	disease,	osteoporosis,	depression,	lung	cancer,	and	diabetes	mellitus	(13).	For	example,	Mapel	et	al.	(24)	estimated	that	COPD	patients	had	a	higher	average	prevalence	of	additional	medical	conditions	(3.7)	compared	to	age	and	sex	matched	controls	(1.7,	p<0.001)	and	that	only	6%	of	COPD	patients	reported	no	additional	medical	condition.	The	role	of	comorbidities	in	COPD	is	an	important	consideration	for	patients'	health	and	as	such,	all-cause	mortality	has	largely	become	the	preferred	metric	for	outcomes	in	COPD	patients	(25).				 8	 Figure	1.1.	Systemic	inflammation	is	associated	with	several	conditions	in	COPD	patients,	as	well	as	a	decreased	quality	of	life.		Cardiovascular	disease	(CVD)	is	among	the	most	common	comorbidities	present	in	COPD	patients.	The	prevalence	of	cardiovascular	disease	in	patients	with	COPD	is	elevated	compared	to	the	general	population	of	similar	age	and	sex	(14,26,27).	Mannino	et	al.	(27)	estimated	that	the	prevalence	of	CVD	in	a	cohort	of	COPD	patients	was	15.2%	in	the	United	States.	A	meta-analysis	by	Chen	et	al.	(28)	reported	that	subjects	with	COPD	were	more	than	two	times	more	likely	to	have	CVD	compared	to	subjects	without	COPD	(OR:	2.46	(95%	CI	2.02–3.00)).	Similar	results	were	reported	by	Curkendall	et	al.	(26)	who	found	comparable	increased	odds	for	several	cardiovascular	diseases	in	COPD	patients	using	a	population-based	cohort	study	design	in	Saskatchewan,	Canada.	In	that	study,	the	estimated	adjusted	odds	ratios	for	subjects	with	COPD	compared	to	subjects	without	COPD	↑	Lung	CancerSYSTEMIC	INFLAMMATION↑	Cardiovascular	Disease↑	AECOPD↑	Diabetes↓	Quality	of	Life	 9	ranged	from	1.11	(95%	CI:	1.02–1.21)	for	stroke	to	5.46	(95%	CI:	4.25–7.02)	for	pulmonary	embolism	(26).	The	presence	of	CVD	in	these	patients	may	largely	be	due	to	their	smoking	history	but	there	is	also	a	suggestion	of	a	relationship	between	lung	inflammation,	systemic	inflammation,	and	vascular	inflammation	(29).	The	additional	risk	of	CVD	results	in	patients	with	COPD	often	being	prescribed	an	HMG-CoA	reductase	inhibitor	(statins).	Statins	may	be	prescribed	for	primary	or	secondary	prevention	of	CVD.	Secondary	prevention	is	typically	delineated	by	a	patient	having	had	a	cardiac	event	prior	to	their	first	statin	prescription,	whereas	primary	prevention	patients	are	those	at	risk	for	such	events	and	are	prescribed	statins,	typically	based	on	levels	of	low-density	lipoprotein	levels	(30).			1.4 Lung	cancer	in	COPD	patients	Lung	cancer	is	the	second	most	common	diagnosed	cancer	among	men	and	women	and	the	most	common	cause	of	cancer-related	mortality	(31).	Although	smoking	is	a	risk	factor	for	both	COPD	and	lung	cancer,	evidence	suggests	that	COPD	itself	is	an	additional	risk	factor	for	lung	cancer	development,	independent	of	smoking	status	or	history	(32–35).	For	example,	Wasswa-Kintu	et	al.	(36)	conducted	a	meta-analysis	that	showed	reduced	lung	function	is	strongly	associated	with	lung	cancer	development.	One	hypothesis	for	the	underlying	mechanism	of	this	association	that	has	been	postulated	is	that	reduced	lung	function	may	result	in	an	inability	to	clear	carcinogens	from	the	airway	(37).	Elevated	markers	for	inflammation	also	appear	to	be	associated	with	lung	cancer	development.	Young	et	al.	(32)	reported	a	six-fold	greater	prevalence	of	COPD	in	lung	cancer	subjects	versus	smoking-status,	sex,	and	age-matched	controls.	Using	data	collected	from	5402		 10	participants	in	the	First	National	Health	and	Nutrition	Examination	Survey	in	the	United	States,	Mannino	et	al.	(33)	found	that	moderate	or	severe	COPD	was	associated	with	an	increased	risk	of	lung	cancer	diagnosis	(HR:	2.8	(95%	CI:	1.8-4.4)),	even	after	adjustment	for	age,	sex,	smoking	status,	duration	of	smoking,	and	smoking	intensity.	A	longitudinal	study	of	176,997	men	yielded	similar	results,	with	both	mild	and	moderate/severe	COPD	associated	with	increased	smoking-adjusted	rates	of	lung	cancer	(RR:	1.5	(95%	CI:	1.2-1.9)	and	RR:	2.2	(95%	CI:	1.8-2.7),	respectively)	(38).	A	longitudinal	study	by	Skillrud	et	al.	(34)	reported	the	smoking-adjusted	probability	of	developing	lung	cancer	was	10.8%	in	COPD	patients	and	2.5%	in	controls	(p=0.023).	Turner	et	al.	(35),	in	a	study	of	448,600	lifelong	non-smokers	in	the	US,	found	that	lung	cancer	mortality	was	significantly	associated	with	combined	emphysema/chronic	bronchitis	(HR:	2.44	(95%	CI:	1.22-4.90)),	over	a	twenty-year	follow-up	period	(35).		The	evidence	from	these	studies	suggests	that	the	association	between	COPD	and	lung	cancer	is	strong	and	while	smoking	is	a	known	risk	factor	for	both	diseases,	this	common	factor	alone	is	not	sufficient	to	explain	all	lung	cancer	cases	(39).	Several	studies	have	linked	increased	markers	of	systemic	inflammation	(specifically	CRP)	to	lung	cancer	development	(see	Table	1.2).	Chaturvedi	et	al.	(40)	showed	that,	using	data	from	the	Prostate,	Lung,	Colorectal,	and	Ovarian	(PLCO)	Cancer	Screening	Trial,	that	subjects	in	the	highest	quartile	of	CRP	levels	had	an	approximately	two	times	greater	risk	of	lung	cancer	than	those	in	the	lowest	quartile,	even	after	adjusting	for	smoking	status	(OR:	1.98	(95%	CI:	1.35-2.89)).	A	study	by	Pine	et	al.	(41),	used	serum	IL-6	and	IL-8	as	biomarkers	for		 11	inflammation	in	two	separate	patient	populations2.	The	results	of	their	analysis	suggest	that	both	biomarkers	were	significantly	associated	with	an	increased	odds	of	lung	cancer.	In	the	first	analysis,	patients	in	the	highest	quartile	of	IL-6	levels	had	three	times	greater	risk	than	those	in	the	lowest	quartile	and	a	two	times	greater	risk	for	IL-8.	In	the	second	analysis,	patients	in	the	highest	quartile	of	both	IL-6	and	IL-8	had	50%	increased	odds	of	lung	cancer	diagnosis	compared	to	patients	in	the	lowest	quartile	(41).	Therefore,	if	it	is	true	that	there	is	a	phenotype	of	COPD	that	is	more	prone	to	systemic	inflammation,	that	phenotype	might	also	be	more	likely	to	develop	lung	cancer.	Identification	of	this	phenotype,	and	delivering	corresponding	treatment	to	reduce	levels	of	systemic	inflammation	might	be	useful	to	reduce	lung	cancer	risk	in	COPD	patients.		The	evidence	presented	above	asserts	that,	after	adjusting	for	smoking	status,	COPD	is	associated	with	higher	levels	of	systematic	inflammation.	Moreover,	further	evidence	demonstrates	that	a	link	exists	between	systemic	inflammation	and	lung	cancer	risk.			1.5 Inhaled	corticosteroids		The	appropriate	use	of	inhaled	corticosteroids	as	therapy	by	COPD	patients	has	been	debated	(42,43),	particularly	for	early	stage	disease;	however,	the	GOLD	guidelines	suggest	that	an	ICS	should	be	used	(as	combination	therapy)	in	treating	COPD	Stage	3	and	4	disease	patients	(5).	The	debate	stems	from	evidence	that	suggests	that	ICS	may	not	provide	                                                2	Pine	et	al.	(41)	used	two	different	study	populations:	the	National	Cancer	Institute	Maryland	study	(NCI-MD)	and	the	Prostate,	Lung,	Cancer,	Ovarian	(PLCO)	screening	trial.			 12	benefits	in	terms	of	improved	lung	function	or	a	reduction	in	all-cause	or	COPD-related	mortality	(44),	but	ICS	have	been	shown	to	reduce	AECOPD	(45)	and	also	to	improve	health-related	quality	of	life3	(46,47).	For	example,	a	meta-analysis	of	24	studies	by	Gartlehner	et	al.	(45)	reported	that	patients	using	ICS	compared	to	placebo	experienced	fewer	exacerbations	(RR:	0.67	(95%	CI:	0.59-0.77)),	but	that	ICS	use	had	no	significant	risk	on	overall	mortality	(RR:	0.81	(95%	CI:	0.60-1.08)).		However,	Sin	et	al.	(48)	found	that	patients	prescribed	an	ICS	after	hospital	discharge	had	a	reduced	risk	of	all-cause	mortality	(RR:	0.75	(95%	CI	:0.68-0.82))	and	that	the	effect	was	more	pronounced	as	the	dosage	increased	(low	dose	(≤	500	µg):	RR:	0.77	(95%	CI:	0.69–0.86);	medium	dose	(501-1000	µg):	RR:	0.48	(95%	CI:	0.37–0.63);	and	high	dose	(≥1000	µg):	RR:	0.55	(95%	CI:	0.44–0.69)).			In	addition	to	the	lack	of	potential	benefit	conferred	by	ICS	use	in	COPD	patients,	there	has	also	been	concern	about	potential	adverse	effects	from	ICS	use.	A	study	by	Suissa	et	al.	(10)	showed	that	current	ICS	increased	the	risk	of	serious	pneumonia	(RR:	1.69	(95%	CI:	1.63-1.75)).	Similarly,	a	population	based	study	by	Eurich	et	al.	(49)	used	a	nested-case	control	design	to	evaluate	the	risk	of	pneumonia	in	ICS	patients.	Patients	were	classified	as	‘current’,	‘past’,	or	‘never’	ICS	users.	The	results	of	the	analysis	showed	a	statistically	significant	increase	in	the	risk	of	pneumonia	from	‘current’	ICS	use	compared	to	‘never’	use	(OR:	1.90	(95%	CI:	1.45-2.50)).		                                                3	In	terms	of	the	St.	George’s	Respiratory	Questionnaire	(SGRQ).		 13	Table	1.2.	Lung	cancer	risk	and	associated	levels	of	systemic	inflammation.			The	benefit	conferred	by	ICS	use	in	terms	of	reduced	risk	of	AECOPD	is	likely	achieved	thorough	a	reduction	in	airway/localized	inflammation.	The	use	of	ICS	to	reduce	localized	inflammation	in	the	airway	and	lungs,	both	as	monotherapy	and	combination	therapy,	is	well-established	(53–55).	More	importantly	for	the	purposes	of	this	dissertation,	evidence	further	suggests	that	ICS	use	reduces	systemic	inflammation	(Table	1.3).	A	proof-of-concept	study	by	Sin	et	al.	(56),	using	a	final	sample	of	41	patients	showed	that	ICS		Study	 Study		Participants		 Biomarker	Smoking	Adjusted	 Estimated	Effect		     Chaturvedi	et	al.	(40)	*	 PLCO	trial	 CRP	 Yes	 OR:	1.98	(95%	CI:	1.35-2.98)		     Pine	et	al.	(41)	*	 PLCO	trial	 IL-6	 Yes	 OR:	1.48	(95%	CI:	1.04-2.10)		 PLCO	trial	 IL-8	 Yes	 OR:	1.57	(95%	CI:	1.10-2.24)		 NCI-MD	 IL-6	 Yes	 OR:	3.29	(95%	CI:	1.88-5.77)		 NCI-MD	 IL-8	 Yes	 OR:	2.06	(95%	CI:	1.19-3.57)		     Allin	et	al.	(50)f	 Danish	General	Population	 CRP	 Yes	 HR:	2.1	(95%	CI:	1.2-3.8)		     Trichopoulos	et	al.	(51)†	 EPIC	(Greece)	 CRP	 Yes	 OR:	1.31	(95%	CI:	1.11-1.53)		     Siemes	et	al.	(52)	‡	 Rotterdam	Study	 CRP	 Yes	 HR:	2.78	(95%	CI:	1.59-4.85)		     		 		 		 		 		*	The	highest	quartile	of	inflammation	level	compared	to	the	lowest	quartile.	f	The	highest	quintile	compared	to	the	lowest	quintile.	†	For	every	1	SD	increment.	‡	For	patients	with	levels	>	3mg/L	compared	to	those	with	≤	3	mg/L.	EPIC:	European	Prospective	Investigation	into	Cancer	and	Nutrition.	NCI-MD:	National	Cancer	Institute-Maryland.				 14	Table	1.3.	Studies	reporting	associations	between	inhaled	corticosteroid	use	and	systemic	inflammation.	Study	 Patients	Systemic	Inflammation		Biomarker	Estimated	Effect		    Pinto-Plata	et	al.	(57)	 COPD	patients	 CRP	 Non-users:	6.3	mg/l;	ICS	users:	3.7	mg/l	(p<0.05)*	Sin	et	al.	(56)	 COPD	patients	 CRP	 Non-users:	-7.6%	(-36.6	to	34.8);	ICSa:	-50.3%		(-72.9	to	-9.0)*		 COPD	patients	 IL-6	 Non-users:	5.0%	(-31.6	to	61.0);	ICSa:	-26.1%		(-43.6	to	-3.2)*	Sin	et	al.	(47)	 COPD	patients	 CRP	 Non-users:	-0.145		(-1.923	to	1.732);		ICSa:	-0.168	(-1.385	to	0.691)		 COPD	patients	 IL-6	 Non-users:	-0.2	(-1.3	to	0.5);	ICSa:	0.1	(-0.6	to	0.9)			 		 		 		*	Statistically	significant.	a	Fluticasone.		IL-6:	Interleukin-6;	CRP:	C-reactive	protein;	Mg/l:	Milligrams	per	litre.				treatment	resulted	in	lower	CRP	and	IL-6	levels	compared	to	controls.	Similarly,	Pinto-Plata	et	al.	(57)	found	that	CRP	levels	were	elevated	in	COPD	patients	relative	to	non-COPD	controls,	but	that	these	levels	were	lower	in	COPD	patients	treated	with	ICS,	compared	to	no	ICS	use	(57).		 15	1.6 Statins	HMG-CoA	reductase	inhibitors	(statins)	are	lipid	lowering	agents	commonly	prescribed	as	primary	or	secondary	prevention	for	cardiovascular	disease.	Patients	with	COPD	are	often	prescribed	statins	because	these	patients	may	face	an	increased	risk	of	comorbid	cardiovascular	disease	(CVD)	(27).	For	example,	Curkendall	et	al.	(26)	estimated	that	the	proportion	of	statin	use	in	a	population-based	cohort	in	COPD	patients	was	11.4%	(26).	Statins	have	been	shown	to	be	effective	in	reducing	all-cause	mortality	in	patients	with	risk	factors	for	CVD4	(58,59)	and	a	recent	report	released	by	the	United	States	Preventative	Services	Task	Force	(USPSTF)	suggests	that	statin	use	is	of	benefit	to	patients	40	to	75	years	of	age	with	at	least	one	risk	factor5	for	CVD	(60).		In	addition,	and	importantly	for	COPD	patients,	statins	may	reduce	the	risk	of	acute	exacerbations	(61)	and	the	risk	of	all-cause	mortality	(62,63).	Miyata	et	al.	(64)	suggest	that	statins	may	reduce	lung	inflammation	by	facilitating	the	clearance	of	particulate	matter	(PM10)	by	limiting	the	activation	of	alveolar	macrophages	and	polymorphonuclear	leukocytes	in	the	lungs.	Lahousse	et	al.	(65)	report	that	sustained	statin	use	(>	two	years)	was	associated	with	decreased	risk	of	all-cause	mortality,	particularly	in	patients	with	higher	levels	of	systemic	inflammation	(CRP).	However,	the	evidence	of	the	beneficial	properties	of	statins	is	not	undisputed.	The	recently	completed	Placebo-Controlled	Trial	of	Simvastatin	in	the	Prevention	of	COPD	Exacerbations	(STATCOPE)	trial	reported	that	a	daily	dose	of	40	mg	of	simvastatin	was	not	associated	with	the	rate	of	acute	exacerbations	per	year	or	time	to	an	                                                4	Without	a	prior	diagnosis	of	CVD.		5	These	risk	factors	are:	dyslipidemia,	diabetes,	hypertension,	and	smoking.		 16	acute	exacerbation	in	patients	with	COPD	(66).	This	result	is	not,	however,	unchallenged	(62,63),	and	perhaps	merely	increases	the	debate	around	the	usefulness	of	statins	in	COPD	patients.	The	STATCOPE	trial	had	several	significant	limitations,	particularly	its	inclusion/exclusion	criteria	(67).	These	criteria	were	designed	to	only	allow	patients	without	any	indication	for	a	statin6,	to	be	included	in	the	study.	The	motivation	for	these	very	restrictive	criteria	was	to	ensure	that	benefits	would	be	observed	in	patients	without	any	reason	to	use	a	statin,	thus	ensuring	that	the	effect	was	as	specific	to	COPD	as	possible.	However,	this	reduces	the	generalizability	of	the	results,	by	potentially	removing	those	patients	in	who	statins	might	be	of	benefit.	This	lack	of	generalizability	may	reveal	the	usefulness	of	observational	studies	to	generate	real-world	evidence	in	evaluating	outcomes	for	COPD	patients	that	use	statins.		The	hypothesized	link	between	health	outcomes	in	COPD	patients	and	the	impact	of	statins	is	the	potential	reduction	of	systemic	inflammation	provided	by	statin	treatment.	Table	1.2	(above)	reports	on	the	increased	likelihood	of	lung	cancer	diagnosis	in	patients	with	higher	levels	of	systemic	inflammation.	Existing	evidence	does	support	the	hypothesis	that	statins	do	reduce	levels	of	systemic	inflammation.	For	example,	the	Pravastatin	Inflammation/CRP	Evaluation	(PRINCE)	trial	found	that	statin	use,	for	both	primary	prevention	and	secondary	prevention	for	CVD	patients,	resulted	in	a	decrease	in	CRP	levels	by	16.9%	(primary	prevention)	and	13.1%	(secondary	prevention)	at	24	weeks	compared	to	a	placebo	group	                                                6	Patients	were	excluded	if	they	met	criteria	for	statin	treatment	according	to	United	States	Adult	Treatment	Panel	III	risk	assessment	from	the	National	Heart,	Lung,	and	Blood	Institute,	if	they	had	a	previous	diagnosis	of	diabetes,	or	were	already	receiving	statins.		 17	(p<0.001	and	p<0.005,	respectively)	(69).	While	the	PRINCE	trial	demonstrated	short	term	effects	of	pravastatin	to	reduce	levels	of	CRP,	longer	term	reductions	were	observed	in	a	randomly	selected	sample	of	patients	from	a	trial	of	patients	with	previous	history	of	myocardial	infarction	(i.e.	secondary	prevention	patients)	(70).	After	five	years	of	follow-up	there	were	statistically	significant	reduction	in	the	mean	levels	of	CRP	(-18.4%)	in	the	statin	treated	group	compared	to	an	increase	in	the	CRP	levels	in	the	placebo	group	(+19.4%)	(70).		Similar	findings	were	reported	in	the	JUPITER	trial7,	which	also	showed	that	statins	were	effective	as	primary	prevention	therapy	(71).	The	trial	results	showed	that	statin	treatment	resulted	in	a	37%	reduction	of	CRP	levels	after	twelve	months	of	use	compared	to	subjects	in	the	placebo	group	(71).	The	results	of	these	studies	suggest	that	statin	use	is	associated	with	decreases	in	the	levels	of	systemic	inflammation	(typically	measured	over	time),	both	over	short	periods	(i.e.	24	weeks)	and	longer	periods	of	time	(five	years).		                                                7	Justification	for	the	Use	of	Statins	in	Primary	Prevention:	An	Intervention	Trial	Evaluating	Rosuvastatin	(JUPITER).		 18	Table	1.4.	Studies	reporting	associations	between	statin	use	and	systemic	inflammation.	Study	 Patients	Systemic	Inflammation	Biomarker	Estimated	Effect		    Albert	et	al.	(64)		 PRINCE	trial	 CRP	 Statin	usersa:	-16.9%	(baseline	level:	0.20	mg/dl,	24	week	level:	0.16	mg/dl)b;	-13.1%	(baseline	level:	0.27	mg/dl,	24	week	level:	0.24	mg/dl)c*		    Bickel	et	al.	(67)	 Patients	with	coronary	artery	disease	 CRP	 Statin	users:	4.3	mg/l;	Non-users:	7.6	mg/l*		  IL-6	 Statin	users:	9.5	pg/ml;	Non-users:	14.4	pg/ml*		    Ridker	et	al.	(65)		 Secondary	prevention	CVD	patients	 CRP	 Statin	usersd:	-18.4%	(baseline	level:	0.38	mg/l,	5-year	level:	0.31	mg/l);	Non-users:	+19.4%	(baseline	level:	0.36	mg/l,	5-year	level:	0.43	mg/l)*		    Ridker	et	al.	(66)		 JUPITER	trial		 CRP	 Statin	usersd:	-37.4%	compared	to	placebo	(median	CRP	level,	statin	users:	1.8	mg/l;	non-users:	3.3	mg/L	after	48	months)*		    Shishehbor	et	al.	(67)		 Primary	prevention	patients		 CRP	 Statin	userse:	-2%	(baseline	level:	2.6	mg/l,	12	week	level:	2.3	mg/l)			 		 		 		*	Results	are	statistically	significant.	a	Rosuvastatin;	b	Primary	prevention	patients;	c	Secondary	prevention	patients;	d	Pravastatin;	e	Atorvastatin.	CRP:	C-reactive	protein;	IL-6:	Interleukin-6;	g:	gram;	mg:	milligram;	pg:	picogram;	l:	litre;	dl:	decilitre;	ml:	millilitre.		 	 19							Figure	1.2.	COPD	is	associated	with	systemic	inflammation	and	systemic	inflammation	is	associated	with	increased	lung	cancer	risk.	Inhaled	corticosteroids	and	statins	have	anti-inflammatory	properties	and,	thus,	have	the	potential	to	reduce	lung	cancer	risk.		1.7 Knowledge	gaps	The	studies	that	comprise	this	dissertation	will	serve	to	address	several	gaps	that	require	further	evidence	regarding	the	treatment	of	COPD	patients.	First,	it	remains	unclear,	in	the	wake	of	evidence	from	the	STATCOPE	trial	(66),	if	statin	treatment	in	COPD	patients	is	of	benefit.	To	this	end,	I	will	evaluate	the	association	between	statin	exposure,	using	a	novel	method	to	define	exposure	in	this	area,	and	mortality.	Second,	there	is	limited	evidence	for	COPD SYSTEMIC	INFLAMMATIONLUNG	CANCERSTATINSINHALED	CORTICOSTEROIDS	 20	whether	ICS	use	in	COPD	patients	might	reduce	lung	cancer	risk.	Given	that	lung	cancer	risk	is	elevated	among	COPD	patients,	further	evidence	is	required	in	order	to	determine	if	ICS	use	among	COPD	patients	might	provide	a	protective	effect.	To	address	this	gap,	a	systematic	review	of	the	evidence	will	be	presented,	followed	by	a	population-based	cohort	study	using	administrative	data	for	the	province	of	British	Columbia	(BC)	linked	to	the	BC	Cancer	Registry	will	be	conducted.	Similarly,	the	evidence	for	the	use	of	statins	and	lung	cancer	risk	has	been	inconclusive.	While	evidence	suggests	that	statin	use	might	reduce	levels	of	systemic	inflammation,	and	systemic	inflammation	has	been	linked	to	lung	cancer	risk,	very	little	evidence	exists	for	statin	use	and	lung	cancer	risk,	particularly	in	COPD	patients.	Therefore,	I	will	conduct	a	population-based	study	evaluating	statin	use	with	lung	cancer	risk	in	a	cohort	of	COPD	patients	to	improve	the	evidence-base	in	this	area.	As	well	as	adding	to	the	evidence-base	for	treatments	for	COPD	patients	outlined	above,	this	dissertation	will	also	improve	knowledge	in	methods	employed	to	answer	these	research	questions.			1.8 Specific	objectives	and	overview	of	this	dissertation		The	objectives	of	this	dissertation	that	will	be	addressed	in	the	following	chapters,	are	as	follows:		1.	to	critically	evaluate	methods	of	defining	exposure	to	medications,	particularly	in	the	context	of	administrative	data	and	chronic	diseases;	2.	to	systematically	appraise	the	evidence	regarding	ICS	use	in	COPD	patients,	and	to	identify	improvements	for	future	studies.			 21	3.	to	evaluate	if	there	is	an	association	between	statin	use	and	health	outcomes,	in	terms	of	mortality,	in	COPD	patients;	4.		to	evaluate	the	use	of	inhaled	corticosteroid	and	statin	use	in	COPD	patients	in	terms	of	lung	cancer	risk;			To	address	the	objectives,	below	is	an	overview	of	each	of	these	chapters:			Chapter	2:	This	chapter	provides	an	overview	and	critical	appraisal	of	methods	of	defining	medication	exposure	that	can	be	used	in	observational	studies,	particularly	in	the	context	of	administrative	data	and	chronic	diseases.	This	chapter	will	provide	a	foundation	for	the	methods	used	in	the	subsequent	chapters	of	this	dissertation.		Chapter	3:	This	chapter	is	a	systematic	review	that	explores	the	evidence	relating	to	ICS	use	in	COPD	patients	and	lung	cancer	risk.	It	discusses	the	results,	and,	more	importantly,	the	methodological	issues	that	arise	from	both	observational	and	trial-based	studies	with	respect	to	ICS	and	lung	cancer.	It	identifies	significant	gaps	that	exist	in	the	current	evidence	base,	which	will	be	addressed	specifically	in	Chapter	5	of	this	dissertation.			Chapter	4:	One	of	the	two	main	objectives	of	this	dissertation	is	to	evaluate	the	association	between	statin	exposure	and	lung	cancer	risk	in	COPD	patients.	As	a	precursor	to	that	analysis,	this	chapter	evaluates	the	association	between	pulmonary	and	all-cause	mortality	and	statin	use,	to	provide	insight	as	to	whether	statins	may	be	of	benefit	in	COPD	patients.	The	analysis	employs	a	novel	study	design	that	has	not	previously	been	used	in	this	area.			 22	Chapter	5:	This	chapter	addresses	the	first	component	of	the	general	research	question	of	this	dissertation:	is	there	an	association	between	ICS	exposure	and	lung	cancer	risk	in	COPD	patients?	To	answer	this	question,	I	use	several	different	methods	of	quantifying	medication	exposure,	building	on	the	narrative	review	of	these	methods	in	Chapter	2,	and	the	existing	evidence	presented	in	Chapter	3.	In	addition	to	using	methods	of	defining	exposure	to	medications	that	are	novel	in	this	area	of	research,	this	chapter	also	employs	a	latency	period	for	lung	cancer	diagnosis	that	is	seldom	used	in	observational	studies,	and	also	accounts	for	lung	cancer	histology	in	the	analysis,	which	is	also	a	novel	contribution.			Chapter	6:	This	chapter	addresses	the	second	component	of	the	overall	research	question	of	this	thesis:	is	there	a	relationship	between	statin	exposure	and	lung	cancer	risk	in	COPD	patients?	Similar	to	the	study	presented	in	Chapter	5,	this	study	builds	on	the	exploration	of	exposure	metrics	presented	in	Chapter	2	and	also	builds	upon	the	results	of	the	analysis	of	statin	exposure	and	all-cause	mortality	presented	in	Chapter	4.			1.9 Closing	remarks	COPD	patients	have	a	considerable	risk	of	morbidity	and	mortality.	This	is	largely	due	to	the	increased	risk	of	severe	AECOPD	and	significant	comorbidities	associated	with	the	disease.	These	acute	exacerbations	and	comorbidities	may	be	the	result	of,	an	often	undetected,	increased	level	systemic	inflammation	that	is	common	in	COPD	patients.	Inhaled	corticosteroids	and	statins	have	anti-inflammatory	properties	and,	as	such,	may		 23	reduce	the	risk	of	exacerbations	and	poor	outcomes	associated	with	these	common	comorbidities.	Further,	the	effects	of	inhaled	corticosteroids	and	statins	may	reduce	lung	cancer	risk,	which	evidence	also	suggests	is	associated	with	systemic	inflammation.	The	high	incidence	and	mortality	rates	associated	with	lung	cancer	mean	that	is	it	a	worthwhile	endeavour	to	identify	any	therapies	that	may	reduce	the	incidence	of	lung	cancer.		 	 24	Chapter	2: A	review	of	methods	for	defining	medication	exposure	in	observational	studies		Summary	The	definition	of	exposure	to	a	medication	and	the	relationship	with	the	outcome	of	interest	in	observational	studies	can	be	complex.	Compared	to	clinical	trials	which	typically	offer	controlled	exposure	environments,	observational	studies	may	more	accurately	reflect	patients’	behaviour	and	physician	practices	in	reality.	However,	observational	studies	still	pose	analytical	problems	when	attempting	to	define	exposure	to	prescribed	medications;	patients’	medication	dose	may	fluctuate,	the	length	of	time	using	the	medication	may	vary,	patients	may	discontinue	taking	their	medications,	or	may	not	adhere	properly	to	the	instructions	given	by	their	physician	or	pharmacist.	This	narrative	review	presents	several	metrics	for	quantifying	medication	exposure	in	observational	studies,	and	critically	evaluates	their	advantages	and	disadvantages.	This	chapter	will	also	inform	the	analytical	approach	used	in	subsequent	chapters	of	this	dissertation.		2.1 Introduction	The	prevalence	of	chronic	diseases	has	been	steadily	increasing	(74)	with	a	concurrent	increase	in	the	number	of	available	medications	and	the	commensurate	use	of	these	medications	(75,76).	The	treatment	of	chronic	disease	and	the	evaluation	of	the	effectiveness	of	these	medications	represents	a	different	scenario	than	evaluating		 25	medications	for	safety	and	efficacy	in	randomized	controlled	trials	(RCTs).	For	medications	used	in	the	treatment	of	chronic	diseases,	where	the	effectiveness	of	a	medication	must	be	evaluated	over	an	extended	duration	of	time,	the	complexity	of	defining	medication	exposure	is	increased.	Moreover,	the	length	of	time	that	a	patient	uses	a	medication	increases	the	potential	for	variability	in	the	usage	of	the	medication	(77).	This	variability	may	influence	the	ability	of	the	medication	to	be	effective	in	a	real-world	setting.			In	observational	studies,	the	classification	of	a	patient	as	exposed	or	unexposed	to	a	medication	can	be	complex.	There	are	a	myriad	of	factors	to	consider,	particularly	in	longitudinal	studies	when	the	duration	of	follow-up	and	exposure	may	extend	over	a	period	of	several	years	(78).	For	example,	the	initial	medication	prescription,	drug	dosage,	and	the	duration	of	the	prescription	all	need	to	be	considered.	Further,	patients	may	switch	between	medications	or	dosages	can	be	altered	by	physicians.	Patients’	behaviour	also	plays	an	important	role	as	they	may	choose	to	discontinue	their	medication,	or	use	it	irregularly	for	a	variety	of	reasons	such	as	cost,	confidence	in	effectiveness	of	the	medication,	or	the	perceived	ability	to	self-treat	(74,75).			Chronic	diseases	may	have	other	associated	complexities.	The	latency	periods	and	induction	periods	(see	Figure	2.1)	that	are	characteristic	of	chronic	diseases	make	the	relationship	between	medication	exposure	and	disease	even	more	complex	(81).	The	induction	period	is	defined	as	the	time	between	the	exposure	to	the	agent	which	causes	the	disease	and	subsequent	disease	initiation.	The	latency	period	is	the	time	from	disease	initiation	until	the	disease	becomes	clinically	detectable	(82).	For	medication	exposure	that		 26	may	provide	a	beneficial	(or	a	detrimental)	effect,	understanding	these	two	periods,	and	how	they	relate	to	medication	exposure	definitions,	is	important.	For	example,	consider	a	disease	that	has	a	latency	period	of	six	months.	A	patient	that	receives	a	medication	(the	exposure),	thought	to	be	protective	for	the	disease,	at	three	months	before	the	disease	has	become	clinically	detectable	could	be	classified	as	exposed.	Although	the	patient	has	received	the	medication	thought	to	prevent	the	disease,	in	reality,	the	disease	process	has	already	begun	and	there	is	little,	if	any,	chance	for	the	medication	to	have	an	effect	on	the	disease	process.	Thus,	appropriately	defining	a	window	of	exposure	which	acknowledges	the	latency	period	of	a	particular	disease,	is	important	in	analyzing	the	effects	of	medications	in	chronic	diseases	or,	indeed,	any	disease	with	long	induction	and/or	latency	periods	(i.e.	cancers).	Analyses	must	also	consider	the	time	that	is	required	for	a	medication	to	be	effective	and	the	time	period	for	which	it	remains	effective.	For	example,	a	patient	that	received	their	first	prescription	of	a	statin	for	the	primary	prevention	of	cardiovascular	disease	and	experienced	a	myocardial	infarction	the	following	day	likely	did	not	experience	any	benefit	from	that	statin	prescription.	In	such	an	instance,	the	patient	should	not	be	classified	as	being	exposed	to	the	medication.	Similarly,	for	a	study	that	has	a	follow-up	time	of	several	years,	a	patient	that	received	their	only	statin	prescription	of	30	days	in	first	year	of	follow-up	and	experienced	an	event	(i.e.	death)	in	the	fifth	year	is	likely	not	have	received	any	sustained	benefit	from	that	initial	statin	prescription,	which	should	be	accounted	for	by	the	exposure	definition.					 27	 Figure	2.1.	Induction	and	latency	periods	associated	with	the	onset	of	a	particular	disease.	In	this	example,	the	‘exposure’	is	that	which	begins	the	disease	process.		Another	relevant	consideration	in	quantifying	exposure	for	patients	with	chronic	disease	is	patients'	adherence	to	medications,	and	discontinuation	or	interruptions,	in	their	medication.	Non-adherence	is	sufficiently	common	and	problematic	that	it	has	become	a	frequently	employed	exposure	metric	(78,79).	As	mentioned	above,	there	are	a	myriad	of	reasons	why	patients	take	their	medication	irregularly	or	cease	to	take	their	medication,	ranging	from	perceptions	(for	example,	confidence	in	the	expertise	of	the	prescribing	physician	(79))	to	negative	effects	of	the	medication	(pain	associated	with	statin	use	(80))	and	these	reasons	are	often	not	captured	in	study	data.			Randomized	controlled	trials	(RCTs)	may	be	better	equipped	to	control	medication	usage,	dosage,	and	sustained	use	by	enforcing	the	protocol	of	the	study.	However,	observational	studies	better	reflect	real-world	conditions	for	post-marketing	surveillance	of	medication	effectiveness,	particularly	in	the	case	for	chronic	diseases	and	cancer	when	the	length	of		 28	study	follow-up	or	the	number	of	events	may	be	prohibitive	to	carry	within	the	context	of	an	RCT.			There	is	an	assortment	of	methods	that	can	be	used	to	define	medication	exposure	including:	ever/never	use,	duration	of	use,	adherence,	discontinuation,	or	cumulative	or	current	dose,	among	others.	Simultaneously	quantifying	the	components	of	medication	exposure	(timing,	duration,	dose,	etc.)	poses	analytic	challenges	and	may	significantly	affect	study	results.	Therefore,	this	chapter	will	provide	an	overview	of	several	different	measures	of	defining	exposure	to	medications	and	critically	evaluate	the	implications	that	each	definition	will	have	on	the	analysis	of	data	and	study	results.	To	do	so,	this	review	will	use	the	context	of	an	observational	cohort	study	that	uses	population-based	administrative	data	to	evaluate	the	association	between	medication	use	and	lung	cancer	risk,	in	addition	to	drawing	on	published	examples	from	the	literature.	Moreover,	it	will	provide	the	foundation	and	context	for	the	decisions	on	exposure	metrics	used	in	analyses	presented	in	subsequent	chapters	of	this	dissertation.				2.2 Measures	of	medication	exposure	2.2.1 Ever/never	use	The	most	basic	method	of	quantifying	medication	exposure	is	simply	to	classify	patients	as	exposed	if	they	have	ever	received	a	prescription	or	were	dispensed	the	medication,	in	the	context	of	administrative	data.	The	advantage	of	this	method	lays	in	its	simplicity,	both	conceptually	and	computationally.	However,	there	are	several	key	issues	that	limit	the		 29	appropriateness	of	this	definition	of	exposure.	For	example,	if	the	follow-up	period	in	a	cohort	study	is	long,	and	exposure	status	if	fixed,	‘exposed’	patients	that	have	contributed	very	little	time	would	be	treated	equally	to	‘exposed’	patients	with	much	longer	follow-up	times.	This	can	be	problematic,	as	the	length	of	a	patients'	follow-up	time	might	not	only	increase	their	probability	of	experiencing	the	outcome	of	interest,	but	also	their	probability	of	being	exposed	to	the	medication.		Since	this	method	of	defining	exposure	is	fixed,	it	also	does	not	take	into	account	improper	use,	interruptions	in	use,	or	discontinuation	of	medication	use.	Moreover,	dose-response	relationships	are	not	considered.	This	method	is	often	used	as	a	basis	for	comparison	within	a	study	where	several	other	metrics	of	medication	exposure	are	used	(80,81).	Using	this	method	of	exposure	classification,	a	patient	that	receives	a	prescription	or	is	dispensed	the	medication	will	always	be	classified	as	‘exposed’	in	the	study	follow-up	period.	This	static	or	fixed	exposure	classification	has	the	potential	to	bias	the	study	results,	particularly	for	studies	with	longer	follow-up	times.				Figure	2.2.	Using	a	fixed,	‘ever/never’	exposure	definition,	a	patient	can	be	considered	exposed	if	ever	having	been	dispensed	the	medication	in	the	follow-up	period.		 30	This	method	of	quantifying	medication	exposure	can	be	improved	by	applying	specific	exposure	windows	in	the	analysis.	In	doing	so,	exposure	is	still	fixed,	but	the	period	for	exposure	assessment	can	be	shortened	to	create	a	more	plausible	link	between	exposure	status	and	the	probability	of	experiencing	the	outcome	of	interest.	In	Chapter	4,	the	analysis	presented	explores	the	association	between	statin	exposure	and	patient	outcomes	(all-cause	mortality	as	the	primary	analysis	and	pulmonary-related	mortality	as	a	secondary	analysis).	To	do	so,	an	exposure	ascertainment	window	was	defined	(one	year	from	the	index	date:	the	diagnosis	of	COPD)	was	used	in	conjunction	with	a	one-year	outcome	ascertainment	window.	Thus,	the	period	in	which	exposure	status	was	defined	did	not	overlap	with	the	period	during	which	the	association	between	exposure	and	the	outcome	of	interest	was	explored	(82,83).	This	analytical	approach	limits	the	chance	of	misclassification	bias	and	immortal	time	bias.	The	potential	for	misclassification	bias	is	reduced	by	defining	a	specific	time	period	during	which	patients	could	be	exposed	and,	also,	by	limiting	the	time	period	during	which	they	could	experience	the	outcome	of	interest.	The	potential	for	immortal	time	bias	is	mitigated	by	creating	the	non-overlapping	exposure	ascertainment	and	outcome	ascertainment	periods.					 31		Figure	2.3.	Exposure	and	outcome	ascertainment	windows.			2.2.2 'Current'	medication	use	Exposure	may	also	be	defined	as	'current'	use,	that	is,	a	specific	period	defined	prior	to	the	event	of	interest.	This	method	is	commonly	used	in	case-control	studies	(86)	and	classifies	a	patient	as	exposed	if	the	medication	was	dispensed	in	a	specified	time	period	preceding	the	event	of	interest.	This	method	may	be	most	appropriate	for	certain	medications	that	are	used	to	prevent	acute	events.	However,	for	diseases	that	develop	slowly	and	that	may	have	poorly	defined	start	times,	it	becomes	less	likely	that	it	is	an	appropriate	measure.	The	limitations	of	this	approach	are	that	it	does	not	appreciate	dose-response	relationships	or	Study Start (Index Date)Exposure Assessment Window Outcome Assessment Window TimePrescriptionDispensedEvent In this case, a patient that receives the medication during the ‘Exposure Assessment Window’ will be considered exposed. Patients that receive their only dispensation of the medication outside of this window will not be considered exposed.Ex s r 	 scertai ment	WindowOutcome	Ascertain t	WindowIn	this	case,	a	patient	that	receives	the	medication	during	the	‘exposure	ascertainment	window’	will	be	considered	exposed.	Patients	that	fill	their	only	prescription	for	the	medication	outside	of	this	window	will	not	be	considered	exposed.	 32	the	aggregate	duration	of	medication	use.	These	two	factors	are	important	for	chronic	diseases	where	medication	usage	may	span	several	years	and	where	dosages	may	also	vary	over	time.	The	intuition,	however,	behind	this	method	is	simple:	that	exposure	in	a	defined	time-window	immediately	prior	to	the	outcome/event	of	interest	is	the	most	important.	Control	patients,	obviously	will	include	those	patients	that	never	received	the	mediation	but	may	also	include	patients	that	received	the	medication	outside	the	defined	exposure	window	(see	Figure	2.4).	The	result	will	be	that	some	patients	that	received	the	medication,	but	outside	of	the	exposure	window,	will	be	counted	as	unexposed,	which	has	the	potential	to	bias	results	towards	the	null,	if	past	usage	affects	the	outcome.				Figure	2.4.	'Current'	medication	use.				 33	An	example	of	this	definition	used	in	practice	is	presented	in	Tournier	et	al.	(89)	where	the	authors	used	three	different	measures	of	exposure	to	assess	the	effect	of	antipsychotic	medications	prior	to	a	metabolic	event.	Among	these	measures,	a	categorical	variable	for	‘current	use’	was	explored,	along	with	ever/never	use,	and	cumulative	duration	of	medication	use.	Current	use	was	defined	using	three	levels:	(i)	current	use	(using	the	medication	at	same	time	as	metabolic	event	occurred),	(ii)	recent	use	(within	six	months	of	the	event),	and,	(iii)	no	use.	These	three	definitions	rendered	conflicting	results:	ever-use	resulted	in	a	protective	hazard	ratio,	current	use	showed	an	increased	risk	of	a	metabolic	event,	and	cumulative	duration	of	use	did	not	show	a	statistically	significant	effect.	One	plausible	explanation	of	this	discrepancy	in	results	suggests	that	patients	prescribed	an	antipsychotic	who	exhibit	signs	of	an	impending	metabolic	event	and	discontinue	use	of	the	antipsychotic	prior	to	the	event	occurring	(89).	As	such,	because	the	exposure	classification	in	the	ever/never	use	scenario	is	fixed,	these	patients	would	remain	considered	as	exposed	in	the	analysis	despite	discontinuing	their	medication	use.	The	authors’	use	of	the	current	use	measure	of	exposure	classification	remedies	this	problem	by	introducing	a	specific	time-window	of	exposure	related	to	the	event.		 		 	 	 	 		 34		Figure	2.5.	Addition	of	a	‘lag’	time	as	a	variation	on	‘current’	use.		Lee	et	al.	(90)	used	current	and	past	exposure	of	inhaled	corticosteroids	(ICS)	to	explore	their	relationship	with	the	development	of	lung	and	laryngeal	cancer.	This	nested-case	control	study	used	current	(within	90	days	of	the	index	date,	the	diagnosis	of	cancer)	and	past	use	(91-365	days	from	the	index	date)	to	define	exposure	to	ICS.	Patients	with	less	than	a	30-day	prescription	for	ICS	were	classified	as	non-users.	Dividing	current	and	past	users	in	this	way	is	a	strength	of	the	study,	however,	it	omits	an	important	characteristic	of	lung	cancer.	That	is,	a	dose	of	ICS	immediately	preceding	a	lung	cancer	diagnosis	should	not	be	considered	as	a	treatment	failure	because	the	ICS	exposure	would	have	occurred	when	the	initiation	of	lung	cancer	had	already	started	(within	the	latency	period).	Lung	cancer	is	often	diagnosed	at	an	advanced	stage,	which	is	why	lung	cancer	screening	programs	have	been	the	subject	of	intense	research	and	the	role	of	such	programs	in	potentially	reducing	the	high	morbidity	and	mortality	is	increasingly	recognized.	Therefore,	when	lung	cancer	is	diagnosed,	it	is	likely	that	the	patient	has	had	lung	cancer	for	a	period	of	time	and	simply		 35	had	not	received	a	diagnosis.	In	this	context,	current	use	is	likely	inadequate	in	evaluating	an	association	between	lung	cancer	development	and	ICS	use.	A	strength	of	this	study,	however,	was	that	two	different	'lag'	periods	were	assigned	in	sensitivity	analysis	(three	and	six	months).	Application	of	these	lag	periods	resulted	in	statistically	significant	odds	ratios	that	demonstrated	a	slightly	reduced	effect	(OR:	0.86	(95%CI:	0.80-0.93)	and	OR:	0.86	(0.80-0.95),	respectively)	compared	to	the	overall	odds	ratio	(OR:	0.79	(95%	CI:	0.69-0.90)	(90).	The	latency	period	of	lung	cancer	is	poorly	understood;	however,	it	is	likely	that	a	six	month	period	may	still	be	too	short	for	ICS	to	have	an	effect	on	lung	cancer	risk	(39).	Moreover,	to	be	considered	exposed,	patients	needed	a	minimum	of	a	30-day	prescription	for	ICS,	but	given	the	proposed	mechanism	of	action	for	ICS	reducing	lung	cancer	risk,	this	length	of	prescription	time	may	also	be	insufficient	(90).	In	this	example,	the	authors	did,	in	fact,	show	that	ICS	provided	a	protective	effect	on	lung	cancer	risk	(OR:	0.79	(95%	CI:	0.69-0.80)).	While	this	result	aligns	with	the	a	priori	study	hypothesis,	further	validation	of	the	effect	would	be	confirmed	if	this	result	persisted	with	the	duration	of	ICS	use,	or	with	longer	definitions	of	a	'lag'	period	(i.e.	twelve	months).	Thus,	patients	might	be	misclassified	as	being	exposed	when	they	have	not	had	a	sufficiently	long	duration	of	use	of	the	medication	for	it	to	confer	benefit.	This	misclassification	would	likely	result	in	a	conservative	bias;	that	is,	it	would	be	expected	that	because	lung	cancer	patients	would	be	defined	as	being	exposed,	the	results	would	be	biased	toward	the	null.		2.2.3 Cumulative	dose	Medication	exposure	can	also	be	defined	using	cumulative	dose,	that	is,	the	total	quantity	of	medication	times	the	dosage	of	a	medication	received	during	a	certain	time	period.	This		 36	measure	is	also	computationally	and	conceptually	simple.	To	calculate	exposure	in	this	way	requires	the	aggregation	of	the	doses	received	by	the	specific	patient	over	the	follow-up	period.	The	most	complicated	issue	arises	when	a	patient	might	switch	between	medications,	that	have	different	strengths,	within	a	specific	class	of	drugs	(i.e.	between	budesonide	and	fluticasone	for	inhaled	corticosteroids)	and	dosages	need	to	be	transformed	into	a	comparable	equivalent	dose	(86,87).	The	strength	of	this	method	is	that	it	appreciates	differences	in	doses	and	quantities	of	medications	received	by	each	patient,	and	that	associations	may	exist	for	cumulative	amounts	of	the	medication	used	by	patients	(i.e.	dose-response	relationships).	As	an	extension,	one	could	calculate	the	mean	daily	dose	over	a	specified	time	period	and,	potentially	account	for	time-dependency	over	specific	time	windows.	The	drawback	of	this	approach	is	that	it	gives	no	weight	to	when,	during	follow-up,	the	actual	exposure	occurs.	For	example,	a	patient	may	use	a	medication	in	high	doses,	but	very	sporadically.	The	calculation	of	cumulative	dose	for	this	patient	would	not	differ	for	a	patient	that	used	their	medication	continuously	(as	directed)	over	the	same	period	of	time,	at	a	lower	dose.	Similarly,	it	does	not	distinguish	between	a	patient	that	used	a	low-dose,	over	a	longer	period	of	time,	and	a	patient	that	used	a	high	dose	for	a	shorter	period	of	time.	Finally,	it	does	not	appreciate	when,	during	the	follow-up	period,	the	medication	was	received	in	relation	to	the	outcome.	Therefore,	to	improve	this	method,	it	would	be	advantageous	to	use	an	approach	that	appreciates	the	cumulative	dose	of	medication	received	and	also	when	it	was	received	during	the	follow-up	period.	One	way	to	accomplish	this	is	to	use	a	series	of	time-windows	during	the	follow-up	period	to	calculate	the	cumulative	dose	of	medication	received	in	each	window	as	they	relate	to	the	outcome		 37	of	interest.	Additional	methods	that	incorporate	cumulative	dose	into	a	more	robust	metric	of	exposure	are	discussed	in	Section	2.2.5	below.			2.2.4 Medication	adherence	and	discontinuation	Adherence	to	medication	can	be	a	useful	metric	for	quantifying	medication	exposure.	This	method	appreciates	that	patients	do	not	always	use	medications	as	directed	by	physicians.	In	observational	studies,	particularly	studies	using	administrative	datasets,	the	medication	possession	ratio	(MPR)	or	proportion	of	days	covered	(PDC)	are	commonly	used	methods	of	capturing	medication	adherence	(93).	These	two	methods	essentially	take	the	aggregate	of	the	prescribed	duration	of	medication	divided	by	the	follow-up	time	of	the	patient.	The	difference	in	the	two	measures	is	reflected	in	how	over-lapping	prescriptions	are	handled;	the	MPR	can	be	greater	than	one,	no	adjustment	is	typically	made	for	prescriptions	that	may	overlap,	whereas,	with	the	PDC,	prescriptions	cannot	overlap,	therefore	the	range	of	possible	values	is	limited	between	‘0’	and	‘1’.	A	threshold	can	then	be	applied	to	classify	the	patients	categorically	as	‘adherent’	or	‘non-adherent’.	For	example,	Blackburn	et	al.	(94)	used	a	threshold	for	a	medication	user	to	be	considered	adherent	of	60%	whereas	Suissa	et	al.	(83)	used	a	threshold	of	80%.	The	threshold	for	adherence	that	is	chosen	will	depend	on	the	medication	under	investigation,	but	should	undoubtedly	be	subject	to	sensitivity	analyses	to	determine	if	the	level	chosen	is	adequate	to	offer	therapeutic	benefit.	A	categorical	measure	of	medication	adherence	might	also	be	useful	to	determine	if	a	gradient	exists	for	each	level	of	adherence.				 38	A	study	by	Ho	et	al.	(95)	evaluated	the	association	between	adherence	(using	the	PDC	method)	to	cardio-protective	medications	(statins,	beta-blockers,	angiotensin	receptor	blockers)	and	all-cause	mortality.	The	study	found	that	cardio-protective	medications	reduced	all-cause	mortality	and,	importantly,	adherent	users8	had	a	more	pronounced	reduction	in	all-cause	mortality.	Interestingly,	this	study	found	that	there	was	no	difference	in	the	risk	of	mortality	between	non-adherent	medication	users	and	‘never’	users.	This	result	highlights	the	importance	of	using	a	threshold	of	adherent	versus	non-adherent	users	as	the	benefit	of	medications	might	only	be	conferred	at	a	certain	level	of	adherence.	Therefore,	this	metric	of	exposure	can	properly	classify	patients	as	exposed	or	unexposed	and	reduce	the	chance	of	bias.					Figure	2.6.	Calculation	of	medication	adherence	using	the	medication	possession	ratio	(MPR).	                                                8	Adherent	patients	had	a	PDC	≥	0.80.	Patients	with	a	PDC	less	than	0.8	were	considered	to	be	non-adherent.			 39	Patients'	discontinuation	of	medication	can	also	be	used	as	a	metric	of	‘exposure’.	Suissa	et	al.	(83)	showed	that	treatment	discontinuation	from	inhaled	corticosteroids	(ICS)	was	associated	with	a	reduction	in	the	risk	of	serious	pneumonia.	Patients	could	‘discontinue’	their	medications	at	any	point	in	time	by	not	refilling	their	prescriptions.			Teichart	et	al.	(84)	provides	another	example	of	discontinuation	whereby	the	authors	examined	the	association	between	discontinuation	of	beta-blockers	and	the	risk	of	MI	using	several	windows	of	discontinuation	(<30	days,	30	days	to	180	days,	>180	days).	The	study	results	suggested	that	the	relative	risk	of	MI	was	increased	(less	than	30	days,	RR:	2.70	(95%	CI:	1.06-6.89);	30	days	to	180	days,	RR:	2.44	(95%	CI:	1.07-5.59)	for	the	first	two	categories	of	beta-blocker	discontinuation	but	that	there	was	no	significant	risk	increase	for	discontinuation	of	greater	than	180	days.	Again,	this	result	emphasizes	the	impact	that	exposure	definition	might	have	on	study	results.		The	foremost	problem	with	this	approach,	when	using	administrative	data,	is	that	the	dispensation	of	the	medication	is	recorded,	but	it	is	unknown	whether	the	patient	has	actually	taken	the	medication.	It	is	necessary,	therefore,	to	make	an	assumption	that	prescriptions	dispensed	are	actually	used	by	the	patient.		2.2.5 Recency-weighted	exposure	measures	Recency-weighted	exposure	definitions	such	as	the	recency-weighted	cumulative	dose	method	is	an	attractive	method	of	defining	exposure	that	can	incorporate	the	duration	of	time	a	patient	has	used	a	medication	and	weights	those	medications	received	proximal	to		 40	the	event	of	interest	more	heavily	than	those	received	earlier	in	the	follow-up	period.	A	study	by	Abrahamowicz	et	al.	(96)	is	the	original	investigation	using	this	method.	The	analysis	builds	on	previous	work	done	by	the	authors’	regarding	benzodiazepine	use	and	adverse	events.	The	focus	of	the	study	was	to	accurately	model	the	cumulative	effects	of	the	duration	and	dose	of	exposure.		While	this	study,	and	subsequent	studies	employing	this	methodology,	focused	primarily	on	acute	events,	this	methodology	can	be	applied	more	generally.	The	original	use	of	the	methodology	in	the	context	of	acute	events	was	by	design,	as	the	methodology	was	developed	to	explore	associations	between	medication	usage	and	adverse	events.	Specifically,	the	authors'	motivation	for	this	study	was	to	find	a	measure	that	would	be	sensitive	enough	to	capture	weak	associations	between	medication	exposure	and	adverse	events	(96).		Avina	et	al.	(97)	used	the	recency-weighted	method	to	explore	whether	it	would	result	in	a	better	model	fit,	compared	to	common	exposure	metrics,	in	predicting	cerebrovascular	accidents	(CVA)	after	exposure	to	oral	glucocorticoids	in	patients	with	rheumatoid	arthritis	(RA).	The	results	of	the	study	suggested,	across	all	measures	of	exposure,	that	there	was	no	statistically	significant	association	between	glucocorticoid	(GC)	exposure	and	CVA,	nor	did	their	results	support	the	incorporation	of	the	recency-weighted	exposure	metric	to	improve	model	fit9.	A	later	study	by	the	same	author	(85)	used	a	similar	approach	(as	a	sensitivity	analysis)	to	estimate	the	risk	of	acute	myocardial	infarction	(AMI)	after	                                                9	In	this	study,	model	fit	was	assessed	by	comparing	the	Akaike	Information	Criterion	(AIC)	value	for	each	model.			 41	exposure	to	glucocorticoids	in	RA	patients.	In	this	study,	again,	the	weighted	cumulative	exposure	method	of	quantifying	medication	exposure	did	not	improve	the	model	fit	(85).			Dixon	et	al.	(86)	conducted	a	study	assessing	the	risk	of	infection	after	oral	glucocorticoids	(86)	using	eleven	different	models	with	various	metrics	for	exposure	to	highlight	the	complexity	of	the	relationship	between	medication	exposure	and	adverse	events.	All	multivariable	models	were	then	compared	to	the	model	incorporating	the	weighted	cumulative	dose	exposure	metric	(using	Akiake	information	criterion	(AIC)).	The	AIC	of	the	weighted	cumulative	dose	model	was	subtracted	from	the	AIC	of	the	conventional	models	to	evaluate	the	difference	between	the	respective	AIC	values	where	a	positive	value	would	indicate	that	the	model	incorporating	the	weighted	cumulative	dose	exposure	definition	had	a	greater	AIC,	and	thus	was	inferior	to	the	conventional	models.	In	every	instance,	the	AIC	for	the	weighted	cumulative	dose	model	was	lower	than	that	of	the	conventional	models	(86)	indicating	that	the	models	containing	the	weighted	metric	for	exposure	provided	better	estimates	of	the	association	between	the	exposure	and	outcome.			A	method	that	attempts	to	similarly	account	for	variation	in	medication	usage	over	the	exposure	period	was	proposed	by	Phadnis	et	al.	(98).	The	authors	suggest	that	using	just	one	covariate	(even	if	the	covariate	was	time-dependent)	was	insufficient	to	measure	variation	in	medication	exposure	over	time.	As	a	proposed	solution,	the	authors	developed	a	method	that	used	three	time-dependent	covariates	simultaneously	in	their	regression	model:	(i)	current	medication	usage	status	(yes/no),	(ii)	the	proportion	of	cumulative	exposure	to	drug	at	a	given	point	in	time,	and	(iii)	patients’	switching	between	taking	and		 42	not	taking	the	medication	(98).	A	study	by	Shireman	et	al.	(99)10	applied	this	method	to	evaluate	the	association	between	antihypertensive	medications	and	cardiovascular	outcomes	in	hemodialysis	patients.	Their	finding	was	that	the	‘...effectiveness	[of	the	drug]	depends	not	only	on	having	a	drug	available	but	is	tempered	by	duration	and	stability	of	use,	likely	reflecting	variation	in	clinical	stability	and	patient	behaviour’	(94,	p.113).	However,	while	this	approach	may	be	attractive	because	it	seems	to	offer	a	comprehensive	account	of	medication	exposure,	the	interpretation	of	the	results	becomes	less	clear.	Instead	of	one	result,	several	hazard	ratios	and	interactions	are	generated	by	the	analysis	and	results	are	presented	in	a	series	of	three-dimensional	graphs	that	are	difficult	to	comprehend.	Therefore,	it	appears	that	the	method	proposed	by	Abrahamowicz	et	al.	(96)	may	be	superior	because	it	appears	to	accomplish	similar	comprehensiveness,	and	is	also	intuitively	attractive	because	it	assigns	different	weights	to	past	exposures,	weighting	those	medications	received	recently	more	heavily	than	those	received	in	earlier	in	the	study	period.	However,	the	method	by	which	weights	are	assigned	needs	to	be	carefully	considered	in	the	context	of	the	medication	and	disease	under	study	when	using	this	approach.	The	exposure	weights	and	the	window	of	relevant	exposure	will	be	different	for	acute	events	versus	chronic	diseases.	The	overall	motivation	for	both	of	these	methods	is	that	definition	of	medication	exposure,	due	to	differences	in	duration	and	time,	and	the	temporal	link	with	the	outcome	being	studied,	is	complex.	These	two	analytical	methods	                                                10	Phadnis	et	al.	(98)	reported	the	development	of	the	method	and	the	Shireman	et	al.	(99)	study	was	the	application	of	the	method	for	antihypertensive	medication	exposure	in	haemodialysis	patients.			 43	attempt	to	appreciate	the	relationship	between	medication	dose,	duration	of	use,	and	proximity	of	exposure	relative	to	the	study	outcome.		 2.3 Discussion	The	objective	of	this	review	was	to	critically	evaluate	methods	used	to	define	exposure	to	medication	in	observational	studies.	The	evaluation	of	these	methods	presented	above	should	reveal	that	there	may	be	choice	of	methods	that	can	be	employed	in	epidemiologic	analysis,	but	these	methods	must	be	carefully	considered,	in	the	context	of	the	medication	and	outcome,	in	a	particular	study.	The	assertion	of	this	review	is	that	basic	metrics	may	be	appealing	due	to	the	ease	of	calculation	and	their	simplicity,	but	they	may	not	capture	the	true	effects	of	medications,	and	have	the	potential	to	misclassify	medication	exposure,	thus	biasing	results,	particularly	in	periods	of	sustained	and/or	interrupted	use.		Common	definitions	of	exposure,	as	described	above,	such	as	ever/never	use,	cumulative	dose,	or	current	use,	do	not	take	into	account	variability	over	time.	The	longer	the	study	follow-up	time,	the	less	appropriate	these	methods	may	be	as	the	potential	for	variability	in	medication	usage	will	increase	over	time.	While	these	methods	of	defining	exposure	may	be	adequate	for	short	exposure	windows,	where	the	outcome	is	always	proximal	to	the	exposure,	very	seldom	will	they	appropriately	quantify	medication	exposure,	particularly	over	a	longer	period	of	time.				 44	As	mentioned,	adherence	to	medications	may	be	an	important	consideration,	but	certain	methods	to	calculate	adherence	may	be	too	coarse,	particularly	over	longer	periods	of	follow-up	time.	Aggregating	days	of	filled	prescriptions	over	the	period	of	follow-up	time	for	longer	studies	may	not	reveal	when	patients	have	significant	interruptions	in	medication	usage.	To	illustrate,	in	a	study	where	a	patient	received	two	30-day	prescriptions	for	a	medication	and	had	a	follow-up	time	of	one	year,	the	resulting	MPR	would	be	17%.	This	specific	patient	used	the	prescribed	medication	for	only	two	months	of	the	potential	twelve	months11.	However,	if,	for	the	same	patient,	the	follow-up	period	was	two	years	and	the	patient	missed	this	same	eight	months	of	therapy	in	the	first	year,	but	was	perfectly	adherent	in	the	second	year	(i.e.	filled	the	full	remaining	twelve	months	of	prescriptions),	the	calculated	MPR	would	be	approximately	58%.	However,	this	does	not	capture	the	fact	that	the	patient	went	eight	months	in	the	first	year	without	using	their	medication.	This	aspect	of	exposure	can	be	improved	by	evaluating	adherence	over	specific	exposure	windows.	In	the	example	presented	above,	using	the	two-year	follow-up	time,	a	better	approach	would	be	to	calculate	medication	adherence	in	the	first	year	(17%)	and	the	second	year	(100%).	However,	this	improvement	still	does	not	account	for	potential	dose-response	relationships.		Discontinuation	is	another	possible	way	to	define	medication	exposure	that	differs	from	non-adherence.	Non-adherence	may	simply	refer	to	a	ratio	of	medication	adherence	less	than	a	pre-defined	threshold.	For	example,	a	threshold	of	medication	adherence	of	80%	                                                11	This	assumes	that	the	patient	did,	in	fact,	use	their	medication,	which	may	not	be	the	case,	and	is	a	drawback	of	all	studies	using	administrative	pharmacy	dispensing	databases.			 45	(83)	has	been	used	to	define	adherent	users	of	a	medication.	Therefore,	patients	with	a	value	of	70%	will	be	considered	non-adherent.	Discontinuation	reflects	a	scenario	whereby	the	patient	has	stopped	taking	the	medication	as	it	was	prescribed.	These	would	be	periods	where	no	medication	was	used	by	the	patient	as	opposed	to	sporadic	usage	for	patients	classified	as	non-adherent.			The	exposure	definitions	presented	in	this	chapter	represent	different	methods	that	might	be	used	to	quantify	exposure	in	an	observational	study.	Certainly,	there	have	many	approaches	to	attempt	quantification	of	medication	exposure	and	the	choice	of	approach	will	largely	depend	on	the	available	data	and	study	objective.	The	objective	of	this	review	was	to	critically	evaluate	the	methods	of	defining	medication	exposure	in	observational	studies,	particularly	in	the	context	of	administrative	data	and	chronic	diseases.			2.4 Concluding	remarks	Defining	medication	exposure	in	observational	studies	of	chronic	diseases	requires	careful	attention	to	the	nuances	of	the	specific	exposure-outcome	associations	under	investigation.	This	review	chapter	sought	to	critically	appraise	methods	that	have	been	used	in	defining	medication	exposure	in	previous	studies	and	bring	forth	the	potential	biases	that	may	result	from	these	various	approaches.	The	review	also	emphasized	the	advantages	to	the	use	of	defined	time-windows	to	enhance	exposure	assessment,	the	use	of	latency	periods	for	diseases	such	as	cancer,	and	the	advantages	of	using	a	recency-weighted	cumulative	dose	or	duration	of	exposure	definition	to	simultaneously	capture	current	and	previous		 46	exposures	to	a	particular	medication.	Several	of	the	methods	that	were	discussed	in	this	chapter	will	be	used	in	the	analyses	presented	in	subsequent	chapters	of	this	dissertation.					 47	Chapter	3: Do	inhaled	corticosteroids	(ICS)	protect	against	lung	cancer	in	patients	with	chronic	obstructive	pulmonary	disease?	A	systematic	review12	 Summary	Chapter	3	systematically	reviews	the	literature	that	evaluates	the	association	between	ICS	use	and	lung	cancer	risk	in	chronic	obstructive	pulmonary	disease	(COPD)	patients.	The	search	strategy	allowed	for	the	inclusion	of	both	observational	and	trial-based	studies.	This	chapter	provides	another	component,	along	with	Chapter	2,	for	the	work	that	will	be	presented	in	Chapter	5	of	this	thesis.		 3.1 Background	Chronic	obstructive	pulmonary	disease	(COPD)	is	a	term	that	includes	subjects	with	chronic	bronchitis	and/or	emphysema	and	is	typically	associated	with	fixed	airflow	obstruction	that	is	progressive	and	seldom	reversible	(100).	The	primary	risk	factor	for	COPD	is	cigarette	smoking,	which	is	associated	with	approximately	85%	of	cases	(32,101–103).	The	economic	burden	associated	with	COPD	is	substantial	and	is	expected	to	increase	                                                12	A	version	of	this	chapter	has	been	accepted	for	publication:	Raymakers	AJN,	McCormick	N,	Marra	CA,	FitzGerald	JM,	Sin	DD,	Lynd	LD.	Do	inhaled	corticosteroids	(ICS)	protect	against	lung	cancer	in	patients	with	chronic	obstructive	pulmonary	disease?	A	systematic	review.	Respirology.	(Electronically	published:	September	2016).		 48	in	coming	years.	Total	direct	medical	costs	attributable	to	COPD	and	its	associated	comorbidities	in	the	United	States	are	expected	to	increase	from	$32	billion	in	2010	to	$49	billion	in	2020	(104).	By	2020,	COPD	is	expected	to	be	the	third-leading	cause	of	death	worldwide	(105).		Chronic	obstructive	pulmonary	disease	is	strongly	associated	with	the	development	of	lung	cancer,	which	is	the	leading	cause	of	cancer	mortality	in	males	and	females	in	developed	countries	(106).	In	2012,	there	were	an	estimated	1.5	million	lung	cancer	deaths	worldwide	(106).	Lung	cancer	is	often	only	detected	at	an	advanced	stage	and	is	typically	associated	with	a	poor	prognosis	(107).	While	smoking	is	the	most	significant	risk	factor	for	lung	cancer,	approximately	10%	of	individuals	who	develop	lung	cancer	are	lifetime	non-smokers	(108).			Although	smoking	is	a	risk	factor	for	both	COPD	and	lung	cancer,	evidence	suggests	that	COPD	itself	is	an	additional	risk	factor	for	lung	cancer	development,	independent	of	smoking	status	or	history	(32,33,35,38).	For	example,	Young	et	al.	(32)	reported	a	six-fold	greater	prevalence	of	COPD	in	lung	cancer	subjects	versus	smoking-status,	sex,	and	age-matched	controls.	Mannino	et	al.	(33)		found	that	moderate	to	severe	COPD	was	associated	with	an	increased	risk	of	lung	cancer	diagnosis	(HR:	2.8	(95%	CI:	1.8-4.4)),	even	after	adjustment	for	age,	sex,	and	smoking	status.		Purdue	et	al.	(38)	showed	in	a	longitudinal	study	of		men	(n=176,997)	rendered	similar	results,	with	both	mild	and	moderate/severe	COPD	associated	with	increased	smoking-adjusted	rates	of	lung	cancer	(RR:	1.5	(95%	CI:	1.2-1.9)	and	RR:	2.2	(95%	CI:	1.8-2.7),	respectively).	A	longitudinal	study	by	Skillrud	et	al.		 49	(34)	reported	the	smoking-adjusted	probability	of	developing	lung	cancer	was	10.8%	in	COPD	patients	and	2.5%	in	controls	(p=0.023).	Turner	et	al.	(35),	in	a	study	of	448,600	lifelong	non-smokers	in	the	US,	found	that	lung	cancer	mortality	was	significantly	associated	with	combined	emphysema/chronic	bronchitis	(HR:	2.44	(95%	CI:	1.22-4.90)),	over	a	follow-up	period	of	twenty	years.	Given	the	evidence	presented	above,	it	seems	apparent	that	the	evidence	strongly	suggests,	after	accounting	for	smoking	history,	COPD	remains	an	independent	risk	factor	for	lung	cancer.			The	association	between	lung	cancer	and	COPD	may	be	well	established,	but	the	specific	mechanism	for	the	link	between	COPD	and	lung	cancer	is	not,	although	several	hypotheses	have	been	postulated.	One	such	hypothesis	is	that	airway	inflammation	associated	with	COPD	can	cause	phenotypic	alterations	in	lung	cells.	For	example,	in	those	COPD	patients	with	a	history	of	smoking,	Skillrud	et	al.	(34)	have	suggested	a	mechanism	whereby	COPD	enables	carcinogens	to	reside	in	the	lungs	for	longer	due	to	decreased	mucociliary	clearance	resulting	from	COPD-related	inflammation,	and	these	substances	tend	to	sit	in	areas	where	lung	tumours	occur	most	often	(34).	Additional	literature	suggests	that	COPD	severity	is	positively	associated	with	the	development	of	lung	cancer	(32).	For	example,	van	Eeden	et	al.	(29)	showed	that	markers	of	systemic	inflammation	such	as	the	level	of	C-reactive	protein	(CRP),	which	are	associated	with	lung	cancer	development	(see	Table	1.2),	fluctuate	with	an	individual’s	FEV1;	that	is,	as	CRP	levels	increase,	lung	function	declines.	A	meta-analysis	by	Wasswa-Kintu	et	al.	(36)	also	reported	that	reduced	FEV1	is	strongly	associated	with	lung	cancer	development,	independent	of	smoking	status,	particularly	in	women.	Therefore,	it	would	appear	that	the	association	between	COPD	and	lung	cancer	is		 50	potentially	enhanced	by	the	severity	or	degree	of	systemic	and	local	inflammation	that	accompanies	COPD.				The	Global	Initiative	for	Chronic	Obstructive	Lung	Disease	(GOLD)	recommend	inhaled	corticosteroids	(ICS)	as	treatment	for	severe	and	very	severe	COPD	patients	especially	in	those	with	a	history	of	exacerbations	(5).	These	patients	typically	have	a	post-bronchodilator	forced	expiratory	volume	in	1	second	(FEV1)	of	<	50%	of	predicted	and	experience	one	or	more	acute	exacerbations	per	year	(5,6).	However,	the	evidence	on	the	benefit	conferred	from	ICS	in	COPD	patients	is	not	unanimous.		Inhaled	corticosteroid	use	in	COPD	patients	appears	to	provide	benefits	for	some	aspects	of	the	disease,	but	not	for	all.	For	example,	ICS	use	has	been	shown	to	improve	patients’	quality	of	life	measured	as	an	improvement	on	the	St	George's	Respiratory	Questionnaire	(109).	Treatment	with	ICS	has	also	been	shown	to	reduce	the	risk	of	acute	exacerbations	of	COPD	(AECOPD)	(44,45).	However,	the	evidence	is	mixed	as	to	whether	ICS	use	is	associated	with	appreciable	gains	in	lung	function.	The	TORCH	trial	(44)	showed	that	there	was	a	small	improvement	in	lung	function,	but	a	meta-analysis	by	Yang	et	al.	(110)	showed	mixed	results	depending	on	the	analytical	approach.	In	terms	of	whether	there	is	benefit	for	reduced	all-cause	mortality	in	COPD	patients,	Sin	et	al.	(111)	showed	that	ICS	use	reduces	the	risk	of	all-cause	mortality.	However,	the	results	of	the	TORCH	trial	(44)	failed		 51	to	show	a	statistically	significant	association	between	ICS	use	and	reduced	mortality13.	Gartlehner	et	al.	(45)	showed	a	similar	result,	reporting	no	statistically	significant	reduction	in	mortality	associated	with	ICS	use.	Finally,	evidence	suggests	that	ICS	use	reduces	markers	of	systemic	inflammation,	which	are	often	present	in	COPD	patients,	and	which	are	also	associated	with	increased	lung	cancer	risk	(56).	The	consequence	of	this	conflicting	evidence	is	contentiousness	as	to	the	most	appropriate	use	of	ICS	in	COPD	patients.			As	the	global	prevalence	of	COPD	increases,	it	can	be	anticipated	that	lung	cancer	incidence	will	commensurately	increase.	A	possible	protective	effect	of	ICS	for	lung	cancer	in	COPD	patients	could	help	add	to	the	evidence	for	the	prescription	of	ICS	in	COPD,	which	could	potentially	help	to	improve	both	COPD-related	outcomes	and	lung	cancer-associated	mortality.	The	objective	of	this	study	was,	therefore,	to	identify	and	critically	appraise	studies	that	have	sought	to	answer	this	research	question,	and	identify	avenues	for	future	research.				                                                13	The	estimated	p-value	for	the	association	between	ICS	use	and	mortality	was	very	near	statistical	significance,	p=0.052.		 52	3.2 Methods	3.2.1 Search	strategy		A	systematic	literature	search	was	conducted	in	the	following	Ovid	databases:	MEDLINE	(1950	to	November	2015),	EMBASE,	EBM	Reviews	-	Cochrane	Central	Register	of	Controlled	Trials	(CENTRAL),	and	International	Pharmaceutical	Abstracts,	Web	of	Science,	and	BIOSIS	Previews.		Database	searches	were	conducted	using	both	subject	headings	(MeSH	headings	in	MEDLINE	and	CENTRAL,	Emtree	in	EMBASE)	and	keywords.		Four	domains	were	incorporated	in	the	search:	1)	COPD	(including	emphysema	and	chronic	bronchitis),	2)	lung	cancer	(including	subtypes),	3)	inhaled	corticosteroids	(including	individual	drug	names),	and	4)	chemoprevention.		Filters	selecting	for	only	RCTs	and	observational	study	designs	were	added	to	some	searches.		In	MEDLINE	and	CENTRAL	all	the	MeSH	headings	and	most	of	the	keywords	applicable	to	the	different	lung	cancer	subtypes	were	included.	A	list	of	MeSH	headings	pertaining	to	generic	drug	name	was	compiled	from	background	literature	searches	and	those	listed	under	the	broader	MeSH	heading	of	glucocorticoid	drugs.	References	of	identified	studies	were	also	searched	for	relevant	publications.	The	search	strategy	is	provided	in	Appendix	A.								 53		Figure	3.1.	Study	selection	process.		a	One	study	(44)	was	included	based	on	the	authors’	prior	knowledge	that	the	desired	data	was	available	from	supplemental	material	provided	by	the	manufacturer	and	sponsor	of	the	clinical	trial.			 54	3.2.2 Study	selection	criteria	To	be	included	in	this	systematic	review,	a	study	was	required	to	have	subjects	with	a	diagnosis	of	COPD	with	some	proportion	of	subjects	treated	with	ICS.	In	an	effort	to	exclude	COPD	cohorts,	which	may	have	included	misclassified	asthma	cases,	COPD	patients	needed	to	be	greater	than	40	years	of	age.	We	retrieved	both	RCTs	and	observational	studies.	A	search	method	was	employed	where	RCTs	examining	the	efficacy	and	other	long-term	outcomes	of	ICS	use	(comparing	ICS	therapy	alone	with	another	therapy,	such	as	β-agonists,	or	combination	therapy)	were	sought	since	they	often	included	secondary	data	on	the	causes	of	subject	withdrawal	and	death,	both	within	the	study	period,	and	over	the	longer	term.	Studies	needed	to	include	lung	cancer	diagnosis	or	lung	cancer-related	mortality	as	a	primary	or	secondary	outcome.	Where	possible,	online	appendices	and	openly	available	RCT	data	were	searched	to	determine	whether	there	was	information	reported	on	lung	cancer	cases	or	mortality.	Study	inclusion	was	based	on	independent	assessment	by	two	reviewers	(AR	and	NM).	Where	necessary,	disagreement	between	selections	was	resolved	through	discussion	with	the	other	authors.		3.2.3 Data	extraction	General	study	details	(including	title,	authors,	year	of	publication	and	data	collection,	country	of	origin,	study	type,	and	study	duration)	were	extracted	from	each	study	using	a	pre-determined	data	abstraction	form.	The	main	outcome	of	interest	was	a	lung	cancer	diagnosis	(Table	3.2),	though	data	were	also	collected	on	lung	cancer	deaths	(Table	3.4).		With	regard	to	ICS	exposure,	details	were	collected	on	the	generic	name	of	each	drug,	dose	(quantity	per	inhalation	and	number	of	inhalations	per	use),	names	and	doses	of	other		 55	medications	given,	and	frequency	and	duration	of	use.	A	decision	was	made	not	to	pool	results	due	to	both	methodological	and	clinical	heterogeneity	between	studies	(112,113).	Studies	varied	substantially	in	patient	populations,	follow-up	time,	ICS	prescribed	and	how	lung	cancer	diagnoses	were	reported.	A	risk	of	bias	assessment	was	completed	in	accordance	with	the	established	guidelines	and	is	presented	in	Appendix	B	(114).		A	checklist	completed	in	accordance	with	the	Preferred	Reporting	Items	for	Systematic	Reviews	and	Meta-Analyses	(PRISMA)	Statement	(115)	is	presented	in	Appendix	C.		 56	Table	3.1.	Characteristics	of	included	studies.				 Calverley	et	al.		(44)	 Kiri	et	al.	(116)	Lung	Health	Study	Research	Group	(117)	Parimon	et	al.	(92)	 Pauwels	et	al.	(118)	Tashkin	et	al.	(119)		 	 	 	 	 	 	Study	Design	 RCT	 Observational;	nested	case-control	 RCT	 Observational;	cohort	 RCT	 RCT	Country	 42	countries	 United	Kingdom	 United	States,	Canada	 United	States	Belgium,	Denmark,	Finland,	Italy,	Netherlands,	Norway,	Spain,	Sweden,	United	Kingdom	United	States,	Czech	Republic,	Netherlands,	Poland,	South	Africa	Years	of	Data	Collection	 2000-2005	 1989-2003	 1994-1999	 1996-2004	 1992-1996	 NR	Follow-Up	Period	 3	years	 ≤	16	years	 4.5	years	 34	months	 3	years	 6	months	Sample	Size	Placebo:	1545;	Salmeterol:	1542;	Fluticasone:	1551;	Fluticasone+	Salmeterol:	1546	=	6184	127	cases	+	1,470	controls=1,597	 559	treated	+	557	ICS-controls=1,116	 517	ICS-exposed	+	9957	ICS-unexposed=10,474	 634	treated	+	643	ICS-controls=1,277	 1120	treated	+	584	ICS-controls=1,704	Age	Parameters	 >=40	years	of	age	 ≥50	years	at	diagnosis	 40-69	years	 ≥40	years	at	study	entry	 30-65	years	at	start	of	follow-up	 ≥40	years	at	study	entry	%	Male	Placebo:	76%;	Salmeterol:	76%;	Fluticasone:	75%;	Fluticasone+	Salmeterol:	75%	Lung	cancer	cases=64%;	Controls=65%	ICS-exposed=64%;	ICS-unexposed=64%	 ICS-exposed=97%;	ICS-unexposed=97%	ICS-exposed=74%;	ICS-unexposed=72%	ICS-exposed=69%;	ICS-unexposed=67%		 57	Table	3.1:	Characteristics	of	included	studies		(…continued).		 Calverley	et	al.		(44)	 Kiri	et	al.	(116)	 Lung	Health	Study	Research	Group	(117)	 Parimon	et	al.	(92)	 Pauwels	et	al.	(118)	 Tashkin	et	al.	(119)		 	 	 	 	 	 	Mean	Age	Placebo:	65.0	(8.2);	Salmeterol:	65.1	(SD:	8.2);	Fluticasone:	65.0	(8.4);	Fluticasone+	Salmeterol:	65.0	(8.3)	Lung	cancer	cases=70.7	years,	Controls=70.8	years	ICS-exposed=56.2,	ICS-unexposed=56.4	 ICS-exposed=66,	mean	ICS-unexposed=64	ICS-exposed=52.5;	ICS-unexposed	mean=52.4	ICS-exposed=63.4;	ICS-unexposed=63.35	Subjects’	Smoking	Status	Placebo:	48.6	(26.9);	Salmeterol:	49.3	(27.7);	Fluticasone:	49.2	(28.6);	Fluticasone+	Salmeterol:	47.0	(26.5)	Former	smokers	(cases	for	mean	2.5	years,	controls	for	mean	2.9	years)	Current	(ICS=90.5%,	placebo=89.8%)	or	quit	within	past	two	years	ICS-exposed=9%	never,	68%	former,	23%	current.	ICS-unexposed=12%	never,	53%	former,	35%	current.	Current	or	former	 Current	(ICS	mean=43.4%,	non-ICS	mean=40.8%)	or	former	Subjects’	Smoking	History/	Quantity	≥	10	pack-years	 n/a	 Current	ICS=22.9	cigarettes/day;	Current	placebo=24.2/day	Exposed=11%	<10/day,	13%	11-15/day,	22%	16-20/day,	22%	21-30/day,	14%	31-40/day,	9%	>40/day.	Non-exposed=14%	<10/day,	12%	11-15/day,	22%	16-20/day,	18%	21-30/day,	11%	31-40/day,	10%	>40/day	ICS	mean	pack-years=39.4,	placebo	mean	pack-years=39.2	ICS	median	pack-years=40.75,	non-ICS	median	pack-years=40		 58	3.3 Results	The	systematic	literature	search	identified	4645	initial	records.	After	removing	duplicates	and	non-English	language	studies,	3339	results	remained.	At	this	stage,	studies	that	either	explicitly	stated	a	different	patient	population	or	that	made	no	reference	to	a	COPD	patient	population	or	an	ICS	exposure	or	treatment	were	removed	(n=2800).	Four-hundred	eighty-six	abstracts	remained	to	be	reviewed,	of	which	85	full-text	studies	warranted	further	review.	At	this	stage,	the	most	common	reason	for	exclusion	was	studies	that	did	not	contain	lung	cancer	as	study	outcome	(n=43).	Six	articles	met	the	criteria	to	ultimately	be	included	in	this	systematic	review:	four	RCTs	and	two	observational	studies	(40,87,110–113)).	The	RCTs,	which	examined	the	efficacy	of	ICS,	included	COPD	patients	(n=1116	to	n=6184	per	study)	of	different	severities	enrolled	at	multiple	centers	throughout	the	United	States,	Canada,	and	Europe.	The	observational	studies	used	administrative	health	databases	from	the	United	States	(n=10,474	COPD	patients	≥40	years	of	age	followed	up	for	up	to	4.6	years)	and	the	United	Kingdom	(n=7079	ex-smokers	≥50	years	of	age	with	incident	COPD	followed	up	for	up	to	sixteen	years)	to	assess	the	impact	of	ICS	prescriptions	on	lung	cancer	development.	Characteristics	of	the	included	studies	are	presented	in	Table	3.1.	Details	of	ICS	use,	including	the	specific	ICS,	dose,	and	frequency	of	administration	in	each	study,	are	presented	in	Table	3.2.	A	flow	chart	illustrating	the	study	selection	process	is	presented	in	Figure	3.1.	The	design	of	each	study	was	assessed	for	potential	risk	of	bias	based	on	six	different	criteria	(114).	The	majority	of	identified	studies	demonstrated	a	‘low	risk’	of	bias.	None	of	the	identified	studies	demonstrated	a	‘high	risk’	of	bias.	Further	details	of	the	risk	of	bias	assessment	are	available	in	Appendix	B.			 59	3.3.1 Randomized	controlled	trials	The	primary	objective	of	each	identified	RCT	was	to	examine	the	efficacy	of	ICS	use	in	COPD.		The	Lung	Health	Study	Research	Group	(117)	studied	the	effect	of	triamcinolone	on	COPD	progression.		The	primary	outcome	was	the	rate	of	decline	in	FEV1,	but	cause-specific	morbidity	and	mortality	were	included	as	secondary	outcomes.	Five	subjects	in	the	ICS	group	(n=559)	and	four	subjects	in	the	placebo	group	(n=557)	died	of	lung	cancer,	producing	a	RR	of	1.25	(95%	CI:	0.34-4.61)	(Table	3.2).	The	mean	follow-up	time	in	this	study	was	40	months,	however,	the	authors	report	that	adherence	to	triamcinolone	may	have	been	as	low	as	53.7%in	the	treatment	arm	(and	58.5%	in	the	placebo	arm,	based	on	canister	weights).				Pauwels	et	al.	(118)	studied	the	long-term	effects	of	budesonide	in	COPD	patients	with	mild	disease.		The	primary	outcome	was	the	change	in	post-bronchodilator	FEV1.		Seven	patients	in	the	ICS	group	(n=634)	and	ten	in	the	placebo	group	(n=643)	withdrew	due	to	a	diagnosis	of	lung	cancer,	resulting	in	a	relative	risk	of	0.71	(95%	CI:	0.27-1.85).	In	addition	to	these	withdrawals,	the	authors	reported	three	deaths	from	lung	cancer	in	the	ICS	treatment	group	and	three	deaths	in	the	controls.		The	associated	relative	risk	for	lung	cancer	mortality	was	1.01	(95%	CI:	0.21-5.01).				Tashkin	et	al.	(119)	studied	the	efficacy	and	safety	of	an	ICS	(budesonide)	and	a	beta-agonist	(formoterol)	combination	inhaler	in	COPD	patients	with	moderate	to	severe	disease.	The	primary	outcome	was	the	change	in	FEV1	before	and	after	receiving	the	medication.	Four	treatment	groups	were	exposed	to	ICS	(n=1120),	one	was	treated	with		 60	formoterol	alone	(n=284),	and	one	received	a	placebo	(n=300;	total	‘unexposed’	subjects	n=584).	The	mean	exposure	duration	among	those	receiving	ICS	(alone	or	as	combination	therapy)	was	between	157.1	(SD:	51.3)	to	168.3	(SD:	37.3)	days	(depending	on	the	treatment	group).	There	were	two	diagnoses	of	squamous	cell	carcinoma	in	the	ICS	exposed	groups	and	none	in	the	unexposed	groups,	resulting	in	an	RR	of	2.61	(0.13-54.24)	(adjusted	for	0	in	the	unexposed	group).	In	addition,	one	person	in	the	ICS	exposed	group	died	from	lung	cancer	but	none	in	the	unexposed,	resulting	in	a	RR	of	1.57	(0.06-38.29)	(adjusted	for	0).				The	Towards	a	Revolution	in	COPD	Health	(TORCH)	trial	(44)	investigated	the	effect	of	combination	therapy	(salmeterol	and	fluticasone	propionate,	as	combination	therapy,	and	each	as	monotherapy)	on	survival	of	COPD	patients14.	This	study	included	subjects	with	a	smoking	history	of	a	minimum	of	ten-pack	years	(current	of	former	smokers)	between	40	and	80	years	of	age.	To	meet	the	inclusion	criteria,	subjects	had	to	have	a	clinical	diagnosis	of	COPD	and	a	pre-bronchodilator	FEV1	<60%	of	predicted	along	with	no	significant	reversibility	of	airflow	obstruction	based	on	post-bronchodilator	change	of	less	than	10%	in	FEV1.	Reported	numbers	for	non-fatal	serious	adverse	events	show	that	63	patients	in	the	control	arm	and	68	patients	in	the	combination	therapy	arm	received	a	lung	cancer	                                                14	While	the	main	publication	from	the	TORCH	study	(44)	did	not	report	lung	cancer	specific	diagnoses	or	mortality,	this	data	was	available	in	supplementary	material	provided	by	the	manufacturer	at	the	address:	https://www.gsk-clinicalstudyregister.com/files2/21083.pdf.	The	relative	risks	reported	in	this	paragraph,	and	presented	in	Table	3.3,	have	been	calculated	by	the	author.	In	the	case	of	this	study,	lung	cancer	diagnosis	or	death	was	neither	a	primary	nor	secondary	outcome	of	the	study;	however,	given	that	the	data	were	available,	the	numbers	have	been	included.			 61	diagnosis.	The	calculated	relative	risk	between	the	two	arms	for	lung	cancer	diagnosis	was	non-significant	(RR:	1.07	(95%	CI:	0.77-1.50)15.	This	study	also	reported	lung	cancer	deaths,	with	34	occurring	in	the	control	arm	and	35	in	the	combination	therapy	arm.	The	relative	risk	for	lung	cancer	death	was	non-significant	(RR:	1.02	(95%	CI:	0.64-1.63)3.		The	results	from	these	identified	RCTs	make	it	difficult	to	make	any	inference	based	on	the	low	numbers	of	lung	cancer	diagnoses	or	deaths.	The	large	width	of	the	confidence	intervals,	which	reflect	the	uncertainty	of	the	estimated	RRs	from	the	RCTs,	demonstrates	that	these	studies	were	likely	underpowered	to	answer	the	specific	research	question	of	this	review.												                                                15	This	is	a	calculation	performed	by	the	author,	based	on	secondary	trial	data.			 62	Table	3.2.	Characteristics	of	ICS	use	among	patients	with	COPD	in	individual	identified	studies.	           Specific	ICS	 ICS	Dose	 Frequency	 Total	ICS/day	         Calverley	et	al.	(44)	Fluticasone	Propionate	 500	µg	 Twice	Daily	 1000	µg	Kiri	et	al.	(116)	 NR	 NRa	 NR	 NR	Lung	Health	Study	Research	Group	(117)	Triamcinolone	 6	x	100ug	inhalations	 Twice	per	day	 1200ug	Parimon	et	al.	(92)	Beclomethasone,	flunisolide,	fluticasone	(converted	to	triamcinolone	equivalents)	NR	 NR	 <1200ug/day,	≥1200ug/day	Pauwels	et	al.	(118)	 Budesonide	 400ug	 Twice	per	day	 800ug	Tashkin	et	al.	(119)	 Budesonide	1=160ug	ICS	/4.5ug	FM,	2=80ug	ICS/4.5ug	FM,	3=160ug	ICS	+	4.5ug	FM	separately,	4=160ug	ICS	(all	x	2	inhalations),	5=FM	only,	6=placebo	Twice	per	day	 1=640ug,	2=320ug,	3=640ug,	4=640ug	          a	This	study	did	report	on	a	'dose-response'	relationship	between	ICS	and	ICS+LABA	use	but	this	was	done	using	the	absolute	number	of	prescriptions	as	opposed	to	the	actual	dose	of	these	prescriptions.	FM=formoterol;	ICS=inhaled	corticosteroid;	LABA=long-acting	beta-agonist;	NR=not	reported;	RCT=randomized	controlled	trial.			 63	3.3.2 Observational	studies	Kiri	et	al.	(116)	used	a	nested	case-control	study	design	to	investigate	whether	ICS	exposure	was	associated	with	a	reduced	risk	of	lung	cancer	in	COPD	patients.	A	cohort	of	7079	incident	COPD	patients	(all	former	smokers)	was	identified	from	the	United	Kingdom’s	General	Practice	Research	Database	(GPRD).	A	COPD	patient	diagnosed	with	lung	cancer	was	matched	to	a	COPD	patient	without	lung	cancer,	with	equal	follow-up	time,	and	their	prescription	records	were	compared.	Subjects	were	categorized	as	users	of	a	combination	of	an	ICS	and	a	long-acting	beta-agonist	(LABA),	an	ICS	alone,	or	short-acting	bronchodilator	(SABD)	alone,	though	no	specific	medications	or	dosages	were	mentioned.			Of	the	entire	cohort	of	COPD	patients	(n=7079)	initially	identified	from	the	database,	127	developed	lung	cancer,	and	they	were	individually	matched	to	1470	controls.	The	mean	duration	of	COPD	(from	diagnosis)	was	1.5	years	(1.6	(cases),	2.1	(controls)).	The	median	time	to	lung	cancer	diagnosis,	from	COPD	diagnosis,	was	2.2	years	(IQR:	1.3-4.0).	This	population	of	COPD	patients	(n=1597)	was	approximately	30%	female,	was,	on	average,	71	years	of	age	at	the	time	of	COPD	diagnosis,	and	had	a	mean	time	since	smoking	cessation	of	more	than	2.5	years.	ICS	use	was	shown	to	have	a	protective	effect	that	was	enhanced	when	combined	with	a	LABA:	the	hazard	ratio	(HR)	for	ICS/LABA	users	was	0.50	(95%	CI:	0.27-0.90)	and	for	ICS	users	alone	was	0.64	(95%	CI:	0.42-0.98).	A	subsequent	analysis	was	performed	to	assess	a	possible	dose-response	relationship	for	this	protective	effect.	While	the	HR	for	ICS/LABA	users	with	one	or	two	prescriptions	per	year	(as	compared	to	users	of	SABD	alone)	was	non-significant	at	0.75	(95%	CI:	0.33-1.75),	there	appeared	to	be	a	dose-response	relationship,	as	measured	by	the	number	of	prescriptions	per	year,	at	higher		 64	levels	of	exposure:	ICS/LABA	users	with	three	or	more	prescriptions	per	year	had	a	HR	of	0.39	(95%	CI:	0.19-0.79),	while	patients	with	three	or	more	prescriptions	per	year	of	ICS	alone	(as	compared	to	users	of	SABD	alone)	had	an	HR	of	0.51	(95%	CI:	0.30-0.84).	Therefore,	this	study	found	that	ICS	may	protect	against	lung	cancer,	but	this	effect	may	only	exist	with	more	frequent	ICS	use,	and	the	protective	effect	may	be	stronger	when	ICS	was	taken	in	combination	with	a	LABA.		Parimon	et	al.	(92)	conducted	an	observational	cohort	study	investigating	whether	ICS	exposure	was	associated	with	a	decreased	risk	of	lung	cancer	in	COPD	patients.	This	study	had	a	very	specific,	predominantly	male,	population	(97%)	from	Veterans	Affairs	(VA)	clinics	in	the	United	States.	The	COPD	cohort	(n=10,474)	consisted	of	subjects	at	least	40	years	of	age	who	satisfied	at	least	one	of	the	following	criteria:	(i)	a	recorded	ICD-9	code	for	COPD;	(ii)	self-reported	chronic	lung	disease;	or	(iii)	prescriptions	for	bronchodilators	(beta-agonists	or	anticholinergics)	in	the	twelve	months	prior	to	enrolment	(the	index	date).		Patients’	medical	and	prescription	history	were	examined	for	ICS	use	and	subsequent	lung	cancer	diagnosis.	Data	on	comorbidities	and	tobacco	use	were	collected	at	baseline	using	a	patient	survey.	Patients’	lung	cancer	diagnosis	had	to	occur	after	the	date	of	study	enrolment.	Within	the	cohort,	423	(4%)	patients	received	a	lung	cancer	diagnosis	at	a	median	time	from	COPD	‘diagnosis’	to	lung	cancer	of	1.4	years	(IQR:	0.7-2.5).	In	the	base-case	analysis,	lung	cancer	diagnosis	could	occur	at	any	time	after	cohort	inclusion,	not	allowing	for	a	latency	period.	Failure	to	incorporate	such	a	period	is	a	major	limitation	of	this	analysis.	Cancer	latency	periods	tend	be	quite	long,	therefore,	it	is	possible	that	subjects	included	in	this	study	could	have	already	had	latent	cancer	when	ICS	exposure		 65	occurred,	limiting	the	possibility	for	this	treatment	to	have	an	effect	(in	either	direction).	ICS	exposure	was	defined	as	being	80%	adherent	in	the	180	days	previous	to	study	inclusion.	The	authors	reported	that	although	20%	of	the	cohort	had	received	an	ICS	in	the	period	prior	to	study	inclusion,	only	5%	of	those	met	this	adherence	criterion.			For	the	primary	analysis,	517	subjects	were	classified	as	ICS	users	and	9,957	non-users.	Exposure	to	ICS	as	a	continuous	variable	(per	100	µg/d)	was	not	found	be	significant	(adjusted	HR:	0.96	(95%	CI:	0.93-1.00).	For	ICS	users	with	mean	daily	dose	of	less	than	1200	µg/day,	there	was	no	association	with	lung	cancer	risk	(adjusted	HR:	1.13	(95%	CI:	0.67-1.90)	compared	to	non-users.	However,	findings	indicated	that	for	higher	doses	of	ICS	(≥	1200	ug/day)	there	was	a	protective	effect	(adjusted	HR:	0.39	(95%	CI:	0.16-0.96))	for	ICS	use	for	lung	cancer	development	compared	to	non-users.	All	analyses	were	adjusted	for	age,	smoking	status	and	intensity,	history	of	non-lung	and	non-skin	cancer,	comorbidity,	and	bronchodilator	use.			The	authors	explored	whether	there	was	an	association	between	COPD	severity	and	the	development	of	lung	cancer.	Since	lung	function	(FEV1)	values	were	not	available,	the	number	of	bronchodilator	canisters	prescribed	each	month	during	the	six	months	prior	to	study	enrolment	was	used	as	a	proxy	of	severity	(92).	Comparisons	of	the	risk	of	lung	cancer	in	beta-agonist	users	and	non-users,	and	ipratropium	users	and	non-users,	suggested	there	was	a	non-significant	trend	towards	an	increased	risk	of	lung	cancer	amongst	cases	with	more	severe	COPD	(heavier	users	of	bronchodilators).					 66	A	secondary	analysis	was	performed	where	patients	were	stratified	according	to	their	smoking	status.	In	this	analysis,	the	HR	for	ICS	use	by	former	smokers	at	<	1200	µg	/day	was	1.22	(95%	CI:	0.65-2.30),	HR	for	former	smokers	at	≥	1200	µg/day	was	0.41	(95%	CI:	0.13-1.30),	HR	for	current	smokers	at	<	1200	µg/day	was	1.01	(95%	CI:	0.36-2.81),	and	a	HR	for	current	smokers	at	≥	1200	µg/day	was	0.39	(95%	CI:	0.10-1.64).	These	results	further	suggest	the	presence	of	a	dose–response	relationship	between	ICS	use	and	lung	cancer	risk.		To	assess	the	potential	impact	of	confounding	by	indication	(whereby	COPD	patients	who	were	prescribed	ICS	are	at	a	higher	risk	of	lung	cancer	than	individuals	who	were	not	prescribed	ICS)	a	propensity	score,	representing	an	individual’s	probability	of	being	prescribed	ICS,	was	calculated	for	each	subject.	After	adjusting	for	the	propensity	score,	the	hazard	ratio	for	the	<	1200	µg/day	group	remained	non-significant	(HR:	1.35	(95%	CI:	0.82-2.23)),	while	that	for	the	≥	1200ug/day	group	was	still	indicative	of	a	protective	effect	but	shifted	from	being	statistically	significant	(HR:	0.39	(95%	CI:	0.16-0.96))	to	non-significant	(HR:	0.57	(95%	CI:	0.24-1.38)).				Results	from	sensitivity	analyses	regarding	the	cohort	definition	included	restricting	the	case	definition	to	those	individuals	(n=3233)	who	met	all	three	of	the	inclusion	criteria:	a	recorded	ICD-9	code	for	COPD,	self-reported	COPD,	and	a	history	of	bronchodilator	prescriptions	(HR,	<	1200ug/day:	0.98	(95%	CI:	0.54-1.80);	HR,	≥	1200ug/day:	0.44	(95%	CI:	0.18-1.09)),	or	restricting	the	case	definition	to	only	those	(n=2493)	with	an	ICD-9	code	for	COPD	and	prescriptions	for	ipratropium	(HR,	<	1200ug/day:	1.09	(95%	CI:	0.55-2.19);		 67	HR,	≥	1200ug/day:	0.37	(95%	CI:	0.13-1.01)).	Interestingly,	excluding	subjects	who	were	diagnosed	with	lung	cancer	within	one	year	of	the	index	date	resulted	in	a	non-significant	association	between	ICS	use	and	lung	cancer	development	(HR	<	1200	µg/day:	0.85	(95%	CI:	0.39-1.84);	HR,	≥	1200	µg	/day:	0.41	(95%	CI:	0.13-1.31)).				Another	subgroup	analysis	examined	the	risk	of	lung	cancer	only	amongst	the	517	members	of	the	cohort	who	were	users	of	ICS.	Findings	from	this	analysis	(HR:	0.90	(95%	CI:	0.82-0.99)	for	every	100	µg	increase	in	dose)	provided	more	evidence	for	a	protective	dose-response	relationship	with	ICS.	As	discussed	by	the	authors,	the	consistency	between	the	primary	analysis	and	subgroup/sensitivity	analyses	suggests	that	the	apparent	protective	effect	of	ICS	is	robust	to	various	potential	confounders	and	analytical	approaches.													 68	Table	3.3.	Results	from	identified	studies	for	the	relative	risk	of	lung	cancer	diagnosis	for	ICS	exposed/treated	and	unexposed/untreated	patients	with	COPD.	Study	 Exposure/Treatment	 Risk	Among	ICS-Exposed/Treated	Risk	Among	ICS-Unexposed/Controls	Relative	Risk	(95%	CI)			 	 	 	 	Calverly	et	al.	(44)	 -	 68/3067=2.22%a	 63/3045=2.07%	 1.07	(0.76-1.50)	Kiri	et	al.	(116)	†	 ICS	alone	(overall)	 61/841=7.3%	 46/421=10.9%	 0.64	(0.42-0.98)b		 ICS	alone:≥3	prescriptions/year	 NR	 NR	 0.51	(0.30-0.84)		 ICS	alone:	1-2	prescriptions/year	 NR	 NR	 0.88	(0.51-1.52)		 ICS	and	LABA	(overall)	 20/335=6.0%	 46/421=10.9%	 0.50	(0.27-0.90)*b		 ICS	and	LABA:	≥3	prescriptions/year	 NR	 NR	 0.39	(0.19-0.79)		 ICS	and	LABA:	1-2	prescriptions/year	 NR	 NR	 0.75	(0.33-1.75)	Parimon	et	al.	(92)	†	 <1,200ug/day	triamcinolone	equivalents	(overall)	 16/298=5.4%	 402/9,957=4.0%	 1.13	(0.67-1.90)		 <1,200ug/day	triamcinolone	equivalents	(former	smokers)	11/201=5.5%	 211/5,188=4.1%	 1.22	(0.65-2.30)		 <1,200ug/day	triamcinolone	equivalents	(current	smokers)	4/64=6.3%	 168/3,349=5.0%	 1.00	(0.36-2.81)		 ≥1,200ug/day	triamcinolone	equivalents	(overall)	 5/219=2.3%	 402/9957=4.0%	 0.39	(0.16-0.96)*		 ≥1,200ug/day	triamcinolone	equivalents	(former	smokers)	3/140=2.1%	 211/5,188=4.1%	 0.41	(0.13-1.30)		 ≥1,200ug/day	triamcinolone	equivalents	(current	smokers)	2/53=3.8%	 168/3,349=5.0%	 0.39	(0.10-1.64)	Pauwels	et	al.	(118)	 -	 7/634=1.1%	 10/643=1.6%	 0.71	(0.27-1.85)	Tashkin	et	al.	(119)	 -	 2/1,120=0.2%	 0/584=0%	 2.61	(0.13-54.24)c			 		 		 		 		*	Statistically	significant.	a	This	is	the	number	of	ICS	treated	patients,	both	as	monotherapy	and	combined	with	salmeterol.	b	Adjusted	for	use	of	other	respiratory	medications,	comorbidities,	duration	of	smoking	cessation,	duration	of	COPD,	and	demographic/clinical	characteristics.	c	Adjusted	for	0	by	adding	0.5	to	the	value	of	each	numerator	and	denominator.	†Observational	study.			 69	3.4 Discussion	This	systematic	review	identified	studies	that	evaluated	the	association	between	ICS	use	and	lung	cancer	risk	in	COPD	patients.	The	results	of	this	review	suggest	that	there	are	several	important	methodological	considerations	that	are	important	in	assessing	the	relationship	between	ICS	use	and	lung	cancer	risk.	Our	search	identified	four	relevant	RCTs,	which	only	reported	small	numbers	of	lung	cancer	diagnoses	and	deaths	as	secondary	data	from	trials	evaluating	the	efficacy	of	ICS	in	COPD.		However,	two	observational	studies	were	identified	that	directly	addressed	this	research	question.	ICS	exposure	in	these	studies	was	associated	with	a	significantly	lower	incidence	of	lung	cancer,	in	their	respective	COPD	populations.	The	results	of	these	two	studies	also	suggest	there	is	a	dose-response	relationship,	with	the	protective	effect	appearing	at	ICS	doses	greater	than	1200	ug/day	(92),	and	when	patients	receive	three	or	more	ICS	prescriptions	per	year	(116).	This	evidence,	however,	should	be	interpreted	cautiously	as	there	are	methodological	concerns	with	both	studies.		Findings	in	identified	studies	were	clearly	dependent	on	the	study	type,	with	the	RCTs	suggesting	ICS	use	had	no	significant	effect	while	the	observational	studies	suggested	ICS	use	might	protect	against	lung	cancer.	It	must	be	noted,	however,	that	the	RCTs	reported	in	this	review	were	not	designed	or	powered	for	looking	at	the	incidence	of	rare	events	so	these	results	are	very	prone	to	a	Type	II	error	(i.e.	failure	to	detect	a	significant	difference	when	one	exists).	It	can	be	argued	that	the	observational	studies	are	more	prone	to	bias	because	the	distribution	of	potentially	confounding	factors	may	not	be	equally	balanced		 70	amongst	the	exposure	groups.		However,	in	the	observational	cohort	study	by	Parimon	et	al.	(92),	the	two	groups	that	were	compared	had	similar	smoking	intensity,	and	the	protective	effect	in	both	observational	studies	remained	after	adjustment	for	known	confounders	like	COPD	duration,	comorbidities,	age,	and	smoking	status.	These	studies	have	the	potential	for	results	to	be	susceptible	to	confounding	by	indication	whereby	patients	that	are	prescribed	ICS	actually	have	more	severe	COPD.	In	that	case,	we	would	expect	those	receiving	ICS	would	have	a	higher	chance	of	developing	lung	cancer.	This	was	not	the	case	in	either	observational	study,	both	studies	showed	protective	effects	of	ICS	in	lung	cancer	development.	Therefore,	it	could	be	the	case	that	these	results	are,	in	fact,	conservative	estimates	of	the	benefits	of	ICS	for	lung	cancer	development	in	COPD	patients.	In	addition,	despite	the	potential	for	bias,	observational	studies	may	be	the	most	adequate	design	for	such	a	study	question.			The	latency	period	associated	with	lung	cancer	also	means	that	protopathic	bias	should	be	an	important	consideration.	Protopathic	bias	can	occur	when	medication	exposure	might	be	based	on	manifestations	of	an	undiagnosed	(i.e.	subclinical)	disease	(120).	Therefore,	in	this	instance,	it	could	be	possible	that	the	symptoms	of	lung	cancer	in	COPD	patients	lead	to	ICS	being	prescribed	prior	to	lung	cancer	diagnosis.	However,	in	that	case	we	would	expect	that	there	would	be	an	increased	risk	of	lung	cancer	(or	at	least	no	significant	and	protective	effect	of	ICS	use),	which	was	not	the	case	in	the	observational	studies.		The	influence	of	protopathic	bias	was	discounted	in	a	recently	published	observational	study	evaluating	the	impact	of	ICS	exposure	on	lung	cancer	incidence	(90).	This	study,	which	was	not	included	in	this	review	because	the	study	population	included	subjects	with	either		 71	COPD,	asthma,	or	other	respiratory	conditions,	reported	a	significant	protective	effect	for	ICS	against	lung	cancer.		Like	the	two	observational	studies	that	were	included	in	this	review,	this	protective	effect	was	dose-dependent.	With	respect	to	protopathic	bias,	those	authors	performed	a	sensitivity	analysis	where	exposure	to	ICS	within	the	three	or	six	months	prior	to	lung	cancer	diagnosis	was	excluded,	and	the	protective	effect	of	ICS	persisted	(90).	Neither	identified	observational	study	(92,116)	used	a	lagged	exposure	period	to	reduce	the	possibility	of	protopathic	bias	in	their	base-case	analysis,	however	Parimon	et	al.	(92)	did	so	in	a	sensitivity	analysis.	The	authors	removed	lung	cancer	cases	that	occurred	in	the	first	year	of	follow-up,	reducing	the	number	of	cases	from	423	to	254,	and	found	no	significant	association	with	ICS	exposure	(either	as	a	continuous	variable	or	stratified	by	low	and	high	dose).	However,	it	is	possible	that	the	reduced	numbers	from	restricting	this	analysis	did	not	provide	sufficient	power	to	detect	an	association.			The	RCTs	included	in	this	review	were	designed	to	study	the	short-term	efficacy	of	ICS	therapy	but	were	underpowered,	both	in	terms	of	the	sample	size	and	follow-up	period,	for	detecting	a	significant	effect	on	lung	cancer	in	either	direction.	Lung	cancer	has	a	latency	period	that	is	likely	longer	than	the	follow	up	time	most	of	these	RCTs.	A	patient	could	feasibly	develop	lung	cancer	at	day	one	of	the	study	and	may	not	become	symptomatic	until	after	the	study	was	completed.	We	must	also	consider	the	likely	duration	of	time	necessary	for	ICS	use	to	make	an	impact	on	lung	cancer	risk.	For	example,	the	six-month	follow-up	period	in	Tashkin	et	al.	(119),	was	likely	far	too	short.	Also,	in	these	trials	lung	cancer	events	were	based	on	adverse	event	reporting	and	withdrawals	during	the	follow-up	period;	therefore,	even	though	the	two	RCTs	had	longer	follow-up	periods	(3	years	(118)		 72	and	4.5	years	(117)),	these	events	may	have	occurred	within	a	shorter	period	of	time	after	the	start	of	ICS	exposure.	Although	COPD	patients	are	at	an	increased	risk	of	lung	cancer,	the	reported	incidence	rates	of	lung	cancer	in	this	population,	48	to	64	per	10,000	individuals	(121),	would	not	result	in	a	substantial	number	of	diagnoses	amongst	the	RCT	study	populations.	While	understandable,	given	the	cost	and	resources	required	to	conduct	an	RCT,	none	of	those	studies	included	in	this	review	enrolled	a	sufficient	number	of	subjects	in	both	the	exposed	and	unexposed	groups	to	detect	a	significant	difference	(see	Table	3.1).		For	example,	to	obtain	the	same	0.56	unadjusted	RR	for	lung	cancer	diagnosis	with	ICS	exposure	reported	by	Parimon	et	al.	(92),	(α=0.05	and	80%	power),	one	would	need	1599	subjects	in	each	study	arm	of	a	RCT.	In	comparison,	the	study	by	the	Lung	Health	Study	Research	Group	(117)	had	less	than	half	this	number	of	subjects	in	each	exposure	group.		Even	more	subjects	would	be	required	when	taking	a	more	conservative	approach	and	anticipating	a	smaller	relative	risk	reduction	from	ICS.	Therefore,	despite	the	potential	for	bias	in	observational	studies,	it	is	likely	the	ideal	design	for	studying	this	research	question	given	the	extended	follow-up	period	required	to	detect	lung	cancer	and	relatively	low	event	rate.			While	observational	studies	are	likely	to	be	superior	to	RCTs	to	answer	this	research	question,	for	reasons	such	as	sample	size	and	feasibility	of	adequate	follow-up	periods,	the	observational	studies	included	in	this	review	have	several	major	limitations.	First,	the	choice	of	patient	populations	was	questionable	in	both	studies.	Parimon	et	al.	(92)	focused	on	a	patient	population	that	was	almost	exclusively	male	(97%).	Kiri	et	al.	(116)	focused	only	on	patients	that	had	quit	smoking.	A	population-based	cohort	analysis	would	be	better		 73	suited	to	answer	this	research	question.	Second,	the	follow-up	time	in	both	studies	was	still	short,	particularly	when	considered	in	conjunction	with	the	latency	period	of	lung	cancer.	Third,	the	complexity	of	the	exposure-outcome	relationship	between	ICS	use	and	lung	cancer	as	an	outcome	was	not	fully	explored.	It	is	likely	that	the	duration	of	ICS	use,	in	addition	to	its	dose,	might	be	crucial	to	whether	or	not	an	effect	can	be	observed.	An	exploration	of	the	relationship	between	medication	exposure	and	lung	cancer	development	warrants	further	investigation,	and	will	be	presented	in	Chapter	5.			3.4.1 Limitations	This	review	has	several	limitations.	First,	a	language	bias	might	exist	because	we	were	only	able	to	assess	studies	written	in	English.	However,	the	majority	of	studies	identified	by	the	search	were	in	written	in	English,	therefore	we	are	confident	in	the	robustness	of	our	results.	Second,	we	did	not	contact	authors	of	all	RCTs	evaluating	ICS	efficacy	for	additional	data	on	ICS	exposure	status	or	lung	cancer	events.	The	rationale	is	that	the	objective	of	this	study	was	to	appraise	methodological	approaches	to	evaluating	the	association	between	ICS	use	and	lung	cancer	risk	in	COPD	patients.	Therefore,	while	we	conducted	a	very	sensitive	search	to	reduce	the	probability	of	missing	relevant	studies,	the	true	aim	was	to	discuss	the	methods	in	the	studies	identified,	particularly	in	the	observational	studies	specifically	addressing	this	relationship.	In	addition,	in	the	case	where	an	RCT	does	exist	that	may	have	secondary	data	on	this	relationship,	it	is	likely	that	it	would	not	have	had	sufficient	power	(given	the	rarity	of	lung	cancer	and/or	short	time	horizons),	nor	would	there	have	been	a	systematic	look	for	lung	cancer.	Finally,	while	often	beneficial,	we	actively	elected	not	to	pool	study	results	due	to	the	heterogeneity	of	studies	included	in		 74	this	review	but	contend	the	value	of	this	review	is	the	identification	of	methodological	considerations	that	are	required	for	future	studies	attempting	to	answer	this	research	question.		Table	3.4.	Results	from	identified	studies	for	the	relative	risk	of	lung	cancer	specific	death	in	for	ICS	treated	and	untreated	patients	with	COPD.	Study	 Risk	Among	ICS-Exposed/Treated	Risk	Among	ICS-Unexposed/	Controls	Relative	Risk	(95%	CI)		 	 	 	Calverley	et	al.	(44)	 35/3067=	1.1%a	 34/3045=1.1%	 1.02	(0.64-1.63)	Lung	Health	Study	Research	Group	(117)	 5/559=0.9%	 4/557=0.7%	 1.25	(0.34-4.61)	Pauwels	et	al.	(118)	 3/634=0.5%	 3/643=0.5%	 1.01	(0.21-5.01)	Tashkin	et	al.	(119)	 1/1,120=0.1%	 0/584=0%	 1.57	(0.06-38.29)b			 		 		 		*	Statistically	significant.	a	This	is	the	number	of	ICS	treated	patients,	both	as	monotherapy	and	combined	with	salmeterol.	b	Adjusted	for	0	by	adding	0.5	to	the	value	of	each	numerator	and	denominator	 				 75	3.5 Closing	remarks	In	conclusion,	this	review	sought	to	critically	appraise	the	literature	on	whether	ICS	use	decreases	the	risk	of	lung	cancer	in	patients	with	COPD.	The	results	from	randomized	controlled	trials	and	observational	studies	were	conflicting,	with	the	RCTs	demonstrating	that	ICS	use	does	not	affect	these	patients’	risk	of	lung	cancer,	while	observational	studies	found	that	ICS	use	did	reduce	this	risk,	particularly	at	high	doses.	However,	the	RCTs	identified	here	were	likely	underpowered	to	answer	the	research	question.	Given	the	substantial	burden	of	lung	cancer,	this	question	has	significant	implications	and	needs	to	be	addressed	with	more	appropriate	research	methods,	using	a	population-based	longitudinal	cohort	design	and	proper	consideration	of	the	complex	exposure	relationship	between	ICS	and	lung	cancer	development.	 76	Chapter	4: Mortality	in	COPD	patients	that	use	statins:	a	population-based	cohort	study16		Summary	This	chapter	presents	an	evaluation	of	the	association	between	statin	use	and	mortality	within	a	cohort	of	COPD	patients.	Patients	were	identified	as	having	COPD	if	they	had	received	three	prescriptions	for	an	anticholinergic	medication	or	a	short-acting	beta	agonist	in	a	365-day	period.	For	this	chapter,	statin	exposure	is	defined	in	the	one-year	period	after	this	COPD	‘diagnosis’.	The	objective	of	this	study	is	to	evaluate	the	association	between	statin	use	by	COPD	patients	with	pulmonary-related	and	all-cause	mortality.	Previous	observational	evidence	suggests	that	statin	treatment	might	offer	a	protective	effect	against	mortality	in	COPD	patients.	However,	a	recently	completed	randomized	controlled	trial	studying	the	efficacy	of	statin	use	and	acute	exacerbations	of	COPD	has	casted	doubt	on	the	beneficial	properties	of	statin	use	in	COPD	patients.			                                                16	A	version	of	this	chapter	has	been	accepted	for	publication	with	the	journal	Chest:	Raymakers	AJN,	Sadatsafavi	M,	Sin	DD,	De	Vera	MA,	and	Lynd	LD.	The	impact	of	statin	use	on	all-cause	mortality	in	patients	with	COPD:	a	population	based	cohort	study.	Chest	(2017).	[Currently	in	press].	Also,	an	earlier	version	of	the	results	of	this	study	were	presented	(as	a	poster)	at	the	Canadian	Association	for	Population	Therapeutics	Conference.	Reference:	Raymakers	AJN,	Sadatsafavi	M,	Lynd	LD.	One-year	survival	among	statin	users	in	a	population-based	cohort	of	COPD	patients.	Canadian	Association	for	Population	Therapeutics	(CAPT).	Toronto,	Ontario	(2015,	2	November).		 77	4.1 	Introduction	Chronic	obstructive	pulmonary	disease	(COPD)	affects	380	million	people	worldwide,	representing	12%	of	adults	over	30	years	of	age	(122)	and approximately	9-10%	of	Canadians	over	40	(1,2).	It	is	a	progressive	and	mostly	irreversible	disease	associated	with	airway	inflammation	and	airflow	limitation.	An	important	aspect	of	COPD	is	that	it	is	often	associated	with	several	comorbidities.	For	example,	the	estimated	prevalence	of	cardiovascular	disease17	(CVD)	in	patients	with	COPD	is	greater	than	two	times	that	of	the	general	population	(14,27,28).	Mannino	et	al.	(27)	reported	that	in	a	cohort	of	20,296	subjects	with	COPD,	3091	(15.2%)	had	concomitant	CVD,	significantly	higher	than	general	population	estimates	(123)18.	Similarly,	COPD	patients	are	at	increased	risk	of	CVD-related	mortality	which	may	increase	as	the	severity	of	patients’	COPD	worsens	(118,119).	The	importance	of	CVD	and	other	comorbidities	in	COPD	is	such	that	all-cause	mortality	has	largely	become	the	most	relevant	metric	for	outcomes	in	these	patients	(25).			Localized	chronic	inflammation	of	the	airways	has	long	been	observed	in	COPD	patients	but	there	is	a	growing	understanding	of	systemic	inflammation	in	a	subset	of	patients	with	COPD.	Specifically,	high	levels	of	C-reactive	protein	(CRP)	and	interleukin-6	(IL6)	have	been	associated	with	poor	outcomes	in	COPD	patients	(18,120).	Sin	et	al.	(14)	reported	on	the	association	between	systemic	inflammation	and	airflow	obstruction,	showing	that	                                                17	Cardiovascular	disease	typically	comprises:	ischemic	heart	disease,	cerebrovascular	disease,	peripheral	vascular	disease,	heart	failure,	rheumatic	heart	disease,	and	congenital	heart	disease.		18	The	Public	Health	Agency	of	Canada	reports	that	general	population	(≥	20)	for	2014	are	6.2%	(95%	CI:	5.9%-6.5%),	with	higher	prevalence	among	males	(7.2%	(95%	CI:	6.7%-7.7%)	than	females	(5.2%	(95%	CI:	4.8%-5.6%)	(119).			 78	patients	with	severe	airflow	obstruction	(FEV	≤	50%	of	predicted)	were	more	than	two	times	more	likely	to	have	elevated	CRP	levels	and	that	these	individuals	were	more	likely	to	report	previous	cardiac	injury	(14).	Agusti	et	al.	(18)	identified	a	specific	COPD	phenotype	that	is	characterized	by	persistent	low-grade	systemic	inflammation	and	increased	risk	of	adverse	cardiovascular	outcomes,	which	may	benefit	from	targeted	therapy	aimed	at	reducing	the	inflammatory	process	(18).	Young	et	al.	(126)	also	suggest	that	COPD	progression	might	be	the	result	of	systemic	inflammation,	rather	than	a	‘spillover’	from	inflammation	in	the	lungs,	and	that	measuring	inflammation	may	provide	valuable	prognostic	information.		Patients	with	cardiovascular	disease,	regardless	of	COPD	status,	are	commonly	treated	with	a	HMG-CoA	reductase	inhibitor	(i.e.,	a	statin).	These	are	lipid-lowering	agents	to	treat	patients	with	elevated	cholesterol	levels	(primary	prevention)	and/or	with	established	cardiovascular	disease	(secondary	prevention).	Statins	have	been	shown	to	be	effective	in	reducing	all-cause	mortality	in	patients	with	risk	factors	for	CVD	(58,59).	In	patients	with	COPD,	observational	evidence	has	shown	statin	use	may	reduce	the	risk	of	acute	exacerbations	of	COPD	(AECOPD)	(57,122,123),	reduce	respiratory-related	mortality	(129),	and	reduce	all-cause	mortality	(62,63).	The	prevailing	hypothesis	is	that	the	reduction	in	inflammation	provided	is	the	mechanism	by	which	statins	can	potentially	reduce	the	risk	of	exacerbations	in	COPD	patients	(57,122)	and	further	evidence	also	suggests	that	statin	use	may	be	associated	with	a	reduced	cancer	risk.	For	example,	Khurana	et	al.	(130)	showed	that	statin	use	was	associated	with	a	45%	reduction		 79	(adjusted)	in	lung	cancer	risk	and	this	effect	was	more	pronounced	in	patients	with	greater	than	six	months	of	statin	use,	where	there	was	a	55%	reduction	in	lung	cancer	risk	(130).			As	stated,	the	retrospective	observational	evidence	for	statin	use	in	COPD	patients	has	supported	the	hypothesis	that	statin	use	may	improve	outcomes	for	COPD	patients,	typically	by	reducing	the	frequency	and	severity	of	AECOPD	(127).	However,	the	recently	completed	Prospective	Randomized	Placebo-Controlled	Trial	of	Simvastatin	in	the	Prevention	of	COPD	Exacerbations	(STATCOPE)	trial	attempted	to	answer	this	same	question	using	a	prospective	randomized	controlled	trial	design	(66).	In	this	study,	there	were	no	significant	differences	in	exacerbation	rates	or	time	to	exacerbation	between	statin	users	and	controls.	While	the	results	of	this	study	may	appear	to	be	strong	because	of	the	prospective	randomized	design,	the	exclusion	criteria	for	study	participants	may	have	limited	the	generalizability	of	results.	Patients	in	this	trial	were	excluded	if	they	had	ever	received	a	statin	previously	or	had	any	indication	for	receiving	a	statin,	including	pre-existing	CVD,	which,	as	stated,	is	more	prevalent	in	COPD	patients.	Exclusion	of	these	patients	may	have	represented	a	scenario	whereby	patients	with	characteristic	systemic	inflammation	often	believed	to	be	the	aetiology	of	COPD-related	comorbidities,	were	excluded	from	the	trial.	Thus,	the	trial	may	have	only	included	patients	that	were	unlikely	to	benefit	from	statins.	A	subsequent	publication	by	the	study	authors	reported	low	levels	of	CRP,	in	both	study	arms,	compared	to	previously	published	studies	of	systemic		 80	inflammation	in	stable	COPD	patients19	(131)	which	may	support	the	hypothesis	that	those	patients	with	higher	levels	of	systemic	inflammation	were	excluded	from	the	STATCOPE	trial	and	questions	the	generalizability	of	the	results.			As	such,	the	relationship	of	statin	exposure	to	pulmonary-related	and	all-cause	mortality	in	COPD	patients	in	a	‘real-world’	setting	remains	unresolved.	Therefore,	the	primary	objective	of	this	study	was	to	evaluate	the	impact	of	statin	use	on	all-cause	and	pulmonary-related	mortality	in	COPD	patients.			4.2 Methods	This	was	a	retrospective	population-based	cohort	study	based	on	administrative	data	from	the	province	of	British	Columbia	(BC),	Canada,	from	1997	to	2007.	Population	Data	BC	(PopDataBC)	is	an	extensive	data	resource	that	contains	health	services	information	from	linkable	databases	(described	below)	for	all	residents	of	British	Columbia	(approximately	4.7	million	residents,	December	201520)	which	provides	the	opportunity	to	conduct	generalizable,	population-based	health	research.			The	PharmaNet	prescription	data	file	includes	prescriptions	filled	for	the	entire	population	of	British	Columbia	regardless	of	insurer	or	payer	(132).	These	data	are	linked	to	the	data	                                                19	The	comparator	is	the	‘Macrolide	Azithromycin	to	Prevent	Rapid	Worsening	of	Symptoms	Associated	With	Chronic	Obstructive	Pulmonary	Disease	(MACRO)’	trial.		20	British	Columbia	Statistics	(BCStats),	Population	Estimates.	http://www.bcstats.gov.bc.ca/StatisticsBySubject/Demography/PopulationEstimates.aspx		 81	on	hospital	separations	(the	Discharge	Abstract	Database	(DAD)	(133),	deaths	(British	Columbia	Vital	Statistics	–	Deaths	(134)),	a	registration	file	containing	the	regional	health	authority	and	census	tract	neighborhood	income	data	(135),	and	physician-billing	data	from	the	provincially	administered	universal	insurance	program	(the	Medical	Services	Plan	(MSP)	data	file	(136)).		4.2.1 Identification	of	the	COPD	cohort	Patients	50	years	of	age	and	older	were	identified	who	had	received	at	least	three	prescriptions	for	an	anticholinergic	or	a	short-acting	beta	agonist	(SABA)	in	a	twelve	month	floating-time	window;	that	is,	over	twelve	continuous	months	between	1998	and	2007.	The	index	date	was	defined	as	the	date	of	receipt	of	the	first	of	these	three	prescriptions.	A	one-year	‘wash-in’	period	for	COPD	was	used	to	identify	incident	COPD	cases.	The	‘wash-in’	period	attempts	to	exclude	prevalent	cases	of	COPD	by	removing	those	patients	that	would	satisfy	the	inclusion	criteria	in	the	first	year	of	which	data	are	available.			4.2.2 Identification	of	statin	users	Statin	use	was	identified	from	PhamaNet	records	using	American	Hospital	Formulary	Service	(AHFS)	codes	(240608;	‘HMG-CoA	Reductase	Inhibitors’)	(137).	The	available	statins	that	had	ever	been	prescribed	to	the	population	of	COPD	patients	during	the	study	period	were:	atorvastatin,	fluvastatin,	lovastatin,	pravastatin,	rosuvastatin,	and	simvastatin	(see	Figure	4.2).	From	the	index	date,	a	one-year	exposure	ascertainment	window	was	constructed	to	classify	COPD	patients	as	exposed	or	unexposed	to	statins.	Patients	that	received	a	statin	within	365	days	of	their	study	index	date	were	classified	as	being	statin		 82	exposed,	and	thus,	exposure	status	was	fixed	over	the	follow-up	period	(the	outcome	ascertainment	period,	365	days)	(see	Figure	4.1).	Only	statins	received	after	the	patients’	index	date	were	considered	to	define	exposure	status.	This	makes	it	possible	for	patients	to	have	received	a	prior	statin	prescription	without	being	classified	as	‘exposed’.	However,	misclassification	as	not	statin	exposed	could	only	occur	if	the	individual	received	a	prescription	prior	to	the	index	date	and	then	a	subsequent	prescription	at	least	366	days	thereafter.	In	this	scenario,	it	is	unlikely	that	such	sporadic	statin	usage	might	confer	any	benefit	and	although	technically	misclassified	as	'unexposed'	this	may	more	accurately	represent	these	patients’	exposure	status.		Further,	the	advantage	of	using	this	defined	window	of	exposure	ascertainment	in	an	observational	cohort	study	is	that	it	reduces	the	probability	than	an	individual	in	the	study	is	more	likely	to	have	been	exposed	due	to	longer	follow-up	time	and	means	that	exposure	periods	and	outcome	periods	are	non-overlapping,	thus	avoiding	complex	time-dependent	biases	(82,83,132).				 83		Figure	4.1.	Exposure	and	outcome	ascertainment	periods	in	the	primary	analysis.		In	sensitivity	analyses,	the	length	of	time	used	to	assess	exposure	to	statins	was	varied	in	to	determine	its	effect	on	the	results.	In	addition,	the	medication	possession	ratio	(MPR)	was	calculated	to	distinguish	between	statin	users	that	had	multiple	or	longer	statin	prescriptions	compared	to	those	that	did	not.	The	MPR	was	calculated	by	identifying	all	the	prescriptions	received	by	an	individual	patient	within	the	exposure	ascertainment	window	and	the	associated	days	supplied	of	each	prescription	was	aggregated	and	divided	by	the	length	of	exposure	ascertainment	window	to	give	a	ratio	representative	of	the	proportion	of	days	in	the	exposure	window	for	which	the	patient	had	a	been	dispensed	a	statin.			Time	(days)PrescriptionDispensedPatients	that	were	dispensed	a	statin	prescription	during	 the	‘exposure	ascertainment	window’	are	considered	exposed.	Patients	dispensed	 their	only statin	prescription	outside	of	this	window	will	not	be	considered	exposed.365 7300Exposure	Ascertainment	Window Outcome	Ascertainment	WindowOutcome	 84	4.2.3 Outcome	ascertainment	The	primary	outcome	for	this	analysis	was	all-cause	mortality	and	the	secondary	outcome	was	pulmonary-related	mortality.	These	were	evaluated	in	the	period	that	began	366	days	after	the	index	date	and	lasted	until	up	to	730	days	after	the	index	date	(i.e.	the	365-day	period	beyond	the	exposure	ascertainment	window;	see	Figure	4.1).	This	approach	prevented	the	overlap	of	the	exposure	and	outcome	periods	(83,132)	and	has	been	used	previously	in	evaluating	medication	exposure	and	hospital	readmissions	in	asthma	patients	(87).	An	advantage	to	this	approach	is	that	it	minimizes	the	potential	for	immortal	time	bias	by	fixing	exposure	status	prior	to	the	outcome	ascertainment	period	(139).	Mortality	was	assessed	as	time-to-death	from	the	end	of	the	exposure	ascertainment	window	to	the	date	of	death	or	to	the	end	outcome	ascertainment	period	for	a	maximal	total	of	365	days	for	the	survivors.	The	cause	of	death	was	identified	using	the	listed	ICD-10	code	in	the	Vital	Statistics	Deaths	file	(134).	Pulmonary-related	deaths	were	those	associated	using	ICD-10	codes:	J15	to	J44	(65).			4.2.4 Adjustment	for	potential	confounders	Covariates	that	were	considered	as	potential	confounders	for	the	association	between	statin	exposure	and	mortality	were	obtained	during	the	exposure	ascertainment	window.	The	demographic	covariates	included:	age,	sex,	neighborhood	income	quintiles	based	place	of	residence,	and	health	authority	(regional	health	service	entity)	in	which	the	patient	resided.	In	addition,	for	each	patient	the	number	of	prescriptions	dispensed,	the	number	of	hospital	encounters,	the	number	of	inpatient	hospital	stays,	and	the	number	of	physician	encounters	were	calculated.	Finally,	to	account	for	comorbidities	at	the	beginning	of	follow-	 85	up,	Charlson	Comorbidity	Index	was	calculated,	excluding	COPD	and	CVD,	based	on	health	services	use	records	during	the	exposure	ascertainment	period	(140,141).			4.2.5 Statistical	analyses	A	Cox	regression	model	was	constructed	to	model	the	time	from	the	end	of	the	exposure	ascertainment	window	to	the	time	to	the	outcome	of	interest	(mortality	in	the	primary	analysis)	or	censoring.	Bivariate	analyses	of	potential	confounders	were	conducted	with	the	outcome	of	interest	such	that	those	with	a	p-value	less	than	0.20	were	considered	for	inclusion	in	the	multivariable	analysis.	The	multivariable	model	was	chosen	by	stepwise	addition	of	candidate	variables	from	the	bivariate	analyses	using	the	aforementioned	criterion	and	compared	at	each	step	using	Akaike	Information	Criterion	(AIC)	(142).	All	analyses	were	conducted	using	SAS	Version	9.4	(SAS	Institute	Inc,	Cary,	NC,	USA).	No	personal	identifying	information	was	made	available	as	part	of	this	study;	procedures	used	were	in	compliance	with	British	Columbia’s	Freedom	of	Information	and	Privacy	Protection	Act.	Ethics	approval	was	obtained	from	the	University	of	British	Columbia.		4.2.6 Sensitivity	analyses	Several	sensitivity	analyses	were	performed	to	test	the	robustness	of	our	results	(presented	in	Table	4.5).	The	exposure	ascertainment	window	was	varied	to	six	months	and	to	eighteen	months.	Alternative	definitions	of	exposure	were	also	used	to	evaluate	the	effect	of	the	specific	definition	on	study	results.	The	medication	possession	ratio	(MPR)	for	statins	was	calculated	by	summing	the	days	supplied	of	statins	received	by	an	individual	patient	over	the	length	of	exposure	ascertainment	window.	We	also	dichotomized	MPR		 86	using	a	commonly	used	threshold	of	0.8	(90,135)	to	compare	‘adherent’	and	‘non-adherent’	statin-users,	as	a	binary	variable,	where	the	reference	category	included	both	non-users	and	those	that	did	not	meet	this	threshold.	Finally,	an	exposure	definition	based	on	the	cumulative	dose	of	statins	received	during	the	exposure	ascertainment	window	was	explored.				4.3 Results	4.3.1 Cohort	of	COPD	patients	The	cohort	of	COPD	patients	consisted	of	39,678	patients	which	were	identified	and	included	in	the	analysis.	The	mean	age	in	the	cohort	of	COPD	patients	on	the	study	index	date	was	71.0	(SD:	11.6)	years	and	54.7%	were	female.	There	were	1446	(4%)	deaths	recorded	within	the	COPD	cohort	in	the	outcome	ascertainment	period.	Further	details	of	the	COPD	cohort	are	available	in	Table	4.1.									 87	Table	4.1.	Patient	characteristics,	stratified	by	exposure	status	to	statins.				 		 		Patient	Characteristic	 Statin	Exposed	 Not	Exposed			 		 			   N	 7566	 32112	Age	 69.9	(SD:	8.8)	 70.5	(SD:	11.4)	%	female	 46.2%	 55.1%	Age	Distribution	(n	(%))	50<60	 1175	(15.5%)	 7292	(22.7%)	60<70	 2460	(32.5%)	 7971	(24.8%)	70<80	 3005	(39.7%)	 9512	(29.6%)	>80	 926	(12.2%)	 7337	(22.8%)	Income	Quintile	(#	(%))	 	  1	 1841	(25.8%)	 7808	(25.4%)	2	 1497	(21.0%)	 6324	(20.6%)	3	 1393	(19.5%)	 5512	(18.0%)	4	 1170	(16.4%)	 5236	(17.1%)	5	 1033	(14.5%)	 4633	(15.1%)	Health	Authority	(#	(%))	Interior		 1504	(19.9%)	 7099	(22.2%)	Fraser		 2370	(31.3%)	 8926	(27.9%)	Vancouver	Coastal		 1341	(17.7%)	 6374	(19.9%)	Vancouver	Island		 1385	(18.3%)	 6109	(19.1%)	Northern		 512	(6.7%)	 1957	(6.1%)	Charlson	Comorbidity	Index	(CCI)	 3	(2-5)	 2	(1-4)	CCI	Category	 	  0	 247	(3.4%)	 1621	(5.2%)	1	 1141	(15.9%)	 8124	(26.2%)	2	 1376	(19.2%)	 5932	(19.1%)	>=3	 4421	(61.5%)	 15351	(49.5%)	Any	Hospitalization	(overnight)		 1621	(21.4%)	 7513	(18.9%)	Number	of	Prescriptions	Receiveda		 39	(26-60)	 31	(18-53)	Number	of	Physician	Encountersa	 14	(8-24)	 12	(6-20)			 		 		CCI:	Charlson	Comorbidity	Index;	SD:	standard	deviation.	Note:	Where	percentages	do	not	add	to	100%	the	reason	is	due	to	rounding.	Except	where	noted,	continuous	variables	are	presented	as	mean	and	standard	deviation.	a	Median	and	interquartile	range.		 88		Figure	4.2.	Distribution	of	statins	prescribed	within	the	one-year	exposure	ascertainment	window	in	the	cohort	of	COPD	patients.		4.3.2 Statin	use	in	the	cohort	of	COPD	patients	Within	this	cohort	of	COPD	patients,	7775	(19.6%)	patients	filled	at	least	one	statin	prescription	in	the	exposure	ascertainment	window.	There	were	a	total	of	41,897	statin	prescriptions	dispensed	during	this	period,	with	an	average	of	five	prescriptions	per	patient-year,	among	patients	who	had	filled	at	least	one	prescription.	The	distribution	of	specific	statins	dispensed	is	shown	in	Figure	4.2.	The	most	prescribed	statin	was	atorvastatin	(calcium)	(49.1%).	Most	patients	received	a	30-day	(15%),	90-day	(23%),	or	100-day	(20%)	supply	of	statins.	0102030405060Atorvastatin	Calcium Fluvastatin	Sodium Lovastatin Pravastatin	Sodium Rosuvastatin	Calcium SimvastatinPercentage	of	Prescriptions	 89	Table	4.2.	Hazard	ratios	from	bivariate	regression	analysis	to	assess	potential	confounders,	with	time	to	all-cause	mortality	as	the	outcome	variable.			 		 		 		 		Parameter		 HR	 95%	Confidence	Interval		 p-value			 		 		 		 			     Statin	Exposure	 0.79	 0.69	 0.91	 0.0012	Age	(continuous)	 1.06	 1.05	 1.06	 <.0001	Age	Category	Age	50-60		 Reference	Age	60-70	 1.69	 1.36	 2.12	 <.0001	Age	70-80	 2.97	 2.43	 3.64	 <.0001	Age	80+	 5.44	 4.46	 6.64	 <.0001	Sex	(Ref=female)	 1.35	 1.22	 1.50	 <.0001	Any	Physician	Encounter	(Ref=None)	 5.06	 2.99	 8.56	 <.0001	Number	of	Physician	Encounters	 1.02	 1.02	 1.02	 <.0001	Any	Hospitalization	(overnight)		 2.52	 2.26	 2.80	 <.0001	Total	Length	of	Hospital	Stay	Over	the	Period	(days)	 1.00	 1.00	 1.01	 <.0001	Number	of	Hospitalizations		 1.46	 1.40	 1.52	 <.0001	Charlson	Comorbidity	Index	(continuous)	 1.14	 1.13	 1.16	 <.0001	Charlson	Comorbidity	Category	0	 Reference	1	 0.81	 0.582	 1.14	 0.2236	2	 1.09	 0.78	 1.56	 0.6189	≥3	 2.33	 1.72	 3.16	 <.0001	Number	of	Prescriptions	Received		 1.00	 1.00	 1.01	 <.0001	Health	Authority		Interior		 Reference	Fraser	 0.93	 0.80	 1.08	 0.3247	Vancouver	Coastal		 1.03	 0.88	 1.20	 0.7266	Vancouver	Island	 1.00	 0.85	 1.17	 0.9557	Northern	 1.04	 0.83	 1.31	 0.7294	Income	Quintiles	 	    5	 Reference	4	 1.20	 0.98	 1.46	 0.0715	3	 1.16	 0.96	 1.42	 0.1277	2	 1.28	 1.06	 1.55	 0.0095	1	 1.38	 1.15	 1.65	 0.0005			 		 		 		 			 90	Table	4.3.	Hazard	ratios	obtained	from	multivariable	regression	analysis	for	association	between	statin	exposure	and	all-cause	mortality	(primary	analysis).			 		 		 		 		Parameter		 HR	95%	Confidence	Interval		 p-value			 		 		 		 			     Statin	Exposure	 0.79	 0.68	 0.92	 0.0016	Age	(continuous)	 1.05	 1.04	 1.06	 <.0001	Sex	(Ref=female)	 1.36	 1.22	 1.51	 <.0001	Physician	Encounter	(Ref=None)	 0.71	 0.41	 1.20	 0.1986	Any	Hospitalization	(overnight)	 1.72	 1.54	 1.93	 <.0001	Charlson	Comorbidity	Index	0	 Reference	1	 0.89	 0.63	 1.26	 0.5079	2	 0.90	 0.64	 1.27	 0.5453	≥3	 1.56	 1.13	 2.15	 0.0068	Number	of	Prescriptions	Received		 1.00	 1.00	 1.01	 <.0001	Income	Quintiles	5	 Reference	4	 1.20	 0.99	 1.47	 0.0681	3	 1.18	 0.97	 1.43	 0.1038	2	 1.26	 1.04	 1.52	 0.0174	1	 1.32	 1.10	 1.58	 0.0026			 		 		 		 					4.3.3 All-cause	and	pulmonary-related	mortality	In	a	bivariate	analysis,	the	hazard	ratio	for	statin	use	associated	with	all-cause	mortality	was	0.79	(95%	CI:	0.69-0.91,	p=0.0012),	indicating	a	protective	effect	of	statin	exposure	(see	Table	4.2).	This	significant	and	protective	effect	of	statin	exposure	for	one-year	all-cause	mortality	was	maintained	in	the	multivariable	analysis	suggesting	a	21%	reduction		 91	in	the	risk	of	mortality	(adjusted	HR:	0.79	(95%	CI:	0.68-0.92,	p=0.0016)	(Table	4.3).	Similar	findings	were	obtained	for	pulmonary-related	mortality	with	an	estimated	bivariate	hazard	ratio	of	0.51	(95%	CI:	0.31-0.85,	p=0.0016)	and	in	multivariable	analysis,	an	adjusted	HR	of	0.57	(95%	CI:	0.34-0.96,	p=0.0254)	indicating	a	greater	than	40%	reduction	in	the	risk	of	mortality	for	pulmonary-specific	causes	(Table	4.4).			4.3.4 Sensitivity	analyses	4.3.4.1 Exposure	ascertainment	window	Several	model	assumptions	were	adjusted	to	determine	their	effect	on	the	results	of	our	study.	First,	the	length	of	the	exposure	ascertainment	window	was	shortened	to	six	months	after	their	COPD	index	date.	In	this	analysis,	there	were	3986	statin	users	identified.	This	resulted	in	134	(3.4%)	all-cause	deaths	among	statin	users	and	1482	(4.5%)	among	non-statin	users	in	the	one-year	follow-up	period.	In	multivariable	analysis,	the	estimated	HR	showed	a	protective	effect	from	statin	exposure	on	all-cause	mortality	(HR:	0.80	(95%	CI:	0.71-0.90,	p=0.0001).		Similarly,	an	analysis	was	conducted	in	which	the	exposure	ascertainment	window	was	extended	to	a	period	of	eighteen	months.	A	protective	effect	was	also	exhibited	in	multivariable	analysis	with	this	lengthened	exposure	ascertainment	window,	with	an	estimated	HR	of	0.66	(95%	CI:	0.54-0.80,	p=0.0002)	for	mortality	within	the	one	year	outcome	ascertainment	period.				 92	Table	4.4.	Multivariable	regression	analysis	for	the	association	between	statin	exposure	and	pulmonary-related	mortality	(secondary	analysis).			 		 		 		 		Parameter		 HR	95%	Confidence	Interval		 p-value			 		 		 		 			     Statin	Exposure	 0.55	 0.32	 0.93	 0.0254	Age	(continuous)	 1.07	 1.05	 1.08	 <.0001	Sex	(Ref=female)	 1.14	 0.82	 1.58	 0.4244	Any	Physician	Encounter	(Ref=None)	 1.06	 0.15	 7.65	 0.9550	Any	Hospitalization	(overnight)		 2.08	 1.49	 2.92	 <.0001	Charlson	Comorbidity	Index	0	 Reference	1	 1.85	 0.56	 6.14	 0.3153	2	 1.96	 0.59	 6.44	 0.2709	≥3	 1.76	 0.55	 5.63	 0.3403	Number	of	Prescriptions	Received		 1.01	 1.00	 1.01	 <.0001	Income	Quintiles	5	 Reference	4	 0.85	 0.46	 1.57	 0.6028	3	 1.19	 0.68	 2.09	 0.5380	2	 0.94	 0.53	 1.66	 0.8371	1	 1.18	 0.70	 1.99	 0.5295			 		 		 		 										 93	Table	4.5.	Multivariable	regression	models	with	time	to	all-cause	mortality	as	the	outcome	using	alternative	specifications	for	the	exposure	variable.			 		 		 		 		Parameter		 Hazard	Ratio	 95%	Confidence	Interval		 p-value			 		 		 		 			     Exposure	Definition		    Medication	Possession	Ratio(MPR)	 0.80	 0.68	 0.94	 0.008	Adherent	(MPR≥0.80)*	 0.85	 0.71	 1.01	 0.0592	Cumulative	Dose		 0.98	 0.96	 1.01	 0.2231		     Exposure	Ascertainment	Window	 	    6	Months	 0.80	 0.71	 0.90	 0.0001	18	Months		 0.74	 0.63	 0.86	 0.0002			 		 		 		 		*	Only	patients	with	an	MPR	greater	than	or	equal	to	0.80	are	considered	exposed.		**Models	are	adjusted	for	covariates	listed	in	Table	4.3.				4.3.4.2 Statin	adherence		The	medication	possession	ratio	(MPR)	was	calculated	for	patients	over	the	course	of	the	exposure	ascertainment	window.	This	gave	some	indication	of	those	patients	that	would	have	received	multiple	prescriptions	or	longer	duration	of	statin	therapy.	Using	the	MPR	as	a	continuous	variable,	the	adjusted	HR	was	0.80	(95%	CI:	0.68-0.94,	p=0.008)	indicating	a	protective	effect	for	statin-use	and	all-cause	mortality,	similar	to	the	20%	reduction	in	risk	in	the	primary	analysis.	An	MPR	threshold	of	0.8	was	then	used	to	distinguish	between	an	adherent	statin	user	and	a	non-adherent	statin	user	(90,135).	There	were	4582	(58.9%)	statin	users	who	met	this	criterion	for	being	considered	adherent.	Statin	users	that	did	not	achieve	the	adherence	level	of	0.8	were	considered	unexposed,	reflecting	a	conservative		 94	estimate	of	the	effect	of	statin	exposure.	The	HR	in	this	analysis	was	in	a	similar	direction	as	previous	results	but	the	estimated	HR	was	not	statistically	significant	(adjusted	HR:	0.85	(95%	CI:	0.71-1.01,	p=0.0592).							Figure	4.3.	Number	of	statin	exposed	patients,	by	MPR	category,	in	the	1-year	exposure	ascertainment	window.	 010203040506070<0.20 [0.20,	0.40) [0.40,	0.60) [0.60,	0.80) >=0.80Percemtage	of	PatientsMPR	Categories	 95	4.3.4.3 Statin	cumulative	dose		The	average	cumulative	dose	(grams)	of	statins	received	during	the	exposure	ascertainment	window	was	3.8	(SD:	3.5).	The	estimated	HR	for	the	association	between	the	risk	of	all-cause	mortality	and	cumulative	dose	(grams)	in	the	adjusted	analysis	was	not	significant	but	in	the	expected	direction	(HR:	0.98	(95%CI:	0.96-1.01,	p=0.	2231)).				4.4 Discussion	This	study	used	population-based	administrative	data	from	the	province	of	British	Columbia,	Canada,	to	evaluate	the	association	of	patients'	outcomes	with	statin	use	in	a	cohort	of	COPD	patients.	The	findings	of	this	study	suggest	that	statin	use	is	associated	with	reduced	pulmonary-related	and	all-cause	mortality	in	COPD	patients.	Further,	the	results	were	robust	across	several	different	definitions	of	statin	exposure.			Statins	have	been	demonstrated	to	reduce	cholesterol	levels	thereby	potentially	reducing	the	risk	of	cardiovascular	disease	(59).	Previous	observational	studies	have	shown	that	statins	might	also	reduce	the	frequency	of	acute	exacerbations	of	COPD	(127).	However,	this	evidence	has	been	disputed	recently	due	to	the	negative	results	of	the	STATCOPE	trial,	which	showed	no	association	between	statin	use	(simvastatin,	40mg	daily)	and	acute		 96	exacerbations	compared	to	placebo	(66)21.	The	results	of	that	study	were	contentious,	largely	due	to	the	study	inclusion	criteria	(68).	Specifically,	patients	in	this	trial	were	excluded	if	they	had	ever	received	a	statin	previously	or	had	any	indication	for	receiving	a	statin	including	those	who	had	glycated	hemoglobin	level	beyond	6.5%	or	mildly	elevated	blood	cholesterol	levels.	Indeed,	nearly	one	in	three	patients	screened	for	this	trial	were	excluded	owing	to	these	and	other	factors.	Exclusion	of	these	patients	may	have	represented	a	scenario	whereby	patients	with	elevated	levels	of	systemic	inflammation	were	excluded	from	the	trial	(67).	Thus,	the	final	sample	might	have	been	comprised	predominantly	of	patients	who	would	not	benefit	from	statins,	casting	doubt	about	the	generalizability	of	results.	Indeed,	a	subsequent	publication	by	the	study	authors	reported	low	levels	of	CRP	compared	to	previously	published	studies	of	systemic	inflammation	in	stable	COPD	patients	(131).	Moreover,	STATCOPE	did	not	have	sufficient	power	to	assess	the	effects	of	statins	on	mortality	of	COPD	patients.	In	the	study	presented	in	this	chapter,	the	association	between	statin	exposure	and	acute	exacerbations	of	COPD	was	not	explored,	however,	a	significant	effect	was	found	in	a	reduction	in	pulmonary-related	and	all-cause	mortality,	underscoring	the	potential	benefits	of	statin	use	in	some	COPD	patients.	Statins	have	also	been	linked	other	benefits,	such	as	reducing	the	risk	of	cancer,	generally	(144),	and	lung	cancer	specifically	(130),	for	which	COPD	patients	are	known	to	have	an	elevated	risk,	not	only	because	COPD	patients	are	typically	current	or	former	smokers,	but	also	because	of	the	increased	systemic	inflammation	associated	with	COPD.		                                                21	This	study	found	no	statistically	significant	difference	in	the	mean	number	of	exacerbations	between	the	statin	arm	and	control	arm	of	the	study,	1.36	+/-	1.61	and	1.39	+/-	1.73,	respectively	(p=0.54).	Moreover,	the	time	to	first	COPD	exacerbation	was	also	similar:	223	days	(95%	CI:	195-275)	and	231	days	(95%	CI:	193-303),	respectively	(p=0.34).		 97		The	proposed	mechanism	by	which	statins	may	improve	outcomes	for	COPD	patients	is	that	statin	exposure	may	reduce	underlying	systemic	inflammation	either	brought	on	by	COPD	or,	indeed,	may	be	manifested	as	COPD	and	its	comorbidities.	The	results	of	this	chapter	may	lend	credence	to	the	idea	that	systemic	inflammation	has	a	negative	effect	on	health	outcomes	and	that	statins	anti-inflammatory	properties	may	reduce	this	inflammation	and	improve	health	outcomes	for	COPD	patients	(145).	It	has	been	postulated	that	there	may	be	a	specific	phenotype	of	COPD	characterized	by	persistent	low	grade	systemic	inflammation,	who	may	benefit	the	most	from	statin	therapy	(18,146).	Several	recently	completed	small	exploratory	trials	have	also	shown	that	statin	use	may	be	associated	with	decreases	in	pulmonary	inflammation	(147–149)	which	supports	the	result	that	statins	may	reduce	pulmonary-related	mortality.	It	is	also	possible	that	while	statins	may	not	modify	the	rate	of	exacerbations	in	COPD,	they	may	have	beneficial	effects	on	mortality,	owing	to	their	pleiotropic	effects	and	their	known	disease-modifying	effects	on	the	cardiovascular	comorbidities,	which	contributes	directly	and	indirectly	to	the	morality	of	patients	with	COPD	(21,142,143).	We	extend	the	previous	work	by	demonstrating	the	salutary	effects	of	statins	on	pulmonary	and	total	mortality	of	COPD	patients	in	the	population.			The	results	of	the	analysis	presented	in	this	chapter	also	support	the	idea	that	randomized	controlled	trials,	while	often	considered	the	ideal	study	design,	may	not	be	adequate	in	certain	circumstances	where	the	inclusion/exclusion	are	not	generalizable	to	the	population	to	which	the	results	would	be	inferred.	COPD	patients	that	may	benefit	from		 98	statins	may	reflect	heterogeneous	patients	with	multiple	comorbidities,	systemic	inflammation,	and	with	more	severe	COPD.	In	Chapter	3	of	this	thesis,	it	was	highlighted	that	conducting	a	randomized	controlled	trial,	where	outcomes	were	rare,	can	be	logistically	difficult	due	to	the	large	number	of	patients	required.	In	the	context	of	the	current	chapter,	the	STATCOPE	trial	was	conducted	in	response	to	observational	findings,	but	did	not	lead	to	conclusive	evidence,	as	would	be	hoped,	for	whether	statin	use	was	beneficial	in	COPD	patients.	In	both	of	these	instances,	the	results	of	well-designed	observational	studies	may	prove	to	be	better	equipped	to	answer	these	important	research	questions.				4.4.1 Strengths	and	limitations	The	most	significant	strength	of	this	study	is	the	use	of	population-based	administrative	data	for	an	entire	Canadian	province	of	approximately	4.5	million	people,	which	enhances	the	generalizability	of	results.	An	additional	strength	of	this	study	is	the	definition	of	exposure	that	has	been	used.	The	exposure	ascertainment	window	equates	the	probability	of	being	exposed	within	the	cohort	due	to	time	in	the	cohort	and	then	evaluates	outcomes	over	the	same	fixed	time	period	for	all	patients.	Using	the	intention-to-treat	principle	in	this	analysis	for	the	exposure	definition	means	that	the	results	of	this	study	are	likely	to	be	conservative	(due	to	the	dilution	of	exposure	over	time)	in	that	patients	may	have	been	exposed	to	statins,	and	received	benefits	from	statin	therapy,	without	being	classified	as	such.	Reciprocally,	patients	who	were	classified	as	exposed	might	have	discontinued	the	medication	shortly	after	the	start	of	follow-up.			 99	This	study	has	several	limitations:	the	cohort	of	COPD	patients	was	defined	based	on	their	prescription	profiles	from	an	administrative	database,	not	a	clinical	diagnosis.	Similar	approaches	have	been	used	previously	(152)	and	it	is	believed	to	be	a	reasonable	approach	to	defining	a	cohort	of	COPD	patients.	Administrative	data	do	not	indicate	how	many	prescriptions	were	ordered	but	never	filled,	nor	how	many	prescriptions	were	filled	but	never	taken.	Patients	having	filled	but	not	taken	their	medication	would	result	in	an	overestimate	of	the	number	of	statin	exposed	patients	in	our	cohort,	potentially	misclassifying	unexposed	patients	as	exposed,	and	result	in	a	conservative	bias.	In	a	sensitivity	analysis,	the	MPR	was	used	as	a	measure	to	determine	adherence	to	statin	therapy.	Patients	with	a	higher	MPR	were	likely	to	have	filled	multiple	prescriptions	and	it	follows	logically	that,	for	most	cases,	having	filled	multiple	prescriptions	means	that	the	previous	prescriptions	had	been	exhausted.	In	addition,	the	possibility	of	the	‘healthy	user’	or	‘healthy	adherer’	effect	cannot	be	ignored	(153).	However,	the	design	of	this	study	and	the	statistical	analysis,	for	example,	by	adjusting	for	the	total	number	of	medications	patients	received	and	the	presence	of	other	comorbid	conditions,	should	minimize	this	effect.	Finally,	I	did	not	have	access	to	patients’	smoking	status	or	lung	function	measurements.	As	such,	I	could	not	adjust	for	these	clinical	characteristics	that	are	important	confounders	and	contributors	to	mortality.	I	did,	however,	adjust	for	other	important	patient	characteristics	such	as	income	level,	which	is	significantly	associated	with	both	smoking	status	and	severity	of	COPD	in	the	real	world.			 100	4.5 Concluding	remarks	In	conclusion,	while	recent	evidence	from	a	randomized	controlled	trial	suggested	that	statin	use	is	of	little	benefit	to	COPD	patients,	this	population-based	analysis	showed	that	statin	use	reduced	pulmonary-related	and	all-cause	mortality	among	COPD	patients.	Moreover,	this	association	persisted	across	several	sensitivity	analyses.	Our	findings,	in	conjunction	with	previously	reported	observational	evidence,	suggests	that	there	may	be	a	specific	subtype	of	COPD	patients,	characterized	by	elevated	levels	of	local	and/or	systemic	inflammation,	that	may	benefit	from	statin	use.		 101	Chapter	5: An	evaluation	of	inhaled	corticosteroid	use	and	lung	cancer	risk	in	chronic	obstructive	pulmonary	disease	patients:	a	population-based	cohort	study22	 Summary		This	chapter	builds	on	the	work	presented	in	Chapter	3	of	this	dissertation	where	observational	evidence	suggested	that	ICS	use	may	be	associated	with	a	reduced	risk	of	lung	cancer	in	COPD	patients.	However,	Chapter	3	also	highlighted	methodological	issues	with	published	studies	that	addressed	this	research	question.	In	this	chapter,	I	will	attempt	to	answer	this	research	question	and	improve	on	the	methods	used	in	previous	studies.	In	doing	so,	this	study	makes	two	valuable	contributions	to	the	literature	in	this	area:	(i)	it	uses	a	number	of	methods	of	quantifying	medication	exposure	using	patients’	prescription	profiles	obtained	from	administrative	data;	and	(ii)	it	uses	a	latency	period	in	the	analysis	which	is	based	on	the	plausible	assumption	of	the	existence	of	a	lag	time	between	the	exposure	to	ICS	and	change	in	the	risk	of	lung	cancer.		                                                22	A	version	of	this	work	has	been	presented	previously,	as	a	poster,	at	the	Canadian	Institutes	for	Health	Research,	Institute	for	Circulatory	and	Respiratory	Health,	Young	Investigators	Forum	in	May	of	2013	(Toronto,	Ontario,	Canada).		 102	5.1 Introduction	Chronic	obstructive	pulmonary	disease	(COPD)	is	a	progressive	and	predominantly	irreversible	disease	characterized	by	airflow	obstruction	(8).	This	disease	may	affect	between	4%	(self-reported)	(154)	to	11%	(1)	of	Canadians	depending	on	the	method	of	identifying	the	disease.	Moreover,	COPD	is	one	of	the	leading	causes	of	death	worldwide,	with	approximately	5%	of	all	deaths	being	attributed	to	the	disease	(155),	with	projections	indicating	that	these	numbers	will	rise	in	the	next	decade	(11).	In	addition	to	the	increased	risk	of	mortality,	patients	with	COPD	report	a	lower	quality	of	life	than	the	general	population,	which	decreases	as	their	disease	progresses	in	severity	(147,148).			Chronic	obstructive	pulmonary	disease	is	a	heterogeneous	disease	often	associated	with	comorbidities	(13,24,25).	While	there	are	several	common	comorbidities,	including	cardiovascular	disease,	diabetes,	depression,	and	osteoporosis,	lung	cancer	is	likely	the	most	severe	due	to	the	poor	prognosis	associated	with	the	diagnosis.	COPD	patients	are	at	an	increased	risk	of	lung	cancer,	which	may	largely	be	due	to	their	smoking	status	or	history	(149,150),	but	while	smoking	is	a	significant	risk	factor	for	lung	cancer,	evidence	also	suggests	that	patients	with	COPD	are,	independent	of	the	role	of	smoking,	at	an	increased	risk	of	lung	cancer.	For	example,	Brennar	et	al.	(160)	showed	that	patients	with	COPD	were	40%	more	likely	to	develop	lung	cancer	after	adjustment	for	patients	smoking	status	(OR:	1.4	(95%	CI:	1.1-1.8)).	Similarly,	a	study	using	a	patient	population	of	UK-based	COPD	patients	in	a	primary	care	setting	estimated	a	relative	risk	of	3.33	(95%	CI:	2.33-4.75)	of	lung	cancer	compared	to	patients	without	a	COPD	diagnosis,	again	controlling	for		 103	patients’	smoking	status23	(161).	Young	et	al.	(32)	presented	similar	results	showing	that	in	patients	with	a	history	of	smoking,	those	with	COPD	experienced	an	increased	risk	of	lung	cancer,	compared	to	individuals	with	a	history	a	smoking	without	COPD.		There	is	further	evidence	to	suggest	that	patients	with	COPD	that	have	no	history	of	smoking	are	also	at	increased	risk	of	developing	lung	cancer	compared	to	individuals	without	COPD,	lending	support	to	the	idea	that	there	are	factors	inherent	in	COPD	that	increase	the	risk	of	lung	cancer	beyond	smoking	status/history.	For	example,	Turner	et	al.	(35)	conducted	a	study	of	448,600	never	smokers	and	found	that	over	a	period	of	twenty	years,	patients	with	emphysema	(HR:	1.66	(95%	CI:	1.06-2.59))	and	combined	emphysema/chronic	bronchitis	(HR:	2.44	(95%	CI:	1.22-4.90))	were	at	an	increased	risk	of	lung	cancer.	Finally,	Kosihol	et	al.	(162)	reported	that	while	the	risk	of	lung	cancer	is	increased	in	patients	with	COPD,	this	increased	risk	is	not	solely	attributable	to	smoking.	After	adjustment	for	smoking	status	lung	cancer	risk	was	elevated	in	patients	with	a	history	of	COPD	(adjusted	OR:	2.5	(95%	CI:	2.0-3.1)	(162).	Therefore,	the	current	evidence	resoundingly	suggests	that	while	smoking	is	an	important	contributor	to	an	increased	risk	of	lung	cancer,	there	also	appears	to	be	an	element	of	COPD	that	increases	lung	cancer	risk,	above	what	can	be	attributable	to	an	individuals’	smoking	status	or	history.			While	the	mechanism	by	which	COPD	is	associated	with		an	increased	risk	of	lung	cancer	is	not	well-established,	there	is	evidence	to	support	the	hypothesis	that	systemic	                                                23	This	study,	interestingly,	showed	that	lung	cancer	incidence	was	positively	associated	with	less	airflow	obstruction	(GOLD	Stages	I	and	II),	whereas	most	evidence	would	suggest	that	lung	cancer	risk	increases	with	greater	disease	severity.		 104	inflammation	may	be	the	primary	etiologic	factor	in	the	causal	pathway (see Table 1.2).	Barnes	et	al.	(13)	report	that	patients	with	COPD,	and	in	particular	patients	with	more	advanced	disease,	have	higher	levels	of	systemic	inflammation,	as	measured	by	circulating	cytokines	(interleukin-6	or	IL6).	Another	study	evaluated	the	association	between	levels	of	systemic	inflammation	using	multiple	biomarkers	(C-reactive	protein,	(CRP),	fibrinogen,	and	leukocyte	count)	and	the	risk	of	lung	cancer	in	COPD	patients	(163).	The	authors	found	a	statistically	significant	increased	risk	of	lung	cancer	when	two	of	these	three	levels	were	elevated	(HR:	2.14	(95%	CI:	1.21-3.77)),	while	controlling	for	smoking	status	(among	other	covariates),	which	increased	to	approximately	four	times	greater	risk	(HR:	4.00	(95%	CI:	2.12-7.54)),	if	all	three	levels	were	elevated,	compared	to	those	with	no	elevated	levels	(163).	Overall,	the	result	of	this	evidence	suggests	that	systemic	inflammation	is	an	important	consideration	in	COPD,	and	needs	to	be	measured	and	considered	when	treating	patients	(126).	Indeed,	several	authors	have	suggested	that	there	may	be	different	phenotypes	of	COPD,	and	that	one	of	these	phenotypes	may	be	specifically	characterized	by	increased	levels	of	systemic	inflammation	(16,139,155,156).	If	this	phenotype	exists,	it	follows	that	these	COPD	patients	would	likely	be	at	an	increased	risk	of	lung	cancer.		Treatment	for	COPD	is	largely	based	on	guidelines	established	by	the	Global	Initiative	for	Lung	Diseases	(GOLD).	Inhaled	corticosteroid	use	is	common	for	patients	with	advanced	stages	of	disease (see	Table	1.1),	for	example,	GOLD	Stages	‘3’	or	‘4’	and/or	for	patients	at	high	risk	of	acute	exacerbations	associated	with	COPD	(AECOPD);	that	is,	≥2	per	AECOPD	per	year,	or	>1	hospitalization	(5).	While	evidence	suggests	that	ICS	use	might	not	necessarily	be	beneficial	in	terms	of	lung	function	and	mortality	in	patients	with	COPD		 105	(44),	ICS	use	may	confer	benefits	in	terms	of	improving	quality	of	life	(46,47)	and	reducing	the	rate	of	AECOPD	(45).	Moreover,	ICS	use	appears	to	reduce	localized	inflammation	in	the	airway	and	lungs	which	is	typical	in	COPD	patients	(43,49,157,158).	Inhaled	corticosteroid	use	has	also	been	associated	with	reducing	systemic	inflammation	which	provides	an	important	mechanism	by	which	ICS	use	might	reduce	lung	cancer	risk	(see	Table	1.3).	For	example,	two	studies	conducted	by	Sin	et	al.	(47,56)	attempted	to	evaluate	the	association	between	ICS	use	and	levels	of	inflammation.	In	the	2004	study,	ICS	use	was	associated	with	a	50%	reduction	in	CRP	levels	and	a	26%	reduction	in	IL-6	levels	compared	to	placebo	treated	controls.	In	the	later	study	(47),	there	were	lower	levels	of	CRP	in	those	patients	treated	with	ICS,	but	the	observed	differences	associated	with	ICS	treatment	were	not	statistically	significant.			Lung	cancer	is	often	diagnosed	at	an	advanced	stage	and	is	typically	accompanied	with	a	poor	prognosis	(168).	As	such,	it	is	important	to	identify	interventions	that	might	reduce	the	risk	of	this	disease,	particularly	in	a	population	of	patients	that	face	an	elevated	risk.	Therefore,	the	objective	of	this	study	was	to	examine	the	association	between	ICS	use	and	lung	cancer	risk,	using	a	variety	of	definitions	of	medication	exposure,	in	a	population-based	cohort	of	COPD	patients.	A	secondary	objective	was	to	explore	whether	ICS	exposure	has	a	significant	association	with	specific	subtypes	of	lung	cancer.						 106			Figure	5.1.	The	basic	conceptual	framework	for	the	study	presented	in	Chapter	5.	COPD	is	associated	with	systemic	and	local	(pulmonary)	inflammation	which	are	also	associated	with	lung	cancer	risk.	Therefore,	the	study	hypothesis	is	that	ICS	use	in	COPD	patients	may	reduce	levels	of	inflammation,	thereby	reducing	the	risk	of	lung	cancer.		5.2 Methods	This	study	uses	population-based	linked	administrative	data	for	the	province	of	British	Columbia,	Canada,	to	identify	a	cohort	of	COPD	patients	based	on	patients’	prescriptions	profile,	linked	to	a	registry	of	cancer	patients	obtained	from	the	British	Columbia	Cancer	Research	Agency.			5.2.1 Database	description	The	population-based	administrative	data	for	this	study	comprised	the	following	data:	LUNG	CANCERCOPDLOCAL	AND	SYSTEMIC	INFLAMMATIONINHALED	CORTICOSTEROIDS	 107	1. Medical	Services	Plan	(MSP)	data	file:	includes	fee-for-service	physician	billings,	for	every	encounter	with	a	physician	in	the	province,	under	the	universal	provincial	health	insurance	scheme,	including	the	date	of	the	encounter	and	the	reason	for	the	encounter	(ICD9	code)	(136).		2. Discharge	Abstracts	Database	(DAD):	comprising	all	hospital	separations	for	the	province	during	the	study	period.	Data	includes	the	date	of	admission,	the	date	of	separation,	primary	and	secondary	reasons	for	admission	(ICD9	code),	and	method	of	admission	(133).	3. PharmaNet	datafile:	is	comprised	of	all	prescriptions	dispensed	in	British	Columbia	over	the	study	period.	This	data	file	provides	the	medication	dispensed	(generic	name),	brand	name,	the	number	of	days	of	the	medication	supplied,	American	Hospital	Formulary	Service	(AHFS)	code,	the	dosage,	the	quantity,	a	unique	medication	identifier	(‘DinPin’),	and	the	date	that	the	medication	was	dispensed	(132).	4. 	British	Columbia	Cancer	Registry	file:	provides	data	on	lung	cancer	diagnosis	date,	death	due	to	cancer,	and	cancer	characteristics	(histology	and	tumour	type)	(169).		5.2.2 Cohort	identification		Patients	were	included	in	the	cohort	if	they	were	50	years	of	age	or	greater	and	had	filled	at	least	three	prescriptions	for	an	anticholinergic	medication	or	short-acting	beta-agonist	(SABA)	in	a	one-year	rolling	time	window	(144,160).	If	the	individual	met	these	criteria,	the	date	of	the	first	of	these	dispensed	prescriptions	was	used	as	their	study	index	date,		 108	which	is	considered	the	date	of	COPD	diagnosis.	To	reduce	some	of	the	heterogeneity	among	COPD	patients,	a	one-year	'wash-in'	period	was	imposed	such	that	patients	were	required	to	have	a	twelve	month	period	prior	to	their	index	where	they	were	not	in	receipt	of	a	SABA	or	anticholinergic	drug.		The	rationale	for	this	decision	is	that	for	those	patients	that	had	already	been	using	one	of	these	medications	used	to	identify	the	cohort,	we	would	not	know	for	how	long	they	may	have	had	COPD.	By	applying	the	'wash-in'	period,	we	increased	the	probability	of	the	patients	we	identified	as	being	incident	COPD	cases.	We	removed	patients	who	could	not	contribute	at	least	two	years	of	follow-up	time	to	ensure	that	patients	would	have	sufficient	time	to	potentially	develop	lung	cancer.			The	PharmaNet	data	file	was	then	linked	with	physician	billings	(Medical	Services	Plan	(MSP)),	hospital	discharges	(Discharge	Abstract	Database),	demographic	information	and	mortality	(Vital	Statistics),	and	the	British	Columbia	Cancer	Agency	Registry	file	using	a	unique	de-identified	personal	health	number	(PHN)	to	conduct	the	analysis.					5.2.3 Inhaled	corticosteroid	use		Inhaled	corticosteroid	users	were	initially	identified	according	to	American	Hospital	Formulary	Service	(AHFS)	codes	520808,	680400,	and	840600	(137).	Identified	prescriptions	meeting	this	criterion	were	further	scrutinized	to	ensure	that	only	inhaled	medications	were	considered	as	potential	ICS	exposure.	For	example,	within	these	three	AHFS	codes,	there	remained	topical	and	oral	formulations;	thus,	these	were	not	considered	as	ICS	exposure.	ICS	monotherapy	and	ICS	combination	therapy	were	both	considered	in	this	analysis.			 109	5.2.4 Latency	period	In	an	attempt	to	provide	sufficient	time	for	ICS	exposure	to	affect	the	pathogenesis	of	lung	cancer,	a	one-year	(365	day)	latency	period	(or	‘lag’	period)	was	applied	in	the	primary	analysis.	That	is,	any	medication	exposure	in	the	one-year	period	prior	to	lung	cancer,	death,	or	censoring,	was	not	counted	as	the	pathogenesis	of	lung	cancer	was	assumed	to	have	already	begun.	The	length	of	the	latency	period	was	chosen	based	on	evidence	by	Henschke	et	al.	(171)	and	Chaturvedi	et	al.	(40)	and	was	subject	to	several	sensitivity	analyses.	A	graphical	representation	of	the	latency	period	associated	with	lung	cancer	diagnosis	and	its	relationship	to	medication	exposure	is	presented	in	Figure	5.2.					 Figure	5.2.	The	latency	period	associated	with	medication	exposure	and	lung	cancer	diagnosis.		End	of	Study	Follow-upIndex	DateLung	Cancer	DiagnosedLatency	Period	(365	days)ICS	Received	(A) ICS	Received	(B)The	application	of	 the	latency	period	 (365	days)	means	that	prescriptions	 received	during	 the	latency	period	 (B),	prior	 to	the	diagnosis	of	lung	cancer	in	this	illustration,	are	not	counted	as	exposures.	Prescriptions	 received	after	the	index	date	and	prior	 to	the	latency	period	are	counted	as	ICS	exposures.	 110	5.2.5 Exposure	measurement	The	method	used	to	measure	medication	exposure	in	observational	studies	using	administrative	data	requires	careful	consideration.	The	choice	of	measure	should	depend	on	the	medication	being	considered	and	the	properties	of	the	study	outcome	(in	this	case,	disease	diagnosis).	To	ensure	the	robustness	of	the	results	of	this	study,	several	different	methods	of	quantifying	drug	exposure	were	used	in	the	analyses.	These	methods	attempted	to	appreciate	the	nuances	of	medication	exposure	over	longer	follow-up	periods	and	a	disease,	lung	cancer,	with	a	substantial	induction	and	latency	periods.	There	were	four	conventional	methods	of	defining	medication	exposure	in	this	analysis:	(i)	time-dependent	exposure;	(ii)	‘current’	use;	(iii)	cumulative	years	of	use;	and,	(iv)	total	cumulative	dose	received.	In	addition,	there	were	two	recency-weighted	exposure	measures:	(v)	recency-weighted	duration	of	medication	use;	and,	(vi)	recency-weighted	cumulative	dose.	Each	of	these	are	described	in	more	detail	below:			(i) Time-dependent	exposure:	The	reference	case	for	this	analysis	was	time-dependent	exposure,	defined	as	a	time-dependent	binary	variable.	For	this	definition	of	medication	exposure,	at	any	specific	time	during	the	follow-up,	a	patient	could	be	considered	as	exposed	or	unexposed	to	ICS,	contingent	on	whether	that	patient	had	filled	an	ICS	prescription	prior	to	that	specific	time	and	after	the	start	of	follow-up.	This	calculation	was	performed	using	the	date	that	the	prescription	was	filled	by	the	patient	and	the	‘days	supply’	of	that	specific	dispensed	prescription.	The	use	of	a	time-dependent	method	of	quantifying	medication	exposure	is	superior	to	a	fixed	‘ever/never’	exposure	definition	whereby	a	patient	receiving	an	ICS	prescription	at		 111	any	time	during	follow-up	would	be	considered	exposed	for	the	entirety	of	the	follow-up	period,	and	has	implications	for	reducing	potential	biases	that	result	from	using	fixed	exposure	definitions.	Immortal-time	bias	may	exist	where	the	time	between	the	index	date	and	the	date	of	first	exposure	is	calculated	as	exposed	time,	which	is	incorrect	because	during	that	time,	the	patient	was	not	actually	unexposed	to	the	medication	under	investigation	(139,172).	Therefore,	using	a	time-dependent	measure	of	medication	exposure	correctly	classifies	the	time	for	which	the	patient	had	not	filled	a	prescription	as	unexposed	time,	and	time	for	which	the	patient	had	filled	a	prescription	as	exposed	time.			(ii) ‘Current	Use’:	In	this	analysis,	current	use	has	been	defined	as	a	patient	that	filled	an	ICS	prescription	in	a	six-month	window	previous	to	the	latency	period	and	those	that	did	not	receive	a	prescription	for	ICS	during	this	period	as	unexposed.		This	definition	of	medication	exposure	implies	there	is	a	period	during	which	exposure	to	the	medication	previous	to	the	outcome	is	especially	important.	This	method	may	be	used	more	frequently	in	the	case	of	acute	events	(i.e.	myocardial	infarction)	whereby	an	exposure	immediately	preceding	an	event	may	be	associated	with	that	event	(85).			(iii) Cumulative	Duration	of	Use:	This	definition,	the	quantity	representing	medication	exposure	is	calculated	by	aggregating	the	length	of	time	for	which	a	patient	has	taken	the	medication.	In	this	analysis,	the	‘days	supply’	of	each	individual	dispensed	prescription	were	aggregated,	for	any	given	point	during	the	study	follow-up	period,	and	divided	by	365,	to	get	the	total	years	of	ICS	use.			 112		(iv) Cumulative	Dose:	Medication	exposure	is	calculated	by	aggregating	the	total	dose	of	medication	prescribed	to	a	specific	patient	over	the	study	period.	Inhaled	corticosteroids	have	different	potencies;	therefore,	each	ICS	prescription	was	converted	to	fluticasone	equivalent	dosages	to	allow	for	comparison	between	different	type	of	ICS	use	(173).	This	method	of	capturing	medication	exposure	is	based	on	the	assumption	that	the	total	amount	of	medication	received	(in	grams,	for	this	analysis)	is	associated	with	the	outcome.	This	is	calculated	using	the	dose	of	the	medication	prescribed	multiplied	by	the	quantity	prescribed.		(v) Recency-Weighted	Cumulative	Duration	of	Use:	The	weighted	cumulative	duration	of	use	approach	was	used	to	account	for	the	duration	of	medication	use	while	also	accounting	for	when,	during	follow-up,	the	medication	use	occurred	(96).	To	accomplish	this,	a	function	was	constructed	that	weighted	those	prescriptions	that	occurred	closer	to	the	date	of	the	event,	or	end	of	follow-up,	higher	than	those	from	earlier	in	the	study	period.	The	assumption	behind	this	method	is	straightforward:	those	prescriptions	that	were	dispensed	more	proximal	to	the	outcome	or	the	end	of	follow-up	are	likely	to	result	in	a	greater	risk	reduction	than	those	prescriptions	that	were	dispensed	earlier	in	the	follow-up	period.	The	assumption	of	this	assigning	weights	in	this	way	is	consistent	with	how	this	method	has	been	used	previously	in	the	literature	(86,96).					 113	(vi) Recency-Weighted	Cumulative	Dose:	The	recency-weighted	cumulative	dose	approach	was	used	to	simultaneously	account	for	cumulative	dose	and	when	during	follow-up	the	dose	was	received	(96).	This	method	implies	that	doses	of	medication	received	across	the	study	period	are	not	equal	and	that	those	that	occur	more	recently	may	have	a	greater	association,	and	in	this	case,	a	greater	risk	reduction,	with	the	study	outcome.	Both	of	the	recency-weighted	approaches	described	above	also	respect	the	latency	period	(discussed	above).	That	is,	both	of	these	methods	weighted	ICS	exposure	until	the	beginning	of	the	latency	period,	and	after	this	period	began,	filled	prescriptions	for	ICS	were	not	counted	toward	exposure.				5.2.6 Adjustment	for	potential	confounders	Covariates	that	were	thought	to	be	potential	confounders	of	the	association	between	ICS	exposure	and	lung	cancer	diagnosis	were	incorporated	into	a	multivariable	model	described	in	the	next	section.	The	demographic	covariates	included:	age,	sex,	neighborhood	income	quintiles	based	residence,	and	health	authority	(regional	health	service)	in	which	the	patient	resided.	In	addition,	for	each	patient	the	number	of	prescriptions	dispensed,	the	number	of	hospital	encounters,	the	number	of	inpatient	hospital	stays,	and	the	number	of	physician	encounters	were	calculated.	Finally,	to	account	for	comorbidities	at	the	beginning	of	follow-up,	the	Charlson	Comorbidity	Index	was	calculated	based	on	health	services	use	(140,141).	Covariates	were	assessed	in	the	one	year	period	immediately	prior	to	the	latency	period.				 114	5.2.7 Statistical	analysis		A	Cox	regression	model	(174,175)	was	used	to	estimate	the	hazard	ratio	associated	with	lung	cancer	diagnosis	based	on	exposure	to	ICS,	using	the	aforementioned	exposure	definitions.	In	the	context	of	this	study,	the	analytical	approach	calculates	the	time	from	COPD	diagnosis	(the	index	date)	to	lung	cancer	diagnosis,	death,	or	end	of	study	follow-up.	A	series	of	bivariate	regression	analyses	were	conducted	to	determine	candidate	covariates	for	the	final	multivariable	model	using	a	threshold	of	p<0.20	to	be	considered	for	inclusion.	Results	from	these	bivariate	regression	models	are	reported	in	Table	5.3.	Each	covariate	that	met	this	threshold	was	added	to	the	multivariable	model	via	stepwise	selection	comparing	Akaike	Information	Criterion	(AIC)	values	(142).	The	AIC	is	a	method	of	comparing	models	which	adds	a	penalty	for	additional	covariates	in	the	model	with	lower	scores	being	considered	superior	to	higher	scores.	Comparison	of	AIC	score	is	a	well-established	method	for	determining	the	most	parsimonious	and	best-fitting	model	(86,96).				Hazard	ratios	(HR)	and	their	associated	95%	confidence	intervals	(CI)	were	reported	for	each	of	the	exposure	metrics	quantifying	ICS	use	as:	(a)	bivariate	analyses,	(b)	multivariable	(age	and	sex	adjusted)	analyses,	and	as	(c)	‘fully’	adjusted	multivariable	analyses,	with	time	to	lung	cancer	diagnosis	as	the	outcome.	Akaike	Information	Criterion	values	are	also	presented	to	show	the	performance	of	each	exposure	metric	with	a	given	set	of	covariates.	Statistical	significance	was	achieved	for	p-values	less	than	alpha	of	0.05.				 115	5.2.8 Secondary	analysis:	medication	possession	ratio	The	medication	possession	ratio	(MPR)	is	typically	used	as	a	measure	of	adherence	as	outlined	in	detail	in	Chapter	2.	It	is	calculated	as	the	ratio	between	the	time	(days,	for	example)	from	the	initial	prescription	of	the	medication	(ICS)	to	the	end	of	follow-up	or	the	outcome	of	interest,	and	the	estimated	total	number	of	days	of	treatment	for	which	the	medication	was	dispensed	(93).	Thus,	it	gives	a	ratio	for	the	amount	of	time	the	patient	had	a	prescription	over	the	course	of	the	study	period,	and	provides	an	estimate	of	sustained	medication	use.	Patients	using	ICS	were	classified	as	adherent	or	non-adherent	based	on	an	MPR	threshold	of	0.8	(78,90).	The	results	of	the	multivariable	regression	using	the	MPR	are	presented	in	Table	5.5.		5.2.9 Sub-group	analysis:	lung	cancer	histology	 	The	BC	Cancer	Agency	Registry	file	provided	information	on	the	histology	of	each	lung	cancer	case	reported.	Therefore,	we	were	able	to	classify	lung	cancers	according	into:	(i)	non-small	cell	lung	cancer	(NSCLC);	(ii)	small	cell	lung	cancer	(SCLC);	and	(iii)	‘other’.	The	distribution	of	these	cancers	within	the	cohort	of	COPD	patients	is	reported	in	Figure	5.5.	Using	the	same	analytical	approach	as	described	in	Section	5.2.6,	individual	multivariable	models	were	estimated	with	the	outcome	variables	SCLC	and	NSCLC.		5.2.10 Sensitivity	analyses	Several	sensitivity	analyses	were	performed	to	test	the	effect	of	assumptions	made	on	the	results	of	this	study.	First,	as	stated	above,	in	the	primary	analysis	a	one-year	period	for	lung	cancer	latency	was	assumed.	In	a	sensitivity	analysis,	this	period	was:	(i)	reduced	to		 116	not	being	present	at	all	(i.e.	zero	days),	(ii)	reduced	to	180	days,	and	(iii)	extended	to	two	years,	to	explore	whether	study	results	were	robust	to	this	assumption.	Second,	to	be	included	in	the	cohort	of	COPD	patients,	individuals	were	required	to	be	50	years	of	age	or	greater.	However,	lung	cancer	incidence	is	quite	low	under	65	years	of	age	(176,177),	therefore,	the	cohort	age	restriction	was	increased	to	greater	than	or	equal	to	65	years	to	be	included.			5.3 Results	A	cohort	of	39,879	patients	was	identified	that	met	the	inclusion	criteria.	The	mean	age	of	the	patients	was	70.6	(SD:	11.2)	years	and	53.5%	of	COPD	patients	were	female.	Mean	follow-up	time	among	the	cohort	was	5.1	years.	More	information	on	the	cohort	of	COPD	patients	is	presented	in	Table	5.1.					 117		Figure	5.3.	Distribution	of	all	ICS	prescriptions	dispensed	to	the	COPD	cohort	within	the	follow-up	period.			      01020304050607080Percentage	of	all	Prescriptions	DispensedICS	Generic	Name	 118	Table	5.1.	Demographics	of	the	COPD	cohort	(n=39,879).	Patient	Characteristic	 Value		  Age	(Mean	(SD)	 70.6	(SD:	11.2)	Age	Distribution		50<60	 8356	(21.0%)	60<70	 10,212	(25.7%)	70<80	 12,436	(31.2%)	>80	 8785	(22.1%)	Female	 21,273	(53.5%)	Income	Quintile		 	1	 9701	(25.6%)	2	 7864	(20.8%)	3	 6854	(18.1%)	4	 6381	(16.8%)	5	 5696	(15.0%)	Health	Authority	Interior		 8569	(21.6%)	Fraser		 11,354	(28.6%)	Vancouver	Coastal		 7740	(19.5%)	Vancouver	Island		 7522	(18.6%)	Northern		 2465	(6.2%)	Hospitalizations		 	Any	Reason	 6651	(16.7%	COPD-related	 1084	(2.7%)	CVD-related		 512	(1.3%)	Charlson	Comorbidity	Score1	 2	(IQR:	1-4)	Charlson	Comorbidity	Category2	 	0	 31,354	(79.0%)	1	 6303	(15.9%)	2	 1176	(3.0%)	3	 843	(2.1%)	Combination	Therapy	(ICS/LABA)	 6585	(16.5%)	Physician	Encounters	(any	reason)1	 11	(3-22)	Number	of	Prescriptions	Filled	(any	reason)1	 26	(6-56)		  Note:	Values	represent	mean	(standard	deviation)	or	number	(percentage)	unless	otherwise	indicated.	Where	percentages	do	not	add	to	100%	the	reason	is	due	to	rounding.				1	-	Median	and	interquartile	range	(IQR).	2	-	Category	0	is	a	Charlson	Score	of	0,	Category	1	is	(0,2],	Category	2	is	(2,	3],	Category	3	is	>3.					 119		5.3.1 ICS	use	in	the	COPD	cohort	There	were	435,991	dispensed	prescriptions	for	an	ICS	within	the	cohort	of	COPD	patients.	When	the	one-year	latency	period	was	applied	and	ICS	prescriptions	filled	during	this	period	were	excluded,	there	were	372,075	filled	ICS	prescriptions	that	remained	for	analysis.	The	resulting	number	of	ICS	users	was	28,314	(71.2%)	within	this	cohort	of	COPD	patients.	Most	patients	filled	more	than	one	prescription	for	ICS,	with	a	median	of	eight	(IQR:	3-19)	ICS	prescriptions	filled	during	the	follow-up	period.	The	most	prescribed	ICS	among	the	patients	was	fluticasone	propionate	(see	Figure	5.3)	and	the	mean	dose	of	ICS	filled	was	1.4	(SD:	0.7)	grams.	The	median	days	of	ICS	supplied	was	60	days	(IQR:	30-90).	Using	a	threshold	to	be	considered	an	adherent	user	of	0.8	(95),	only	2.4%	of	the	cohort	was	considered	adherent	to	ICS.	Figure	5.4	shows	the	distribution	of	the	MPR	for	ICS	users	in	the	COPD	cohort.			 Figure	5.4.	Medication	Possession	Ratio	(MPR)	for	ICS	use	in	the	COPD	cohort	over	the	study	follow-up	period,	by	category.	050001000015000200002500030000MPR=0 0<MPR<0.8 MPR≥0.8Number	of	PatientsMedication	Possession	Ratio	 120	5.3.2 	Lung	cancer	among	COPD	patients	Among	the	cohort	of	COPD	patients,	there	were	initially	1966	cases	of	lung	cancer	that	occurred	within	the	study	follow-up	period.	Lung	cancers	that	were	diagnosed	within	one	year	of	the	index	date	were	removed	which	resulted	in	994	cases	of	lung	cancer	that	remained.	The	median	age	at	lung	cancer	diagnosis	was	71.3	(IQR:	65.6-76.4)	years	and	46.2%	of	lung	cancer	diagnoses	occurred	in	females.	Consistent	with	the	evidence	for	lung	cancer,	diagnosis	was	associated	with	a	poor	prognosis.	There	were	401	deaths	in	patients	diagnosed	with	lung	cancer	and	the	median	time	from	diagnosis	to	death	from	any	cause	was	112	days	(IQR:	24-267).	Of	the	identified	lung	cancer	cases,	854	(85.9%)	were	classified	as	non-small	cell	lung	cancer	(NSCLC)	which	is	consistent	with	population	estimates	(177,178).	The	distribution	of	lung	cancer	histology	is	presented	in	Figure	5.5.			5.3.3 Bivariate	results:	potential	confounders		Several	covariates	showed	statistically	significant	associations	with	the	outcome	of	interest,	time	to	lung	cancer	diagnosis	(see	Table	5.3).	As	expected,	age	was	significantly	associated	with	lung	cancer	risk,	showing	an	approximately	1%	increased	risk	per	additional	year	(HR:	1.01	(95%	CI:	1.00-1.01,	p=0.0011)).	Men	had	a	higher	risk	of	lung	cancer	compared	to	women	(HR:	1.39	(95%	CI:	1.24-1.56,	p<0.0001)).	The	number	of	comorbidities	and	the	COPD-related	hospitalizations	was	also	associated	with	an	increased	risk	of	lung	cancer	(HR:	1.29	(95%	CI:	1.27-1.30,	p<0.0001)	and	HR:	2.56	(2.00-3.27,	p<0.001),	respectively)	which	also	may	be	indicative	of	the	elevated	levels	of	systemic	inflammation	or	pre-existing	lung	cancer,	or	both.			 121	Table	5.2.	Bivariate	regression	results	for	covariates	considered	for	inclusion	in	the	multivariable	model,	with	time	to	lung	cancer	diagnosis	as	the	outcome	variable.			 		 		 		 		Covariate	 Hazard	Ratio		 95%	Confidence	Interval	 p-value			 		 		 		 			     Age	 1.01	 1.00	 1.01	 0.0011	Age	Categories	 		 		 		 		<60	 Ref	 Ref	 Ref	 Ref	[60,	70)	 2.02	 1.67	 2.43	 <.0001	[70,	80)	 2.33	 1.95	 2.80	 <.0001	≥80	 1.29	 1.03	 1.61	 0.241	Sex	(Male)	 1.39	 1.24	 1.56	 <.0001	Health	Authority	 	    Interior	 1.29	 0.98	 1.71	 0.0744	Fraser	 1.23	 0.93	 1.62	 0.1485	Vancouver	Coastal	 1.04	 0.78	 1.40	 0.7687	Vancouver	Island	 1.46	 1.10	 1.94	 0.0084	Northern	 Ref	 Ref	 Ref	 Ref	Income	Quintile	 	    5	 Ref	 Ref	 Ref	 Ref	4	 1.27	 1.03	 1.57	 0.0246	3	 1.14	 0.92	 1.41	 0.2152	2	 1.23	 1.00	 1.50	 0.0491	1	 1.24	 1.02	 1.51	 0.0305	Total	Number	of	Prescriptions		 1.00	 0.99	 1.00	 <.0001	Charlson	Comorbidity	Score	(Continuous)	 1.06	 0.97	 1.16	 0.1853	Charlson	Comorbidity	Score	(Categorical)	 	    0	 Ref	 Ref	 Ref	 Ref	1	 1.15	 0.98	 1.36	 0.0931	2	 0.94	 0.62	 1.40	 0.7459	≥3	 0.90	 0.55	 1.48	 0.6814	Inpatient	Stay	 3.57	 3.16	 4.03	 <.0001	Number	of	hospitalizations	 1.66	 1.64	 1.68	 <.0001	COPD-related	hospitalization	 2.56	 2.00	 3.27	 <.0001	CVD-related	hospitalization	 1.04	 0.58	 1.88	 0.8958	Combination	Therapy	(ICS/LABA)	 1.27	 1.11	 1.47	 0.007	Number	of	physician	encounters	 1.02	 1.02	 1.02	 <.0001	Oral	glucocorticoid	use	 1.09	 0.91		 1.30	 	0.3399	HR:	Hazard	Ratio;	AIC:	Akaike	Information	Criterion;	CI:	Confidence	Interval. 	 122	5.3.4 Bivariate	results:	exposure	definitions	Table	5.3	presents	the	results	of	bivariate	analyses,	as	well	as	multivariable	(age	and	sex-adjusted)	analyses,	using	each	of	the	aforementioned	exposure	metrics	with	the	time	to	lung	cancer	diagnosis	as	the	outcome.	All	of	the	exposure	metrics	capturing	ICS	use	were	significantly	associated	with	lung	cancer	diagnosis	in	bivariate	and	multivariable	analyses,	in	the	a	priori	hypothesized	direction,	showing	a	protective	effect	for	ICS	use	on	lung	cancer	risk.	No	bivariate	or	multivariable	analyses	indicated	that	ICS	exposure	might	increase	lung	cancer	risk.	In	the	reference	case	(time-dependent	exposure),	the	estimated	bivariate	HR	was	0.70	(95%	CI:	0.61-0.80,	p<0.0001).	The	magnitude	of	the	protective	effect	varied,	with	largest	risk	reduction	for	time-dependent	exposure	(see	above)	and	the	smallest	effect	size	for	cumulative	duration	of	ICS	use	(HR:	0.89	(95	%	CI:	0.84-0.95,	p=0.0004))	per	gram	of	ICS	dispensed.	The	use	of	both	recency-weighted	metrics	(cumulative	years	of	use	and	cumulative	dose)	in	bivariate	regression	analyses	resulted	in	a	lower	risk	of	lung	cancer	(HR:	0.76	per	year	of	exposure	(95%	CI:	0.68-0.82,	p<0.0001)	and	HR:	0.62	per	gram	of	fluticasone	equivalent	dose	(95%	CI:	0.48-0.80,	p=0.0002),	respectively).					   	 123	Table	5.3.	Bivariate,	and	age	and	sex	adjusted,	regression	results	(hazard	ratio	and	95%	CI)	for	each	ICS	exposure	definition	with	time	to	lung	cancer	diagnosis	as	the	outcome	variable.	       Exposure	Metrics	 Bivariate	 Age	and	Sex	Adjusted			 HR	 95%	CI	 HR	 95%	CI		       Time-Dependent	ICS	Exposure		 0.70	 0.61	 0.80	 0.725	 0.63	 0.83		       Current	Usea	 0.78	 0.68	 0.88	 0.80	 0.71	 0.91		       Cumulative	Durationb	 0.89	 0.84	 0.95	 0.	90	 0.84	 0.96		       Cumulative	Doseb	 0.79	 0.66	 0.95	 0.80	 0.67	 0.95		       Recency-Weighted	Duration	of	Use	 0.75	 0.68	 0.82	 0.76	 0.68	 0.83		       Recency-Weighted	Cumulative	Dose	 0.62	 0.48	 0.80	 0.62	 0.49	 0.80		 	 	 	 	 		 		HR:	Hazard	Ratio;	Ref:	Reference	Category;	CI:	Confidence	Interval. a	Current	use	is	defined	as	having	filled	a	prescription	in	the	6-	month	period	immediately	prior	to	the	latency	period.	b	Measured	as	a	continuous	variable.	 		5.3.5 Multivariable	analysis:	main	results	In	multivariable	analyses,	after	adjustment	for	potential	confounders,	there	remained	a	reduction	in	lung	cancer	risk	from	ICS	use	compared	to	no	ICS	use	in	patients	with	COPD.	Although	the	magnitude	of	the	effect	varied	according	to	the	specific	exposure	metric		 124	employed	in	the	multivariable	regression	analyses,	the	protective	effect	was	consistent	across	all	metrics	of	exposure.	In	the	reference	case,	classifying	ICS	exposure	using	time-dependent	ICS	exposure	metric,	the	resulting	adjusted	HR	was	0.70	(95%	CI:	0.61-0.80,	p<0.0001),	suggesting	a	30%	reduction	in	lung	cancer	risk	associated	with	ICS	use.	Based	on	AIC	values,	this	exposure	metric	was	superior	to	the	other	conventional	metrics	of	measuring	exposure	(current	use,	cumulative	duration	of	use,	and	cumulative	dose).		Exposure	to	ICS,	as	measured	by	the	two	recency-weighted	metrics,	were	statistically	significantly	associated	with	reduced	lung	cancer	risk.	The	recency-weighted	duration	of	use	exposure	metric	showed	an	approximately	25%	reduction	in	lung	cancer	risk	from	ICS	use	(HR:	0.74	(95%	CI:	0.66-0.87,	p<0.0001)	per	year	of	exposure,	and	use	of	the	recency-weighted	cumulative	dose	metric	resulted	in	an	HR	of	0.57	(95%CI:	0.43-0.74,	p<0.0001)	indicated	a	43%	reduction	in	the	risk	of	lung	cancer	per	gram	(fluticasone	equivalent)	of	ICS	use.			For	each	of	the	multivariable	analyses	using	distinct	exposure	metrics,	AIC	values	were	compared	to	determine	the	metric	that	produced	the	best	model	fit.	Of	the	time-dependent	exposure	metrics	presented	in	Table	5.4	the	best	(lowest)	AIC	value	was	19116	for	recency-weighted	duration	of	use,	approximately	15	points	lower	than	the	next	lowest	value	(time-dependent	exposure).	This	indicated	that	the	use	of	the	recency-weighted	duration	of	use	exposure	metric	provided	the	best	model	fit,	given	the	study	data.			 125	Table	5.4.	Multivariable	results	(hazard	ratio,	95%	CI,	p-value)	for	each	ICS	exposure	metric	and	associated	Akaike	Information	Criterion	values,	with	time	to	lung	cancer	diagnosis	as	the	outcome	variable.			 		 		 		 		 		Exposure	Metrics	 Multivariable		Regression†			 HR	 95%	CI	 p-value	 AIC		 	 	 	 	 	Time-Dependent	ICS	Exposure	 0.70	 0.61	 0.80	 <0.0001	 19132		 	 	 	 	 	Current	Usea	 0.76	 0.67	 0.87	 <0.0001	 19140		 	 	 	 	 	Cumulative	Years	of	Use	 0.89	 0.83	 0.95	 0.0003	 19141		 	 	 	 	 	Cumulative	Doseb	 0.83	 0.72	 0.97	 0.0201	 19149		 	 	 	 	 	Recency-Weighted	Duration	of	Use	 0.74	 0.66	 0.82	 <0.0001	 19116		 	 	 	 	 	Recency-Weighted	Cumulative	Dose	 0.57	 0.43	 0.74	 <0.0001	 19133			 		 		 		 		 		HR:	Hazard	Ratio;	AIC:	Akaike	Information	Criterion;	CI:	Confidence	Interval.	†Multivariable	regression	analysis	was	adjusted	for	the	following	covariates:	age,	sex,	region,	income	quintile,	inpatient	hospitalization,	number	of	physician	encounters,	COPD	hospitalization,	the	year	of	cohort	entry,	Charlson	Comorbidity	Score,	the	total	number	of	prescriptions	received,	oral	glucocorticoid	use,	and	time-dependent	statin	exposure.	a	Current	use	is	defined	as	receiving	having	received	a	prescription	in	the	6-month	period	immediately	prior	to	the	defined	latency	period.	b	Measured	as	a	continuous	variable	(grams).					 126	5.3.6 Secondary	analysis:	medication	possession	ratio	In	multivariable	analysis	incorporating	the	MPR	to	account	for	exposure	to	ICS,	the	resulting	HR	indicated	that	a	higher	MPR	resulted	in	a	reduction	in	lung	cancer	risk	(Table	5.5).	Because	the	effect	of	MPR	measured	as	a	continuous	variable	was	unlikely	to	be	linear,	ICS	users	were	categorized	as	adherent	to	their	medication	if	their	individual	MPR	was	greater	than	or	equal	to	0.8.	A	significant	effect	was	observed	in	adherent	ICS	users	where	there	was	a	greater	than	50%	reduction	in	lung	cancer	risk	compared	those	who	did	not	use	any	ICS	(HR:	0.46	(95%	CI:	0.27-0.80,	p<0.0001))	after	adjustment	for	potential	confounders.			Table	5.5.	Multivariable	regression	analysis	using	the	medication	possession	ratio	to	capture	exposure	to	ICS.			 		 		 		 		Exposure	Metrics	 Multivariable		Regression			 HR	 95%	CI	LL	 95%CI	UL	 p-value		 	 	 	 	Adherent	vs	Non-Adherenta	 0.46	 0.27	 0.80	 0.0057		 	 	 	 	MPR	Categoryb	 	 	 	 	1	 Reference	 	 	 	2	 0.55	 0.48	 0.64	 <0.0001	3	 0.29	 0.16	 0.50	 <0.0001			 		 		 		 		a	Adherent	users	are	defined	as	having	a	MPR	≥	0.8.	b	MPR	categories	are	as	follows:	the	reference	category	is	MPR=0	(reference	category);	MPR	category	‘2’	is	a	MPR	>	0	and	<	0.8;	MPR	category	‘3’	is	≥	0.8.			 127	5.3.7 Lung	cancer	histology		Two	subgroup	analyses	were	conducted	with	specific	lung	cancer	histology.	The	results	of	these	analyses	also	suggested	that	the	protective	effect	of	ICS	use	occurs	in	the	two	main	types	of	lung	cancer	histology.	The	results	were	also	consistent	for	both	the	reference	case	exposure	metric	(time-dependent	exposure)	and	the	recency-weighted	duration	of	use	metric.	In	multivariable	analysis,	ICS	use,	using	the	reference	case	exposure	definition,	was	associated	with	an	almost	30%	reduction	in	the	risk	of	NSCLC	(HR:	0.70	(95%	CI:	0.60-0.82,	p<0.0001))	and	for	the	recency-weighted	duration	of	use	metric,	there	was	a	24%	reduction	in	risk	associated	with	ICS	use	(HR:	0.76	(95%	CI:	0.68-0.84,	p<0.0001).	In	the	case	of	SCLC,	ICS	use,	for	the	reference	case,	was	also	associated	with	a	risk	reduction,	although	there	was	more	uncertainty	around	the	estimates	(HR:	0.59	(95%	CI:	0.40-0.87,	p=0.0084)	and	HR:	0.56	(95%	CI:	0.39-0.80,	p=0.0018)	for	the	recency-weighted	duration	of	use	metric),	likely	due	to	the	small	number	of	SCLC	cases	(n=117).				 Figure	5.5.	Distribution	of	lung	cancer	cases,	according	to	histology.	01002003004005006007008009001000NSCLC SCLC OtherNumber	of	CasesLung	Cancer	Type	 128	Table	5.6.	Sub-group	analyses	based	on	lung	cancer	histology.	Multivariable	regression	analysis	with	time	to	NSCLC	or	SCLC	diagnosis	as	the	outcome	variables.		 			 Multivariable		Regression			 HR	 95%	Confidence	Interval	 p-value		 	 	 	 		 	 	 	 	NSCLC	 	 	 	 	Time-Dependent	ICS	Exposurea	 0.70	 0.60	 0.82	 <0.0001	Recency-Weighted	Duration	of	Useb	 0.76	 0.68	 0.84	 <0.0001		 	 	 	 	SCLC	 	 	 	 	Time-Dependent	ICS	Exposure	 0.59	 0.40	 0.87	 0.0084	Recency-Weighted	Duration	of	Use	 0.56	 0.39	 0.80	 0.0018			 		 		 		 		HR:	Hazard	Ratio;	CI:	Confidence	Interval;	NSCLC:	Non-small	cell	lung	cancer;	SCLC:	Small	cell	lung	cancer.	a	This	is	the	reference-case	for	the	analysis.	b	The	recency-weighted	duration	of	use	exposure	metric	is	presented	because	it	was	selected	as	the	best	model	based	on	AIC	values	(an	a	priori	criterion).			5.3.8 Sensitivity	analyses	In	sensitivity	analyses,	the	length	of	the	latency	period	associated	with	lung	cancer	was	varied	(Table	5.7)	to	evaluate	whether	the	assumption	of	a	one-year	latency	period	had	an	effect	on	the	relationship	between	ICS	exposure	and	lung	cancer	risk.	Interestingly,	when	the	latency	period	was	removed	altogether,	using	the	time-dependent	exposure	status	metric	and	the	recency-weighted	duration	of	use	metric,	the	multivariable	hazard	ratio	was	statistically	significant	and	in	the	opposite	direction	(HR:	1.12	(95%	CI:	1.02-1.40,	p=0.0242	and	HR:	1.19	(95%	CI:	1.11-1.28,	p<0.0001),	respectively).	While	this	result	contrasts	with	the	a	priori	hypothesis	of	this	study	and	results	from	the	primary	analysis,	given	the	latency	period	associated	with	lung	cancer	development,	this	result	may	actually	illustrate	the	potential	for	protopathic	bias	when	a	latency	period	is	not	incorporated	into		 129	the	analysis	(179).		In	the	next	analysis,	a	180-day	latency	period	was	assumed.	The	estimated	HR	for	ICS	use,	using	both	exposure	metrics	were	in	the	expected	direction,	indicating	a	protective	effect	of	ICS	use	for	lung	cancer	risk,	but	the	results	were	not	statistically	significant.	Finally,	the	latency	period	was	extended	to	a	period	of	two	years,	and	the	resulting	adjusted	HR	for	time-dependent	ICS	use	and	recency-weighted	duration	of	ICS	use	both	indicated	a	substantial	risk	reduction	in	lung	cancer	risk	from	ICS	use	(HR:	0.32	(95%	CI:	0.28-0.37,	p<0.0001)	and	HR:	0.31	(95%	CI:	0.26-0.37,	p<0.0001),	respectively).			5.4 Discussion	This	study	explored	the	relationship	between	ICS	exposure	and	lung	cancer	risk	using	a	population-based	cohort	of	chronic	obstructive	pulmonary	disease	patients.	Additionally,	to	my	knowledge,	this	is	the	first	study	that	has	done	so	using	a	variety	of	medication	exposure	definitions,	with	the	intention	of	exploring	the	relationship	by	which	ICS	exposure	may	be	associated	to	lung	cancer	risk	in	COPD	patients.	A	priori,	and	based	on	the	existing	literature,	the	hypothesis	of	this	study	was	that	ICS	exposure	would	be	associated	with	a	reduced	risk	of	lung	cancer.	The	results	of	this	study	aligned	with	this	hypothesis:	ICS	use,	using	all	exposure	definitions,	and	adjusted	for	a	wide	range	of	potential	confounders,	was	associated	with	a	reduced	risk	of	lung	cancer	diagnosis.	The	results	of	the	analysis	suggest	that,	according	to	the	AIC	criterion,	the	recency-weighted	duration	of	use	metric,	a	method	that	accounts	for	the	duration	of	ICS	use,	while	simultaneously	accounting		 130	for	when	during	follow-up	that	use	occurred,	was	the	best	method	for	measuring	this	association,	and	resulted	in	a	greater	than	25%	reduced	risk	of	lung	cancer.			Table	5.7.	Sensitivity	analyses	of	different	lengths	of	the	latency	period	and	a	cohort	age	restriction,	using	the	time-dependent	ever	metric	of	ICS	medication	exposure	(the	reference	case).		 	 	 			 Multivariable		Regression			 HR	 95%	CI	LL	 95%	CI	UL	 p-	value		 	 	 	 	Latency	Period	 	 	 	 	(i)	None	 	 	 	 	Time-Dependent	ICS	Exposurea	 1.20	 1.02	 1.40	 0.0242	Recency-Weighted	Duration	of	Useb	 1.19	 1.11	 1.28	 <0.0001		 	 	 	 	(ii)	6	months	 	 	 	 	Time-Dependent	ICS	Exposure	 0.91	 0.78	 1.05	 0.1974	Recency-Weighted	Duration	of	Use	 0.97	 0.89	 1.05	 0.4758		 	 	 	 	(iii)	1	yearc	 	 	 	 	Time-Dependent	ICS	Exposure	 0.70	 0.61	 0.80	 <0.0001	Recency-Weighted	Duration	of	Use	 0.74	 0.66	 0.82	 <0.0001		 	 	 	 	(iv)	2	years	 	 	 	 	Time-Dependent	ICS	Exposure	 0.32	 0.28	 0.37	 <0.0001	Recency-Weighted	Duration	of	Use	 0.31	 0.26	 0.37	 <0.0001		 	 	 	 	Cohort	(Age>=	65	years)	 	 	 	 	Time-Dependent	ICS	Exposure	 0.66	 0.56	 0.77	 <0.0001	Recency-Weighted	Duration	of	Use	 0.70	 0.62	 0.79	 <0.0001			 		 		 		 		a	This	is	the	reference-case	for	the	analysis.	b	The	recency-weighted	duration	of	use	exposure	metric	is	presented	because	it	was	selected	as	the	best	model	based	on	AIC	values	(an	a	priori	specified	criterion).	c	A	1-year	latency	period	was	assumed	in	the	primary	analysis	and	is	presented	here	for	comparison.				 131	This	study	presented	in	this	chapter	had	three	main	objectives.	First,	the	broadest	objective	of	this	analysis	was	to	evaluate	the	association	between	ICS	use	and	lung	cancer	risk	using	population-based	administrative	data.	Prior	observational	evidence	suggested	a	protective	effect	of	ICS	use	(92,116);	however,	these	studies	had	significant	limitations,	such	as	patient	populations	that	were	not	representative	of	the	COPD	population.	Moreover,	no	clinical	trial	evidence	exists	that	specifically	addresses	this	research	question,	nor	should	it	be	expected	given	the	nature	of	the	research	question.	Therefore,	the	use	of	high	quality	population-based	administrative	data	that	facilitates	the	accurate	classification	of	both	exposure	and	outcomes	at	the	population	level,	coupled	with	sophisticated	epidemiologic	analyses,	can	provide	the	best	evidence-base	in	this	area.	Second,	in	addressing	this	research	question,	I	sought	to	explore	the	association	between	ICS	use	and	lung	cancer	risk	using	a	variety	of	medication	exposure	definitions	to	ensure	the	robustness	of	the	study	results	and	to	add	to	the	methodological	literature	regarding	medication	exposure	using	administrative	data.	The	third	objective	was	to	explore	the	use	of	a	latency-period,	and	variations	thereof,	to	add	the	methodologic	literature	of	observational	studies	with	lung	or	other	cancer	as	the	outcome	of	interest.				Implementation	of	a	latency	period	in	the	primary	analysis	is	a	valuable	contribution	of	this	study	to	the	literature.	The	intuition	for	the	use	of	such	a	period	is	simple.	Medication	received,	for	example,	on	the	day	immediately	prior	to	a	lung	cancer	diagnosis	is	unlikely	to	have	an	impact	on	whether	or	not	lung	cancer	develops.	Indeed,	lung	cancer	is	often	diagnosed	at	an	advanced	stage	and	is	likely	to	have	been	present	(sub-clinically	or	undiagnosed)	for	quite	some	period	prior	to	actually	being	clinically	diagnosed.	The		 132	sensitivity	analyses	altering	the	length	of	this	latency	period	also	produced	interesting	results	(see	Table	5.6).	Removal	of	the	latency	period	altogether	resulted	in	a	statistically	significant	association	between	ICS	use	and	increased	lung	cancer	risk.	While	this	result	does	not	align	with	the	study	hypothesis,	as	stated	above,	given	what	is	known	about	the	pathogenesis	of	lung	cancer,	this	is	a	likely	example	of	protopathic	bias	(179).	However,	as	the	latency	period	increased,	first	to	six	months,	then	to	one	year,	and	then	two	years,	the	HR	associated	with	use	ICS	decreased	and	became	statistically	significant.	Given	the	current	knowledge	about	of	tumour	growth	time	in	lung	cancer,	the	assumption	of	a	one	year	latency	period	seems	appropriate,	but	this	requires	further	research.				The	recency-weighted	approaches	were	intuitively	attractive	as	they	simultaneously	accounted	for	the	duration	of	use,	the	dosage,	and	the	point	in	time	during	follow-up	of	when	the	prescription	was	filled.	This	method	has	been	used	in	several	previous	studies	(80,81),	but	it	has	been	exclusively	used	for	acute	medical	events,	never	for	a	disease	such	as	lung	cancer	that	has	induction	and	latency	periods.	Both	methods	of	recency-weighting	ICS	exposure	resulted	in	HRs	that	indicated	a	reduction	in	lung	cancer	risk	and	the	corresponding	AIC	value	for	recency-weighted	duration	of	use	model	was	superior	to	the	other	approaches	of	defining	medication	exposure.	While	it	is	possible	that	this	method	of	defining	medication	exposure	simply	would	not	be	relevant	for	diseases	with	long	latency	periods,	the	results	of	this	study	suggest	otherwise.	Therefore,	the	recency-weighted	approach	may	be	a	valuable	method	of	quantifying	medication	exposure	in	studies	evaluating	cancer	risk	and	medication	use.	At	minimum,	the	results	of	this	study	suggest		 133	that	this	method	should	be	considered	as	a	sensitivity	analysis	for	evaluating	cancer	risk	associated	with	medication	use.			To	illustrate	the	impact	that	immortal	time	bias	(180)	can	have	on	analysis	results,	a	bivariate	regression	model	was	estimated	with	ever/never	use	as	a	fixed	covariate.	This	analysis	produced	a	hazard	ratio	of	0.56	(95%	CI:	0.49-0.64,	p<0.0001).	If	this	estimated	HR	is	contrasted	with	the	bivariate	analysis	which	used	the	time-dependent	method	of	exposure	(HR:	0.70	(95%	CI:	0.68-0.88,	p<0.0001),	the	magnitude	of	the	effect	from	an	incorrect	specification	of	the	exposure	definition	becomes	immediately	apparent,	with	an	overestimate	in	the	risk	reduction	of	approximately	14%.	This	highlights	the	importance	of	using	time-dependent	covariates	in	regression	analyses.			Medication	adherence	to	inhaled	medications	is	typically	poor	(181).	Thus,	providing	evidence	on	the	relation	between	adherence	to	ICS	and	lung	cancer	is	an	important	contribution	of	this	study.	Results	from	this	analysis	suggests	that	adherent	patients	have	a	lower	risk	of	lung	cancer	diagnosis	than	non-adherent	patients.	Moreover,	the	study	results	show	that	the	total	cumulative	dose	of	medication	received	plays	an	important	role	in	the	risk	of	lung	cancer.	As	such,	this	should	provide	evidence	and	thus,	motivation	for	patients	to	take	their	inhaled	medication,	as	prescribed,	if	doing	so	can	confer	benefits	that	extend	to	a	reduction	in	lung	cancer	risk.			The	fact	that	the	impact	of	ICS	use	on	lung	cancer	risk	did	not	differ	between	lung	cancer	histology	is	unsurprising	given	the	existing	literature.	If	the	mechanism	by	which	ICS	use		 134	may	reduce	lung	cancer	risk	in	COPD	patients	is	via	a	reduction	in	systemic	inflammation,	then	this	finding	aligns	with	Chaturvedi	et	al.	(40).	In	that	study,	increased	CRP	levels,	while	significantly	associated	with	lung	cancer	risk,	were	not	statistically	significantly	associated	with	a	particular	lung	cancer	histology24.	Similarly,	the	results	of	this	study	suggest	that	the	mechanism	by	which	ICS	might	reduce	lung	cancer	risk	is	not	specific	to	either	histology	of	lung	cancer.			5.4.1 Strengths	and	limitations	This	study	has	several	key	strengths.	First,	the	population-based	nature	the	data,	comprising	approximately	4.3	million	residents	of	British	Columbia,	Canada,	is	a	definite	strength	as	it	protects	against	selection	biases	that	may	occur	when	subjects	are	recruited.	Thus,	using	these	data	improves	on	previously	reported	studies	that	used	very	specific	patient	populations	(87	110).	Second,	this	is	the	first	study	that	has	been	able	to	incorporate	specific	types	of	lung	cancer,	from	accurate	clinical	level	data,	in	evaluating	whether	ICS	use	is	associated	with	lung	cancer	risk	in	COPD	patients.	Third,	incorporating	latency	period	associated	with	lung	cancer	development	into	the	primary	analysis	is	also	an	important	contribution	of	this	study.	Including	this	latency	period	respects	the	time	during	which	lung	cancer	develops,	where	medication	exposures	may	no	longer	have	an	effect.	Future	observational	studies	must	consider	this	aspect	of	(lung)	cancer	when	defining	medication	exposure.	Fourth,	to	reduce	the	chance	of	survivorship	bias,	a	cohort	of	incident	COPD	patients	was	identified	using	a	one-year	wash-in	period.	However,	this	may	have	also	                                                24	Similar	to	the	present	study,	lung	cancer	was	divided	into	3	categories:	(1)	small	cell	lung	cancer,	(2)	non-small	cell	lung	cancer,	and	(3)	other.			 135	resulted	in	a	cohort	of	patients	with	less	severe	COPD,	and	as	such,	a	lower	rate	of	lung	cancer	diagnosis,	particularly	if	this	resulted	in	a	cohort	of	younger	patients.	In	a	sensitivity	analysis,	however,	the	age	of	the	cohort	was	restricted	to	65	years	and	over,	and	the	results	of	the	analysis	were	consistent	with	the	main	analysis	of	the	entire	cohort.	Finally,	this	study	used	an	extensive	list	of	methods	to	define	exposure	to	ICS	using	pharmacy	records	from	an	administrative	database.	Importantly,	the	finding	that	ICS	reduced	lung	cancer	risk	was	consistent	across	each	of	these	methods	of	defining	exposure.	The	consistency	across	each	of	these	methods	is	important	given	the	inherent	limitations	of	administrative	data	in	measuring	medication	exposure,	variability	in	use	of	inhaled	medications,	and	the	heterogeneity	of	COPD	patients.			This	study	has	several	limitations	which	must	be	acknowledged.	First,	while	administrative	data	is	a	valuable	source	of	information	and	can	be	extremely	useful	to	answer	research	questions,	it	is	limited	in	the	scope	of	variables	that	could	inform	exposure-outcome	associations.	For	example,	while	filled	prescriptions	are	recorded,	there	is	no	data	on	whether	or	not	patients	actually	use	their	medication.	Moreover,	in	the	context	of	inhaled	medications,	it	is	well-established	that	patients	often	use	an	incorrect	technique	and	seldom	use	their	medications	as	directed	(182,183)	and	administrative	data	do	not	provide	this	type	of	information.	Second,	no	clinical	data	were	available	for	these	patients	and	the	classification	of	patients	as	having	COPD	is	based	solely	on	their	prescription	profiles.	For	example,	no	data	on	lung	function,	nor	any	other	clinical	marker	of	disease	(or	disease	severity)	was	available.	However,	the	definition	that	has	been	used	to	identify	these	COPD	patients	has	been	used	previously	(144,160)	and	is	likely	a	sensitive	definition	rather	than		 136	a	specific	definition.	Moreover,	the	importance	of	this	study	is	not	that	these	results	are	applicable	for	COPD	patients,	but	rather	that	ICS	may	offer	a	protective	effect	for	lung	cancer	risk	in	those	with	poor	lung	function	(or	lung	inflammation),	regardless	of	whether	the	patient	has	diagnosed	COPD.	Third,	this	study	is	subject	to	the	limitations	of	all	observational	studies,	where	unmeasured	confounding	may	be	present.	It	is	possible	that	ICS	users	may	have	differed	from	non-users,	in	a	manner	that	was	unmeasured	in	the	study	data,	which	affected	their	risk	of	lung	cancer.	However,	the	population-based	nature	of	the	study	data,	the	systematic	approach	to	inclusion	of	potential	confounders,	and	the	use	of	a	broad	set	of	exposure	metrics,	should	have	minimized	the	potential	for	bias.	Moreover,	the	magnitude	of	the	association	between	ICS	exposure	and	lung	cancer	risk,	and	also,	the	consistency	of	this	association	across	all	of	the	exposure	metrics,	enhances	the	validity	of	the	results	of	this	study.	Lastly,	no	data	on	patients	smoking	status	was	available	for	this	analysis.	While	an	obvious	limitation,	given	the	literature	on	COPD,	it	can	be	reasonably	assumed	that	the	majority	of	these	patients	do	have	a	history	of	smoking	or	may	indeed	be	current	smokers	with	likely	as	little	as	15%	of	the	cohort	being	‘never’	smokers	(97,98).		5.5 Concluding	remarks	In	conclusion,	this	study	provides	evidence	for	the	protective	effect	of	ICS	exposure	on	lung	cancer	development	in	patients	with	COPD.	In	doing	so,	this	chapter	makes	several	important	contributions	to	the	literature,	including:	(i)	the	use	of	alternative	methods	of	quantifying	medication	exposure,	including	the	application	of	the	recency-weighted	method	of	quantifying	medication	exposure,	(ii)	the	incorporation	of	a	latency	period	in	the		 137	analysis	to	appreciate	the	characteristics	of	lung	cancer	development,	and	(iii)	the	inclusion	of	specific	lung	cancer	histology,	which	allowed	for	evaluation	of	whether	medication	exposure	had	differential	effects	for	lung	cancer	subgroups.				The	appropriate	use	of	ICS	in	COPD	patients	is	often	debated	and	not	all	patients	might	benefit	from	the	use	of	ICS.	The	clinical	benefits	and	risk	of	use	in	an	individual	patient	must	be	weighed	by	the	physician.	This	study,	however,	does	indicate	that	potential	benefits	may	accrue	from	ICS	use	in	COPD	patients	and	that	sustained	use	may	be	associated	with	reduced	risk	of	lung	cancer.	Moreover,	ICS	use	may	also	reduce	acute	exacerbations	of	COPD	and	improve	quality	of	life.	These	results	highlight	the	importance	of	properly	identifying	which	patients	might	be	at	the	highest	risk	of	lung	cancer,	to	enhance	the	therapeutic	benefits	in	COPD	patients.											 138	Chapter	6: An	evaluation	of	the	association	between	statin	use	and	lung	cancer	risk	in	chronic	obstructive	pulmonary	disease	patients:	a	population-based	cohort	study25		Summary:		In	this	chapter,	I	expand	on	the	results	of	the	analysis	presented	in	Chapter	4	of	this	thesis.	In	Chapter	4,	I	showed	that	statin	use	appears	to	be	associated	with	a	reduction	in	the	risk	of	mortality	when	compared	against	no	statin	use	in	COPD	patients.	In	Chapter	5,	the	results	of	the	analysis	showed	that	inhaled	corticosteroid	(ICS)	use	was	associated	with	a	reduction	in	lung	cancer	risk	in	COPD	patients,	across	several	definitions	of	medication	exposure.	Previously,	Nielsen	et	al.	(144)	have	reported	that	statins	may	reduce	lung	cancer	risk	in	COPD	patients.	In	this	chapter,	I	expand	on	the	work	presented	in	Chapter	4	and	will	evaluate	whether	the	pleiotropic	effects	of	statins	extend	to	reducing	the	risk	of	lung	cancer	in	COPD	patients,	using	a	similar	analytic	framework	as	presented	in	Chapter	5.	In	addition,	this	chapter	explores	whether	or	not	there	may	be	a	synergistic	effect	for	concurrent	ICS	and	statin	use,	and	also	employs	a	negative	control	exposure	to	test	the	robustness	of	the	results	presented	in	Chapters	5	and	6.	                                                25	A	presentation	based	on	this	chapter	has	been	accepted	as	an	oral	presentation	at	the	Canadian	Centre	for	Applied	Research	in	Cancer	Control	(ARCC)	Conference	to	be	held	in	Toronto,	Ontario,	Canada,	in	May	2017.			 139	6.1 Introduction	Chronic	obstructive	pulmonary	disease	(COPD)	is	a	disease	that	is	associated	with	considerable	morbidity	and	mortality	(8).	The	disease	is	characterized	by	pulmonary	and	systemic	inflammation	and	is	typically	associated	with	several	comorbidities	(24,25).	In	particular,	the	prevalence	of	cardiovascular	disease	(CVD)	is	approximately	two	to	three	times	greater	in	COPD	patients	than	in	the	general	population	(28).	As	such,	HMG-CoA	reductase	inhibitors,	or	statins,	which	are	indicated	for	treatment	for	hypercholesterolemia	in	patients	with	established	CVD	(184),	or	thought	to	be	at	risk	for	CVD	(185),	are	a	commonly	used	medication	by	COPD	patients.	Evidence	suggests	that	patients	with	COPD	are	also	at	increased	risk	of	lung	cancer	(32–35).	There	may	be	multiple	factors	that	increase	COPD	patients’	lung	cancer	risk,	one	of	which	is	that	COPD	patients	typically	have	a	history	of	smoking	(186).	However,	it	would	appear	that	in	a	subset	of	patients,	the	increased	risk	of	lung	cancer	extends	beyond	what	can	be	attributed	to	their	smoking	status	or	history.	In	these	patients,	it	is	thought	that	the	additional	risk	of	lung	cancer	is	due	to	increased	systemic	inflammation	that	may	increase	their	risk	of	cancer	over	and	above	what	is	attributable	to	their	smoking	history	or	status	(see	Table	1.2).	Patients	with	concomitant	CVD	and	COPD	exhibit	even	higher	levels	of	systemic	inflammation	than	COPD	patients	without	CVD	(187)	and,	as	such,	are	likely	at	an	even	greater	risk	of	lung	cancer.			 140	While	there	are	a	great	deal	of	studies	that	evaluate	whether	statin	use	is	associated	with	cancer	risk	(175,176)	there	are	a	limited	number	of	studies	that	have	evaluated	statin	use	and	lung	cancer	risk,	specifically.	For	example,	Marelli	et	al.	(190)	conducted	an	analysis	using	electronic	medical	records	of	45,857	matched	pairs	of	adult	Americans	with	an	average	of	4.6	years	of	follow-up,	and	found	no	statistically	significant	relationship	between	exposure	to	statins	and	risk	of	any	cancer	compared	with	no	statin	exposure	(HR:	1.04	(95%	CI:	0.99-1.09)).	Van	Gestel	et	al.	(191)	evaluated	the	association	between	statin	use	and	cancer	mortality,	and,	more	specifically,	the	association	between	statin	use	with	lung	cancer	specific	mortality	in	COPD	patients.	The	authors	reported	a	two-fold	greater	risk	of	lung	cancer	mortality	among	COPD	patients	(HR:	2.06	(95%	CI:	1.32-3.20))	but	found	no	statistically	significant	association	between	statin	use	and	lung	cancer	mortality	in	these	patients	(HR:	0.75	(95%	CI:	0.28-2.05))26.	A	meta-analysis	by	Bonovas	et	al.	(192)	attempted	to	evaluate	the	association	between	statin	use	and	cancer	risk	using	seven	large	randomized	controlled	trials	(RCTs)	–	each	with	more	than	3000	participants.	The	results	of	their	study	showed	no	significant	association	between	statin	use	and	development	of	any	cancer	–	nor	did	their	study	specifically	show	any	significant	association	between	statin	use	and	the	development	of	respiratory	cancer.	In	their	meta-regression	analysis,	there	was	some	evidence	that	suggested	statin	use	may	decrease	cancer	incidence	in	younger	patients	(192).	However,	similar	to	what	was	reported	in	Chapter	3	with	respect	to	RCTs	for	ICS	use	and	lung	cancer	risk,	the	analysis	presented	in	this	study	used	RCTs	that	evaluated	statin	use	for	cardiovascular	outcomes,	not	cancer.	Therefore,	the	results	of	this	meta-analysis	                                                26	For	cancer	mortality,	the	results	showed	borderline	significance	for	a	protective	effect	of	statin	use	versus	non-use	in	COPD	patients	(HR:	0.57	(95%	CI:	0.20-1.01)).			 141	should	be	interpreted	cautiously,	as	the	results	may	be	an	instance	of	Type	II	error	(that	is,	a	failure	to	detect	a	significant	effect	when	one	exists).	A	study	by	Setoguchi	et	al.	(193)	focusing	on	an	elderly	population	suggested	no	statistically	significant	reduction	in	lung	cancer	incidence	associated	with	statin	use.	Conversely,	a	meta-analysis	of	case-control	studies	produced	an	overall	pooled	OR	of	0.71	(95%	CI:	0.56-0.89)	for	any	cancer	and	a	non-significant	result	of	0.75	(95%	CI:	0.50-1.11)	specifically	for	lung	cancer	development	(194).	The	authors	acknowledged,	however,	that	the	studies	included	in	this	meta-analysis	had	short	time	horizons	and	were	not	all	sufficiently	powered	to	address	outcomes	such	as	lung	cancer	development	and	mortality,	specifically,	thereby	rendering	the	non-significant	results	questionable.	Another	study	by	Khurana	et	al.	(130)	focused	on	statin	use	prior	to	lung	cancer	diagnosis	in	the	US	(in	a	VA	setting).		In	this	study,	statin	use	was	defined	as	a	categorical	variable	based	on	years	of	statin	use	(either	any	use	or	use	longer	than	six	months)	to	account	for	different	effects	based	on	duration	of	use.	The	primary	analysis	presented	in	the	study	reported	an	estimated	OR	of	0.55	(95%	CI:	0.52-0.59)	suggesting	a	significant	protective	effect	for	statin	use	versus	no	use.	The	same	study	found	that	the	duration	of	statin	use	also	impacted	the	likelihood	of	lung	cancer	development.	That	is,	patients	that	had	used	a	statin	for	greater	than	six	months	had	an	OR	of	0.45	(95%	CI:	0.42-0.48)	compared	to	patients	with	shorter	duration	of	use.	Indeed,	the	results	were	similar	for	smokers	who	used	statins	for	greater	than	six	months	with	an	OR	of	0.47	(95%	CI:	0.43-0.51)	compared	to	patients	with	no	statin	use.	One	of	the	limitations	to	this	study,	much	like	that	of	Parimon	et	al.	(92)	study	for	ICS	exposure	reported	on	in	Chapters	3	and	5,	is	that	the	study	population	was	almost	exclusively	male	(97.9%).			 142	Understanding	the	association	between	lung	cancer	risk	and	statin	use	in	COPD	patients	can	have	important	implications	for	the	management	of	COPD.	The	significant	costs,	morbidity,	and	mortality	associated	with	lung	cancer	means	that	any	reduction	in	the	number	of	incident	cases	will	mean	a	significant	gain	to	patients	and	for	health	services	provision.	Given	the	potential	role	of	statins	in	improving	outcomes	in	COPD	patients,	the	evidence	base	in	support	of	using	statins	in	COPD	patients	will	be	stronger	if	it	is	shown	that	statin	use	also	reduces	lung	cancer	risk.	The	objective	of	the	analysis	presented	in	this	chapter,	therefore,	was	to	evaluate	the	association	between	statin	use	and	lung	cancer	development	in	a	population-based	cohort	of	COPD	patients.				Figure	6.1.	Conceptual	framework	for	the	analysis	presented	in	Chapter	6.	Systemic	inflammation	resulting	from,	or	as	a	cause	of	COPD,	is	significantly	associated	with	lung	cancer	risk.	The	hypothesis	of	this	study	is	that	statin	use	might	reduce	levels	of	systemic	inflammation	thereby	reducing	lung	cancer	risk.	LUNG	CANCERCOPDSYSTEMIC	INFLAMMATIONSTATINS	 143	6.2 Methods	This	study	used	population-based	administrative	data	for	the	province	of	British	Columbia,	Canada,	to	identify	a	cohort	of	COPD	patients	based	on	individuals’	administrative	prescription	records.	The	databases	used	for	this	study	and	a	description	of	the	criteria	to	be	included	in	the	study	cohort	are	presented	in	Chapter	5	(Section	5.2.1	and	Section	5.2.2,	respectively).		6.2.1 Latency	period	To	attempt	to	appropriately	classify	statin	exposure	with	respect	to	the	pathogenesis	of	lung	cancer,	a	one-year	(365	day)	latency	period	(or	‘lag’	period)	was	applied	in	the	primary	analysis.	That	is,	any	dispensed	statin	prescription	that	was	filled	in	the	one-year	period	prior	to	lung	cancer	diagnosis,	death,	or	censoring,	was	not	considered	as	an	exposure	because	lung	cancer	pathogenesis	is	likely	have	already	been	initiated	during	this	period,	and	any	exposure	during	this	period	was	assumed	to	not	confer	an	effect	on	lung	cancer	development.	Each	of	the	approaches	to	defining	medication	exposure	described	below	incorporated	this	latency	period,	and	the	assumption	of	a	one-year	latency	period	was	subjected	to	several	sensitivity	analyses.					 144		Figure	6.2.	A	graphical	representation	of	the	latency	period,	and	how	medication	exposure	is	considered	with	respect	to	this	latency	period.	 6.2.2 Exposure	measurement	Statin	users	were	initially	identified	according	to	American	Hospital	Formulary	Service	(AHFS)	code	‘240608’:	‘HMG-CoA	Reductase	Inhibitors’	(137).	The	distribution	of	specific	statins	prescribed	within	the	cohort	of	COPD	patients	is	presented	in	Figure	6.3.	While	there	were	several	different	statins	prescribed	to	the	cohort	of	COPD	patients,	a	class	effect	was	assumed	for	all	statins	(195).	Similar	to	the	approach	taken	in	Chapter	5	of	this	thesis,	the	analytic	approach	to	evaluate	the	association	between	statins	and	lung	cancer	employed	an	array	of	methods	of	defining	medication	exposure.	The	reason	behind	taking	this	approach	is	to	explore	whether	the	effect	of	statins	on	lung	cancer	risk	is	robust	across	different	methods	of	defining	exposure.	Moreover,	this	approach	allows	one	to	see	if	the	results	of	the	analysis	are	consistent	across	End	of	Study	Follow-upIndex	DateLung	Cancer	DiagnosedLatency	Period	(365	days)StatinReceived	(A)StatinReceived	(B)The	application	of	 the	latency	period	 (365	days)	means	that	prescriptions	 received	during	 the	latency	period	 (B),	prior	 to	the	diagnosis	of	lung	cancer	in	this	illustration,	are	not	counted	as	exposures.	Prescriptions	 received	after	the	index	date	and	prior	 to	the	latency	period	are	counted	as	statin	exposures.	 145	all	exposure	definitions.	This	analysis	used	four	(i-iv)	conventional	measures	for	medication	exposure	and	two	recency-weighted	approaches	(v,	vi),	as	follows:	(i)	time-dependent	exposure;	(ii)	‘current’	use;	(iii)	cumulative	years	of	use;	(iv)	cumulative	dose	received;	(v)	recency-weighted	duration	of	use;	and	(vi)	recency-weighted	cumulative	dose.		All	exposure	measures	were	time-dependent	and	are	described	in	more	detail	below:	(i) Time-dependent	exposure:	This	method	of	defining	medication	exposure	is	considered	as	the	reference	case	for	this	analysis	and	classifies	those	patients	who	had	filled	a	statin	prescription	during	the	follow-up	period	(after	the	index	date,	or	COPD	‘diagnosis’)	as	‘exposed’,	and	those	that	did	not	as	‘unexposed’.	Importantly,	this	method	was	‘time-dependent’;	that	is,	whether	or	not	a	patient	was	considered	exposed	or	not	exposed	to	statins	could	vary	over	the	course	of	the	follow-up	time,	depending	on	whether	or	not	the	patient	had	been	dispensed	a	statin	prescription	at	that	time.	Allowing	for	a	patients’	exposure	status	to	vary	over	the	study	follow-up	time	has	important	implications	for	minimizing	bias,	such	as	immortal	time	bias,	which	is	not	uncommon	in	observational	studies	using	administrative	data	to	evaluate	the	effectiveness	of	medications	(196)	and	has	the	potential	to	render	significantly	biased	results	(180).	This	bias	may	be	present	when	exposure	status	is	fixed,	and	the	time	prior	to	actual	exposure	is	counted	as	exposed	time,	despite	the	patient	not	yet	being	exposed	to	the	medication	(139,172).	(ii) ‘Current	Use’:	‘Current’	use	is	defined	as	a	patient	having	filled	at	least	one	prescription	for	a	statin	in	the	six	month	period	prior	to	the	beginning	of	the	latency	period.	This	definition	of	medication	exposure	implies	that	there	may	be	a	period		 146	during	which	exposure	to	the	medication,	previous	to	the	event,	is	especially	important.		(iii) Cumulative	Duration	of	Use:	For	this	approach,	the	days	supplied	of	each	prescription	was	aggregated	during	the	follow-up	period	for	each	individual	patient	and	divided	by	365	to	get	the	total	years	of	statin	use.	This	definition	of	medication	exposure	assumes	that	the	length	of	time	for	which	a	patient	has	taken	the	medication	will	be	associated	with	the	study	outcome.	(iv) Cumulative	Dose:	This	definition	of	medication	exposure	is	calculated	by	summing	the	actual	amount	of	medication	prescribed	to	a	specific	patient	over	the	study	period.	This	method	of	capturing	medication	exposure	assumes	that	the	total	amount	of	medication	received	(in	grams,	for	this	analysis)	is	associated	with	the	outcome.	The	cumulative	dose	is	calculated	using	the	dose	of	medication	prescribed	multiplied	by	the	quantity	prescribed.	Statins	prescriptions	were	converted	to	equivalent	doses	of	atorvastatin	(the	most	frequently	prescribed	statin	in	the	cohort)	(197).	(v) Recency-Weighted	Cumulative	Duration	of	Use:	The	weighted	cumulative	duration	of	use	method	was	employed	to	account	for	the	duration	of	medication	use	while	also	accounting	for	when,	during	follow-up,	the	prescription	was	filled	with	respect	to	the	outcome	(96).	To	account	for	when	during	follow-up	the	prescription	was	filled,	a	function	was	estimated	that	weighted	prescriptions	filled	closer	to	the	date	of	the	event,	or	end	of	follow-up,	greater	than	those	filled	earlier	in	the	follow-up	period,	while	also	respecting	the	latency	period.	This	shape	of	the	weighting	function	is	based	on	previous	literature	employing	this	method	(80,81,91).	This	method	assumes	that	while	the	duration	of	use	is	important,	the	exposures	occurring	immediately	previous		 147	to	the	event,	again	respecting	the	latency	period,	confer	a	greater	protective	effect	than	exposures	that	occurred	earlier	in	the	follow-up	period.		(vi) Recency-Weighted	Cumulative	Dose:	Similar	to	the	approach	taken	above	in	(v),	the	recency-weighted	cumulative	dose	exposure	definition	was	used	to	simultaneously	account	for	both	the	cumulative	dose	and	the	time	point	at	which	the	prescription	was	filled	(96).	The	same	weighting	function	was	employed	which	assigned	a	greater	weight	to	doses	of	prescriptions	of	medications	filled	more	proximal	to	the	outcome,	while	respecting	the	latency	period.	The	underlying	assumption	with	this	method	of	defining	exposure	is	that	the	cumulative	dose	of	the	medication	is	important,	but	it	is	also	important	when	that	dose	is	received,	relative	to	the	study	outcome.		6.2.3 Adjustment	for	potential	confounders	Covariates	that	were	identified	as	potential	confounders	of	the	association	between	statin	exposure	and	lung	cancer	diagnosis	were	incorporated	into	the	multivariable	model,	based	on	the	procedure	described	in	the	next	section.	Potential	confounders	were	assessed	in	the	one	year	period	preceding	the	latency	period.	The	demographic	covariates	considered	included:	age,	sex,	neighborhood	income	quintiles	based	neighborhood	of	residence,	and	the	health	authority	(regional	health	service)	where	the	patient	resided.	For	each	patient,	the	number	of	prescriptions	dispensed	(excluding	statins),	the	number	of	hospital	encounters,	the	number	of	inpatient	hospital	stays,	and	the	number	of	physician	encounters	were	calculated.	Moreover,	to	account	for	comorbidities	experienced	by	the	patient	during	the	follow-up	period,	the	Charlson	Comorbidity	Index	was	calculated	based	on	health	services	records,	excluding	COPD,	CVD,	and	cancer	(185,186).		 148	6.2.4 Statistical	analysis		A	Cox	regression	model	was	used	to	estimate	the	hazard	of	lung	cancer	diagnosis	based	on	statin	exposure.	In	the	context	of	this	study,	the	analytical	approach	calculates	the	time	from	COPD	‘diagnosis’	(the	index	date)	to	lung	cancer	diagnosis,	death,	or	end	of	study	follow-up.		To	identify	potential	confounders	to	be	included	the	multivariable	model,	a	series	of	bivariate	regression	analyses	were	carried	out	to	determine	candidate	covariates	for	the	final	multivariable	model	using	a	threshold	of	p<0.20	for	consideration.	The	results	of	these	bivariate	regressions	with	potential	covariates	are	reported	in	Table	6.3.	Each	covariate	that	met	this	threshold	in	a	bivariate	regression	was	then	added	to	the	multivariable	model	via	stepwise	selection	comparing	Akaike	Information	Criterion	(AIC)	values,	with	lower	AIC	values	indicating	better	model	fit	(142).	Comparison	of	AIC	values	is	a	method	of	evaluating	model	fit	which	incorporates	a	penalty	for	the	addition	of	each	covariate,	thereby	favoring	more	parsimonious	models.	Hazard	ratios	and	associated	95%	confidence	intervals	are	reported	for	each	of	the	exposure	metrics	as:	(a)	bivariate	analyses;	(b)	multivariable	age	and	sex	adjusted	analyses;	and	(c)	‘fully’	adjusted	multivariable	analyses,	with	time	to	lung	cancer	diagnosis	as	the	outcome.	An	interaction	term	for	statin	and	ICS	use	was	also	evaluated	given	the	evidence	presented	both	in	Chapter	5	as	well	as	the	a	priori	hypothesis	of	this	study.		6.2.5 Secondary	analysis:	medication	possession	ratio	The	medication	possession	ratio,	typically	used	as	a	measure	of	adherence,	is	the	ratio	between	the	time	from	the	initial	prescription	of	a	statin	to	the	end	of	follow-up,	or	the	event	of	interest,	and	the	number	of	days	supplied	of	the	medication	(93).	Thus,	it	gives	a		 149	ratio	for	which	the	patient	had	a	prescription	over	the	course	of	the	follow-up	period,	and	provides	an	estimate	of	medication	use.	The	relationship	between	MPR	and	lung	cancer	risk	is	unlikely	to	be	linear,	therefore	the	MPR	was	used	to	categorize	COPD	patients	as	‘adherent’	if	the	value	of	MPR	was	greater	than	or	equal	to	0.8	(95)	over	the	follow-up	period.	However,	whether	or	not	a	patient	was	‘adherent’	to	statins	is	not	focus	of	this	study.	Rather,	the	MPR	is	used	as	measure	which	estimates	the	proportion	of	follow-up	time	for	which	a	COPD	patient	was	dispensed	a	statin,	and	is	simply	another	method	of	capturing	medication	use.	6.2.6 Sub-group	analysis:	lung	cancer	histology	 	The	British	Columbia	Cancer	Agency	Registry	file	provided	information	on	each	lung	cancer	case	to	classify	lung	cancers	as:	(i)	non-small	cell	lung	cancer	(NSCLC);	(ii)	small	cell	lung	cancer	(SCLC);	and	(iii)	‘other’.	The	distribution	of	these	cancers	within	the	cohort	of	COPD	patients	is	reported	in	Figure	6.5.	Using	the	same	model	developed	for	the	main	multivariable	analysis,	the	lung	cancer	cases	were	restricted	to	either	SCLC	or	NSCLC,	and	the	association	between	statin	use	and	these	specific	cancer	types	was	evaluated	and	is	presented	in	Table	6.5	below.		6.2.7 Sensitivity	analyses	Several	sensitivity	analyses	were	performed	to	test	the	importance	of	assumptions	on	the	results	of	this	study.	First,	in	the	primary	analysis	a	one-year	period	for	lung	cancer	latency	was	assumed.	Therefore,	to	test	this	assumption,	the	latency	period	was:	(i)	reduced	to	not	present	at	all	(i.e.	zero	days);	(ii)	reduced	to	six	months;	and	(iii)	extended	to	two	years.		 150	Second,	lung	cancer	incidence	is	low	for	patients	less	than	65	years	of	age	(163,164).	Therefore,	a	sensitivity	analysis	was	conducted	in	which	the	cohort	of	COPD	patients	was	restricted	to	65	years	and	over	to	evaluate	whether	statin	exposure	resulted	in	a	similar	effect	on	lung	cancer	risk	under	this	restriction.			6.2.8 Negative	control	exposure		To	explore	whether	or	not	the	association	between	medication	exposure	and	lung	cancer	risk	observed	in	this	study	might	be	due	to	confounding,	a	further	analysis	was	conducted	employing	a	negative	control	as	an	exposure.	In	this	approach,	an	alternative	medication	class	was	identified	for	which	there	was	no	evidence	of	an	association	between	exposure	to	the	medication	and	a	reduction	in	the	risk	of	developing	cancer.	This	approach	is	similar	to	using	a	placebo	in	a	trial-setting;	that	is,	the	placebo,	similar	to	the	negative	control	exposure,	should	have	no	association	with	the	study	outcome	(198).	Therefore,	if	an	association	exists	between	the	negative	control	exposure,	similar	to	what	was	found	in	the	primary	analysis,	it	is	likely	that	the	original	result	was	due	to	confounding	and	provides	evidence	for	the	absence	of	a	true	association	(198–200).	To	conduct	the	negative	control	exposure	analysis,	calcium	channel	blockers	(CCB)	were	chosen	based	on	a	review	of	the	literature	which	suggested	there	was	no	association	between	CCB	and	lung	cancer	development,	nor	evidence	supporting	an	association	with	cancer,	generally	(201–203).	Consistent	with	existing	literature	that	has	employed	negative	control	exposures	in	observational	research,	time-dependent	CCB	exposure	was	first	included	in	a	bivariate	Cox	regression	model,	then	in	the	fully-adjusted	multivariable	model,	and	then	also	included	in		 151	a	multivariable	Cox	regression	model	with	exposures	of	the	interest,	statin	and	ICS	exposure	(188,193).			6.3 Results	A	cohort	of	39,678	COPD	patients	was	identified	that	met	the	study	inclusion	criteria.	The	mean	age	of	the	patients	was	70.6	(SD:	11.2)	years	and	53.5%	were	female.	Mean	follow-up	time	among	patients	in	the	COPD	cohort	was	5.1	years.	There	were	994	cases	of	lung	cancer	identified	within	the	COPD	cohort.	Further	characteristics	of	the	cohort	of	COPD	patients	is	presented	in	Table	6.1.		6.3.1 Statin	use	in	the	COPD	cohort	There	were	12,469	COPD	patients	that	received	at	least	one	prescription	for	a	statin.		Among	these	statin	users,	there	were	258,458	statins	prescriptions	dispensed	during	the	study	follow-up	time,	resulting	in	an	average	of	approximately	21	prescriptions	per	patient	and	an	average	cumulative	dose	of	11.9	(SD:	9.2)	grams.	Figure	6.3	shows	that	the	most	commonly	prescribed	statin	was	atorvastatin	(>55%)	with	simvastatin	as	the	second	most	prescribed	(24%).	The	MPR	for	statin	users	is	shown	in	Figure	6.4;	of	those	prescribed	a	statin,	only	10.3%	had	an	MPR	greater	than	0.80,	which	is	a	common	threshold	to	be	considered	as	adherent	(95).					 152	Table	6.1.	Demographics	of	the	COPD	cohort	(n=39,879).		 		Patient	Characteristic	 Value			 			 	Age	 70.6	(SD:	11.2)	Age	Distribution		50<60	 8356	(21.0%)	60<70	 10,212	(25.7%)	70<80	 12,436	(31.2%)	≥80	 8785	(22.1%)	Female	 21,273	(53.5%)	Income	Quintile		 	1	 9701	(25.6%)	2	 7864	(20.8%)	3	 6854	(18.1%)	4	 6381	(16.8%)	5	 5696	(15.0%)	Health	Authority	Interior		 8569	(21.6%)	Fraser		 11,354	(28.6%)	Vancouver	Coastal		 7740	(19.5%)	Vancouver	Island		 7522	(18.6%)	Northern		 2465	(6.2%)	Hospitalizations		 6651	(16.7%	Any	Reason	 6651	(16.7%	COPD-related	 1084	(2.7%)	CVD-related		 512	(1.3%)	Charlson	Comorbidity	Category1	 	0	 31,354	(79.0%)	1	 6303	(15.9%)	2	 1176	(3.0%)	≥3	 843	(2.1%)	Combination	Therapy	(ICS/LABA)	 6585	(16.5%)	Physician	Encounters	(any	reason)2	 11	(3-22)	Number	of	Prescriptions	Received	(any	reason)2	 21	(7-44)		 	Note:	Values	represent	mean	(standard	deviation)	or	number	(percentage)	unless	otherwise	indicated.	Where	percentages	do	not	add	to	100%	the	reason	is	due	to	rounding.				1	–	Category	0	is	a	Charlson	Score	of	0,	Category	1	is	(0,2],	Category	2	is	(2,	3],	Category	3	is	>3.	This	calculation	excludes	COPD,	CVD,	and	cancer.	2	–	Median	and	interquartile	ranges.			 153		Figure	6.3.	Distribution	of	all	statins	dispensed	among	statin	users	in	the	COPD	cohort.	 	Figure	6.4.	Medication	possession	ratio	(MPR)	for	statin	users	in	the	COPD	cohort.	0102030405060Atorvastatin Fluvastatin Lovastatin Pravastatin Rosuvastatin SimvastatinPercetnage	of	All	Statin		PrescriptionsStatin	Generic	Name05000100001500020000250003000035000MPR=0 0<MPR<0.8 MPR≥0.8Number	of	COPD	patientsMedication	Possession	Ratio	 154	6.3.2 Bivariate	and	age/sex	adjusted	results:	statin	exposure	definitions	In	the	bivariate	analyses,	statin	exposure	was	not	significantly	associated	with	lung	cancer	risk	using	any	of	the	conventional	methods	(time-dependent	exposure,	current	use,	cumulative	years	of	use,	cumulative	dose)	of	defining	medication	exposure,	though	the	direction	of	the	effect	was	in	the	a	priori	expected	direction,	that	is,	showing	a	reduction	in	the	risk	of	lung	cancer	associated	with	statin	exposure.	The	two	recency-weighted	exposure	metrics,	however,	both	resulted	in	a	reduction	in	the	risk	of	lung	cancer	by	13%	and	2%	(recency-weighted	duration	of	use	(HR:	0.87	(95%	CI:	0.79-0.95))	per	year	of	statin	use,	and	recency-weighted	cumulative	dose	(HR:	0.98	(95%	CI:	0.96-0.99))	per	gram	of	statin	use,	respectively).			The	results	of	the	age	and	sex	adjusted	analysis	were	similar	to	the	bivariate	results	in	that	none	of	the	conventional	time-dependent	exposure	definitions	for	statin	use	exhibited	statistical	significance,	though	hazard	ratios	were	all	less	than	one,	suggesting	a	protective	effect	from	statin	use,	as	expected.	Similar	to	the	bivariate	analysis,	both	recency-weighted	approaches	produced	statistically	significant	hazard	ratios.	For	the	recency-weighted	duration	of	use	metric,	statin	use	was	associated	with	a	13%	reduction	in	lung	cancer	risk	per	year	of	exposure	(HR:	0.86	(95%	CI:	0.79-0.94)	and	a	2%	reduction	in	lung	cancer	risk	for	each	additional	gram	of	statin	received	(HR:	0.98	(95%	CI:	0.96-0.99).	Full	results	of	the	bivariate	and	age/sex	adjusted	analyses	are	presented	in	Table	6.2.				 155	Table	6.2.	Bivariate,	and	age/sex	adjusted	regression	results	(hazard	ratios	and	95%	confidence	intervals)	for	each	exposure	definition	with	time	to	lung	cancer	diagnosis	as	the	outcome.			 		 		 		 		 		 		Exposure	Metrics	 Bivariate	 Age	and	Sex	Adjusted			 HR	 95%	CI	LL	 95%	CI	UL	 HR	 95%	CI	LL	 95%	CI	UL		 	 	 	 	 	 	Time-dependent	statin	exposure	 0.88	 0.76	 1.02	 0.87	 0.74	 1.01		 	 	 	 	 	 	Current	Usea	 0.91	 0.78	 1.06	 0.90	 0.77	 1.05		 	 	 	 	 	 	Cumulative	Years	of	Use	 0.97	 0.91	 1.02	 0.96	 0.91	 1.02		 	 	 	 	 	 	Cumulative	Doseb	 0.99	 0.98	 1.00	 0.99	 0.98	 1.00		 	 	 	 	 	 	Recency-Weighted	Cumulative	Duration	of	Use	 0.87	 0.79	 0.95	 0.86	 0.79	 0.94		 	 	 	 	 	 	Recency-Weighted	Cumulative	Dose	 0.98	 0.96	 0.99	 0.98	 0.96	 0.99			 		 		 		 		 		 		HR:	Hazard	Ratio;	Ref:	Reference	Category;	CI:	Confidence	Interval;	LL:	Lower	Limit;	UL:	Upper	Limit	a	Current	use	is	defined	as	having	filled	a	prescription	in	the	6-	month	period	immediately	prior	to	the	latency	period.														b	Measured	as	a	continuous	variable	(grams).									 156	Table	6.3.	Bivariate	regression	model	results,	with	time	to	lung	cancer	diagnosis	as	the	outcome,	for	covariates	to	be	considered	for	inclusion	in	the	multivariable	model.			 		 		 		 		Covariate	 Hazard	Ratio		 95%	Confidence	Interval	 p-value			 		 		 		 			     Age	 1.01	 1.00	 1.01	 0.0011	Age	Categories	 		 		 		 		<60	 Ref	 Ref	 Ref	 Ref	[60,	70)	 2.02	 1.67	 2.43	 <.0001	[70,	80)	 2.33	 1.95	 2.80	 <.0001	≥80	 1.29	 1.03	 1.61	 0.241	Sex	(Male)	 1.39	 1.24	 1.56	 <.0001	Health	Authority	 	    Interior	 1.29	 0.98	 1.71	 0.0744	Fraser	 1.23	 0.93	 1.62	 0.1485	Vancouver	Coastal	 1.04	 0.78	 1.40	 0.7687	Vancouver	Island	 1.46	 1.10	 1.94	 0.0084	Northern	 Ref	 Ref	 Ref	 Ref	Income	Quintile	 	    5	 Ref	 Ref	 Ref	 Ref	4	 1.27	 1.03	 1.57	 0.0246	3	 1.14	 0.92	 1.41	 0.2152	2	 1.23	 1.00	 1.50	 0.0491	1	 1.24	 1.02	 1.51	 0.0305	Total	Number	of	Prescriptions		 1.00	 0.99	 1.00	 <.0001	Charlson	Comorbidity	Score	(Continuous)	 1.06	 0.97	 1.16	 0.1853	Charlson	Comorbidity	Score	(Categorical)	 	    0	 Ref	 Ref	 Ref	 Ref	1	 1.15	 0.98	 1.36	 0.0931	2	 0.94	 0.62	 1.40	 0.7459	≥3	 0.90	 0.55	 1.48	 0.6814	Inpatient	Stay	 3.57	 3.16	 4.03	 <.0001	Number	of	hospitalizations	 1.66	 1.64	 1.68	 <.0001	COPD-related	hospitalization	 2.56	 2.00	 3.27	 <.0001	CVD-related	hospitalization	 1.04	 0.58	 1.88	 0.8958	Combination	Therapy	(ICS/LABA)	 1.27	 1.11	 1.47	 0.007	Number	of	physician	encounters	 1.02	 1.02	 1.02	 <.0001	Oral	glucocorticoid	use	 1.09	 0.91		 1.30	 	0.3399	HR:	Hazard	Ratio;	AIC:	Akaike	Information	Criterion;	CI:	Confidence	Interval;	LL:	Lower	Limit;	UL:	Upper	Limit. 	 157	 6.3.3 Multivariable	analysis		In	multivariable	analysis,	after	adjusting	for	potential	confounders,	the	hazard	ratio	for	statin	exposure	was	in	the	expected	direction	but	was	not	statistically	significant	for	the	conventional	exposure	metrics	(metrics	(i)-(iv)	listed	above	in	Section	6.2.2).	For	these	exposure	metrics,	the	largest	reduction	of	risk	resulted	from	the	reference	case	exposure	definition,	where	time-dependent	statin	exposure	was	associated	with	a	14%	decrease	in	lung	cancer	risk,	but	was	not	statistically	significant	(HR:	0.85	(95%	CI:	0.73-1.00,	p=0.050)),	at	an	absolute	threshold	for	statistical	significance	of	alpha	equal	to	0.05.	Although	AIC	values	were	similar	among	these	conventional	measures	of	exposure,	the	model	employing	time-dependent	statin	exposure	metric	produced	the	smallest	AIC	value	among	these	metrics	(Table	6.4).	Exposure	classified	based	on	the	cumulative	dose	of	statin	received	had	the	poorest	AIC	value,	and	the	estimated	HR	was	not	statistically	significant	for	lung	cancer	risk	(HR:	0.99	(95%	CI:	0.96-1.00,	p=0.128)).	The	two	recency-weighted	approaches	had	superior	AIC	values	compared	to	the	conventional	time-dependent	exposure	definitions,	exhibiting	approximately	a	ten-point	difference.	The	estimated	multivariable	hazard	ratio	for	the	recency-weighted	duration	of	use	exposure	metric	showed	a	15%	reduction	in	lung	cancer	risk	from	per	year	of	statin	use	(HR:	0.85	(95%	CI:	0.77-0.93,	p=0.0006))	and	the	recency-weighted	cumulative	dose	measure	showed	a	5%	reduction	in	lung	cancer	risk	per	gram	of	statin	(HR:	0.97	(95%	CI:	0.96-0.99,	p=0.0019)).	Of	all	the	models	incorporating	time-dependent	covariates	in	multivariable	analysis,	the	recency-weighted	duration	of	use	exposure	definition	had	the		 158	best	AIC	value	(19122).	Full	results	of	multivariable	analysis,	along	with	AIC	values,	are	presented	in	Table	6.4.		Table	6.4.	Multivariable	regression	results	for	each	statin	exposure	metric	with	time	to	lung	cancer	diagnosis	as	the	outcome	variable.				 		 		 		 		 		Exposure	Metric	 Multivariable	Regression†			 HR	 95%	CI	LL	 95%CI	UL	 p-value	 AIC		 	 	 	 	 	Time-Dependent	Statin	Exposure	 0.85	 0.73	 1.00	 0.050	 19132		 	 	 	 	 	Current	Usea	 0.88	 0.75	 1.04	 0.145	 19133		 	 	 	 	 	Cumulative	Years	of	Use	 0.95	 0.90	 1.01	 0.118	 19133		 	 	 	 	 	Cumulative	Doseb	 0.99	 0.98	 1.00	 0.128	 19133		 	 	 	 	 	Recency-Weighted	Duration	of	Use	 0.85	 0.77	 0.93	 0.001	 19122		 	 	 	 	 	Recency-Weighted	Cumulative	Dose	 0.97	 0.96	 0.99	 0.002	 19124			 		 		 		 		 		HR:	Hazard	Ratio;	AIC:	Akaike	Information	Criterion;	CI:	Confidence	Interval;	LL:	Lower	Limit;	UL:	Upper	Limit	†Multivariable	regression	analysis	was	adjusted	for	the	following	covariates:	age,	sex,	region,	income	quintile,	inpatient	hospitalization,	number	of	physician	encounters,	COPD	hospitalization,	the	year	of	cohort	entry,	Charlson	Comorbidity	Score,	the	total	number	of	prescriptions	received,	oral	glucocorticoid	use,	and	time-dependent	ICS	exposure.	a	Current	use	is	defined	as	receiving	having	received	a	prescription	in	the	6-month	period	immediately	prior	to	the	defined	latency	period.	b	Measured	as	a	continuous	variable	(grams).			 159	Given	the	results	from	Chapter	5,	where	ICS	use	was	associated	with	a	reduced	risk	of	lung	cancer,	an	interaction	term	was	added	to	the	multivariable	model	to	evaluate	whether	a	synergistic	effect	between	statin	and	ICS	use	might	reduce	the	risk	of	lung	cancer.	The	addition	of	this	term	into	the	multivariable	model,	however,	resulted	in	a	non-significant	hazard	ratio	for	the	interaction	term	(HR:	1.01	(95%	CI:	0.72-1.42,	p=0.9427),	therefore	not	lending	to	support	to	the	idea	of	a	synergistic	effect	of	concurrent	statin	and	ICS	use.	Addition	of	statin	use	to	the	multivariable	model	including	ICS	use	did	not	alter	the	significant	association	of	ICS,	suggesting	an	independent	protective	effect.			Table	6.5.	Evaluation	of	the	medication	possession	ratio	to	capture	exposure	to	statins	over	the	study	follow-up	period,	and	lung	cancer	risk.			 		 		 		 		Exposure	Metric	 Multivariable		Regression			 HR	 95%	CI	LL	 95%	CI	UL	 p-value		 	 	 	 	Adherent	vs	Non-Adherenta	 0.46	 0.29	 0.73	 0.0011		 	 	 	 	MPR	Categoryb	 	 	 	 	1	 Reference	 	 	 	2	 0.69	 0.58	 0.81	 <0.0001	3	 0.41	 0.26	 0.66	 0.0002			 	 	 	 	a	Adherent	users	are	defined	as	having	a	MPR	≥	0.8.	b	MPR	categories	are	as	follows:	the	reference	category	‘1’	is	MPR	=	0;	MPR	category	‘2’	is	a	MPR	>	0	and	<	0.8,	and	MPR	category	‘3’	is	≥0.8.			 160	6.3.4 Lung	cancer	histology		Of	the	994	cases	of	lung	cancer	identified	within	the	COPD	cohort,	854	were	classified	as	non-small	cell	lung	cancer	(NSCLC)	and	117	were	classified	as	small	cell	lung	cancer	(SCLC).	The	distribution	of	these	classifications	align	with	estimates	from	other	jurisdictions	that	report	approximately	15%	of	lung	cancer	cases	are	classified	as	SCLC	(177,178).	The	estimated	hazard	ratios	for	the	association	between	statin	use	the	development	of	NSCLC	was	0.83	(95%	CI:	0.70-0.99,	p=0.0349)	for	time-dependent	exposure	and	0.83	(95%	CI:	0.75-0.92,	p=0.0004)	for	the	recency-weighted	duration	of	use	metric.	For	SCLC,	the	estimated	HRs	for	each	metric	of	statin	exposure	were	not	statistically	significant	(Table	6.6),	however,	this	result	may	have	been	due	to	the	lower	number	of	SCLC	cases	observed	(n=117).		Table	6.6.	Evaluation	of	association	between	statin	exposure	and	lung	cancer	histology.			 Multivariable		Regression			 HR	 95%	CI	LL	 95%CI	UL	 p	value		 	 	 	 	NSCLC	 	 	 	 	Time-Dependent	Statin	Exposurea	 0.83	 0.70	 0.99	 0.0349	Recency-Weighted	Duration	of	Useb	 0.83	 0.75	 0.92	 0.0004		 	 	 	 	SCLC	 	 	 	 	Time-Dependent	Statin	Exposure	 1.18	 0.77	 1.80	 0.4542	Recency-Weighted	Duration	of	Use	 1.04	 0.82	 1.32	 0.7561			 		 		 		 		HR:	Hazard	Ratio;	CI:	Confidence	Interval;	LL:	Lower	Limit;	UL:	Upper	Limit;	NSCLC:	Non-small	cell	lung	cancer;	SCLC:	Small	cell	lung	cancer.	a	This	is	the	reference-case	for	the	analysis.	b	The	recency	weighted	duration	of	use	exposure	metric	is	presented	because	it	was	selected	as	the	best	model	based	on	AIC	values	(an	a	priori	criterion).				 161	6.3.5 Sensitivity	analyses		Several	sensitivity	analyses	were	performed	to	explore	how	different	specifications	of	the	latency	period	might	affect	the	results	of	this	study.	These	results	are	presented	in	Table	6.7.	When	the	latency	period	was	eliminated	altogether,	statin	use	was	not	significantly	associated	with	lung	cancer	risk	using	the	time-dependent	statin	exposure	metric.	When	a	six-month	latency	period	was	applied,	the	estimated	HR	for	each	exposure	metric	was	not	statistically	significant	but,	again,	was	in	the	expected	direction.	When	the	latency	period	was	extended	to	two	years,	the	association	between	lung	cancer	risk	and	statin	use	was	statistically	significant,	suggesting	an	almost	40%	reduction	in	lung	cancer	risk	from	statin	use	compared	to	non-use	(HR:	0.62	(95%	CI:	0.52-0.73,	p<0.0001).	For	the	recency-weighted	duration	of	use	exposure	metric,	the	results	were	similar	under	this	assumption,	where	the	estimated	multivariable	HR	suggested	a	greater	than	40%	reduction	in	lung	cancer	risk	conferred	from	statin	use	(HR:	0.57	(95%	CI:	0.49-0.66,	p<0.0001).		In	another	analysis,	the	cohort	of	COPD	patients	was	then	restricted	to	those	65	years	of	age	and	over	to	reflect	the	fact	the	lung	cancer	has	a	typically	occurs	in	patients	near	this	age.	The	estimated	HR	for	time-dependent	statin	exposure,	under	the	reference	latency	period,	was	0.72	(95%	CI:	0.60-0.86,	p=0.0004)	showing	a	protective	effect	from	statin	use	on	lung	cancer	risk.	Similarly,	using	the	recency-weighted	duration	of	use	exposure	metric,	statin	use	was	associated	with	an	22%	decrease	in	lung	cancer	risk	per	year	of	statin	use	(HR:	0.77	(95%	CI:	0.70-0.87,	p<0.0001).				 162	A	further	sensitivity	analysis	was	conducted,	using	a	negative	control	exposure,	to	detect	whether	the	results	of	the	primary	analysis	were	inherently	biased	or	confounded.	To	do	so,	the	association	between	time-dependent	CCB	exposure	and	lung	cancer	risk	was	explored	in	the	multivariable	model.	In	the	multivariable	(without	statin	and	ICS	exposure),	and	in	multivariable	analysis	including	time-dependent	ICS	and	statin	exposure,	no	association	between	the	negative	control	exposure	and	lung	cancer	risk	was	found	using	any	definition	of	medication	exposure	(for	time-dependent	CCB	exposure,	the	estimated	HRs	were	0.89	(95%	CI:	0.75-1.10,	p=0.2102)	and	0.92	(95%	CI:	0.77-1.11,	p=0.385)	when	included	with	statin	and	ICS	exposure).	This	result	lends	further	credibility	to	there	being	a	true	association	between	statin	use	and	a	reduction	in	lung	cancer	risk.			6.4 Discussion	This	study	evaluated	the	association	between	lung	cancer	risk	and	statin	exposure	in	a	population-based	cohort	of	COPD	patients.	The	analysis	employed	an	array	of	metrics	for	quantifying	medication	exposure,	adding	to	the	pharmacoepidemiologic	literature	about	the	use	of	exposure	metrics	in	studies	using	administrative	data.	While	statin	exposure	was	not	statistically	significantly	associated	with	a	reduction	in	lung	cancer	risk	across	all	exposure	metrics,	the	overall	results	of	this	study	do	suggest	that	statin	use	in	COPD	patients	potentially	reduces	the	risk	of	lung	cancer	and	lends	support	to	the	hypothesis	that	patients	with	COPD	might	benefit	from	statin	therapy	in	this	regard.	Comparison	of	AIC	values	to	determine	the	model	that	best	fit	the	study	data	suggested	that	the	recency-weighted	cumulative	duration	of	use	exposure	metric	was	best.	Beyond	superior	AIC		 163	values,	the	recency-weighted	approach	is	intuitive	as	well;	it	implies	that	the	duration	of	statin	use	is	important,	but	also	when	that	use	occurs,	with	respect	to	the	outcome,	is	also	important.	The	decreasing	gradient	of	risk	observed	for	higher	categories	of	the	MPR,	and	also	the	magnitude	of	the	protective	effect	in	multivariable	analysis	using	this	measure	of	exposure,	strengthen	the	plausibility	of	these	results.	The	results	of	this	analysis	also	suggest	that	statin	use	may	reduce	the	risk	of	lung	cancer	in	COPD	patients	aged	sixty-five	or	greater,	and	that	the	protective	effect	of	statins	might	be	greater	for	NSCLC.		Table	6.7.	Sensitivity	analyses:	evaluation	of	different	lengths	of	the	latency	period	and	a	cohort	age	restriction,	using	time-dependent	exposure	and	the	recency-weighted	duration	of	exposure	metrics,	with	time	to	lung	cancer	as	the	outcome.				 Multivariable		Regression			 HR	 95%	CI	LL	 95%CI	UL	 p-value		 	 	 	 	Latency	Period	 	 	 	 	None	 	 	 	 	Time-Dependent	Statin	Exposurea	 1.00	 0.87	 1.16	 0.9698	Recency-Weighted	Duration	of	Useb	 1.04	 0.97	 1.12	 0.2754		 	 	 	 	6	months	 	 	 	 	Time-Dependent	Statin	Exposure	 0.96	 0.82	 1.12	 0.5808	Recency-Weighted	Duration	of	Use	 1.00	 0.92	 1.08	 0.9214		 	 	 	 	1	yearc	 	 	 	 	Time-Dependent	Statin	Exposure	 0.85	 0.73	 1.00	 0.05	Recency-Weighted	Duration	of	Use	 0.85	 0.77	 0.93	 0.0006		 	 	 	 	2	years	 	 	 	 	Time-Dependent	ICS	Exposure	 0.62	 0.52	 0.73	 <0.0001	Recency-Weighted	Duration	of	Use	 0.57	 0.49	 0.66	 <0.0001		 	 	 	 	Cohort	(Age	≥	65	years)	 	 	 	 	Time-Dependent	Statin	Exposure	 0.72	 0.60	 0.86	 0.0004	Recency-Weighted	Duration	of	Use	 0.78	 0.70	 0.87	 <0.0001			 		 		 		 		a	This	is	the	reference-case	for	the	analysis.	b	The	recency-weighted	duration	of	use	exposure	metric	is	presented	because	it	was	selected	as	the	best	model	based	on	AIC	values.	c	A	1-year	latency	period	was	assumed	in	the	primary	analysis	and	is	presented	here	for	comparison.		 164	The	results	of	this	analysis	are	strengthened	by	the	existence	of	a	plausible	biological	mechanism	by	which	statin	use	might	reduce	lung	cancer	risk.	Systemic	inflammation,	which	may	result	from	COPD	or,	indeed,	may	be	a	cause	of	COPD	(126),	is	associated	with	increased	lung	cancer	risk.	Evidence	suggests	that	elevated	levels	of	markers	for	systemic	inflammation	are	associated	with	a	one	to	three	times	greater	likelihood	of	lung	cancer,	independent	of	smoking	status	(see	Table	1.2).	Previously	conducted	studies	have	reported	that	statin	use	also	appears	to	be	associated	with	reducing	levels	of	systemic	inflammation.	There	have	been	several	trials	conducted	which	demonstrated	that	statin	use	is	associated	with	reduced	levels	of	systemic	inflammation	(see	Table	1.3).	For	example,	in	the	JUPITER	trial,	statin	use	was	associated	with	a	37.4%	reduction	in	systemic	inflammation	after	48	months	relative	to	placebo	(71).	In	addition	to	their	effect	on	systemic	inflammation,	there	is	recently	published	evidence	from	pilot	studies	that	suggest	statin	use	may	actually	reduce	markers	for	local/pulmonary	inflammation	(139,141)	which	has	also	been	associated	with	increased	lung	cancer	risk.	While	the	evidence	is	far	from	unanimous,	previously	completed	observational	studies	support	the	idea	that	statins	might	reduce	cancer	risk	(205)	but	only	one	study	previously	found	statistically	significant	reductions	in	lung	cancer	risk,	specifically	associated	with	statin	use	(130).	The	results	of	this	study,	however,	are	not	generalizable;	the	analysis	used	data	from	the	United	States	Veterans	Health	Administration	which	was	almost	exclusively	male	(97.9%).		Evidence	also	suggests	statins	have	been	shown	to	improve	survival	for	those	who	continued	statin	therapy	after	cancer	diagnosis	(206).	Therefore,	this	study	provides	an	important	contribution	to	the	evidence	for	statin	use	and	lung	cancer	risk.			 165	In	addition	to	the	results	of	this	study	showing	a	robust	protective	effect	of	statin	use	and	a	reduction	in	lung	cancer	risk,	the	results	also	suggested	that	statin	use	reduced	the	risk	of	NSCLC,	specifically.	The	results	for	SCLC,	however,	were	not	statistically	significant.	The	inclusion	of	lung	cancer	histology	into	this	analysis	is	a	novel	contribution	and	the	results	presented	in	this	chapter	were	also	consistent	with	the	results	presented	in	Chapter	5.	Prior	to	this	analysis,	I	had	no	a	priori	hypothesis	as	to	whether	statin	exposure	would	offer	differential	effects	on	specific	lung	cancer	histology.	However,	the	absence	of	a	differential	effect	is	supported	by	the	work	of	Chaturvedi	et	al.	(40)	which	found	that	elevated	levels	of	systemic	inflammation	were	associated	with	lung	cancer,	generally,	and	there	was	no	statistically	significant	difference	in	the	levels	of	systemic	inflammation	were	elevated	in	patients	that	developed	SCLC	or	NSCLC.	That	is,	systemic	inflammation	appeared	to	be	significantly	associated	with	both	types	of	lung	cancer.	Therefore,	if	the	mechanism	by	which	statins	are	thought	to	reduce	the	risk	of	lung	cancer	in	COPD	patients	is	via	a	reduction	in	systemic	inflammation,	and	elevated	levels	of	systemic	inflammation	are	associated	with	both	SCLC	and	NSCLC,	the	results	of	the	analysis	suggesting	statins	reduce	the	risk	of	NSCLC	are	consistent.	While	the	results	of	the	analysis	with	SCLC	and	statin	use	were	not	statically	significant,	this	is	likely	due	to	the	low	number	of	observed	SCLC	cases.				Previously	published	evidence	suggests	that	statin	use	in	COPD	patients	is	associated	with	a	slowing	in	the	decline	in	lung	function	(207),	a	reduction	in	the	risk	of	acute	exacerbations	of	COPD	(208),	and	a	reduction	in	the	risk	of	all-cause	mortality	(61).	In	this	dissertation,	the	results	from	the	analyses	presented	in	Chapter	4	aligned	with	previously	published	observational	evidence	and	demonstrated	that	statin	use	might	reduce	the	risk		 166	of	all-cause	and	pulmonary-related	mortality.	In	this	chapter,	the	results	presented	show	a	potential	additional	benefit	to	statin	use	in	COPD	patients,	in	terms	of	a	reduction	in	lung	cancer	risk.	Several	recently	published	studies	have	estimated	that	the	burden	associated	with	COPD	will	increase	substantially	in	coming	years,	thereby	necessitating	the	identification	of	therapies	that	can	reduce	this	burden	(4,11).	Therefore,	the	culmination	of	this	evidence	suggests	that	statins	have	pleiotropic	effects	in	COPD	patients	and	should	be	considered	as	a	potential	therapy	beyond	hypercholesterolemia	(209).	This	evidence	might	also	suggest	that	identification	of	patients	with	elevated	levels	of	systemic	inflammation	might	offer	prognostic	information	and	allow	for	targeted	statin	treatment.		6.4.1 Strengths	and	limitations	The	study	has	several	strengths.	First,	it	uses	population-based	administrative	data	for	an	entire	Canadian	province	which	significantly	enhances	the	generalizability	of	its	findings	compared	to	other	existing	studies.	Moreover,	it	was	possible	to	link	this	administrative	data	to	high-quality	registry	data	from	the	British	Columbia	Cancer	Agency	to	accurately	identify	the	diagnosis	date	and	histology	of	lung	cancer.	As	such,	the	data	used	in	this	study	provides	the	highest	level	of	real-world	effectiveness	evidence.	Moreover,	where	previous	studies	may	have	lacked	adequate	power	to	identify	an	association	between	statin	exposure	and	lung	cancer	risk,	this	study	did	not.	Second,	it	is	the	first	study	that	has	used	an	extensive	list	of	medication	exposure	definitions	to	address	the	question	of	whether	statins	might	confer	benefit,	in	terms	of	reduced	lung	cancer	risk,	in	COPD	patients.	Statin	exposure	was	statistically	significant	in	several	adjusted	analyses	and	for	all	exposure	definitions	when	the	cohort	was	restricted	to	patients	aged	65	and	over,	which	enhances		 167	the	robustness	of	these	results.	In	addition,	the	use	of	recency-weighted	approaches,	which	showed	superior	model	fit	to	the	conventional	exposure	definitions,	is	also	a	key	strength,	and	may	provide	a	useful	methodological	approach	in	future	studies	evaluating	cancer	risk	associated	with	medication	use.	This	method	of	defining	medication	exposure	implies	that	the	duration	of	statin	use	is	important,	but	also	that	when	the	use	occurs,	proximal	to	the	outcome,	is	important.	Third,	the	incorporation	of	a	latency	period	associated	with	lung	cancer,	and,	therefore,	not	classifying	medication	exposures	immediately	preceding	lung	cancer	diagnosis	as	relevant	exposures,	is	a	strength	of	this	study	and	should	inform	future	observational	studies	in	cancer	research.	Finally,	the	use	of	a	negative	control	(in	this	case,	CCB	exposure)	to	detect	if	the	results	of	the	analysis	were	due	to	bias	or	confounding	is	a	major	strength	of	this	study,	which	provides	support	for	a	true	association	between	statin	exposure	and	lung	cancer	risk,	and	significantly	enhances	the	robustness	of	the	results	(and	also	the	results	presented	in	Chapter	5	which	used	the	same	analytic	approach).		There	are	several	limitations	to	this	study	which	also	require	acknowledgement.	First,	the	administrative	data	used	in	this	study	did	not	include	any	clinical	variables	that	would	be	useful	in	the	analysis	(for	example,	level	of	systemic	inflammation	or	lung	function).	As	such,	COPD	patients	were	not	identified	according	to	spirometry	but	rather	based	on	their	individual	prescription	records.	However,	previous	studies	have	used	a	similar	method	for	identifying	COPD	patients	(144,160)	and	this	is	believed	to	be	a	sensitive	approach	to	identifying	these	patients.	It	is	possible	that	this	approach	identified	some	patients	as	having	COPD,	when,	in	reality,	they	did	not.	To	attempt	to	provide	a	more	specific	definition	of	COPD,	by	also	adding	in	the	requirement	for	a	physician	encounter	with	an	ICD-9	code		 168	(491,	492,	and	496)	consistent	with	COPD	within	one	year	of	the	index	date.	This	reduced	the	size	of	our	cohort	by	less	than	approximately	10%	and	is	an	approach	is	similar	that	used	by	Curkendall	et	al.	(26)	and	Gershon	et	al.	(210).	Using	a	more	restrictive	approach,	for	example	by	requiring	a	hospitalization	for	COPD,	may	have	only	identified	patients	with	more	severe	disease.	Therefore,	this	approach	should	enhance	the	generalizability	of	the	study	results.	In	addition,	the	focus	of	this	research	was	not	to	study	patients	with	COPD,	but	rather	to	evaluate	lung	cancer	risk	in	patients	at	high	risk	for	developing	the	disease.	As	such,	while	I	attempted	to	identify	a	cohort	of	COPD	patients,	it	is	less	important	that	patients	have	COPD,	and	more	important	that	patients	were	at	an	increased	risk	of	lung	cancer	in	the	cohort.	Second,	smoking	status	and	smoking	history	were	also	not	captured	in	our	administrative	data.	However,	previous	literature	does	suggest	that	the	majority	of	the	cohort	will	have	a	history	of	smoking	(97,98)	so	it	would	be	expected	that	the	majority	of	this	cohort	did,	indeed,	have	a	history	of	smoking.	For	example,	estimates	suggest	that	approximately	85%	of	COPD	patients	have	a	history	smoking	(211–213);	therefore,	we	could	crudely	assume	that	approximately	34,000	patients	identified	in	the	cohort	of	COPD	patients	had	a	history	of	smoking.	It	might	also	be	the	case	that	smokers	may	have	more	severe	COPD,	and	that	this	group	of	patients	would	be	more	likely	to	develop	lung	cancer,	which	would	conservatively	bias	the	results	for	the	effect	of	statins.	Third,	while	a	strong	effect	was	observed	for	statin	use	and	lung	cancer	risk,	this	study	is	subject	to	the	limitations	of	all	observational	studies	whereby	we	cannot	be	certain	that	there	is	an	element	of	unmeasured	or	residual	confounding	that	explains	the	study	results.	In	order	to	mitigate	this	possibility,	a	systematic	approach	to	identifying	potential	confounders	to	be	included	in	the	multivariable	analysis	was	adopted.	The	magnitude	of	the	effect	size	also		 169	reduces	the	likelihood	that	the	protective	effect	of	statin	use	could	be	explained	by	residual	confounding.	Moreover,	the	use	of	a	one	year	latency	period	reduces	the	likelihood	that	the	study	results	could	be	explained	by	protopathic	bias.	Finally,	the	study	results	were	also	consistent	for	a	variety	of	medication	exposure	definitions	and	in	several	sensitivity	analyses.			6.5 Conclusions	This	analysis	presented	in	this	chapter	demonstrated	that	statin	use	in	COPD	patients	may	reduce	the	risk	of	lung	cancer.	While	the	association	between	statin	exposure	and	lung	cancer	risk	was	not	consistently	statistically	significant	across	all	specified	exposure	metrics,	the	direction	of	the	hazard	ratios	was	consistent	with	the	a	priori	study	hypothesis.	Using	the	recency-weighted	approaches	to	capture	statin	exposure	resulted	in	statistically	significant	hazard	ratios,	and	these	two	models	were	deemed	superior	based	on	an	a	priori	specified	criterion.	In	sub-group	analyses,	statin	exposure	was	significantly	associated	with	a	reduction	in	lung	cancer	risk	for	patients	sixty-five	years	of	age	or	greater,	across	all	exposure	metrics,	and	also	for	NSCLC.	These	results	were	further	strengthened	by	an	analysis	incorporating	a	negative	control	exposure	to	detect	residual	confounding	or	bias.	In	combination	with	the	results	of	Chapter	4,	where	statins	were	shown	to	reduce	mortality	in	COPD	patients,	on	balance,	the	results	presented	in	this	analysis	strengthen	the	hypothesis	that	there	may	be	a	segment	of	COPD	patients,	likely	characterized	by	elevated	levels	systemic	inflammation,	that	could	benefit	substantially	from	statin	therapy.	 		 170	Chapter	7: Discussion	and	conclusions		Summary		In	this	chapter	I	summarize	and	reiterate	the	results	from	Chapters	2-6,	and	discuss	both	the	methodological	and	empirical	implications	of	these	findings	in	a	broader	context.	I	also	acknowledge	the	limitations	of	my	research,	identify	avenues	for	future	research,	and	offer	final	conclusions.			7.1 Research	findings	and	implications	The	overall	objective	of	the	studies	that	form	this	dissertation	was	to	evaluate	the	association	between	medication	exposure,	both	to	statins	and	inhaled	corticosteroids,	with	lung	cancer	development	in	a	population-based	cohort	of	chronic	obstructive	pulmonary	disease	patients	(Figure	7.1).	Under	that	over-arching	objective,	there	were	several	specific	ancillary	objectives.		The	first	was	to	present	and	critically	evaluate	different	measures	of	medication	exposure	in	observational	studies	and	to	illustrate	the	potential	biases	that	could	be	created	if	due	consideration	was	not	given	to	this	aspect	of	the	analysis.	In	doing	so,	I	highlighted	methods	of	defining	medication	exposure	that	would	be	useful	in	the	subsequent	analyses	presented	in	this	disseration.	These	measures	could	be	classified	in	three	different	general	categories:	(i)	conventional	methods	of	defining	medication	exposure;	(ii)	recency-	 171	weighted	approaches	to	defining	medication	exposure;	and	(iii)	medication	adherence	approaches	to	defining	exposure.	These	categories	of	defining	medication	exposure	are	used	in	Chapters	5	and	6	of	this	dissertation	and	have	implications	for	the	understanding	of	the	relationship	between	medication	exposure	and	the	study	outcome.		The	second	was	to	identify	and	assess	the	literature	that	attempted	to	evaluate	whether	ICS	use	in	COPD	patients	was	associated	with	lung	cancer	risk.	Several	RCTs	were	identified	that	evaluated	ICS	treatment	in	COPD	patients,	but	none	of	these	addressed	the	study	question	directly.	In	these	studies,	lung	cancer	diagnosis	was	recorded	as	a	secondary	outcome	(i.e.	adverse	events),	which	made	it	difficult	to	assess	whether	or	not	ICS	use	was	associated	with	lung	cancer	risk.	The	lack	of	a	trial	to	answer	this	important	research	question	is	unsurprising;	adequately	powering	a	study	to	detect	lung	cancer	outcomes	would	require	a	large	number	of	participants,	an	extensive	follow-up	period,	and	thus	substantial	costs	and	resources.	As	such,	it	is	likely	that	the	best	approach	to	generate	the	real-world	evidence	required	to	answer	this	research	question	would	be	a	well-designed	observational	study.	The	studies	identified	in	Chapter	3	also	showed	that	existing	evidence	from	observational	studies	implied	that	there	might	be	a	significant	and	protective	effect	between	ICS	exposure	and	lung	cancer	risk	(87,110).	However,	the	findings	of	Chapter	3	also	suggested	that	these	results	should	be	interpreted	cautiously.	Both	identified	observational	studies	dealt	with	specific	patient	populations	and	used	metrics	to	capture	medication	exposure	that	led	to	questions	about	the	reliability	of	the	study	results.	Thus,	the	question	of	whether	there	was	an	association	between	ICS	use	and	lung	cancer	risk,	in	my	opinion,	remained	unanswered.			 172			Figure	7.1.	The	conceptual	framework	for	this	thesis.	COPD	may	be	the	result	of	and/or	a	cause	of	systemic	inflammation	which	has	been	linked	to	poor	health	outcomes.	Statin	and	ICS	use	have	been	shown	to	reduce	levels	of	systemic	inflammation	and	may	therefore	reduce	levels	of	systemic	inflammation	and	may	therefore	reduce	mortality	and	lung	cancer	risk.	 Third,	it	was	important	to	evaluate	whether	statin	use,	generally,	was	of	benefit	to	COPD	patients.	Therefore,	the	analysis	presented	in	Chapter	4	evaluated	whether	there	was	an	association	between	statin	use	and	health	outcomes,	specifically	both	all-cause	and	pulmonary-related	mortality.	The	impetus	for	this	analysis	stemmed	from	the	results	presented	from	the	STATCOPE	trial	(66)	which	questioned	the	clinical	utility	of	statins	in	COPDALL-CAUSE	MORTALITYSYSTEMIC	INFLAMMATIONLUNG	CANCERHEALTH	OUTCOMESPULMONARY-RELATED	MORTALITYINHALED	CORTICOSTEROIDS STATINSREDUCED	INFLAMMATION	 173	treating	COPD	patients.	However,	it	is	my	contention	that	there	were	several	problematic	issues	with	the	STATCOPE	trial.	The	inclusion	and	exclusion	criteria	used	in	the	STATCOPE	trial,	I	believe,	was	overly	restrictive	and	therefore	rendered	the	trial	unable	to	generate	meaningful	answers	to	the	clinical	question	of	whether	statins	should	be	prescribed	to	COPD	patients,	for	benefits	beyond	improving	CVD	outcomes.	In	reality,	the	STATCOPE	trial	likely	identified	a	subset	of	COPD	patients	that	would	not	benefit	from	statins.	That	is,	in	an	attempt	to	identify	a	homogeneous	group	of	COPD	patients	without	comorbidity	to	enroll	in	the	trial,	STATCOPE	likely	excluded	patients	with	elevated	levels	of	systemic	inflammation,	thereby	minimizing	the	chance	of	statins	showing	any	clinical	utility.	The	results	of	the	analysis	using	real-world	data	presented	in	Chapter	4,	however,	did	indeed	show	that	statin	use	confers	benefits	to	COPD	patients	in	terms	of	a	reduction	in	the	risk	of	all-cause	and	pulmonary-related	mortality,	and	bolsters	the	case	for	statin	use	in	these	patients.	I	believe	these	findings	to	be	significant,	as	they	demonstrate	the	importance	of	real-world	evidence	in	comparison	to	results	obtained	from	a	controlled	trial	setting.	Thus,	the	results	of	the	study	conducted	in	Chapter	4	provide	evidence	for	the	beneficial	properties	of	statins	that	is	much	more	relevant	and	informative	than	the	results	of	the	STATCOPE	trial.		Finally,	the	objective	of	the	analyses	presented	in	Chapters	5	and	6	was	to	answer	the	overall	research	questions	of	this	thesis,	and	also	to	contribute	to	the	methodological	literature	as	to	how	such	a	research	question	might	be	addressed.	The	results	presented	in	these	chapters	demonstrated	the	protective	effect	of	statins	and	ICS	in	terms	of	reduced	risk	of	lung	cancer,	and	these	results	were	consistent	for	a	variety	of	measures	of	defining		 174	medication	exposure.	While	the	results	for	statin	use	were	less	conclusive	than	the	results	for	ICS	use,	the	magnitude	of	the	estimated	protective	effect	from	ICS	and	statin	use	on	lung	cancer	diagnosis,	coupled	with	the	systemic	approach	to	model	construction	with	a	host	of	potential	confounders,	also	minimized	the	likelihood	of	these	results	being	attributable	to	residual	confounding.	The	use	of	a	negative	control	exposure	to	attempt	to	detect	bias	or	confounding	was	an	additional	strength	of	the	analysis	presented	in	these	two	chapters.		Moreover,	the	population-based	administrative	data	which	were	linked	to	a	registry	of	lung	cancer	patients	used	in	this	study	was	superior	to	data	used	in	previous	studies	in	this	area.	The	use	of	cancer	registry	data	limits	the	possibility	of	measurement	error	of	the	outcome,	and	also	minimizes	the	possibility	of	misclassification	of	lung	cancer	cases.	In	addition,	the	use	of	population-based	administrative	data	increases	the	generalizability	of	my	results,	minimizes	the	likelihood	of	selection	bias	which	is	a	common	concern	with	observational	data,	and	provided	both	the	power	and	the	length	of	follow-up	time,	required	to	conduct	such	a	study.		The	results	presented	in	these	chapters	have	important	implications	for	the	treatment	of	COPD	patients.	Lung	cancer	is	often	diagnosed	at	a	late	stage	and	as	such,	the	diagnosis	is	accompanied	with	a	poor	prognosis	of	survival	(176).	For	example,	the	1-year	survival	associated	with	a	diagnosis	of	Stage	IV	lung	cancer	is	39.4%	(95%	CI:	36.2-42.2%)	(214).	The	poor	survival,	low	quality	of	life,	and	substantial	costs	of	therapy	associated	with	a	lung	cancer	diagnosis	should	place	a	considerable	emphasis	on	the	need	for	interventions	to	reduce	the	risk	of	lung	cancer,	particularly	in	patients	with	COPD	who	are	already	at	a	higher	risk	of	the	disease.	The	results	of	these	studies	suggest	that	statins	and	ICS	may		 175	reduce	the	risk	of	lung	cancer	in	COPD	patients.	Both	statins	and	ICS	have	a	relatively	low	cost,	a	low	user	burden,	and	appear	to	provide	a	reduction	in	the	risk	of	lung	cancer.	For	COPD	patients	that	already	use	these	medications,	the	findings	reported	in	Chapter	5	and	6,	that	the	duration	of	use	for	both	ICS	and	statins	is	associated	with	a	reduced	risk	of	lung	cancer,	should	highlight	the	need	for	COPD	patients	to	use	their	medications	as	prescribed.	Moreover,	inhaled	medications	are	often	used	inappropriately	by	users	(215).	Therefore,	this	could	also	provide	motivation	for	patients	to	receive	proper	training	from	clinicians	to	maximize	the	effectiveness	of	the	medication.		The	economic	burden	of	COPD	is	also	substantial	and	this	is	likely	to	increase	in	coming	years.	Najafzadeh	et	al.	(4)	reported	that	the	incidence	of	COPD	is	due	to	increase	until	2035,	and	that	while	smoking	cessation	might	reduce	the	burden	of	disease,	interventions	to	reduce	acute	exacerbations	of	COPD	(AECOPD)	would	have	the	most	significant	impact	on	reducing	this	burden.	Similarly,	Khakban	et	al.	(11)	reported	on	the	increasing	burden	of	COPD-related	hospitalizations,	which	consumes	considerable	health	care	costs	and	resources,	thereby	posing	a	significant	problem	to	health	service	provision	and	increased	pressures	on	a	limited	healthcare	budget.	With	respect	to	lung	cancer,	a	recent	study	conducted	by	Fitzmaurice	et	al.	(216)	reported	that	lung	cancer	was	the	leading	cause	of	cancer-related	mortality	in	men,	and	the	third	leading	cause	in	women,	which	resulted	in	an	estimated	1.2	million	deaths	worldwide.	These	studies	highlight	the	need	for	evidence	to	inform	interventions	to	reduce	the	substantial	burden	caused	by	both	COPD	and	lung	cancer.			 176	Table	7.1.	Key	findings	for	each	specific	chapter	of	this	dissertation.	Chapter	 Key	Finding		 	Chapter	2	 Methods	to	measure	medication	exposure	require	careful	consideration	in	observational	studies	in	the	context	of	chronic	diseases.	Methods	that	incorporate	both	the	timing	and	duration	of	prescriptions	have	theoretical	and	practical	advantages.		Chapter	3	 Observational	evidence	supports	the	hypothesis		that	lung	cancer	risk	might	be	decreased	by	inhaled	corticosteroid	use	but	there	are	concerns	about	the	generalizability	of	the	results.	Chapter	4	 The	use	of	statins	in	COPD	patients	is	contentious,	but	this	study	shows	that	statin	use	might	reduce	all-cause	and	pulmonary-related	mortality.		Chapter	5	 Inhaled	corticosteroid	use	appears	to	be	associated	with	reduced	risk	of	lung	cancer,	using	a	variety	of	medication	exposure	definitions.	This	highlights	the	role	ICS	therapy	in	COPD	patients.			Chapter	6	 Statin	use	appears	to	reduce	lung	cancer	risk	in	COPD	patients,	however	the	results	are	less	conclusive	than	for	ICS.	In	conjunction	with	the	results	of	Chapter	4,	it	appears	there	is	a	segment	of	the	COPD	population	that	would	benefit	substantially	from	statin	use.	This	chapter	also	explored	if	there	was	a	potential	synergistic	effect	of	statins	and	ICS	use,	and	found	no	synergistic	effect.		 			 177	7.2 Research	contributions	This	dissertation	makes	several	valuable	contributions	to	the	literature.	These	contributions	are	both	empirical,	which	will	inform	the	evidence	base	of	the	therapeutic	benefits	that	can	be	obtained	in	COPD	patients	from	statin	and	ICS	use,	and	also	methodological,	in	that	they	will	inform	future	pharmacoepidemiologic	analyses	using	administrative	data	and	non-acute	outcomes.		The	empirical	contributions	of	this	dissertation	should	improve	understanding	and	management	of	patients	with	COPD.	The	results	presented	in	Chapter	4	showed	that	statins	may	confer	additional	beneficial	effects	in	COPD	patients,	and	reduce	the	risk	of	mortality,	compared	to	non-users.	As	noted	above,	this	is	not	to	argue	that	statins	should	be	prescribed	to	all	COPD	patients,	but	rather	that	there	may	exist	a	sub-group	of	COPD	patients	with	significant	levels	of	systemic	inflammation,	that	might	benefit	in	terms	of	reduced	mortality	from	statin	use.	Similarly,	Chapter	6	further	demonstrated	the	potential	pleiotropic	effects	of	statins	and	showed	a	significant	reduction	in	lung	cancer	risk	associated	with	the	use	of	a	statin,	particularly	for	patients	over	65	and	those	at	risk	of	developing	NSCLC.	Moreover,	the	results	were	consistent	for	a	number	of	exposure	metrics,	and	in	several	sensitivity	analyses	which	reduces	the	likelihood	of	the	results	being	explained	by	residual	confounding.	Therefore,	statin	use	appears	to	reduce	all-cause	and	pulmonary-related	mortality,	in	addition	to	potentially	reducing	the	risk	of	lung	cancer.	Finally,	the	appropriate	stage	at	which	ICS	should	be	prescribed	in	COPD	patients	is	contentious,	but	this	study	builds	on	previous	evidence	reported	in	observational	studies		 178	that	showed	ICS	use	reduces	the	risk	of	lung	cancer.	This	evidence	would	suggest	that	ICS	could	potentially	be	initiated	earlier	in	COPD	patients;	however,	it	must	also	be	balanced	against	and	individual	patients’	risk	of	adverse	effects	(10,49).		One	further	observation	from	the	research	presented	in	this	dissertation	is	the	deficiency	of	randomized	controlled	trials	to	answer	‘real-world’	questions.	Suissa	(180)	stated:		‘The	randomized	controlled	trial	design	is	essential	to	evaluate	the	effectiveness	and	safety	of	medications	and	to	obtain	regulatory	approval	for	their	use	in	clinical	practice.	Yet,	it	rarely	provides	information	on	their	pragmatic	benefit	in	terms	of	major	disease	outcomes.’	(180).	Randomized	controlled	trials	are	the	current	‘gold	standard’	as	the	highest	level	of	evidence	but	the	design	does	not	always	capture	the	data	that	is	required	for	real-world	decisions.	Moreover,	RCTs	are	not	always	feasible	if	the	outcome	in	question	is	rare.	For	example,	no	RCT	has	been	conducted	to	address	whether	ICS	use	reduces	the	risk	of	lung	cancer	in	COPD	patients,	despite	COPD	patients	being	at	an	increased	risk	of	lung	cancer,	ICS	use	being	contentious,	and	observational	evidence	to	suggest	such	an	association.	It	is	well-established	that	RCTs	are	expensive,	and	may	result	in	additional	costs	and	resource	use	to	health	service	(217).	It	is	also	becoming	increasingly	well-understood	that	RCTs	might	not	be	able	to	generate	the	evidence	that	is	required	to	improve	real-world	treatment	decisions	(206,207).	In	instances	where	outcomes	are	rare	and	long-term	follow-up	is	required	(such	as	lung	cancer	diagnoses)	or	where	there	are	significant		 179	comorbidities	resulting	in	considerable	patient	heterogeneity	(such	as	COPD),	it	can	be	difficult	to	conduct	a	traditional	RCT.	Under	these	conditions,	an	RCT	would	need	enroll	a	substantial	number	of	patients	and	have	an	extensive	follow-up	period.	At	the	other	end	of	the	spectrum,	restrictive	inclusion	or	exclusion	criteria	in	a	trial	can	easily	remove	all	external	validity	from	the	trial	results	(220),	often	at	a	significant	cost	(both	financial	and	in	terms	of	resources).	The	STATCOPE	trial	provides	an	excellent	example27	of	trial	evidence	that	may	not	offer	the	type	of	real-world	evidence	required	to	inform	clinical	decisions.	While	the	results	might	be	internally	valid,	the	inclusion/exclusion	criteria	resulted	in	non-generalizable	conclusions	about	all	COPD	patients.	In	the	trial,	statin	therapy	was	evaluated	in	COPD	patients	to	validate	previous	observational	evidence	that	suggested	statins	reduced	acute	exacerbations	(61,122).	Despite	this	observational	evidence,	the	trial	found	no	benefit	from	statin	therapy	in	COPD	patients,	but	these	results	were	met	with	objections	of	how	the	study	was	designed,	specifically	in	terms	of	the	inclusion	and	exclusion	criteria	(62,63,139).	As	stated	in	Section	7.1	above,	by	excluding	patients	with	comorbidities,	it	is	likely	that	the	STATCOPE	trial	excluded	those	patients	with	elevated	levels	of	systemic	inflammation,	and	thus,	we	would	not	expect	that	the	patients	included	in	the	STATCOPE	trial	would	benefit,	in	terms	of	COPD-related	outcomes,	from	statin	therapy.	In	reality,	the	usefulness	of	the	STATCOPE	trial	may	be	that	it	identified	those	patients	that	may	not	benefit	from	statin	therapy,	as	opposed	to	                                                27	There	are	many	examples	of	RCTs	that	have	overly	restrictive	inclusion	or	exclusion	criteria	such	that	patients	included	in	the	trial	are	not	indicative	of	the	overall	patient	population,	thus	rendering	results	not	generalizable.	For	example,	a	study	conducted	by	Travers	et	al.	(215)	of	asthma	patients	suggested	that	the	vast	majority	(>90%)	of	asthma	patients	would	not	be	have	met	the	criteria	to	be	selected	for	recently	completed	trials.		 180	determining	the	effectiveness	of	statin	therapy	in	a	specific	subset	of	COPD	patients	who	may	be	at-risk	for	poor	health	outcomes	due	to	increased	levels	of	systemic	inflammation.	The	scarcity	of	available	funding	for	health	research	means	that	decisions	to	conduct	such	trials	should	be	scrutinized,	particularly	when	the	analysis	of	observational	data	(population-based	data)	might	be	able	to	produce	similar	results	with	a	much	lower	cost	and	resource	burden.	As	such,	further	research	to	develop	approaches	to	clinical	trials	that	would	generate	tangible	real-world	evidence	(i.e.	pragmatic	trials)	that	can	inform	treatment	decisions	would	be	a	valuable	endeavour.	There	is	also	a	need	to	ensure	that	observational	studies	are	well-designed	to	mitigate	the	potential	impact	of	bias	and	confounding.	Certainly,	there	has	been	a	move	toward	using	this	type	of	evidence	more	prominently	in	the	context	of	regulatory	decision-making	(221).	For	example,	the	United	States	Food	and	Drug	Administration	has	a	produced	a	document,	in	the	context	of	medical	devices	where	conventional	randomized	controlled	trials	are	difficult	to	conduct,	to	describe	situations	where	real-world	evidence	may	be	used	for	regulatory	approval	(221).	Similarly,	observational	research	using	administrative	data	could	benefit	from	improved	data	collection	and	the	ability	to	link	this	data	to	clinical	data	that	could	mitigate	potential	biases	and	generate	better	evidence	to	inform	decisions.	The	methodological	contributions	of	this	dissertation	are	the	following:	(i)	the	use	of	an	extensive	catalogue	of	exposure	metrics;	(ii)	the	use	of	a	recency-weighted	approach	that	has	not	previously	been	used	in	the	analysis	of	cancer	diagnosis;	(iii)	the	incorporation	of	a		 181	latency	period	in	the	primary	analysis	coupled	with	the	recency-weighted	exposure	metrics,	and	(iv)	the	use	of	a	negative	control	exposure.	The	use	of	a	variety	of	exposure	metrics	allowed	for	the	possibility	of	evaluating	whether	or	not	results	differed	by	the	way	medication	exposure	was	quantified	and	subsequently	incorporated	into	regression	analyses.	The	use	of	several	measures	also	allowed	for	evaluating	whether	results	were	robust	across	each	of	these	exposure	metrics.	Limiting	the	analysis	to	just	one	of	these	metrics	of	defining	exposure	may	have	led	to	false	conclusions.	Therefore,	I	believe	it	may	be	important	for	studies	to	incorporate	several	different	exposure	metrics	based	on	the	existing	clinical	literature	and	biological	plausibility	to	ensure	the	robustness	of	their	results28.	These	metrics	can	then	be	evaluated	according	to	an	a	priori	defined	criterion	to	determine	which	best	suits	the	study	data.		Second,	the	use	of	the	recency-weighted	approach	in	determining	the	association	between	medication	exposure	and	lung	cancer	risk	has	not	been	used	previously,	to	my	knowledge.	This	method	is	very	intuitive,	particularly	for	studies	with	long	follow-up	time,	and	could	likely	be	used	for	any	non-acute	event.	The	theory	behind	using	this	method	is	that	medication	exposures	distal	to	the	outcome	should	be	considered	differently	than	exposures	more	proximal	to	the	outcome	(96).	I	estimated	models	using	this	approach	in	two	ways:	recency-weighted	duration	of	use,	and	recency-weighted	cumulative	dose.	The	recency-weighted	approach	resulted	in	superior	model	fit	in	both	Chapters	5	and	6.		The	                                                28	This	includes	the	use	of	time-dependent	covariates	where	applicable	to	reduce	the	possibility	of	bias	(i.e.	immortal	time	bias).	In	Section	5.4	I	exhibited	the	potential	impact	that	this	bias	could	have	on	the	study	results.			 182	choice	of	exposure	metric(s)	should	reflect	the	nature	of	the	relationship	between	the	exposure	and	the	outcome,	where	possible	incorporating	previous	knowledge	of	this	relationship	from	published	literature,	clinical	expertise,	and	biological	plausibility.	In	the	absence	of	a	solid	understanding	of	the	relationship	between	the	exposure	and	outcome,	the	best	approach	is	to	use	a	variety	of	exposure	metrics	to	explore	which	might	best	fit	the	data.		The	application	of	a	latency	period	in	the	analysis	with	lung	cancer	as	the	study	outcome	is	a	significant	contribution,	as	it	was	not	incorporated	in	the	primary	analysis	of	either	observational	study	evaluating	ICS	use	and	lung	cancer	risk.	By	not	incorporating	a	latency	period,	ICS	exposure	which	occurred	close	to	the	outcome	or	end	of	follow-up,	would	be	misclassified	as	exposed	time	when	it	is	likely	the	pathogenesis	of	lung	cancer	had	already	begun.	Misclassification	of	exposure	during	this	time	might	result	in	a	bias	toward	a	null	effect	or	protopathic	bias,	whereby	those	at	early	stages	of	lung	cancer	development	are	prescribed	more	medication	(an	ICS	or	a	statin,	or	both).	As	stated,	previous	studies	did	not	include	such	a	period,	which	is	a	significant	limitation	which	was	improved	upon	by	the	work	contained	in	this	thesis.	One	caveat	to	inclusion	of	the	latency	period	is	that	the	exact	onset	of	lung	cancer	prior	to	actual	diagnosis	is	not	known,	thus	making	the	choice	of	appropriate	latency	period	difficult.	In	my	analysis,	the	length	of	the	latency	period	was	assumed,	based	on	work	by	Henschke	et	al.	(171)	and	Chaturvedi	et	al.	(40),	and	was	subjected	to	several	sensitivity	analyses.	The	use	of	the	latency	period	had	significant	effects	on	the	association	between	both	ICS	and	statin	therapy	and	lung	cancer	risk	in	COPD	patients.	In	Chapters	5	and	6,	several	sensitivity	analyses	were	conducted	to	explore	the		 183	impact	of	the	assumed	latency	period	on	the	association	between	lung	cancer	risk	and	ICS	or	statin	exposure.	Removal	of	the	latency	period	resulted	in	statin	and	ICS	exposure	increasing	the	risk	of	lung	cancer.	However,	given	that	it	is	well-understood	that	carcinogenesis	will	have	begun	prior	to	lung	cancer	diagnosis,	this	result	may	be	an	example	of	protopathic	bias.	As	the	latency	period	was	increased,	from	six	months	to	two	years,	the	protective	effect	of	statins	and	ICS	increased	and	became	statistically	significant	(see	Tables	5.7	and	6.7).		Finally,	the	use	of	a	negative	control	exposure	provides	another	example	of	an	epidemiologic	analysis	to	evaluate	the	effect	of	unmeasured	or	unknown	confounding	or	bias.	Lipsitch	et	al.	(200)	argued	that	this	approach	should	be	widely	adopted	in	epidemiology	to	minimize	the	chance	of	spurious	results.	While	not	prolifically	used	in	epidemiologic	analyses,	the	method	of	using	negative	control	exposures	is	reasonably	intuitive,	and	is	akin	to	using	a	placebo	treatment	in	a	randomized	trial	(187,	188).	The	use	of	a	negative	control	exposure	to	detect	whether	or	not	the	results	found	in	Chapters	5	and	6	were	due	an	inherent	bias,	increases	one’s	confidence	that	the	observed	reduction	in	lung	cancer	risk	is	can	be	correctly	attributed	to	the	medication	exposure.				7.3 Limitations	of	this	research	The	main	limitation	of	this	research	is	the	absence	of	clinical	variables	in	the	cohort	of	COPD	patients.	Patients	with	COPD	were	identified	according	to	their	individual		 184	prescription	records.	Ideally,	data	would	be	available	that	would	include	patients’	level	of	airflow	obstruction	and	levels	of	inflammation	(systemic	and	local).	Patients	with	COPD	are	a	heterogeneous	group	of	patients;	therefore,	in	the	absence	of	clinical	data	that	might	be	included	to	adjust	for	confounding,	I	have	used	the	health	services	usage,	obtained	from	the	linked	administrative	data	(other	prescriptions	filled,	physician	encounters,	and	hospitalizations),	to	attempt	to	adjust	for	the	heterogeneity	(and	the	presence	of	comorbidities)	in	these	patients.			While	it	would	be	ideal	to	have	physician-diagnosed	COPD	via	spirometry,	the	cohort	definition	employed	in	this	dissertation	used	has	been	used	previously	(144,160).	Moreover,	the	primary	concern	of	this	study	was	not	to	identify	a	cohort	of	COPD	patients,	but	rather	to	identify	a	cohort	of	patients	who	were	at	an	increased	risk	of	lung	cancer,	and	determine	the	association	with	medication	exposure	and	lung	cancer	risk.	Thus,	I	acknowledge	that	I	cannot	be	certain	that	patients	who	satisfied	the	cohort	definition	did	indeed	have	COPD;	however,	it	is	likely,	due	to	the	receipt	of	the	prescriptions	used	to	identify	the	patients	to	be	included	in	the	cohort,	that	each	patient	does	have	some	degree	of	obstructive	respiratory	disease.	This	inclusion	of	patients	without	COPD	would	also	likely	result	in	a	conservative	bias.	That	is,	if	COPD	patients	are	at	an	increased	risk	of	lung	cancer,	and	might	benefit	from	statins	or	an	ICS,	inclusion	of	patients	at	lower	risk	for	lung	cancer	and	also	less	severe	COPD,	that	did	not	use	these	medications	and	did	not	develop	lung	cancer,	would	likely	bias	results	toward	the	null,	showing	no	benefit	from	exposure	to	an	ICS	or	statins.		 185	In	addition	to	the	absence	of	clinical	data	for	COPD	diagnosis,	smoking	status	and	smoking	history	were	not	available	within	the	linked	administrative	data	used	in	this	thesis.	I	acknowledge	this	as	a	limitation	given	the	role	that	smoking	has	in	both	COPD	and	lung	cancer.	The	literature	in	this	area	suggests	that	approximately	85%	of	COPD	patients	have	a	history	of	smoking	(103).	Moreover,	the	data	would	likely	only	be	improved	if	they	included	very	high	quality	and	timely	data	with	regard	to	smoking	behaviour	to	truly	capture	the	heterogeneity	of	smokers,	which	is	seldom	available.	While	this	is	a	definite	limitation,	levels	of	systemic	inflammation	have	been	reported	to	be	higher	in	COPD	patients	that	continue	to	smoke,	compared	both	to	former	smokers,	and	never	smokers	(222).	Therefore,	it	might	be	that	the	identification	of	these	COPD	patients	might	represent	a	group	that	had	greatest	risk	of	lung	cancer	and	could	the	most	benefit	from	statin	or	ICS	therapy.		Therefore,	the	inclusion	of	never-smokers	and	former	smokers,	both	who	would	likely	have	lower	levels	of	systemic	inflammation	and	lower	lung	cancer	risk,	might	result	in	a	conservative	bias.			7.4 Future	direction	of	research	This	dissertation	has	shown	that	ICS	and	statins	are	both	associated	with	a	reduction	in	lung	cancer	risk	in	COPD	patients.	To	do	this,	a	population-based,	longitudinal,	retrospective	design	was	used.	These	results	were	robust	across	a	variety	of	metrics	to	define	exposure	status	for	both	medications,	and	also	stable	over	several	sensitivity	analyses	performed.	There	are	several	avenues	that	I	have	identified	for	future	research.		 186	The	most	obvious	would	be	to	conduct	a	prospective	study	that	evaluates	statin	and	ICS	use	and	lung	cancer	risk.	However,	as	noted	above,	lung	cancer	is	a	rare	outcome,	even	among	COPD	patients.	As	such,	a	prospective	study	would	need	a	very	large	number	of	participants,	and	these	participants	would	need	to	be	followed	over	an	extended	period	of	time,	which	is	unlikely	to	be	feasible.	Therefore,	I	would	argue	that	the	work	contained	in	this	dissertation	should	provide	an	impetus	for	more	well-designed	observational	studies	to	generate	evidence	where	RCTs	are	problematic	or	simply	not	feasible.		The	weighting	function	that	was	chosen	based	on	previous	literature	and	on	the	assumption	that	prescriptions	received	by	an	individual	patient	nearer	the	study	outcome	should	be	weighted	more	heavily	than	those	received	earlier	in	follow-up	time	(96).	However,	the	effect	of	these	medications	in	the	early	phases	of	lung	tumor	carcinogenesis	is	not	well-understood	and	future	research	to	better	understand	medication	exposure	during	the	induction	period	of	lung	cancer	could	be	extremely	valuable.	Similarly,	more	research	to	provide	a	better	clinical	and/or	biological	understanding	of	the	duration	of	the	induction	and	latency	periods	would	be	a	significant	contribution	of	future	research	in	lung	cancer.		The	rationale	behind	much	of	the	work	presented	in	this	dissertation	is	that	local	and	systemic	inflammation	is	associated	with	poor	health	outcomes	–	specifically	mortality	and	lung	cancer	diagnosis.	Therefore,	a	future	study	that	could	evaluate	the	use	of	ICS	and	statins	while	adjusting	for	measured	levels	of	inflammation	would	be	of	interest.	Moreover,	a	better	understanding	of	the	link	between	local	and	systemic	inflammation	might	improve		 187	the	ability	to	stratify	patients	according	to	which	patients	would	benefit	the	most	from	statin	and	ICS	therapy.				7.5 Knowledge	translation	This	dissertation	adds	to	the	knowledge	base	about	methods	of	quantifying	exposure	in	in	pharmacoepidemiolgic	studies,	specifically	using	administrative	data	in	the	context	of	chronic	diseases.	The	results	of	this	work	also	support	the	hypothesis	that	systemic	inflammation	is	the	responsible	causal	mechanism	between	COPD	and	lung	cancer	incidence,	and	also	the	role	of	ICS	and	statins	to	reduce	systemic	inflammation.	These	findings	of	this	dissertation	should	have	implications	for	the	management	of	COPD	patients,	both	in	terms	of	better	understanding	different	phenotypes	of	the	disease	and,	potentially,	treatment	for	patients.	This	could	further	have	implications	for	improving	the	methods	of	understanding	different	manifestations	or	phenotypes	COPD,	specifically	in	terms	of	a	phenotype	characterised	by	the	presence	of	elevated	levels	of	systemic	inflammation,	and	bolster	the	case	for	ICS	and	statins	to	be	prescribed	for	patients	that	match	the	phenotype	of	those	who	might	derive	the	most	clinical	benefit.		The	results	of	individual	chapters	of	this	dissertation	will	be	submitted	to	peer	reviewed	journals	to	be	published	in	order	to	studied	by	researchers	and	clinicians.	Moreover,	results	have	been	and	will	continue	to	be	presented	at	research	conferences	worldwide	in	hopes	of	disseminating	the	knowledge	gained	through	this	research	as	widely	as	possible.				 188	7.6 Conclusions		The	analyses	contained	within	this	dissertation	represent	important	contributions	to	the	field	of	pharmacoepidemiology	and	respiratory	disease.	The	use	of	an	extensive	array	of	exposure	definitions,	including	a	recency-weighted	approach,	the	incorporation	of	a	latency	period,	and	the	use	of	a	negative	control	exposure,	are	important	contributions	to	the	methodological	literature.	The	results	of	this	dissertation	also	have	important	implications	for	the	management	of	patients	with	COPD,	a	disease	that	is	associated	with	considerable	morbidity	and	mortality.	Moreover,	the	findings	support	the	hypothesis	that	systemic	inflammation	may	be	the	cause	of	poor	health	outcomes	in	COPD	patients,	and	that	a	phenotype	of	COPD,	characterized	by	systemic	inflammation,	may	exist.	Future	research	to	better	understand	this	link	and	to	possibly	stratify	patients	according	to	levels	of	inflammation	may	offer	specific	treatment	options	for	these	patients.	Certainly,	the	results	will	also	raise	the	question	of	whether	or	not	a	trial	is	warranted	to	evaluate	the	effect	of	inhaled	corticosteroids	and	statins	on	lung	cancer	risk.	Although,	as	suggested	above,	such	a	trial	would	be	resource-intensive,	and	would	need	to	be	designed	very	carefully	in	order	for	the	results	to	generate	answers	to	the	relevant	clinical	questions.	As	such,	the	body	of	work	contained	in	this	dissertation	highlights	the	importance	of	well-designed	observational	studies	to	provide	real-world	evidence	to	inform	treatment	decisions.	 			 189	Bibliography	1.  Gershon AS, Wang C, Wilton AS, Raut R, To T. Trends in chronic obstructive pulmonary disease prevalence, incidence, and mortality in ontario, canada, 1996 to 2007: A population-based study. Arch Intern Med. 2010 Mar 22;170(6):560–5.  2.  Buist AS, McBurnie MA, Vollmer WM, Gillespie S, Burney P, Mannino DM, et al. International variation in the prevalence of COPD (the BOLD Study): a population-based prevalence study. The Lancet. 2007;370(9589):741–750.  3.  Gershon A, Hwee J, Victor JC, Wilton A, Wu R, Day A, et al. Mortality trends in women and men with COPD in Ontario, Canada, 1996-2012. Thorax. 2015 Feb;70(2):121–6.  4.  Najafzadeh M, Marra CA, Lynd LD, Sadatsafavi M, FitzGerald JM, McManus B, et al. Future Impact of Various Interventions on the Burden of COPD in Canada: A Dynamic Population Model. de Torres JP, editor. PLoS ONE. 2012 Oct 11;7(10):e46746.  5.  Global Strategy for Diagnosis, Management, and Prevention of COPD - 2016 [Internet]. Global Initiative for Chronic Obstructive Lung Disease - GOLD. [cited 2016 Dec 4]. Available from: http://goldcopd.org/global-strategy-diagnosis-management-prevention-copd-2016/ 6.  Vestbo J, Hurd SS, Agustí AG, Jones PW, Vogelmeier C, Anzueto A, et al. Global Strategy for the Diagnosis, Management, and Prevention of Chronic Obstructive Pulmonary Disease: GOLD Executive Summary. Am J Respir Crit Care Med. 2013 Feb 15;187(4):347–65.  7.  Han MK, Agusti A, Calverley PM, Celli BR, Criner G, Curtis JL, et al. Chronic Obstructive Pulmonary Disease Phenotypes: The Future of COPD. Am J Respir Crit Care Med. 2010 Sep;182(5):598–604.  8.  Viegi G, Pistelli F, Sherrill DL, Maio S, Baldacci S, Carrozzi L. Definition, epidemiology and natural history of COPD. Eur Respir J. 2007 Nov 1;30(5):993–1013.  9.  Mannino DM. The natural history of chronic obstructive pulmonary disease. Eur Respir J. 2006 Mar 1;27(3):627–43.  10.  Suissa S, Patenaude V, Lapi F, Ernst P. Inhaled corticosteroids in COPD and the risk of serious pneumonia. Thorax. 2013 Nov 1;68(11):1029–36.  11.  Khakban A, Sin DD, FitzGerald JM, McManus B, Ng R, Hollander Z, et al. The Projected Epidemic of COPD Hospitalizations Over the Next 15 Years: A Population Based Perspective. Am J Respir Crit Care Med [Internet]. 2016 Sep 14 [cited 2016 Dec 11]; Available from: http://www.atsjournals.org/doi/abs/10.1164/rccm.201606-1162PP 12.  MacNee W. Systemic inflammatory biomarkers and co-morbidities of chronic obstructive pulmonary disease. Ann Med. 2013 May;45(3):291–300.  	 190	13.  Barnes PJ, Celli BR. Systemic manifestations and comorbidities of COPD. Eur Respir J. 2009 May 1;33(5):1165–85.  14.  Sin DD. Why Are Patients With Chronic Obstructive Pulmonary Disease at Increased Risk of Cardiovascular Diseases?: The Potential Role of Systemic Inflammation in Chronic Obstructive Pulmonary Disease. Circulation. 2003 Mar 25;107(11):1514–9.  15.  van Eeden SF, Yeung A, Quinlam K, Hogg JC. Systemic Response to Ambient Particulate Matter. Proc Am Thorac Soc. 2005 Apr 1;2(1):61–7.  16.  Agusti A. Systemic Effects of Chronic Obstructive Pulmonary Disease: What We Know and What We Don’t Know (but Should). Proc Am Thorac Soc. 2007 Oct 1;4(7):522–5.  17.  Fabbri LM, Rabe KF. From COPD to chronic systemic inflammatory syndrome? The Lancet. 2007 Sep 7;370(9589):797–9.  18.  Agusti A, Faner R. Systemic inflammation and comorbidities in chronic obstructive pulmonary disease. Proc Am Thorac Soc. 2012 May;9(2):43–6.  19.  Wouters EFM, Groenewegen KH, Dentener MA, Vernooy JHJ. Systemic Inflammation in Chronic Obstructive Pulmonary Disease: The Role of Exacerbations. Proc Am Thorac Soc. 2007 Dec 1;4(8):626–34.  20.  Ballaz S, Mulshine JL. The Potential Contributions of Chronic Inflammation to Lung Carcinogenesis. Clin Lung Cancer. 2003 Jul 1;5(1):46–62.  21.  Brenner DR, Boffetta P, Duell EJ, Bickeboller H, Rosenberger A, McCormack V, et al. Previous lung diseases and lung cancer risk: a pooled analysis from the International Lung Cancer Consortium. Am J Epidemiol. 2012 Oct 1;176(7):573–85.  22.  Sin DD, Man SF, McWilliams A, Lam S. Surfactant protein D and bronchial dysplasia in smokers at high risk of lung cancer. Chest. 2008 Sep;134(3):582–8.  23.  Barnes PJ, Adcock IM. Chronic obstructive pulmonary disease and lung cancer: a lethal association. Am J Respir Crit Care Med. 2011 Oct 15;184(8):866–7.  24.  Mapel DW, Hurley JS, Frost FJ, Petersen HV, Picchi MA, Coultas DB. Health Care Utilization in Chronic Obstructive Pulmonary Disease: A Case-Control Study in a Health Maintenance Organization. Arch Intern Med. 2000 Sep 25;160(17):2653–8.  25.  Sin DD, Anthonisen NR, Soriano JB, Agusti AG. Mortality in COPD: Role of comorbidities. Eur Respir J. 2006 Dec;28(6):1245–57.  26.  Curkendall SM, deLuise C, Jones JK, Lanes S, Stang MR, Goehring E, et al. Cardiovascular Disease in Patients with Chronic Obstructive Pulmonary Disease, Saskatchewan Canada. Ann Epidemiol. 2006 Jan;16(1):63–70.  	 191	27.  Mannino DM, Thorn D, Swensen A, Holguin F. Prevalence and outcomes of diabetes, hypertension and cardiovascular disease in COPD. Eur Respir J. 2008 May 14;32(4):962–9.  28.  Chen W, Thomas J, Sadatsafavi M, FitzGerald JM. Risk of cardiovascular comorbidity in patients with chronic obstructive pulmonary disease: a systematic review and meta-analysis. Lancet Respir Med. 2015 Aug;3(8):631–9.  29.  Van Eeden S, Leipsic J, Paul Man SF, Sin DD. The relationship between lung inflammation and cardiovascular disease. Am J Respir Crit Care Med. 2012 Jul 1;186(1):11–6.  30.  Kopjar B, Sales AE, Piñeros SL, Sun H, Li Y-F, Hedeen AN. Adherence with statin therapy in secondary prevention of coronary heart disease in veterans administration male population. Am J Cardiol. 2003;92(9):1106–1108.  31.  Siegel R, Ma J, Zou Z, Jemal A. Cancer statistics, 2014: Cancer Statistics, 2014. CA Cancer J Clin. 2014 Jan;64(1):9–29.  32.  Young RP, Hopkins RJ, Christmas T, Black PN, Metcalf P, Gamble GD. COPD prevalence is increased in lung cancer, independent of age, sex and smoking history. Eur Respir J. 2009 Feb 5;34(2):380–6.  33.  Mannino DM, Aguayo SM, Petty TL, Redd SC. Low lung function and incident lung cancer in the United States: data From the First National Health and Nutrition Examination Survey follow-up. Arch Intern Med. 2003 Jun 23;163(12):1475–80.  34.  Skillrud DM. COPD: causes, treatment, and risk for lung cancer. Compr Ther. 1986 Nov;12(11):13–6.  35.  Turner MC, Chen Y, Krewski D, Calle EE, Thun MJ. Chronic obstructive pulmonary disease is associated with lung cancer mortality in a prospective study of never smokers. Am J Respir Crit Care Med. 2007 Aug 1;176(3):285–90.  36.  Wasswa-Kintu S. Relationship between reduced forced expiratory volume in one second and the risk of lung cancer: a systematic review and meta-analysis. Thorax. 2005 Jul 1;60(7):570–5.  37.  Kuller LH, Ockene J, Meilahn E, Svendsen KH. Relation of forced expiratory volume in one second (FEV1) to lung cancer mortality in the Multiple Risk Factor Intervention Trial (MRFIT). Am J Epidemiol. 1990 Aug;132(2):265–74.  38.  Purdue MP, Gold L, Järvholm B, Alavanja MCR, Ward MH, Vermeulen R. Impaired lung function and lung cancer incidence in a cohort of Swedish construction workers. Thorax. 2007 Jan;62(1):51–6.  	 192	39.  Alberg AJ. Cigarette smoking: health effects and control strategies. Drugs Today. 2008 Dec;44(12):895–904.  40.  Chaturvedi AK, Caporaso NE, Katki HA, Wong H-L, Chatterjee N, Pine SR, et al. C-Reactive Protein and Risk of Lung Cancer. J Clin Oncol. 2010 Jun 1;28(16):2719–26.  41.  Pine SR, Mechanic LE, Enewold L, Chaturvedi AK, Katki HA, Zheng Y-L, et al. Increased Levels of Circulating Interleukin 6, Interleukin 8, C-Reactive Protein, and Risk of Lung Cancer. J Natl Cancer Inst. 2011 Jul 20;103(14):1112–22.  42.  Postma DS, Calverley P. Inhaled corticosteroids in COPD: a case in favour. Eur Respir J. 2009 Jul 1;34(1):10–2.  43.  Suissa S, McGhan R, Niewoehner D, Make B. Inhaled Corticosteroids in Chronic Obstructive Pulmonary Disease. Proc Am Thorac Soc. 2007 Oct 1;4(7):535–42.  44.  Calverley PMA, Anderson JA, Celli B, Ferguson GT, Jenkins C, Jones PW, et al. Salmeterol and Fluticasone Propionate and Survival in Chronic Obstructive Pulmonary Disease. N Engl J Med. 2007 Feb 22;356(8):775–89.  45.  Gartlehner G, Hansen RA, Carson SS, Lohr KN. Efficacy and Safety of Inhaled Corticosteroids in Patients With COPD: A Systematic Review and Meta-Analysis of Health Outcomes. Ann Fam Med. 2006 May 1;4(3):253–62.  46.  van der Valk P, Monninkhof E, van der Palen J, Zielhuis G, van Herwaarden C. Effect of Discontinuation of Inhaled Corticosteroids in Patients with Chronic Obstructive Pulmonary Disease: The COPE Study. Am J Respir Crit Care Med. 2002 Nov 15;166(10):1358–63.  47.  Sin DD, Man SF, Marciniuk DD, Ford G, FitzGerald M, Wong E, et al. The effects of fluticasone with or without salmeterol on systemic biomarkers of inflammation in chronic obstructive pulmonary disease. Am J Respir Crit Care Med. 2008 Jun 1;177(11):1207–14.  48.  Sin DD, Man SFP. Inhaled corticosteroids and survival in chronic obstructive pulmonary disease: does the dose matter? Eur Respir J. 2003 Feb;21(2):260–6.  49.  Eurich DT, Lee C, Marrie TJ, Majumdar SR. Inhaled Corticosteroids and Risk of Recurrent Pneumonia: A Population-Based, Nested Case-Control Study. Clin Infect Dis. 2013 Oct 15;57(8):1138–44.  50.  Allin KH, Bojesen SE, Nordestgaard BG. Baseline C-Reactive Protein Is Associated With Incident Cancer and Survival in Patients With Cancer. J Clin Oncol. 2009 May 1;27(13):2217–24.  51.  Trichopoulos D. Plasma C-Reactive Protein and Risk of Cancer: A Prospective Study from Greece. Cancer Epidemiol Biomarkers Prev. 2006 Feb 1;15(2):381–4.  	 193	52.  Siemes C, Visser LE, Coebergh J-WW, Splinter TAW, Witteman JCM, Uitterlinden AG, et al. C-Reactive Protein Levels, Variation in the C-Reactive Protein Gene, and Cancer Risk: The Rotterdam Study. J Clin Oncol. 2006 Jan 3;24(33):5216–22.  53.  Jen R, Rennard SI, Sin DD. Effects of inhaled corticosteroids on airway inflammation in chronic obstructive pulmonary disease: a systematic review and meta-analysis. Int J Chron Obstruct Pulmon Dis. 2012;7:587–95.  54.  Barnes NC, Qiu YS, Pavord ID, Parker D, Davis PA, Zhu J, et al. Antiinflammatory effects of salmeterol/fluticasone propionate in chronic obstructive lung disease. Am J Respir Crit Care Med. 2006 Apr 1;173(7):736–43.  55.  Hattotuwa KL, Gizycki MJ, Ansari TW, Jeffery PK, Barnes NC. The Effects of Inhaled Fluticasone on Airway Inflammation in Chronic Obstructive Pulmonary Disease: A Double-Blind, Placebo-controlled Biopsy Study. Am J Respir Crit Care Med. 2002 Jun 15;165(12):1592–6.  56.  Sin DD, Lacy P, York E, Man SF. Effects of fluticasone on systemic markers of inflammation in chronic obstructive pulmonary disease. Am J Respir Crit Care Med. 2004 Oct 1;170(7):760–5.  57.  Pinto-Plata VM. C-reactive protein in patients with COPD, control smokers and non-smokers. Thorax. 2005 Oct 21;61(1):23–8.  58.  Brugts JJ, Yetgin T, Hoeks SE, Gotto AM, Shepherd J, Westendorp RGJ, et al. The benefits of statins in people without established cardiovascular disease but with cardiovascular risk factors: meta-analysis of randomised controlled trials. BMJ [Internet]. 2009 [cited 2014 Jul 14];338. Available from: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2714690/ 59.  Taylor F, Ward K, Moore TH, Burke M, Davey Smith G, Casas JP, et al. Statins for the primary prevention of cardiovascular disease. In: The Cochrane Collaboration, editor. Cochrane Database of Systematic Reviews [Internet]. Chichester, UK: John Wiley & Sons, Ltd; 2011 [cited 2016 Feb 22]. Available from: http://doi.wiley.com/10.1002/14651858.CD004816.pub4 60.  Bibbins-Domingo K, Grossman DC, Curry SJ, Davidson KW, Epling JW, García FAR, et al. Statin Use for the Primary Prevention of Cardiovascular Disease in Adults: US Preventive Services Task Force Recommendation Statement. JAMA. 2016 Nov 15;316(19):1997–2007.  61.  Soyseth V, Brekke PH, Smith P, Omland T. Statin use is associated with reduced mortality in COPD. Eur Respir J. 2006 Sep 27;29(2):279–83.  62.  Mancini GBJ, Etminan M, Zhang B, Levesque LE, FitzGerald JM, Brophy JM. Reduction of Morbidity and Mortality by Statins, Angiotensin-Converting Enzyme Inhibitors, and 	 194	Angiotensin Receptor Blockers in Patients With Chronic Obstructive Pulmonary Disease. J Am Coll Cardiol. 2006 Jun;47(12):2554–60.  63.  Lawes CM, Thornley S, Young R, Hopkins R, Marshall R, Chan WC, et al. Statin use in COPD patients is associated with a reduction in mortality: a national cohort study. Prim Care Respir J. 2012 Jan 4;21(1):35–40.  64.  Miyata R, Bai N, Vincent R, Sin DD, Van Eeden SF. STatins reduce ambient particulate matter-induced lung inflammation by promoting the clearance of particulate matter < 10 µm from lung tissues. CHEST J. 2013 Feb 1;143(2):452–60.  65.  Lahousse L, Loth DW, Joos GF, Hofman A, Leufkens HGM, Brusselle GG, et al. Statins, systemic inflammation and risk of death in COPD: The Rotterdam study. Pulm Pharmacol Ther. 2013 Apr;26(2):212–7.  66.  Criner GJ, Connett JE, Aaron SD, Albert RK, Bailey WC, Casaburi R, et al. Simvastatin for the Prevention of Exacerbations in Moderate-to-Severe COPD. N Engl J Med. 2014 Jun 5;370(23):2201–10.  67.  Young RP, Hopkins RJ, Agusti A. Statins as adjunct therapy in COPD: how do we cope after STATCOPE? Thorax. 2014 Oct 1;69(10):891–4.  68.  Mancini GBJ, Road J. Are Statins out in the COLD? The STATCOPE Trial. Can J Cardiol. 2015 Aug;31(8):970–3.  69.  Albert MA, Danielson E, Rifai N, Ridker PM, Investigators  for the P. Effect of Statin Therapy on C-Reactive Protein Levels: The Pravastatin Inflammation/CRP Evaluation (PRINCE): A Randomized Trial and Cohort Study. JAMA. 2001 Jul 4;286(1):64–70.  70.  Ridker PM, Rifai N, Pfeffer MA, Sacks F, Braunwald E, Cholesterol, et al. Long-term effects of pravastatin on plasma concentration of C-reactive protein. Circulation. 1999;100(3):230–235.  71.  Ridker PM, Danielson E, Fonseca FA, Genest J, Gotto Jr AM, Kastelein JJ, et al. Rosuvastatin to prevent vascular events in men and women with elevated C-reactive protein. N Engl J Med. 2008;359(21):2195.  72.  Bickel C, Rupprecht HJ, Blankenberg S, Espinola-Klein C, Rippin G, Hafner G, et al. Influence of HMG–CoA reductase inhibitors on markers of coagulation, systemic inflammation and soluble cell adhesion. Int J Cardiol. 2002;82(1):25–31.  73.  Shishehbor MH. Statins Promote Potent Systemic Antioxidant Effects Through Specific Inflammatory Pathways. Circulation. 2003 Jul 29;108(4):426–31.  74.  Ward BW, Schiller JS. Prevalence of Multiple Chronic Conditions Among US Adults: Estimates From the National Health Interview Survey, 2010. Prev Chronic Dis [Internet]. 	 195	2013 Apr 25 [cited 2016 Jan 15];10. Available from: http://www.cdc.gov/pcd/issues/2013/12_0203.htm 75.  Gu Q, Dillon CF, Burt VL. Prescription drug use continues to increase: U.S. prescription drug data for 2007-2008. NCHS Data Brief. 2010 Sep;(42):1–8.  76.  Rotermann M, Sanmartin C, Hennessy D, Arthur M. Prescription medication use by Canadians aged 6 to 79 [Internet]. Statistics Canada; 2014 [cited 2015 Jul 14]. Available from: http://www.statcan.gc.ca/pub/82-003-x/2014006/article/14032-eng.pdf 77.  Bartlett G, Abrahamowicz M, Tamblyn R, Grad R, Čapek R, Berger R du. Longitudinal patterns of new Benzodiazepine use in the elderly. Pharmacoepidemiol Drug Saf. 2004 Oct;13(10):669–82.  78.  Vacek PM. Assessing the Effect of Intensity When Exposure Varies Over Time. Stat Med. 1997 Mar 15;16(5):505–13.  79.  Cecere LM, Slatore CG, Uman JE, Evans LE, Udris EM, Bryson CL, et al. Adherence to Long-Acting Inhaled Therapies among Patients with Chronic Obstructive Pulmonary Disease (COPD). COPD J Chronic Obstr Pulm Dis. 2012 May 23;9(3):251–8.  80.  Wei MY, Ito MK, Cohen JD, Brinton EA, Jacobson TA. Predictors of statin adherence, switching, and discontinuation in the USAGE survey: Understanding the use of statins in America and gaps in patient education. J Clin Lipidol. 2013 Sep;7(5):472–83.  81.  Kuh D, Ben-Shlomo Y, Lynch J, Hallqvist J, Power C. Life course epidemiology. J Epidemiol Community Health. 2003 Oct;57(10):778–83.  82.  Stricker BHC, Stijnen T. Analysis of individual drug use as a time-varying determinant of exposure in prospective population-based cohort studies. Eur J Epidemiol. 2010 Apr;25(4):245–51.  83.  Suissa S, Coulombe J, Ernst P. Discontinuation of inhaled corticosteroids in COPD and the risk reduction of pneumonia. CHEST J. 2015;148(5):1177–1183.  84.  Teichert M, de Smet PA, Hofman A, Witteman JC, Stricker BHC. Discontinuation of β-blockers and the risk of myocardial infarction in the elderly. Drug Saf. 2007;30(6):541–549.  85.  Avina-Zubieta JA, Abrahamowicz M, De Vera MA, Choi HK, Sayre EC, Rahman MM, et al. Immediate and past cumulative effects of oral glucocorticoids on the risk of acute myocardial infarction in rheumatoid arthritis: a population-based study. Rheumatology. 2013 Jan 1;52(1):68–75.  86.  Dixon WG, Abrahamowicz M, Beauchamp M-E, Ray DW, Bernatsky S, Suissa S, et al. Immediate and delayed impact of oral glucocorticoid therapy on risk of serious infection 	 196	in older patients with rheumatoid arthritis: a nested case-control analysis. Ann Rheum Dis. 2012 Jul 1;71(7):1128–33.  87.  Sadatsafavi M, Lynd LD, De Vera MA, Zafari Z, FitzGerald JM. One-year outcomes of inhaled controller therapies added to systemic corticosteroids after asthma-related hospital discharge. Respir Med. 2015 Mar;109(3):320–8.  88.  FDA Guidance on Pharmacoepidemiology [Internet]. [cited 2016 Jun 30]. Available from: http://www.fda.gov/downloads/drugs/guidancecomplianceregulatoryinformation/guidances/ucm243537.pdf 89.  Tournier M, Bégaud B, Cougnard A, Auleley G-R, Deligne J, Blum-Boisgard C, et al. Influence of the drug exposure definition on the assessment of the antipsychotic metabolic impact in patients initially treated with mood-stabilizers. Br J Clin Pharmacol. 2012 Jul;74(1):189–96.  90.  Lee CH, Hyun MK, Jang EJ, Lee NR, Kim K, Yim JJ. Inhaled corticosteroid use and risks of lung cancer and laryngeal cancer. Respir Med. 2013 Aug;107(8):1222–33.  91.  Lynd LD, Guh DP, Paré PD, Anis AH. Patterns of Inhaled Asthma Medication Use: A 3-Year Longitudinal Analysis of Prescription Claims Data From British Columbia, Canada. Chest. 2002 Dec;122(6):1973–81.  92.  Parimon T, Chien JW, Bryson CL, McDonell MB, Udris EM, Au DH. Inhaled Corticosteroids and Risk of Lung Cancer among Patients with Chronic Obstructive Pulmonary Disease. Am J Respir Crit Care Med. 2007 Apr 1;175(7):712–9.  93.  Sikka R, Xia F, Aubert R. Estimating medication persistency using administrative claims data. Am J Manag Care. 2005 2005;11(7):449–57.  94.  Blackburn DF, Dobson RT, Blackburn JL, Wilson TW, Stang MR, Semchuk WM. Adherence to statins, beta-blockers and angiotensin-converting enzyme inhibitors following a first cardiovascular event: a retrospective cohort study. Can J Cardiol. 2005 May 1;21(6):485–8.  95.  Ho PM, Magid DJ, Shetterly SM, Olson KL, Maddox TM, Peterson PN, et al. Medication nonadherence is associated with a broad range of adverse outcomes in patients with coronary artery disease. Am Heart J. 2008 Apr;155(4):772–9.  96.  Abrahamowicz M, Bartlett G, Tamblyn R, du Berger R. Modeling cumulative dose and exposure duration provided insights regarding the associations between benzodiazepines and injuries. J Clin Epidemiol. 2006 Apr;59(4):393–403.  97.  Avina-Zubieta JA, Abrahamowicz M, Choi HK, Rahman MM, Sylvestre M-P, Esdaile JM, et al. Risk of cerebrovascular disease associated with the use of glucocorticoids in patients with incident rheumatoid arthritis: a population-based study. Ann Rheum Dis. 2011 Jun 1;70(6):990–5.  	 197	98.  Phadnis MA, Shireman TI, Wetmore JB, Rigler SK, Zhou X, Spertus JA, et al. Estimation of Drug Effectiveness by Modeling Three Time-Dependent Covariates: An Application to Data on Cardioprotective Medications in the Chronic Dialysis Population. Stat Biopharm Res. 2014 Jul 3;6(3):229–40.  99.  Shireman TI, Phadnis MA, Wetmore JB, Zhou X, Rigler SK, Spertus JA, et al. Antihypertensive Medication Exposure and Cardiovascular Outcomes in Hemodialysis Patients. Am J Nephrol. 2014;40(2):113–22.  100.  Raherison C, Girodet P-O. Epidemiology of COPD. Eur Respir Rev. 2009 Nov 30;18(114):213–21.  101.  Lokke A, Ulrik CS, Dahl R, Plauborg L, Dollerup J, Kristiansen LC, et al. Detection of previously undiagnosed cases of COPD in a high-risk population identified in general practice. Copd J Chronic Obstr Pulm Dis. 2012 Aug;9(5):458–65.  102.  Lundbäck B, Lindberg A, Lindström M, Rönmark E, Jonsson AC, Jönsson E, et al. Not 15 but 50% of smokers develop COPD?--Report from the Obstructive Lung Disease in Northern Sweden Studies. Respir Med. 2003 Feb;97(2):115–22.  103.  Tan WC, Sin DD, Bourbeau J, Hernandez P, Chapman KR, Cowie R, et al. Characteristics of COPD in never-smokers and ever-smokers in the general population: results from the CanCOLD study. Thorax. 2015 Sep 1;70(9):822–9.  104.  Ford ES. Hospital discharges, readmissions, and ED visits for COPD or bronchiectasis among US adults: findings from the nationwide inpatient sample 2001-2012 and Nationwide Emergency Department Sample 2006-2011. Chest. 2015 Apr;147(4):989–98.  105.  Murray CJ, Lopez AD. Alternative projections of mortality and disability by cause 1990–2020: Global Burden of Disease Study. The Lancet. 1997 May 24;349(9064):1498–504.  106.  Torre LA, Bray F, Siegel RL, Ferlay J, Lortet-Tieulent J, Jemal A. Global cancer statistics, 2012: Global Cancer Statistics, 2012. CA Cancer J Clin. 2015 Mar;65(2):87–108.  107.  Fry JS, Hamling JS, Lee PN. Systematic review with meta-analysis of the epidemiological evidence relating FEV1 decline to lung cancer risk. BMC Cancer. 2012;12(1):498.  108.  Sun S, Schiller JH, Gazdar AF. Lung cancer in never smokers--a different disease. Nat Rev Cancer. 2007 Oct;7(10):778–90.  109.  Szafranski W, Cukier A, Ramirez A, Menga G, Sansores R, Nahabedian S, et al. Efficacy and safety of budesonide/formoterol in the management of chronic obstructive pulmonary disease. Eur Respir J. 2003 Jan 1;21(1):74–81.  110.  Yang IA, Clarke MS, Sim EHA, Fong KM. Inhaled corticosteroids for stable chronic obstructive pulmonary disease. Cochrane Database Syst Rev. 2012;7:CD002991.  	 198	111.  Sin DD, Wu L, Anderson JA, Anthonisen NR, Buist AS, Burge PS, et al. Inhaled corticosteroids and mortality in chronic obstructive pulmonary disease. Thorax. 2005 Dec;60(12):992–7.  112.  Ryan R. Heterogeneity and subgroup analyses in Cochrane Consumers and Communication Review Group reviews: planning the analysis at protocol stage [Internet]. Cochrane Consumers and Communication Review Group; 2014 [cited 2016 Jun 23]. Available from: http://cccrg.cochrane.org/sites/cccrg.cochrane.org/files/uploads/Heterogeneity_subgroup_analyses.pdf 113.  Edwards DJG, Anderson I. Systematic Review and Guide to Selection of Selective Serotonin Reuptake Inhibitors. Drugs. 1999;57(4):507–33.  114.  Higgins J, Green S. Cochrane Handbook for Systematic Reviews of Interventions [Internet]. The Cochrane Collaboration; 2011 [cited 2016 Jul 12]. Report No.: Version 5.1.0. Available from: http://training.cochrane.org/handbook 115.  Moher D, Liberati A, Tetzlaff J, Altman DG, Group TP. Preferred Reporting Items for Systematic Reviews and Meta-Analyses: The PRISMA Statement. PLOS Med. 2009 Jul 21;6(7):e1000097.  116.  Kiri VA, Fabbri LM, Davis KJ, Soriano JB. Inhaled corticosteroids and risk of lung cancer among COPD patients who quit smoking. Respir Med. 2009 Jan;103(1):85–90.  117.  Lung Health Study Group. Effect of Inhaled Triamcinolone on the Decline in Pulmonary Function in Chronic Obstructive Pulmonary Disease. N Engl J Med. 2000 Dec 28;343(26):1902–9.  118.  Pauwels RA, Löfdahl C-G, Laitinen LA, Schouten JP, Postma DS, Pride NB, et al. Long-term treatment with inhaled budesonide in persons with mild chronic obstructive pulmonary disease who continue smoking. N Engl J Med. 1999;340(25):1948–1953.  119.  Tashkin DP, Rennard SI, Martin P, Ramachandran S, Martin UJ, Silkoff PE, et al. Efficacy and safety of budesonide and formoterol in one pressurized metered-dose inhaler in patients with moderate to very severe chronic obstructive pulmonary disease: results of a 6-month randomized clinical trial. Drugs. 2008;68(14):1975–2000.  120.  Delgado-Rodríguez M, Llorca J. Bias. J Epidemiol Community Health. 2004 Aug 1;58(8):635–41.  121.  Kiri VA, Soriano J, Visick G, Fabbri L. Recent trends in lung cancer and its association with COPD: an analysis using the UK GP Research Database. Prim Care Respir J. 2010 Mar;19(1):57–61.  122.  Adeloye D, Chua S, Lee C, Basquill C, Papana A, Theodoratou E, et al. Global and regional estimates of COPD prevalence: Systematic review and meta–analysis. J Glob 	 199	Health [Internet]. 2015 [cited 2016 Aug 25];5(2). Available from: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4693508/ 123.  Public Health Agency of Canada: Chronic Disease and Injury Indicator Framework Measures [Internet]. Prevalence of cardiovascular diseases, adults (self-reported). 2014 [cited 2016 Dec 5]. Available from: http://infobase.phac-aspc.gc.ca:9600/PHAC/dimensionMembers.jsp?l=en&rep=i3212B12F133F4CE88AD13DB60CA37237&s# 124.  Sidney S, Sorel M, Quesenberry CP, DeLuise C, Lanes S, Eisner MD. COPD and incident cardiovascular disease hospitalizations and mortality: Kaiser Permanente Medical Care Program. CHEST J. 2005;128(4):2068–2075.  125.  Sin DD, Man SFP. Chronic obstructive pulmonary disease as a risk factor for cardiovascular morbidity and mortality. Proc Am Thorac Soc. 2005;2(1):8–11.  126.  Young RP, Hopkins RJ. Interleukin-6 and statin therapy: potential role in the management of COPD. Respir Res. 2013;14:74.  127.  Blamoun AI, Batty GN, DeBari VA, Rashid AO, Sheikh M, Khan MA. Statins may reduce episodes of exacerbation and the requirement for intubation in patients with COPD: evidence from a retrospective cohort study. Int J Clin Pract. 2008 Sep 1;62(9):1373–8.  128.  Wang M-T, Lo Y-W, Tsai C-L, Chang L-C, Malone DC, Chu C-L, et al. Statin Use and Risk of COPD Exacerbation Requiring Hospitalization. Am J Med. 2013 Jul;126(7):598–606.e2.  129.  Robert P. Young, Raewyn J. Hopkins, Wing Cheuk Chan, Simon Thornley, Greg D. Gamble. Effect Of Statin Therapy On Mortality In COPD: Analysis Of Cause Specific Deaths In A National Cohort Study. In: A110 LATE BREAKING ABSTRACTS IN OBSTRUCTIVE LUNG DISEASES [Internet]. American Thoracic Society; 2013 [cited 2016 Oct 19]. p. A6017–A6017. (American Thoracic Society International Conference Abstracts). Available from: http://www.atsjournals.org/doi/abs/10.1164/ajrccm-conference.2013.187.1_MeetingAbstracts.A6017 130.  Khurana V, Bejjanki HR, Caldito G, Owens MW. Statins Reduce the Risk of Lung Cancer in Humans. Chest. 2007 May;131(5):1282–8.  131.  Criner GJ, Rogers TJ, Cornwell WD, Voelker H, Albert RK, Bailey WC, et al. Changes in HS-CRP Levels and Rates of Acute Exacerbations of COPD in Current and Former Smokers in MACRO and STATCOPE. In: B38 TALKING ABOUT COPD BIOMARKERS [Internet]. American Thoracic Society; 2015 [cited 2016 Dec 5]. p. A2918–A2918. (American Thoracic Society International Conference Abstracts). Available from: http://www.atsjournals.org/doi/abs/10.1164/ajrccm-conference.2015.191.1_MeetingAbstracts.A2918 	 200	132.  British Columbia Ministry of Health (2011): PharmaNet. BC Ministry of Health. Data Extract. Data Stewardship Committee (2011). http://www.popdata.bc.ca/data.  133.  Canadian Institute for Health Information (2011): Discharge Abstract Database (Hospital Separations). Population Data BC. Data Extract. MOH (2010). http://www.popdata.bc.ca/data.  134.  British Columbia Vital Statistics Agency (2010): Vital Statistics Deaths. Population Data BC. Data Extract BC Vital Statistics Agency (2009). http://www.popdata.bc.ca/data.  135.  British Columbia Ministry of Health (2011). Consolidation file (MSP registration and Premium Billing). Population Data BC. Data Extract. MOH (2010). http://www.popdata.bc.ca/data.  136.  British Columbia Ministry of Health (2011): Medical Services Plan (MSP) Payment Information File. Population Data BC. Data Extract. MOH (2010). http://www.popdata.bc.ca/data.  137.  AHFS Drug Information [Internet]. AHFS Drug Information. [cited 2017 May 24]. Available from: http://www.ahfsdruginformation.com/ 138.  Weinberg CR. Toward a clearer definition of confounding. Am J Epidemiol. 1993;137(1):1–8.  139.  Velentgas P, Dreyer NA, Nourjah P, Smith SR, Torchia MM, others. Developing a protocol for observational comparative effectiveness research: a user’s guide [Internet]. Government Printing Office; 2013 [cited 2016 Nov 9]. Available from: http://books.google.com/books?hl=en&lr=&id=785i9BdQR7AC&oi=fnd&pg=PT13&dq=%22comparative+effectiveness%22+%22of+interest+in+which+treatments+are%22+%22treatment+comparisons,%22+%22experimental+methods+(or+a%22+%22not+assigned+by+the+investigator+but+rather%22+%22validity+is+defined+as+the+absence+of%22+&ots=fpa9oTwEnz&sig=dK4bx-46t5xPbufj4i4ZZQW3DyQ 140.  Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: Development and validation. J Chronic Dis. 1987 Jan 1;40(5):373–83.  141.  Romano PS, Roos LL, Jollis JG. Presentation adapting a clinical comorbidity index for use with ICD-9-CM administrative data: Differing perspectives. J Clin Epidemiol. 1993 Oct 1;46(10):1075–9.  142.  Akaike H. A new look at the statistical model identification. IEEE Trans Autom Control. 1974 Dec;19(6):716–23.  143.  Wei L, Wang J, Thompson P, Wong S, Struthers AD, MacDonald TM. Adherence to statin treatment and readmission of patients after myocardial infarction: a six year follow up study. Heart. 2002 Sep;88(3):229–33.  	 201	144.  Nielsen SF, Nordestgaard BG, Bojesen SE. Statin Use and Reduced Cancer-Related Mortality. N Engl J Med. 2012 Nov 8;367(19):1792–802.  145.  Antonopoulos A, Margaritis M, Lee R, Channon K, Antoniades C. Statins as anti-inflammatory agents in atherogenesis: molecular mechanisms and lessons from the recent clinical trials. Curr Pharm Des. 2012;18(11):1519–1530.  146.  Agusti A, Bel E, Thomas M, Vogelmeier C, Brusselle G, Holgate S, et al. Treatable traits: toward precision medicine of chronic airway diseases. Eur Respir J. 2016 Feb;47(2):410–9.  147.  Maneechotesuwan K, Wongkajornsilp A, Adcock IM, Barnes PJ. Simvastatin suppresses airway IL-17 and upregulates IL-10 in patients with stable COPD. Chest. 2015 Nov 1;148(5):1164–76.  148.  Maneechotesuwan K, Kasetsinsombat K, Wongkajornsilp A, Barnes PJ. Simvastatin up-regulates adenosine deaminase and suppresses osteopontin expression in COPD patients through an IL-13-dependent mechanism. Respir Res. 2016;17:104.  149.  Mroz RM, Lisowski P, Tycinska A, Bierla J, Trzeciak PZ, Minarowski L, et al. Anti-inflammatory effects of atorvastatin treatment in chronic obstructive pulmonary disease. A controlled pilot study. J Physiol Pharmacol Off J Pol Physiol Soc. 2015 Feb;66(1):111–28.  150.  Zvezdin B, Milutinov S, Kojicic M, Hadnadjev M, Hromis S, Markovic M, et al. A postmortem analysis of major causes of early death in patients hospitalized with COPD exacerbation. Chest. 2009 Aug;136(2):376–80.  151.  Campo G, Pavasini R, Malagù M, Punzetti S, Napoli N, Guerzoni F, et al. Relationship between Troponin Elevation, Cardiovascular History and Adverse Events in Patients with acute exacerbation of COPD. COPD. 2015;12(5):560–7.  152.  Huiart L, Ernst P, Suissa S. Cardiovascular morbidity and mortality in COPD. Chest. 2005 Oct 1;128(4):2640–6.  153.  Shrank WH, Patrick AR, Brookhart MA. Healthy User and Related Biases in Observational Studies of Preventive Interventions: A Primer for Physicians. J Gen Intern Med. 2011 Jan 4;26(5):546–50.  154.  Government of Canada SC. Canadian Community Health Survey - Annual Component (CCHS) [Internet]. 2015 [cited 2017 May 30]. Available from: http://www23.statcan.gc.ca/imdb/p2SV.pl?Function=getSurvey&SDDS=3226&Item_Id=118913 155.  WHO | Burden of COPD [Internet]. WHO. [cited 2016 Dec 5]. Available from: http://www.who.int/respiratory/copd/burden/en/ 	 202	156.  Mahler DA, Mackowiak JI. Evaluation of the Short-Form 36-Item Questionnaire to Measure Health-Related Quality of Life in Patients With COPD. Chest. 1995 Jun;107(6):1585–9.  157.  Ståhl E, Lindberg A, Jansson S-A, Rönmark E, Svensson K, Andersson F, et al. Health-related quality of life is related to COPD disease severity. Health Qual Life Outcomes. 2005;3:56.  158.  Thun MJ, Lally CA, Calle EE, Heath CW, Flannery JT, Flanders WD. Cigarette smoking and changes in the histopathology of lung cancer. J Natl Cancer Inst. 1997;89(21):1580–1586.  159.  Flanders WD, Lally CA, Zhu B-P, Henley SJ, Thun MJ. Lung Cancer Mortality in Relation to Age, Duration of Smoking, and Daily Cigarette Consumption. Cancer Res. 2003 Oct 1;63(19):6556–62.  160.  Brenner AV, Wang Z, Kleinerman RA, Wang L, Zhang S, Metayer C, et al. Previous pulmonary diseases and risk of lung cancer in Gansu Province, China. Int J Epidemiol. 2001 Feb;30(1):118–24.  161.  García Rodríguez LA, Wallander M-A, Martín-Merino E, Johansson S. Heart failure, myocardial infarction, lung cancer and death in COPD patients: A UK primary care study. Respir Med. 2010 Nov;104(11):1691–9.  162.  Koshiol J, Rotunno M, Consonni D, Pesatori AC, De Matteis S, Goldstein AM, et al. Chronic Obstructive Pulmonary Disease and Altered Risk of Lung Cancer in a Population-Based Case-Control Study. Vij N, editor. PLoS ONE. 2009 Oct 8;4(10):e7380.  163.  Thomsen M, Dahl M, Lange P, Vestbo J, Nordestgaard BG. Inflammatory Biomarkers and Comorbidities in Chronic Obstructive Pulmonary Disease. Am J Respir Crit Care Med. 2012 Nov 15;186(10):982–8.  164.  Punturieri A, Croxton TL, Weinmann GG, Kiley JP. Chronic Obstructive Pulmonary Disease. Am J Respir Crit Care Med. 2008 Sep 1;178(5):441–3.  165.  Punturieri A, Szabo E, Croxton TL, Shapiro SD, Dubinett SM. Lung cancer and chronic obstructive pulmonary disease: needs and opportunities for integrated research. J Natl Cancer Inst. 2009 Apr 15;101(8):554–9.  166.  Barnes PJ. How corticosteroids control inflammation: Quintiles Prize Lecture 2005. Br J Pharmacol. 2006 Jun;148(3):245–54.  167.  Gizycki MJ, Hattotuwa KL, Barnes N, Jeffery PK. Effects of fluticasone propionate on inflammatory cells in COPD: an ultrastructural examination of endobronchial biopsy tissue. Thorax. 2002 Sep 1;57(9):799–803.  	 203	168.  Coleman M, Forman D, Bryant H, Butler J, Rachet B, Maringe C, et al. Cancer survival in Australia, Canada, Denmark, Norway, Sweden, and the UK, 1995–2007 (the International Cancer Benchmarking Partnership): an analysis of population-based cancer registry data. The Lancet. 2011 Jan 14;377(9760):127–38.  169.  British Columbia Cancer Agency Registry Data (2011). V2. Population Data BC. Data Extract. BC Cancer Agency (2011). http://www.popdata.bc.ca/data.  170.  Suissa S, Assimes T, Ernst P. Inhaled short acting β agonist use in COPD and the risk of acute myocardial infarction. Thorax. 2003 Jan 1;58(1):43–6.  171.  Henschke CI, Yankelevitz DF, Yip R, Reeves AP, Farooqi A, Xu D, et al. Lung Cancers Diagnosed at Annual CT Screening: Volume Doubling Times. Radiology. 2012 May 1;263(2):578–83.  172.  Gail MH. Does cardiac transplantation prolong life? A reassessment. Ann Intern Med. 1972 May;76(5):815–7.  173.  World Health Organization Collaborating Center for Drug Statistics  Methodology. ATC classification index with DDDs, 2003; Oslo, Norway.  174.  Allison PD. Survival Analysis Using the SAS System: A Practical Guide. Cary, North Carolina, USA: SAS Institute; 1995.  175.  Cox DR. Regression Models and Life-Tables. J R Stat Soc Ser B Methodol. 1972;34(2):187–220.  176.  British Columbia Cancer Agency - Facts & Figures [Internet]. [cited 2016 Dec 6]. Available from: http://www.bccancer.bc.ca/health-info/disease-system-statistics/bc-cancer-statistics/facts-and-figures 177.  Cancer Research UK - Lung cancer incidence statistics [Internet]. Cancer Research UK. 2015 [cited 2016 Dec 6]. Available from: http://www.cancerresearchuk.org/health-professional/cancer-statistics/statistics-by-cancer-type/lung-cancer/incidence 178.  Lung Cancer | American Cancer Society [Internet]. [cited 2016 Dec 6]. Available from: http://www.cancer.org/cancer/lungcancer/index 179.  Tamim H, Monfared AAT, LeLorier J. Application of lag-time into exposure definitions to control for protopathic bias. Pharmacoepidemiol Drug Saf. 2007 Mar;16(3):250–8.  180.  Suissa S. Immortal Time Bias in Pharmacoepidemiology. Am J Epidemiol. 2008 Jan 7;167(4):492–9.  181.  Han MK. Medication adherence in COPD: what have we learned? Thorax. 2009;64(11):922–923.  	 204	182.  Lavorini F, Magnan A, Christophe Dubus J, Voshaar T, Corbetta L, Broeders M, et al. Effect of incorrect use of dry powder inhalers on management of patients with asthma and COPD. Respir Med. 2008 Apr;102(4):593–604.  183.  Melani AS. Inhalatory therapy training: a priority challenge for the physician. Acta Bio-Medica Ateneo Parm. 2007 Dec;78(3):233–45.  184.  LaRosa JC, Grundy SM, Waters DD, Shear C, Barter P, Fruchart J-C, et al. Intensive lipid lowering with atorvastatin in patients with stable coronary disease. N Engl J Med. 2005 Apr 7;352(14):1425–35.  185.  Chou R, Dana T, Blazina I, Daeges M, Jeanne TL. Statins for Prevention of Cardiovascular Disease in Adults: Evidence Report and Systematic Review for the US Preventive Services Task Force. JAMA. 2016 Nov 15;316(19):2008–24.  186.  Alberg AJ. Epidemiology of Lung Cancer: Looking to the Future. J Clin Oncol. 2005 May 10;23(14):3175–85.  187.  King PT. Inflammation in chronic obstructive pulmonary disease and its role in cardiovascular disease and lung cancer. Clin Transl Med. 2015 Jul 29;4(1):26.  188.  Boudreau DM, Yu O, Johnson J. Statin Use and Cancer Risk: A Comprehensive Review. Expert Opin Drug Saf. 2010 Jul;9(4):603–21.  189.  Vinogradova Y, Coupland C, Hippisley-Cox J. Exposure to statins and risk of common cancers: a series of nested case-control studies. BMC Cancer. 2011;11:409.  190.  Marelli C, Gunnarsson C, Ross S, Haas S, Stroup DF, Cload P, et al. Statins and Risk of CancerA Retrospective Cohort Analysis of 45,857 Matched Pairs From an Electronic Medical Records Database of 11 Million Adult Americans. J Am Coll Cardiol. 2011 Jul 26;58(5):530–7.  191.  van Gestel YRBM van, Hoeks SE, Sin DD, Hüzeir V, Stam H, Mertens FW, et al. COPD and cancer mortality: the influence of statins. Thorax. 2009 Nov 1;64(11):963–7.  192.  Bonovas S, Filioussi K, Tsavaris N, Sitaras NM. Statins and Cancer Risk: A Literature-Based Meta-Analysis and Meta-Regression Analysis of 35 Randomized Controlled Trials. J Clin Oncol. 2006 Sep 25;24(30):4808–17.  193.  Setoguchi S, Glynn RJ, Avorn J, Mogun H, Schneeweiss S. Statins and the Risk of Lung, Breast, and Colorectal Cancer in the Elderly. Circulation. 2007 Jan 2;115(1):27–33.  194.  Taylor ML, Wells BJ, Smolak MJ. Statins and cancer: a meta-analysis of case–control studies. Eur J Cancer Prev. 2008;17(3):259–268.  	 205	195.  Rinfret S, Behlouli H, Eisenberg MJ, Humphries K, Tu JV, Pilote L. Class effects of statins in elderly patients with congestive heart failure: A population-based analysis. Am Heart J. 2008 Feb;155(2):316–23.  196.  Suissa S, Azoulay L. Metformin and the risk of cancer: time-related biases in observational studies. Diabetes Care. 2012 Dec;35(12):2665–73.  197.  Smith MB, Lee NJ, Haney E, Carson S. Drug Class Review: HMG-CoA Reductase Inhibitors (Statins) and Fixed-dose Combination Products Containing a Statin: Final Report Update 5 [Internet]. Portland (OR): Oregon Health & Science University; 2009. (Drug Class Reviews). Available from: http://www.ncbi.nlm.nih.gov/books/NBK47273/ 198.  Arnold BF, Ercumen A. Negative Control Outcomes: A Tool to Detect Bias in Randomized Trials. JAMA. 2016 Dec 27;316(24):2597–8.  199.  Arnold BF, Ercumen A, Benjamin-Chung J, Colford JM. Brief Report: Negative Controls to Detect Selection Bias and Measurement Bias in Epidemiologic Studies. Epidemiology. 2016 Sep;27(5):637–41.  200.  Lipsitch M, Tchetgen Tchetgen E, Cohen T. Negative Controls: A Tool for Detecting Confounding and Bias in Observational Studies. Epidemiology. 2010 May;21(3):383–8.  201.  Sørensen HT, Olsen JH, Mellemkjær L, Thulstrup AM, Steffensen FH, McLaughlin JK, et al. Cancer risk and mortality in users of calcium channel blockers. Cancer. 2000 Jul 1;89(1):165–70.  202.  Fryzek JP, Poulsen AH, Lipworth L, Pedersen L, Nørgaard M, McLaughlin JK, et al. A Cohort Study of Antihypertensive Medication Use and Breast Cancer Among Danish Women. Breast Cancer Res Treat. 2006 Jun 1;97(3):231–6.  203.  Grimaldi-Bensouda L, Klungel O, Kurz X, Groot MCH de, Afonso ASM, Bruin ML de, et al. Calcium channel blockers and cancer: a risk analysis using the UK Clinical Practice Research Datalink (CPRD). BMJ Open. 2016 Jan 1;6(1):e009147.  204.  Smith GD. Negative control exposures in epidemiologic studies. Epidemiol Camb Mass. 2012 Mar;23(2):350-351-352.  205.  Friis S, Poulsen AH, Johnsen SP, McLaughlin JK, Fryzek JP, Dalton SO, et al. Cancer risk among statin users: A population-based cohort study. Int J Cancer. 2005 Apr 20;114(4):643–7.  206.  Cardwell CR, Menamin ÚM, Hughes CM, Murray LJ. Statin Use and Survival from Lung Cancer: A Population-Based Cohort Study. Cancer Epidemiol Biomarkers Prev. 2015 May 1;24(5):833–41.  207.  Alexeeff SE, Litonjua AA, Sparrow D, Vokonas PS, Schwartz J. Statin Use Reduces Decline in Lung Function. Am J Respir Crit Care Med. 2007 Oct 15;176(8):742–7.  	 206	208.  Ingebrigtsen TS, Marott JL, Nordestgaard BG, Lange P, Hallas J, Vestbo J. Statin use and exacerbations in individuals with chronic obstructive pulmonary disease. Thorax. 2015 Jan 1;70(1):33–40.  209.  Liao J, Laufs U. Pleiotropic Effects of Statins. Annu Rev Pharmacol Toxicol. 2005;45(1):89–118.  210.  Gershon AS, Wang C, Guan J, Vasilevska-Ristovska J, Cicutto L, To T. Identifying individuals with physcian diagnosed COPD in health administrative databases. COPD. 2009 Oct;6(5):388–94.  211.  Davis RM, Novotny TE. The Epidemiology of Cigarette Smoking and Its Impact on Chronic Obstructive Pulmonary Disease. Am Rev Respir Dis. 1989 Sep;140(3_pt_2):S82–4.  212.  Kornmann O, Beeh KM, Beier J, Geis UP, Ksoll M, Buhl R. Newly Diagnosed Chronic Obstructive Pulmonary Disease. Respiration. 2003 Feb 17;70(1):67–75.  213.  Vestbo J, Lange P. Natural history of COPD: Focusing on change in FEV: Natural history of COPD. Respirology. 2016 Jan;21(1):34–43.  214.  Cancer Surveillance and Outcomes, Survival Statistics 2011 [Internet]. British Columbia Cancer Agency; 2012 [cited 2016 Dec 22]. Available from: http://www.bccancer.bc.ca/statistics-and-reports-site/Documents/Regional_Survival_Report_2012.pdf 215.  Lavorini F, Magnan A, Christophe Dubus J, Voshaar T, Corbetta L, Broeders M, et al. Effect of incorrect use of dry powder inhalers on management of patients with asthma and COPD. Respir Med. 2008 Apr;102(4):593–604.  216.  Fitzmaurice C, Allen C, Barber RM, Barregard L, Bhutta ZA, Brenner H, et al. Global, Regional, and National Cancer Incidence, Mortality, Years of Life Lost, Years Lived With Disability, and Disability-Adjusted Life-years for 32 Cancer Groups, 1990 to 2015: A Systematic Analysis for the Global Burden of Disease Study. JAMA Oncol [Internet]. 2016 Dec 3 [cited 2016 Dec 21]; Available from: http://jamanetwork.com/journals/jamaoncology/fullarticle/2588797 217.  Liniker E, Harrison M, Weaver JMJ, Agrawal N, Chhabra A, Kingshott V, et al. Treatment costs associated with interventional cancer clinical trials conducted at a single UK institution over 2 years (2009–2010). Br J Cancer. 2013 Oct 15;109(8):2051–7.  218.  Booth CM, Tannock IF. Randomised controlled trials and population-based observational research: partners in the evolution of medical evidence. Br J Cancer. 2014 Feb 4;110(3):551–5.  	 207	219.  Sherman RE, Anderson SA, Dal Pan GJ, Gray GW, Gross T, Hunter NL, et al. Real-World Evidence — What Is It and What Can It Tell Us? N Engl J Med. 2016 Dec 8;375(23):2293–7.  220.  Travers J, Marsh S, Williams M, Weatherall M, Caldwell B, Shirtcliffe P, et al. External validity of randomised controlled trials in asthma: to whom do the results of the trials apply? Thorax. 2007 Mar 1;62(3):219–23.  221.  United States Food and Drug Administration (FDA). United States Food and Drug Administration - Draft Guidance on the Use of Real-World Evidence to Support Regulatory Decision-Making for Medical Devices [Internet]. US Food and Drug Administration; 2016 [cited 2016 Dec 8]. Available from: http://www.fda.gov/MedicalDevices/ResourcesforYou/Industry/ucm513478.htm 222.  Serapinas D, Narbekovas A, Juskevicius J, Sakalauskas R. Systemic inflammation in COPD in relation to smoking status. Multidiscip Respir Med. 2011 Aug 31;6(4):214–9.  		 208	Appendices	Appendix A  	Primary search strategy 	Search	Strategy:	--------------------------------------------------------------------------------	1					Bronchitis,	Chronic/		2					bronchitis,	chronic.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		3					(chronic	adj3	bronchitis).mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		4					or/1-3		5					bronchitis.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		6					bronchitis/	or	bronchiolitis/		7					bronchitides.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		8					bronchiolitides.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		9					Lung	Diseases,	Obstructive/		10					(obstructive	adj4	lung	disease?).mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		11					(obstructive	adj3	pulmonary	disease?).mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		12					or/9-11		13					Lung	Diseases/		14					((pulmonary	or	lung)	adj3	disease?).mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		15					disease,	lung.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		16					diseases,	lung.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		17					lung	disease.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		18					pulmonary	disease.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]			 209	19					disease,	pulmonary.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		20					diseases,	pulmonary.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		21					pulmonary	diseases.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		22					or/13-21	23					Airway	Obstruction/		24					(airway	adj3	obstruction?).mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		25					23	or	24		26					Respiratory	Tract	Diseases/		27					(respiratory	tract	adj3	disease$).mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]	28					26	or	27		29					Cough/		30					26	or	27		31					cough$.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		32					29	or	31		33					Bronchial	Diseases/	34					(bronchial	adj3	disease$).mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		35					33	or	34		36					Dyspnea/		37					dyspnea$.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		38					(breath	adj2	shortness$).mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		39					or/36-38		40					Pulmonary	Disease,	Chronic	Obstructive/		41					(pulmonary	disease?	adj3	chronic	obstructive).mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		42					copd.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		43					coad.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		44					(chronic	obstructi$	adj3	airway).mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]			 210	45					chronic	obstructive	lung	disease$.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		46					or/40-45		47					Pulmonary	Emphysema/		48					Emphysema/		49					emphysema.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		50					or/47-49		51					or/1-39		52					Lung	Neoplasms/		53					exp	Lung	Neoplasms/		54					Carcinoma,	Bronchogenic/		55					Carcinoma,	Non-Small-Cell	Lung/	56					Small	Cell	Lung	Carcinoma/		57					Adenocarcinoma/		58					Carcinoma,	Squamous	Cell/		59					Carcinoma,	Large	Cell/		60					Adenocarcinoma,	Bronchiolo-Alveolar/		61					Carcinoma,	Small	Cell/		62					Multiple	Pulmonary	Nodules/		63					Pancoast	Syndrome/		64					Pulmonary	Blastoma/	65					Solitary	Pulmonary	Nodule/		66					(lung	adj3	(cancer?or	neoplasm?	or	carcinoma?)).mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		67					57	or	58	or	54	or	63	or	61	or	62	or	64	or	52	or	60	or	55	or	56	or	59	or	65		68					or/52-66		69					2	or	1	or	3		70					Bronchitis/		71					70	or	5	or	7		72					Bronchiolitis/		73					bronchiolitis.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		74					72	or	73	or	8		75					10	or	9	or	11		76					14	or	13		77					24	or	23		78					26	or	27		79					31	or	29			 211	80					34	or	33		81					36	or	38	or	37		82					43	or	44	or	42	or	45	or	40	or	41		83					Glucocorticoids/		84					glucocorticoid?.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		85					corticosteroid$.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		86					glucocorticoid$.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		87					Administration,	Inhalation/		88					86	or	83	or	85		89					"Nebulizers	and	Vaporizers"/		90					Inhalation	Spacers/		91					Metered	Dose	Inhalers/		92					Aerosols/		93					or/87,89-92		94					88	and	93		95					88		96					Beclomethasone/		97					BECLOMETHASONE.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		98					Betamethasone/		99					BETAMETHASONE.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		100					Budesonide/		101					BUDESONIDE.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		102					Clobetasol/		103					CLOBETASOL.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		104					Dexamethasone/		105					DEXAMETHASONE.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		106					Dexamethasone	Isonicotinate/		107					DEXAMETHASONE	ISONICOTINATE.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		108					Fluprednisolone/			 212	109					FLUPREDNISOLONE.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		110					Melengestrol	Acetate/		111					MELENGESTROL	ACETATE.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		112					Methylprednisolone/		113					METHYLPREDNISOLONE.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		114					Triamcinolone/		115					TRIAMCINOLONE.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		116					or/96-115		117					67	and	82	and	116		118					67	and	82	and	95		119					67	and	82	and	88		120					67	and	82	and	94		121					pulmonary	emphysema$.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		122					47	or	48	or	121	or	49		123					69	or	122	or	82		124					69	or	75	or	122	or	82		125					123	and	67	and	116		126					67	and	124	and	116		127					123	and	67	and	95		128					67	and	95	and	124		129					Chemoprevention/		130					chemoprevention.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		131					chemoprophylaxis.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		132					or/129-131		133					67	and	132		134					123	and	67	and	132		135					132	and	116		136					132	and	95		137					volon.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		138					aristocort.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]			 213	139					metipred.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		140					6-methylprednisolone.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]	141					6	methylprednisolone.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		142					urbason.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		143					medrol.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		144					acetate,	melengestrol.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		145					melengestrol.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		146					isonicotinate,	dexamethasone.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		147					he-111.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		148					he	111.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		149					he111.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		150					auxison.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		151					pulmicort.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		152					budesonide,	s-isomer.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		153					budesonide,	r-isomer.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		154					horacort.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		155					rhinocort.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		156					flubenisolone.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		157					betadexamethasone.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]			 214	158					cellestoderm.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		159					celestone.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		160					celeston.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		161					celestona.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		162					clofenazon.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		163					clobetasol	propionate.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		164					clobetasol	17-propionate.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		165					clobetasol	17	propionate.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		166					clobex.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		167					cormax.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		168					olux.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		169					dermovate.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		170					embeline.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		171					embeline	e.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		172					temovate.mp.	[mp=title,	original	title,	abstract,	name	of	substance	word,	subject	heading	word,	unique	identifier]		173					or/137-172		174					173	or	116		175					123	and	67	and	174		176					67	and	174	and	124		177					67	and	82	and	174		178					132	and	174		179					fluticasone.mp.		180					Flunisolide.mp.		181					179	or	180			 215	182					181	or	174		183					182	and	123	and	67		184					182	and	67	and	124		185					182	and	67	and	82		186					((pulmonary	or	panacinar	or	panlobular	or	centriacinar	or	centrilobular	or	focal)	adj3	emphysema?).tw.		187					186	or	122		188					limit	133	to	(clinical	trial,	all	or	clinical	trial,	phase	i	or	clinical	trial,	phase	ii	or	clinical	trial,	phase	iii	or	clinical	trial,	phase	iv	or	clinical	trial	or	controlled	clinical	trial	or	randomized	controlled	trial)		189					Randomized	Controlled	Trial/		190					Randomized	Controlled	Trials	as	Topic/		191					Random	Allocation/		192					Double-Blind	Method/		193					single-blind	method/		194					clinical	trial/		195					exp	Clinical	Trials	as	Topic/		196					or/189-195		197					(clinic$	adj	trial$1).tw.		198					((singl$	or	doubl$	or	treb$	or	tripl$)	adj	(blind$3	or	mask$3)).tw.		199					Placebos/		200					Placebo$.tw.		201					Randomly	allocated.tw.	202					(allocated	adj2	random).tw.		203					or/197-202		204					196	or	203		205					Case	report.tw.		206					case	reports/		207					letter/		208					historical	article/		209					Review	of	reported	cases.pt.		210					Review,	multicase.pt.		211					or/205-210		212					204	not	211		213					epidemiologic	studies/		214					exp	case	control	studies/		215					exp	cohort	studies/		216					Case	control.tw.		217					(cohort	adj	(study	or	studies)).tw.			 216	218					Cohort	analy$.tw.		219					(Follow	up	adj	(study	or	studies)).tw.		220					(observational	adj	(study	or	studies)).tw.		221					Longitudinal.tw.	222					Retrospective.tw.	223					Cross	sectional.tw.		224					Cross-Sectional	Studies/		225					or/213-224		226					133	and	204		227					133	and	212		228					133	and	225		 		 217	Appendix B  Risk of bias assessment  Sub-appendix:	B1	Summary	table			Random	Sequence	Generation	(selection	bias)	Allocation	Concealment	(selection	bias)	Blinding	of	Participants	and	Personnel	(performance	bias)	Blinding	of	Outcome	Assessment	(detection	bias)	Incomplete	Outcome	Data	(attrition	bias)	Selective	Reporting	(reporting	bias)	Other	bias	   Calverly		 		 		 		 		 		 		 		 		 High	Risk	of	Bias	Kiri	 		 NA	 		 		 		 		 		 		 Unclear	Risk	of	Bias	LHSRG		 		 	 		 		 		 		 		  		 Low	Risk	of	Bias	Parimon		 		 NA	 		 		 		 		 		 NA	 Not	Applicable	Pauwels		 		 		 		 		 		 		 		   Tashkin		 		 		 		 		 		 		 		   	 		 218	Sub-appendix:	B2	Detail	table	of	bias	assessment	(randomized	controlled	trial	studies	and	observational	studies)	(114)		 Random	Sequence	Generation	(selection	bias)	Allocation	Concealment	(selection	bias)	Blinding	of	Participants	and	Personnel	(performance	bias)	Blinding	of	Outcome	Assessment	(detection	bias)	Incomplete	Outcome	Data	(attrition	bias)	 Selective	Reporting	(reporting	bias)	Calverley	(2007)		 Low	Risk	As	stated	in	protocol,	treatment	assignment	generated	via	computer	program	Unclear	As	stated	in	the	Methods	of	this	study,	patients	were	randomly	assigned,	in	permuted	blocks	Low	Risk	Double-blind	study;	outcome	(vital	status)	was	objective	Low	Risk	Double-blind	study;	independent	committee,	whose	members	were	unaware	of	treatment	assignments,	determined	primary	cause-of-death	and	whether	death	was	attributable	to	COPD		Low	Risk	Slightly	more	discontinuations	in	placebo	group	during	the	original	study,	but	vital	status	at	3	years	was	known	for	all	but	one	participant		Low	Risk	Primary	outcome	matches	that	listed	on	clinicaltrials.gov	listing	and	study	protocol	Kiri	(2009)		 Unclear	Not	applicable	–	non-randomized	study	Not	Applicable	Not	applicable	–	non-randomized	study	Low	Risk	Observational	retrospective	study	using	administrative	data:	participants	were	aware	of	their	treatment,	but	were	unaware	of	the	study		Low	Risk	Observational	retrospective	study	using	administrative	data:	lung	cancer	diagnoses	were	made	during	routine	care	by	individuals	unaware	of	the	study												Unclear	Duration	of	treatment	the	same	for	cases	and	controls,	but	data	were	collected	retrospectively	Unclear	No	protocol	available		 219	Sub-appendix:	B2	Detail	table	of	bias	assessment	(randomized	controlled	trial	studies	and	observational	studies)	(114)	(….continued)		 Random	Sequence	Generation	(selection	bias)	Allocation	Concealment	(selection	bias)	Blinding	of	Participants	and	Personnel	(performance	bias)	Blinding	of	Outcome	Assessment	(detection	bias)	Incomplete	Outcome	Data	(attrition	bias)	 Selective	Reporting	(reporting	bias)	Lung	Health	Study	Research	Group	(2000)		Unclear	Text	only	states	that	participants	were	“randomly	assigned”	Unclear	Only	reports	that	participants	were	“randomly	assigned”	Low	Risk	Participants	and	clinical	centre	staff	were	unaware	of	the	study-drug	assignments	Low	Risk	Participants	and	clinical	centre	staff	were	unaware	of	the	study-drug	assignments	Low	Risk	Overall	91%	of	PFTs	and	95%	of	questionnaires	were	completed,	similar	rates	of	satisfactory	adherence	in	treatment	and	placebo	group,	slightly	more	discontinuations	in	placebo	group	(38	vs.	28)		Unclear	No	protocol	available	Parimon	(2007)		Unclear	Not	applicable	–	non-randomized	study	Not	Applicable	Not	applicable	–	non-randomized	study	Low	Risk	Observational	retrospective	study	of	administrative	data;	participants	were	aware	of	their	treatment,	but	unaware	of	the	study	objective												Low	Risk	Observational	retrospective	study	of	administrative	data:	lung	cancer	diagnoses	were	made	during	routine	care	Unclear	Unlikely	to	be	a	problem	since	exposure	and	outcome	data	were	obtained	from	administrative	databases,	but	didn’t	stratify	median	follow-up	time	by	exposure	or	outcome	Unclear	No	protocol	available		 220	Sub-appendix:	B2	Detail	table	of	bias	assessment	(randomized	controlled	trial	studies	and	observational	studies)	(114)	(….continued)		 Random	Sequence	Generation	(selection	bias)	Allocation	Concealment	(selection	bias)	Blinding	of	Participants	and	Personnel	(performance	bias)	Blinding	of	Outcome	Assessment	(detection	bias)	Incomplete	Outcome	Data	(attrition	bias)	 Selective	Reporting	(reporting	bias)	Pauwels	(1999)		 Unclear	Reports	that	subjects	were	“randomly	assigned”	Unclear	Reports	that	subjects	were	“randomly	assigned”	Low	Risk	Double-blind	study;	subjects	in	placebo	group	used	a	dry-powder	inhaler	Low	Risk	Double-blind	study;	primary	outcome	(change	in	FEV1)	was	measured	objectively	through	spirometry	Low	Risk	71%	remained	in	the	study	for	the	whole	3	years,	similar	withdrawal	rates	and	reasons	for	withdrawal	in	treatment	and	control	group	(though	reasons	not	broken	down	by	study	arm)	Unclear	No	protocol	available	Tashkin	(2008)	(32)	 Low	Risk	Computer-generated	randomization	scheme	was	used	at	each	site	Low	Risk	Computer-generated	randomization	scheme	was	used	at	each	site	Low	Risk	Double-blind	study;	all	participants	received	both	types	of	inhalers	containing	either	active	treatment,	placebo,	or	combinations	Low	Risk	Double-blind	study;	primary	outcome	(FEV1)	was	measured	objectively	through	spirometry	Low	Risk	Discontinuation	rates	were	lower	in	the	combination	therapy	groups,	but	rates	in	placebo	and	monotherapy	groups	were	similar	Low	Risk	Co-primary	efficacy	variables	match	those	listed	on	clinicaltrials.gov	listing	  	 		 221	Appendix C  Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Checklist  TITLE   Title  1 Identify the report as a systematic review, meta-analysis, or both.  Title, Pgs 1,2,6 ABSTRACT   Structured summary  2 Provide a structured summary including, as applicable: background; objectives; data sources; study eligibility criteria, participants, and interventions; study appraisal and synthesis methods; results; limitations; conclusions and implications of key findings; systematic review registration number.  Not applicable in the context of a PhD thesis INTRODUCTION   Rationale  3 Describe the rationale for the review in the context of what is already known.  Pgs 47-48 Objectives  4 Provide an explicit statement of questions being addressed with reference to participants, interventions, comparisons, outcomes, and study design (PICOS).  Pg 48 METHODS   Protocol and registration  5 Indicate if a review protocol exists, if and where it can be accessed (e.g., Web address), and, if available, provide registration information including registration number.  Not available Eligibility criteria  6 Specify study characteristics (e.g., PICOS, length of follow-up) and report characteristics (e.g., years considered, language, publication status) used as criteria for eligibility, giving rationale.  Pgs 50-52 Information sources  7 Describe all information sources (e.g., databases with dates of coverage, contact with study authors to identify additional studies) in the search and date last searched.  Pg 50 Search  8 Present full electronic search strategy for at least one database, including any limits used, such that it could be repeated.  Appendix A Study selection  9 State the process for selecting studies (i.e., screening, eligibility, included in systematic review, and, if applicable, included in the meta-analysis).  Pgs 50-52 Data collection process  10 Describe method of data extraction from reports (e.g., piloted forms, independently, in duplicate) and any processes for obtaining and confirming data from investigators.  Pgs 52,53 Data items  11 List and define all variables for which data were sought ( PICOS, funding sources) and any assumptions and simplifications made.  Pgs 52,53 Risk of bias in individual studies  12 Describe methods used for assessing risk of bias of individual studies (including specification of whether this was done at the study or outcome level), and how this information is to be used in any data synthesis.  Pgs 56,59,67 Summary measures  13 State the principal summary measures (e.g., risk ratio, difference in means).  Pg 52 (Tables 3.3,3.4) Synthesis of results  14 Describe the methods of handling data and combining results of studies, if done, including measures of consistency (e.g., I2) for each meta-analysis.  Not applicable 	 222	Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Checklist (…continued) Risk of bias across studies  15 Specify any assessment of risk of bias that may affect the cumulative evidence (e.g., publication bias, selective reporting within studies).  Pgs 56,59,67 Additional analyses  16 Describe methods of additional analyses (e.g., sensitivity or subgroup analyses, meta-regression), if done, indicating which were pre-specified.  Not applicable  RESULTS   Study selection  17 Give numbers of studies screened, assessed for eligibility, and included in the review, with reasons for exclusions at each stage, ideally with a flow diagram.  Pg 56; Figure 3.1 Study characteristics  18 For each study, present characteristics for which data were extracted (e.g., study size, PICOS, follow-up period) and provide the citations.  Tables 3.1,3.2 Risk of bias within studies  19 Present data on risk of bias of each study and, if available, any outcome level assessment (see item 12).  Pg 56 Results of individual studies  20 For all outcomes considered (benefits or harms), present, for each study: (a) simple summary data for each intervention group (b) effect estimates and confidence intervals, ideally with a forest plot.  Pgs 57-65; Tables 3.3,3.4 Synthesis of results  21 Present results of each meta-analysis done, including confidence intervals and measures of consistency.  Not applicable Risk of bias across studies  22 Present results of any assessment of risk of bias across studies (see Item 15).  Pgs 67-69 Additional analysis  23 Give results of additional analyses, if done (e.g., sensitivity or subgroup analyses, meta-regression [see Item 16]).  Not applicable DISCUSSION   Summary of evidence  24 Summarize the main findings including the strength of evidence for each main outcome; consider their relevance to key groups (e.g., healthcare providers, users, and policy makers).  Pgs 67-70 Limitations  25 Discuss limitations at study and outcome level (e.g., risk of bias), and at review-level (e.g., incomplete retrieval of identified research, reporting bias).  Pg 71 Conclusions  26 Provide a general interpretation of the results in the context of other evidence, and implications for future research.  Pg 73 FUNDING   Funding  27 Describe sources of funding for the systematic review and other support (e.g., supply of data); role of funders for the systematic review.  Not applicable; no funding for this study  	

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0354577/manifest

Comment

Related Items