UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Beyond full-time equivalents : gender differences in activity and practice patterns for BC's primary… Hedden, Lindsay Kathleen 2015

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

Item Metadata

Download

Media
24-ubc_2016_february_hedden_lindsay.pdf [ 4.17MB ]
Metadata
JSON: 24-1.0221359.json
JSON-LD: 24-1.0221359-ld.json
RDF/XML (Pretty): 24-1.0221359-rdf.xml
RDF/JSON: 24-1.0221359-rdf.json
Turtle: 24-1.0221359-turtle.txt
N-Triples: 24-1.0221359-rdf-ntriples.txt
Original Record: 24-1.0221359-source.json
Full Text
24-1.0221359-fulltext.txt
Citation
24-1.0221359.ris

Full Text

	BEYOND	FULL-TIME	EQUIVALENTS:	GENDER	DIFFERENCES	IN	ACTIVITY	AND	PRACTICE	PATTERNS	FOR	BC’S	PRIMARY	CARE	PHYSICIANS		by	Lindsay	Kathleen	Hedden	BSc	(hons),	The	University	of	Waterloo,	Canada,	2004	MSc,	The	University	of	British	Columbia,	Canada,	2008					A	THESIS	SUBMITTED	IN	PARTIAL	FULFILLMENT	OF	THE	REQUIREMENTS	FOR	THE	DEGREE	OF		DOCTOR	OF	PHILOSOPHY	in	THE	FACULTY	OF	GRADUATE	AND	POSTDOCTORAL	STUDIES	(Population	and	Public	Health)	THE	UNIVERSITY	OF	BRITISH	COLUMBIA	(Vancouver)		December	2015	©Lindsay	Kathleen	Hedden,	2015		 ii	Abstract	There	is	widespread	sentiment	that	British	Columbia	(BC)	is	facing	a	substantial	shortage	and	a	maldistribution	of	primary	care	physicians	(PCPs).	The	increasing	proportion	of	PCPs	who	are	women	has	been	cited	as	contributing	to	this	complex	problem,	based	largely	on	assumptions	that	female	physicians	work	less,	take	time	off	to	raise	children,	and	retire	earlier	compared	to	their	male	counterparts.		However,	Canadian	evidence	supporting	these	assertions	is	lacking.	This	thesis	uses	population-based,	administrative	data	resources	to	undertake	a	comprehensive	assessment	of	the	potential	impact	of	the	increasing	feminization	of	BC’s	PCP	workforce,	focusing	on	gender	differences	in	career	trajectories,	billing	patterns,	activity,	patient	and	service	mix,	and	scopes	of	practice.		It	consists	of	four	components:	a	systematic	review	of	existing	literature;	a	longitudinal	analysis	of	gender-specific	remuneration	and	activity;	a	comparison	of	patient	populations	seen	by	male	and	female	PCPs;	and	an	examination	of	gender-driven	differences	in	selected	aspects	of	primary	care	practice.				 The	results	of	these	analyses	suggest	that	female	PCPs	have	lower	total	remuneration,	see	fewer	patients,	and	deliver	fewer	services	annually	compared	to	male	PCPs;	however,	this	gender-related	activity	gap	is	narrowing	over	time	as	male	physicians	are	reducing	their	activity	levels.		Female	physicians	derive	more	of	their	income	from	direct	clinical	care	delivery	(rather	than	from	clinical	or	non-clinical	incentive	payments)	when	compared	to	male	physicians.	The	proportion	of	physicians’	total	compensation	derived	from	direct	care	delivery	is	declining,	particularly	for	male	physicians.			 iii	Results	also	show	that	female	physicians	have	smaller	overall	practices	and	disproportionately	treat	younger,	healthier	patients.	They	are	less	likely	to	provide	off-site	and	after-hours	care,	but	more	likely	to	include	obstetrical	care	in	their	practices.		This	thesis	demonstrates	that	gender	differences	in	primary	care	practice	go	beyond	salary	and	service	volumes.	More	robust	measures	of	physician	supply	that	address	the	implications	of	gender	differences	in	patient	mix,	service	mix,	and	practice	style	need	to	be	developed	as	more	evidence	in	these	areas	becomes	available.		The	gender	division	of	unpaid	care	work,	household	responsibilities,	and	work-life	balance,	and	the	implications	for	health	human	resource	planning	all	deserve	careful	attention	in	future	work.			 		 iv	Preface	This	thesis	is	my	original	and	independent	work.	All	projects	and	associated	methods	were	approved	by	the	UBC	Behavioural	Ethics	Board	(Certificate	number	H11-03367).	I	was	the	lead	investigator	for	the	projects	outlined	in	Chapter	3,	and	conducted	in	Chapters	2,	4,	5,	6	and	7,	responsible	for	design,	data	analysis	and	interpretation,	and	manuscript	(and	chapter)	composition	and	revision.		Sandra	Peterson	and	Kerry	Kerluke	performed	the	initial	dataset	preparation.	Morris	Barer	provided	general	oversight	on	the	completion	of	all	projects.	Morris	Barer,	Kimberlyn	McGrail,	Michael	Law	and	Ivy	Bourgeault	provided	feedback	on	the	objectives	and	analysis	strategy,	and	provided	feedback	and	edits	on	the	chapter	and	manuscript	drafts.			A version of Chapter 2 has been published: Hedden,	L.,	Barer,	M.	L.,	Cardiff,	K.,	McGrail,	K.	M.,	Law,	M.	R.,	&	Bourgeault,	I.	L.	(2014).	The	implications	of	the	feminization	of	the	primary	care	physician	workforce	on	service	supply:	a	systematic	review.	Human	Resources	for	Health,	12(1),	32.	 Some	sections	of	Chapter	8	have	been	published	as	an	editorial	in	several	media	outlets	through	Evidence	Network.	For	example:			• Hedden,	L.,	&	Barer.	(2014a,	September	8).	Sex,	Lies	and	Physician	Supply.	Ottawa	Life	Magazine.	Ottawa	ON.	Retrieved	on	August	27,	2015	from	http://www.ottawalife.com/2014/09/sex-lies-and-physician-supply-why-female-doctors-are-not-to-blame/					 v	• Hedden,	L.,	&	Barer,	M.	(2014b,	September	7).	Sex,	Lies	and	Physician	Supply:	Why	Female	Doctors	Are	Not	to	Blame.	The	Huffington	Post.	Retrieved	on	August	27,	2015	from	http://www.huffingtonpost.ca/lindsay-hedden/female-doctors-canada_b_5775298.html		• Hedden,	L.,	&	Barer,	M.	(2014c,	September	7).	Sexe,	mensonge	et	effectif	des	médecins.	Le	Huffington	Post	Quebec.	Retrieved	on	August	27,	2015	from	http://quebec.huffingtonpost.ca/morris-barer/sexe-mensonge-et-effectif-des-medecins_b_5754654.html		• Hedden,	L.,	&	Barer,	M.	L.	(2014d,	August	26).	Female	physicians	not	to	blame	for	Canada’s	doctor	shortage.	Guelph	Mercury.	Guelph	ON.	Retrieved	on	August	27,	2015	from	http://www.guelphmercury.com/opinion-story/4777752-female-physicians-not-to-blame-for-canada-s-doctor-shortage/			• Hedden,	L.,	&	Barer,	M.	L.	(2014e,	August	30).	Comment:	Women	not	to	blame	for	doctor	shortage.	Victoria	Times	Colonist.	Victoria	BC.	Retrieved	on	August	27,	2015	from	http://www.timescolonist.com/opinion/op-ed/comment-women-not-to-blame-for-doctor-shortage-1.1334924			 				 		 vi	Table	of	contents		Abstract	..........................................................................................................................................................	ii	Preface	...........................................................................................................................................................	iv	Table	of	contents	.......................................................................................................................................	vi	List	of	tables	................................................................................................................................................	xi	List	of	figures	...........................................................................................................................................	xiii	List	of	abbreviations	..............................................................................................................................	xiv	Acknowledgements	................................................................................................................................	xvi	CHAPTER	1:	Introduction	......................................................................................................................	1	1.1	Core	terminology	...........................................................................................................................	3	1.1.1	Gender	and	sex	.......................................................................................................................	3	1.1.2	Practice	specialty	..................................................................................................................	5	1.2	Primary	care	physician	remuneration	in	British	Columbia	........................................	6	1.2.1	Payments	for	clinical	care	.................................................................................................	6	1.2.2	Incentives	and	non-clinical	payments	.........................................................................	8	1.2.3	Current	pattern	of	physician	payments	......................................................................	9	1.3	The	feminization	of	British	Columbia’s	primary	care	physician	workforce	.....	11	1.3.1	Physician	supply	.................................................................................................................	11	1.3.2	Medical	students	and	residents	...................................................................................	13	1.4	Health	human	resources	planning	......................................................................................	15	1.4.1	Gender	and	health	human	resources	policy	and	planning	..............................	17	1.4.2	Conceptual	model	..............................................................................................................	19	1.5	Research	objectives	and	hypotheses	.................................................................................	27	1.6	Thesis	roadmap	...........................................................................................................................	29	1.7	Publication	of	thesis	chapters	...............................................................................................	31	CHAPTER	2:	Systematic	literature	review	...................................................................................	32	2.1	Methodology	.................................................................................................................................	33	2.1.1	Search	strategy	and	inclusion	criteria	......................................................................	33	2.1.2	Data	abstraction	and	article	typology	.......................................................................	34	2.2	Results	.............................................................................................................................................	37	2.2.1	Search	results	......................................................................................................................	37	2.2.2	Thematic	results	.................................................................................................................	40		 vii	2.3	Discussion	......................................................................................................................................	50	2.3.1	Consistency	of	results	......................................................................................................	52	2.3.2	Quality	of	included	studies	............................................................................................	54	2.3.3	Remaining	knowledge	gaps	and	future	research	.................................................	55	2.3.4	Limitations	............................................................................................................................	57	2.3.5	Implications	for	health	human	resource	planners	..............................................	58	2.4	Conclusions	...................................................................................................................................	58	CHAPTER	3:	Data	sources,	file	preparation,	and	study	variables	......................................	60	3.1	Outline	of	data	sources	and	linkage	procedures	..........................................................	60	3.2	Datasets	..........................................................................................................................................	62	3.2.1	MSP	Practitioner	File	and	CPSBC	Registry	..............................................................	62	3.2.2	MSP	Consolidation	File	....................................................................................................	63	3.2.3	MSP	Payment	Information	File	....................................................................................	64	3.2.4	APP	Database	.......................................................................................................................	65	3.2.5	Discharge	Abstracts	Database	and	Vital	Statistics	..............................................	66	3.3	Analytic	file	creation	.................................................................................................................	67	3.4	Creation	of	physician	cohort	.................................................................................................	68	3.5	Variables	and	indicators	..........................................................................................................	69	3.5.1	Physician	characteristics	................................................................................................	71	3.5.2	Physician	activity	...............................................................................................................	72	3.5.3	Patient	characteristics	.....................................................................................................	76	3.5.4	Patient	care	approaches	..................................................................................................	78	CHAPTER	4:	Remuneration	and	Activity	......................................................................................	81	4.1	Introduction	..................................................................................................................................	81	4.2	Methods,	variables,	and	data	sources	................................................................................	83	4.2.1	Dependent	variables	.........................................................................................................	84	4.2.2	Explanatory	variables	......................................................................................................	84	4.2.3	Statistical	analyses	............................................................................................................	85	4.2.4	Some	notes	on	model	interpretation	.........................................................................	92	4.3	Results	.............................................................................................................................................	94	4.3.1	Cohort	description	and	demographics	.....................................................................	94	4.3.2	Unadjusted	payments	and	activity	.............................................................................	97	4.3.3	Question	1.1:	Multivariate	results	............................................................................	101	4.3.4	Question	1.2:	Multivariate	results	............................................................................	109		 viii	4.4	Discussion	....................................................................................................................................	113	4.4.1	Limitations	..........................................................................................................................	116	4.4.2	Implications	........................................................................................................................	118	CHAPTER	5:	Clinical	activities	and	incentives	.........................................................................	119	5.1	Introduction	................................................................................................................................	119	5.2	Methods,	variables,	and	data	sources	..............................................................................	122	5.2.1	Dependent	variables	.......................................................................................................	122	5.2.2	Explanatory	variables	....................................................................................................	123	5.2.3	Statistical	analyses	..........................................................................................................	124	5.3	Results	...........................................................................................................................................	134	5.3.1	Descriptive	statistics	and	bivariate	results	..........................................................	134	5.3.2	Question	2.1:	Multivariate	results	............................................................................	140	5.3.3	Question	2.2:	Multivariate	results	............................................................................	142	5.3.4	Question	2.3:	Multivariate	results	............................................................................	146	5.3.5	Question	2.4:	Multivariate	results	............................................................................	150	5.4	Discussion	....................................................................................................................................	152	5.4.1	Limitations	..........................................................................................................................	155	5.4.2	Implications	........................................................................................................................	157	CHAPTER	6:	Patient	characteristics	.............................................................................................	158	6.1	Introduction	................................................................................................................................	158	6.2	Methods	........................................................................................................................................	160	6.2.1	Dependent	variables	.......................................................................................................	161	6.2.2	Explanatory	variables	....................................................................................................	162	6.2.3	Statistical	analyses	..........................................................................................................	164	6.3	Results	...........................................................................................................................................	171	6.3.1	Descriptive	statistics	and	bivariate	results	..........................................................	172	6.3.2	Question	3.1:	Multivariate	results	............................................................................	177	6.3.3	Question	3.2:	Multivariate	results	............................................................................	181	6.3.4	Question	3.3:	Multivariate	results	............................................................................	184	6.4	Discussion	....................................................................................................................................	191	6.4.1	Limitations	..........................................................................................................................	195	6.4.2	Implications	........................................................................................................................	197	CHAPTER	7:	Practice	patterns	........................................................................................................	199	7.1	Introduction	................................................................................................................................	199		 ix	7.1.1	Clinical	scope	.....................................................................................................................	200	7.1.2	Referrals	...............................................................................................................................	201	7.1.3	Accessibility:	off-site	and	after-hours	care	delivery	.........................................	202	7.2	Methods	........................................................................................................................................	204	7.2.1	Dependent	variables	.......................................................................................................	205	7.2.2	Explanatory	variables	....................................................................................................	207	7.2.3	Statistical	analyses	..........................................................................................................	208	7.3	Results	...........................................................................................................................................	225	7.3.1	Descriptive	statistics	and	bivariate	results	..........................................................	225	7.3.2	Question	4.1:	Multivariate	results	............................................................................	238	7.3.3	Question	4.2:	Multivariate	results	............................................................................	241	7.3.4	Questions	4.3	and	4.4:	Multivariate	results	..........................................................	255	7.3.5	Question	4.5:	Multivariate	results	............................................................................	264	7.4	Discussion	....................................................................................................................................	266	7.4.1	Referral	patterns	..............................................................................................................	267	7.4.2	Out-of-office	and	after-hours	care	provision	.......................................................	268	7.4.3	Obstetrics	............................................................................................................................	270	7.4.4	Mental	health	.....................................................................................................................	271	7.4.5	Patient	contact	patterns	................................................................................................	273	7.4.6	Cross-cutting	trends	.......................................................................................................	274	7.4.7	Limitations	..........................................................................................................................	275	7.4.8	Implications	........................................................................................................................	277	CHAPTER	8:	Conclusions	...................................................................................................................	278	8.1	Summary	of	results	and	contributions	...........................................................................	279	8.1.1	Feminization	of	PCP	workforce:	Systematic	review	.........................................	279	8.1.2	Remuneration	and	activity	..........................................................................................	280	8.1.3	Clinical	payments,	clinical	incentives,	and	non-clinical	incentives	............	283	8.1.4	Patient	characteristics	...................................................................................................	285	8.1.5	Practice	patterns	..............................................................................................................	286	8.2	Beyond	physician	gender:	Cross-cutting	trends	.........................................................	290	8.2.1	Challenges	related	to	changes	in	accessibility	and	comprehensiveness	of	primary	care	..................................................................................................................................	290	8.2.2	Internationally-trained	physicians	...........................................................................	292	8.2.3	Patient	socioeconomic	status	.....................................................................................	295		 x	8.3	Work,	family	and	HHR	planning	........................................................................................	297	8.3.1	Family	responsibilities	..................................................................................................	297	8.3.2	HHR	policy	and	planning:	The	importance	of	gender	......................................	300	8.4	Strengths	and	limitations	......................................................................................................	304	8.5	Recommendations	for	further	inquiry	............................................................................	309	8.6	Recommendations	for	HHR	planning	and	policy	action	.........................................	311	8.7	Final	reflections	........................................................................................................................	316	References	...............................................................................................................................................	318	Appendix	1:	Systematic	review	search	strategies	..................................................................	332	A1.1	Medline	search	strategy	.....................................................................................................	332	A1.2	Embase	search	strategy	......................................................................................................	334	A1.3	Web	of	Science	........................................................................................................................	336	Appendix	2:	Summary	of	included	studies	................................................................................	337	Appendix	3:	Alternative	Payment	Program	payment	types	..............................................	354	Appendix	4:	Variables	and	indicators	..........................................................................................	357	Appendix	5:	Computation	of	Adjusted	Clinical	Groups	........................................................	361			 		 xi	List	of	tables	Table	3.1:	Data	Sources	and	Relevant	Outputs	........................................................................	62	Table	3.2:	Variable	Matrix	.................................................................................................................	70	Table	3.3:	Practice	rurality	................................................................................................................	71	Table	3.4:	Fee	items	in	clinical	and	non-clinical	payment	categories	............................	75	Table	4.1:	Cohort	demographics	by	physician	gender	..........................................................	96	Table	4.2:	Physician	payments	and	activity	averaged	across	all	study	years	.............	98	Table	4.3:	Mean	physician	activity,	by	demographic	characteristics	(for	2011/12)	.............................................................................................................................................................	100	Table	4.4:	Multivariate	modeling	results:	total	compensation,	patient	contacts,	services	and	unique	patients	.................................................................................................	102	Table	4.5:	Multivariate	modeling	results:	binary	APP	uptake	.........................................	107	Table	4.6:	Multivariate	results:	ordinal	logistic	APP	uptake	............................................	109	Table	5.1:	Proportion	of	physicians	who	billed	or	any	clinical	or	non-clinical	incentives,	by	gender,	2005-2012	.......................................................................................	134	Table	5.2:	Payment	types	as	a	percentage	($)	of	total	compensation,	by	gender,	averaged	across	2005-2012	...................................................................................................	135	Table	5.3:	Number	(%)	of	male	and	female	physicians	who	billed	for	time	on	call	or	rural	and	remote	incentives,	2005-2012	..........................................................................	135	Table	5.4:		Primary	care	physician	payments,	by	demographic	characteristics	(for	2011/12)	........................................................................................................................................	137	Table	5.5:	Physician	payments,	by	demographic	characteristics	(for	2011/12)	....	140	Table	5.6:	Multivariate	logit-normal	results:	Percent	clinical	income	.........................	141	Table	5.7:	Multivariate	binary	results:	clinical	and	non-clinical	incentives	..............	143	Table	5.8:	Multivariate	logit-normal	results:	Clinical	and	non-clinical	incentives	.	145	Table	5.9:	Multivariate	binary	results:	on-call	payments	and	rural	and	remote	incentives	........................................................................................................................................	151	Table	6.1:	Patient	population	characteristics,	averaged	across	all	study	years	by	physician	gender	.........................................................................................................................	173	Table	6.2:	Patient	population	demographics	(weighted	by	contacts),	by	physician	demographics	(2011/12)	........................................................................................................	175	Table	6.3:	Patient	population	morbidity,	by	physician	demographics	(2011/12)	.	176	Table	6.4:	Multivariate	results	for	patient	demographics	.................................................	178	Table	6.5:	Multivariate	results	for	patient	morbidity	.........................................................	182	Table	6.6:	Multivariate	results:	Proportion	of	income	from	direct	clinical	care	.....	185		 xii	Table	6.7:	Multivariate	results:	binary	clinical	and	non-clinical	incentives	(patient	characteristics	included)	.........................................................................................................	187	Table	6.8:	Multivariate	results:	proportion	clinical	and	non-clinical	incentives	(patient	characteristics	included)	........................................................................................	189	Table	7.1:	Unadjusted	referral	patterns	averaged	across	all	study	years	by	gender	.............................................................................................................................................................	226	Table	7.2:	Out-of-office	and	after-hours	provision	averaged	across	all	study	years	by	gender	........................................................................................................................................	227	Table	7.3:	Clinical	scope	(obstetrics	and	mental	health)	averaged	across	all	study	years	by	gender	............................................................................................................................	227	Table	7.4:	Patient	contact	frequency	averaged	across	all	study	years	by	physician	gender	..............................................................................................................................................	228	Table	7.5:	Unadjusted	referral	patterns	(rates)	by	physician	demographics	(for	2011/12)	........................................................................................................................................	230	Table	7.6:	Percentage	of	physicians	who	provided	out-of-office	care	by	physician	demographics	(for	2001/12)	.................................................................................................	232	Table	7.7:	After-hours	incentives	and	phone	consultations	(for	2011/12)	..............	233	Table	7.8:	Practice	areas	by	physician	demographics	(for	2011/12)	..........................	235	Table	7.9:	Frequency	of	patient	contacts	by	physician	demographics	(for	2011/12)	.............................................................................................................................................................	237	Table	7.10:	Multivariate	models	for	referrals	per	contact	................................................	239	Table	7.11:	Multivariate	models	for	care	provision	in-office-only	................................	242	Table	7.12:	Multivariate	models	for	the	odds	of	provided	care	outside	of	office	...	243	Table	7.13:	Multivariate	model	for	the	proportion	of	care	provided	out-of-office	249	Table	7.14:	Multivariate	models	for	the	odds	of	providing	off-hours	care	or	telephone	consultations	...........................................................................................................	251	Table	7.15:	Multivariate	models	for	the	odds	of	providing	obstetrical	care	.............	256	Table	7.16:	Multivariate	models	for	the	proportion	of	billings	related	to	obstetrical	care	....................................................................................................................................................	259	Table	7.17:	Multivariate	results	for	the	proportion	of	patient	contacts	for	which	mental	health	incentives	or	counselling	visit	were	billed	.........................................	262	Table	7.18:	Multivariate	Results	for	Frequency	of	Patient	Contacts	............................	265				 		 xiii	List	of	figures	Figure	1.1:	Method	of	remuneration	for	physicians	in	British	Columbia		....................	10	Figure	1.2:	Number	of	primary	care	physicians	in	Canada	by	gender,	1998-2013	.	12	Figure	1.3:	Proportion	of	primary	care	physicians	in	BC	health	service	delivery	areas	who	are	female	...................................................................................................................	13	Figure	1.4:	Enrolment	in	Canadian	medical	schools	by	gender,	1968-2011	..............	14	Figure	1.5:	Proportion	of	new	Canadian	residents	choosing	family	medicine	and	other	specialties,	by	gender,	2002-2012	............................................................................	15	Figure	1.6:	Health	Human	Resources	Conceptual	Model	.....................................................	20	Figure	3.1:	Selection	of	physician	cohort	....................................................................................	69	Figure	3.2:	Physician	compensation,	divided	into	clinical	and	non-clinical	payments	...............................................................................................................................................................	74	Figure	4.1:	Five-year	multivariate	GLMs,	total	compensation	........................................	110	Figure	4.2:	Five-year	multivariate	GLMs,	patient	contacts	...............................................	111	Figure	4.3:	Five-year	multivariate	GLMs,	services	...............................................................	112	Figure	4.4:	Five-year	multivariate	GLMs,	unique	patient	counts	...................................	112	Figure	5.1:	Annual	fixed	effects	GLMs	for	percent	clinical	care	delivery	....................	147	Figure	5.2:	Annual	fixed	effects	GLMs	for	percent	clinical	incentives	..........................	148	Figure	5.3:	Annual	fixed	effects	GLMs	for	percent	non-clinical	incentives	................	149	Figure	7.1:	Percent	of	physicians	who	billed	for	one	or	more	home	visits,	by	year	.............................................................................................................................................................	245	Figure	7.2:	Percent	of	physicians	who	billed	for	one	or	more	LTC	visits,	by	year	..	246	Figure	7.3:	Percent	of	physicians	who	billed	for	one	or	more	hospital	visits,	by	year	.............................................................................................................................................................	247	Figure	7.4:	Percent	of	physicians	who	billed	for	one	or	more	ER	visits,	by	year	....	247	Figure	7.5:	Percent	of	physicians	who	billed	for	one	or	more	off-hours	incentives,	by	year	.............................................................................................................................................	252	Figure	7.6:	Percent	of	physicians	who	billed	for	one	or	more	telephone	consultations,	by	year	...............................................................................................................	254	Figure	7.7:	Percent	of	physicians	who	billed	for	one	or	more	deliveries,	by	year	..	257		 xiv	List	of	abbreviations	ACG	 Adjusted	Clinical	Group	ADG	 Aggregated	Diagnosis	Group	AIC	 Akaike’s	Information	Criterion	APP	 Alternative	Payment	Program	BC	 British	Columbia	BCMA	 British	Columbia	Medical	Association	BIC	 Bayesian	Information	Criterion	CCHS	 Canadian	Community	Health	Survey	CPSBC	 College	of	Physicians	and	Surgeons	of	British	Columbia	DAD	 Discharge	Abstract	Database	FFS	 Fee	for	Service		FTE	 Full-Time	Equivalents	GLM	 Generalized	linear	model	GP	 General	Practitioner	GPSC	 General	Practice	Services	Committee	HA	 Health	authority		HHR	 Health	Human	Resources	HSDA	 Health	Services	Delivery	Area	ICD	 International	Classification	of	Diseases	LHA	 Local	Health	Area	MOCAP	 Medical	On-Call	Availability	Program		 xv	MSC	 Medical	Services	Commission	MSP		 Medical	Services	Plan	PCP	 Primary	Care	Physician	OECD	 Organization	for	Economic	Co-operation	and	Development	SAS	 Statistical	Analysis	System			SES	 Socioeconomic	Status	UK	 United	Kingdom	US	 United	States			 		 xvi	Acknowledgements	First	and	foremost,	I	would	like	to	express	my	utmost	appreciation	to	my	supervisor,	Dr.	Morris	Barer,	for	his	guidance	and	support,	and	his	detail-oriented	and	thoughtful	feedback.	I	am	also	grateful	for	the	support	of	my	other	committee	members,	Drs.	Kimberlyn	McGrail,	Michael	Law,	and	Ivy	Bourgeault,	all	of	whom	were	generous	with	their	time	and	provided	me	with	thoughtful	input.	I	would	like	to	thank	Dr.	McGrail,	in	particular,	for	supporting	me	through	a	difficult	few	months	following	an	unexpected	problem	with	my	data	preparation.			 I	would	like	to	thank	my	friends	and	colleagues	at	the	Centre	for	Health	Services	and	Policy	Research,	with	whom	I	shared	ideas,	successes	and	frustrations.	Ruth	Lavergne,	Dimi	Panagiotoglou,	Saskia	Sivananthan	and	Scally	Chu	provided	much-needed	help	with	the	seemly	endless	stream	of	statistical	problems	I	stumbled	through.	Sincere	thanks	also	to	Megan	Engelhardt	for	being	a	wonderful	friend	and	colleague	during	the	last	years	of	my	program,	and	for	listening	patiently	and	without	judgment	when	I	had	yet	another	thesis-related	rant	to	share.								 I	want	to	also	acknowledge	the	contributions	made	by	Sandra	Peterson,	who	prepared	my	datasets	at	lightning	speed;	Rachael	McKendry,	who	helped	me	to	prepare	my	Data	Access	Request;	Dawn	Mooney,	who	made	me	beautiful	maps	and	posters	to	display	my	work;	and	Dr.	Rick	White,	who	provided	guidance	and	support	in	the	development	of	my	statistical	analysis	plan,	and	helped	me	to	find	solutions	some	difficult	statistical	problems.		I	wish	to	extend	my	gratitude	to	the	organizations	that	financially	supported	me	and	my	research:	The	Canadian	Institutes	of	Health	Research;	the		 xvii	Transdisciplinary	Understanding	and	Training	on	Research	in	Primary	Health	Care	program;	and	the	Western	Regional	Training	Centre.			 Finally,	I	would	like	to	thank	my	family	and	friends	for	their	love,	encouragement	and	support	especially	during	the	last	two	years	of	this	project.	Thank	you	for	putting	up	with	my	withdrawal	from	many	social	activities	over	the	last	years	when	my	thesis	had	to	be	the	priority.	A	special	thank	you	to	my	cousin	Jen,	whose	stories	about	her	own	experiences	as	a	graduate	student	made	me	realize	that	my	challenges	were	not	mine	alone,	and	that	they	were	all	surmountable	with	some	hard	work	and	a	little	humour.		To	Robin,	thank	you	for	being	the	most	supportive	partner	I	could	imagine	through	what	has	been	a	very	difficult	journey	for	both	of	us.	Thank	you	for	keeping	me	on	track,	when	“on	track”	was	the	last	place	I	wanted	to	be.	Your	unspoken	faith	in	me	was	frequently	the	only	thing	that	kept	me	from	giving	up.	I	am	so	excited	that	we	are	both	finally	at	the	end	of	this	“adventure”,	and	I	can’t	wait	to	see	what	the	next	stage	of	our	lives	will	bring.						 1	CHAPTER	1:	Introduction		 There	is	widespread	sentiment	that	British	Columbia	(BC),	and	indeed	Canada	as	a	whole,	is	facing	both	a	substantial	shortage	and	a	maldistribution	of	primary	care	physicians	(Hildebrand,	2013;	Luk,	2013;	Nicholson	&	Levy,	2009;	Snadden,	2008;	Tyrrell,	Dauphinee,	&	Canadian	Medical	Forum	Task	Force,	1999).	The	situation	has	been	referred	to	as	a	“crisis”	(Nicholson	&	Levy,	2009)	that	is	“crippling	the	healthcare	infrastructure	in	Canada”	(Chew,	Amirthalingam,	Firoz,	Goyal,	&	Singh,	2013).	The	increasing	proportion	of	the	physician	workforce	who	are	women	–	referred	to	as	workforce	“feminization”	–	has	been	commonly	cited	as	a	source	of	this	supply	problem	under	the	assumption	that	female	physicians	are	more	likely	engage	in	part-time	work	arrangements	(see	for	example	Esmail,	2007).	A	maldistribution	of	physicians	could	be	driven	by	factors	on	either	the	supply	side	–	for	example,	direct	reductions	in	the	physician	to	population	ratio,	or	changes	in	physicians’	number	and	mix	of	activities	–	or	on	the	need/demand	side	–	for	example,	increases	in	population	need	for	health	care	services.		One	commonly	cited	driver	of	increased	need	for,	and	use	of,	primary	care	services	is	the	aging	of	the	baby-boomer	generation.	Recent	studies	have	demonstrated,	however,	that	the	aging	population	is	not	generally	the	largest	predictor	of	increasing	expenditures,	and	has	a	reasonably	predictable	impact	on	health	services	use	(McGrail,	Evans,	Barer,	Kerluke,	&	McKendry,	2011).	On	the	supply	side,	BC’s	per	capita	supply	of	primary	care	physicians	has	been	steadily	increasing	since	1986,	rising	from	100	to	123	per	100,000	population		 2	(Canadian	Institute	for	Health	Information,	2013).	Given	that	there	has	not	been	any	reduction	in	the	physician	to	population	ratio,	any	shortages	resulting	from	supply-side	deficiencies	would	need	to	be	the	result	of	some	combination	of	changes	in	physicians’	activity	mix	(number	of	patients	seen	or	numbers	of	services	delivered,	for	example)	or	practice	patterns	(scope	of	practice	or	specific	service	delivery)	which,	in	turn,	may	be	driven	by	changes	in	the	physician	workforce	demographic.		An	important	demographic	shift,	and	the	subject	of	this	thesis,	is	the	growing	representation	of	women	in	the	primary	care	physician	workforce.		Female	physicians,	on	average,	work	fewer	hours	and	deliver	fewer	services	than	their	male	counterparts,	thus	potentially	reducing	effective	service	supply	(Crossley,	Hurley,	&	Jeon,	2009;	Watson,	Slade,	Buske,	&	Tepper,	2006;	Weizblit,	Noble,	&	Baerlocher,	2009).	If	the	proportion	of	physicians	who	are	women	were	to	continue	to	grow,	more	physicians	could	be	needed	in	order	to	maintain	current	levels	of	service	provision	(Esmail,	2007).		There	are	a	number	of	problems	with	the	argument	that	workforce	feminization	will	result	in	service	shortages.		First,	the	proportion	of	physicians	who	are	women	will	not	continue	to	grow	forever;	the	proportion	of	female	physicians	in	Canadian	medical	schools	appears	to	have	leveled	off,	hovering	between	58%	and	60%	since	2004	(The	Association	of	Faculties	of	Medicine	of	Canada,	2013).	Second,	there	are	many	assumptions	but	little	empirical	data	that	assess	the	differences	in	how	male	and	female	physicians	practice,	and	the	effects	of	those	practice	differences	on	patterns	of	use	and,	ultimately,	on	health	outcomes	for	patients.		Existing	literature	tends	to	focus	heavily	on	self-reported	time	spent	working		 3	(which	itself	is	differentially	reported	by	men	and	women),	without	looking	more	deeply	into	actual	activities,	patient	and	service	mix	and	patterns	of	service	delivery.	Examinations	of	work	trajectories,	including	leaves	of	absence,	pre-retirement	and	retirement,	are	virtually	non-existent.	Additionally,	much	of	the	work	in	this	area	has	been	based	on	retrospective	surveys,	frequently	plagued	by	small,	unbalanced	samples,	which	create	risk	of	recall	and	selection	biases.		The	purpose	of	this	thesis	is	to	undertake	a	comprehensive	empirical	assessment	of	the	potential	impact	of	the	feminization	of	BC’s	primary	health	care	physician	workforce	by	examining	gender-driven	differences	in	career	trajectories,	billing	patterns	(for	clinical	and	non-clinical	services,	incentives	and	on-call	payments),	activity,	patient	and	service	mix,	and	scopes	of	practice.			In	this	Chapter,	I	define	the	core	terminology	that	will	be	used	throughout	the	thesis,	describe	the	extent	of	the	feminization	of	BC’s	primary	care	physician	workforce,	and	outline	a	conceptual	model	that	informs	my	subsequent	review	of	the	literature	and	analytic	work.	I	conclude	with	a	roadmap	of	the	remaining	thesis	chapters	and	an	outline	of	my	main	objectives	and	hypotheses.			1.1	Core	terminology	1.1.1	Gender	and	sex	There	is	a	complex	and	lengthy	literature	that	attempts	to	define	and	disentangle	sex	and	gender	terminology.	My	definitions	of	these	terms	are	intended	not	to	attempt	to	contribute	to	that	discussion,	but	solely	to	clarify	how	the	terms	are	used	in	the	thesis.				 4			 “Sex”	refers	to	the	biological	and	physiological	characteristics	of	being	male	or	female	(Brannon,	2010;	Deaux,	1985).	“Gender”,	in	contrast,	refers	to	the	psychological,	socially-	and	societally-constructed	roles,	behaviors,	attitudes,	and	attributes	that	are	commonly	associated	with	being	either	a	man	or	a	woman	(Brannon,	2010;	CIHR	Institute	of	Gender	and	Health,	2012;	Deaux,	1985).	The	effects	of	gender	and	sex	are	frequently	difficult,	if	not	impossible	to	disentangle,	and	many	agencies	refer	to	their	combined	effects	under	an	umbrella	term	such	as	sex/gender	(CIHR	Institute	of	Gender	and	Health,	2012).		In	this	thesis,	I	examine	the	combined	effects	of	the	biological	characteristics	(sex)	and	socially-constructed	roles	(gender)	on	activity,	patient	and	service	mix,	and	scope	of	practice.	For	example,	maternity	leaves	–	temporary	leaves	of	absence	or	reductions	in	activity	level	for	the	purpose	of	child	birth	and/or	child	care	–	are	a	function	of	both	the	biological	capacity	for	pregnancy	and	birth	(sex),	but	also	the	greater	responsibility	for	childcare	that	is	still	placed	upon	women	(gender)	(Brannon,	2010;	Deaux,	1985).		I	make	no	attempts	to	disentangle	the	two	since,	in	most	cases,	it	would	be	impossible	to	do	so.	Also,	even	in	cases	where	a	sex	effect	and	gender	effect	would	be	distinguishable	from	each	other,	attempts	to	disentangle	those	effects	would	be	beyond	the	scope	of	this	work,	would	not	possible	using	administrative	data,	and	are	not	relevant	from	a	health	human	resources	policy	perspective.	Although	the	effects	I	am	reporting	are	a	function	of	both	sex	(biology)	and	gender	(social	construction),	for	simplicity,	I	refer	these	effects	consistently	as	“gender	effects”.			 5	1.1.2	Practice	specialty		This	thesis	focuses	on	physicians	who	practice	primary	care,	which	is	defined	by	Health	Canada	as	follows:			“Primary	care	…	[includes]	health	promotion,	illness	and	injury	prevention,	and	the	diagnosis	and	treatment	of	illness	and	injury.	Primary	care	serves	a	dual	function	in	the	health	care	system:		• direct	provision	of	first-contact	services	(by	providers	such	as	family	physicians,	nurse	practitioners,	pharmacists,	and	telephone	advice	lines);	and	• a	coordination	function	to	ensure	continuity	and	ease	of	movement	across	the	system,	so	that	care	remains	integrated	when	Canadians	require	more	specialized	services	(with	specialists	or	in	hospitals,	for	example)	(Health	Policy	Branch	Health	Canada,	2005)” Primary	care	physicians	(PCPs)	are	often	referred	to	interchangeably	as	general	practitioners,	family	physicians,	or	family	medicine	specialists.	The	term	“general	practitioner”	is	typically	used	to	describe	a	physician	who	completed	a	generalized	one-year	junior	rotating	internship	under	training	requirements	in	Canada	pre-1992,	after	which	the	internship	was	replaced	with	a	two-year	family	medicine	specialty	residency	program.	The	term	is	no	longer	used	to	describe	new	physicians;	family	medicine	appears	to	be	the	mostly	commonly	used	replacement	term.		Family	medicine	specialists	will	have	completed	a	specific	accredited	two-year	family	medicine	residency	program	in	Canada	(or	similar	training	program	in	a	jurisdiction	approved	by	the	College	of	Family	Physicians	of	Canada).	Both	general	practitioners	and	family	medicine	specialists	provide	primary	care	in	BC	according	to	the	Health	Canada	definition	above.		 6	Henceforth,	and	for	the	purposes	of	my	own	empiric	work,	when	I	refer	to	PCPs,	I	am	referring	to	both	general	practitioners	and	those	who	have	the	family	medicine	specialty	designation.		1.2	Primary	care	physician	remuneration	in	British	Columbia	1.2.1	Payments	for	clinical	care	1.2.1.1	Fee-for-service	Since	the	introduction	of	publicly-funded	insurance	for	physician	services	across	Canada,	the	majority	of	BC’s	primary	care	physicians	have	been	paid	using	fee-for-service	(FFS)	compensation	arrangements,	whereby	they	are	reimbursed	a	specific	amount	for	providing	a	specific	service	to	a	specific	patient.	A	physician’s	FFS	income	depends	on	the	number	and	types	of	services	provided	(Office	of	the	Auditor	General	of	British	Columbia,	2003).	The	process	for	determining	fee	levels	is	described	in	detail	elsewhere	(Doctors	of	BC,	n.d.).	In	brief,	under	a	Master	Agreement	between	Doctors	of	BC,	and	the	Medical	Services	Commission	(MSC),	proposals	for	fee	increases,	or	for	the	addition	or	removal	of	a	fee	item,	are	made	(and	implemented)	by	the	MSC	and	are	ratified	by	Doctors	of	BC	and	the	Medical	Services	Plan	(Doctors	of	BC,	n.d.).	FFS	remuneration	has	been	cited	as	a	source	of	inefficiency	wherever	it	is	used,	since	the	inception	of	its	use.	It	creates	financial	incentives	for	over-use	of	health	care	services	since	physicians	are	paid	more	when	their	patients	use	more	care	(Blomqvist	&	Busby,	2012;	Léger,	2011),	raising	overall	healthcare	costs.	This	has	been	confirmed	in	the	empirical	literature:	a	Cochrane	review	of	physician		 7	payment	systems	found	that	physicians	remunerated	using	a	FFS	model	provide	more	consultations	and	diagnostic	testing	than	physicians	paid	using	alternative	reimbursement	models	(Gosden	et	al.,	2006).		That	same	review	also	noted	that	FFS	remuneration	was	associated	with	greater	continuity	of	care	but	lower	patient	satisfaction	with	access	to	physicians	(Gosden	et	al.,	2006).			1.2.1.2	Alternative	payments	In	1968,	the	BC	Ministry	of	Health	introduced	alternative	payment	programs	(APP)	in	order	to	pay	physicians	for	activities	that	were	not	adequately	being	compensated	through	FFS,	or	for	circumstances	in	which	FFS	dos	not	adequately	support	delivery	of	physician	services	or	patient	access	to	health	care	(Office	of	the	Auditor	General	of	British	Columbia,	2003).	Rather	than	paying	physicians	directly,	APP	pays	agencies	that	in	turn	employ	physicians.		APP	generally	takes	one	of	three	forms:	• Service	contracts	(service-based	payment);	• Sessional	agreements	(time-based	payment);	and	• Salary	(employee-based	payment)	(Office	of	the	Auditor	General	of	British	Columbia,	2003).	The	proportion	of	physician	expenditures	through	APP	arrangements	has	been	increasing	steadily	since	the	program’s	inception.	The	program	made	up	11%	of	total	physician	expenditures	in	2001-02;	this	increased	to	29%	by	2011-12	(Canadian	Institute	for	Health	Information,	2013).		 8	1.2.2	Incentives	and	non-clinical	payments		1.2.2.1	General	Practice	Services	Committee:	Incentive	payments	The	General	Practices	Services	Committee	(GPSC)	is	a	joint	initiative	of	the	BC	Ministry	of	Health,	the	BC	Medical	Association,	and	the	Society	of	General	Practitioners	of	BC	that	began	in	2003.	The	collaboration	has	spearheaded	several	recent	attempts	at	reforms	of	primary	care	in	the	province,	including	the	introduction	of	a	large	basket	of	incentive	payments	for	particular	types	of	patient	care:	mental	health,	obstetrics,	care	of	the	frail	elderly,	palliative	care,	and	complex	and	chronic	care	(Lavergne	et	al.,	2013).	These	incentive	fee	items	can	be	billed	by	physicians	on	top	of	their	regular	fee-for-service	activity	billings.	The	largest,	the	complex	and	chronic	care	fee	item,	is	$314	per	eligible	patient	seen,	per	year.	Because	of	the	size	and	scope	of	these	incentive	payments,	they	are	providing	approximately	$32,000	in	additional	gross	income	per	physician	per	year,	totaling	$700	million	in	spending	in	the	province	since	2006	(Lavergne,	Peterson,	McKendry,	Sivananthan,	&	McGrail,	2014).	1.2.2.2	Medical	On-Call	/	Availability	Program		The	Medical	On-Call	/	Availability	Program	(MOCAP)	was	established	in	a	2001	working	agreement	between	the	BC	Medical	Association	and	the	BC	Ministry	of	Health.	The	purpose	of	the	program	is	to	ensure	that	the	medical	needs	of	patients	who	do	not	have	a	regular	source	of	primary	care	(and	who	may	therefore	otherwise	seek	treatment	at	emergency	rooms,	acute	care	hospitals	or	Diagnostic	and	Treatment	centers)	are	met	(British	Columbia	Ministry	of	Health	Services,		 9	2004).		The	program	compensates	physicians	who	provide	coverage	as	part	of	an	established	call	rotation.	Levels	of	payment	are	based	on	response	time	within	which	the	on-call	physician	is	required	to	respond,	and	by	the	clinical	status	of	the	patient	(British	Columbia	Ministry	of	Health	Services,	2004),	and	range	from	$250	per	call	back	(to	a	maximum	of	$26,000	per	annum)	to	$225,000	annually	for		coverage	24	hours	per	day,	seven	days	per	week	and	365	days	per	year	(British	Columbia	Ministry	of	Health	Services,	2004).	The	program	cost	the	province	close	to	$130	million	per	year	between	2007	and	2012	(MOCAP	Review	Team,	2007).	1.2.2.3	Rural	and	remote	programs		The	Rural	Practice	Subsidiary	Agreement	is	an	agreement	between	the	BC	Ministry	of	Health,	BCMA	and	MSC	that	was	first	signed	in	2007,	and	attempts	to	enhance	recruitment	and	retention	of	physicians	to	underserviced	rural	and	remote	areas	primarily	through	financial	incentives	(Joint	Standing	Committee	on	Rural	Issues,	2012).	These	incentive	payments	include	the	Rural	Retention	Program,	Recruitment	Incentive	and	Contingency	Funds,	Isolation	allowances	and	others.	Physicians	are	eligible	to	receive	these	incentives	if	they	agree	to	practice	in	an	underserviced	region	for	a	set	period	of	time	(Joint	Standing	Committee	on	Rural	Issues,	2012).		1.2.3	Current	pattern	of	physician	payments	Despite	significant	growth	in	alternative	payments	as	a	proportion	of	overall	physician	spending	since	the	program’s	inception	in	1968,	the	majority	of	BC’s	primary	care	physicians	continue	to	be	paid	primarily	by	FFS	remuneration	(Figure		 10	1.1).	Forty	eight	percent	of	primary	care	physicians	receive	90%	or	more	of	their	income	as	fees	for	specific	services.	Also,	physicians	who	receive	blended	remuneration	report	55%	of	their	income	comes	from	FFS	sources	(The	College	of	Family	Physicians	of	Canada,	The	Canadian	Medical	Association,	&	The	Royal	College	of	Physicians	and	Surgeons	of	Canada,	2010).		Figure	1.1:	Method	of	remuneration	for	physicians	in	British	Columbia	(the	College	of	Family	Physicians	of	Canada	et	al.,	2010)				0%10%20%30%40%50%60%70%80%90%100%Male Female <35 35-44 45-54 55-64 65+Sex Age	Group All	Physicians90%+	Fee-For-Service 90%+Salary 90%+Capitation90%+	Sessional 90%+	Service	Contract 90%+otherBlended Not	reported	 11	1.3	The	feminization	of	British	Columbia’s	primary	care	physician	workforce	1.3.1	Physician	supply	As	noted	previously,	BC’s	primary	care	physician	to	population	ratio	has	been	increasing	steadily	since	the	mid-1980s	(Canadian	Institute	for	Health	Information,	2012).	At	122	per	100,000	population,	it	is	also	higher	than	the	Canadian	average	of	109	per	100,000	population	(Canadian	Institute	for	Health	Information,	2012).	From	2009-2013,	the	supply	of	primary	care	physicians	in	BC	grew	13%,	while	the	general	population	only	grew	by	4.9%	(Canadian	Institute	for	Health	Information,	2012).	The	total	number	and	per-capita	supply	of	general	practice	and	primary	care	physicians	in	Canada	has	also	been	steadily	increasing,	rising	from	57,462	in	1998	to	76,518	in	2012	(Canadian	Collaborative	Centre	for	Physician	Resources	&	The	Canadian	Medical	Association,	2012).	Over	this	same	period,	the	primary	care	workforce	became	increasingly	feminized	(Figure	1.2).	Between	2008	and	2012,	the	number	of	female	primary	care	physicians	increased	23%,	compared	to	an	increase	of	8%	for	male	primary	care	physicians	(Canadian	Institute	for	Health	Information,	2012).				 12			Figure	1.2:	Number	of	primary	care	physicians	in	Canada	by	gender,	1998-2013	(Canadian	Collaborative	Centre	for	Physician	Resources	&	The	Canadian	Medical	Association,	2012)		 The	proportion	of	primary	care	physicians	who	are	women	has	not	increased	uniformly	across	BC.	In	some	health	service	delivery	areas	(HSDAs),	this	proportion	is	almost	0.5,	while	in	others,	it	remains	as	low	as	0.25	(Figure	1.3).	As	of	2012,	women	outnumbered	men	in	family	medicine,	and	also	in	four	other	specialties	(endocrine/metabolism,	geriatric	medicine,	medical	genetics,	and	paedatrics)	(Sullivan,	2012).		05000100001500020000250001998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013Number	of	PhysiciansYearMales,	General	PracticeMales,	Family	MedicineFemales,	General	PracticeFemales,	Family	MedicineMales,	Total	General	PracticeFemales,	Total	General	Practice	 13		Figure	1.3:	Proportion	of	primary	care	physicians	in	BC	health	service	delivery	areas	who	are	female	1.3.2	Medical	students	and	residents		Between	1968	and	2011,	the	percentage	of	female	students	in	Canadian	medical	schools	increased	from	14%	to	58%	(Figure	1.4)	(The	Association	of	Faculties	of	Medicine	of	Canada,	2013).		Over	that	same	period,	total	enrollment	more	than	doubled,	from	4681	to	10853	(The	Association	of	Faculties	of	Medicine	of	Canada,	2013).					 14			Figure	1.4:	Enrolment	in	Canadian	medical	schools	by	gender,	1968-2011	(The	Association	of	Faculties	of	Medicine	of	Canada,	2013)		The	number	(and	proportion)	of	residents	who	are	female	is	also	increasing,	having	climbed	from	506	(45%	of	residents)	in	2002	to	1536	(57%	of	residents)	in	2012	(Figure	1.5)	(Canadian	Resident	Matching	Service,	2012).	The	number	(and	proportion)	of	both	male	and	female	graduates	choosing	family	medicine	has	also	increased	climbing	from	331	(30%)	in	2002	to	923	(35%)	in	2012	(Canadian	Resident	Matching	Service,	2012).		More	women	choose	family	medicine	(23%)	compared	to	men	(12%);	the	proportion	of	family	medicine	residents	who	are	female	has	increased	from	59%	in	2002	to	65%	in	2012	(Canadian	Resident	Matching	Service,	2012).			010002000300040005000600070001968/691970/711972/731974/751976/771978/791980/811982/831984/851986/871988/891990/911992/931994/951996/971998/992000/012002/032004/052006/072008/092010/11Number	of	StudentsYearMenWomen	 15		Figure	1.5:	Proportion	of	new	Canadian	residents	choosing	family	medicine	and	other	specialties,	by	gender,	2002-2012	(Canadian	Resident	Matching	Service,	2012)		 Taken	together,	these	data	indicate	that	the	primary	care	physician	workforce	in	Canada	has	become	increasingly	feminized,	and	that	this	demographic	shift	will	continue	for	the	foreseeable	future.	This	trend	results	primarily	from	a	dramatic	increase	in	the	proportion	of	medical	school	enrollees	who	are	women,	and	a	modest	increase	in	the	proportion	of	women	who	are	choosing	family	medicine	as	their	specialty	(compared	to	their	male	counterparts)	(Canadian	Resident	Matching	Service,	2012;	The	Association	of	Faculties	of	Medicine	of	Canada,	2013).		1.4	Health	human	resources	planning	Despite	increases	in	research	dollars	and	attention	from	the	policy	community	and	the	public,	most	current	approaches	to	physician	supply	forecasting	051015202530354045PercentageYearMen,	Other	SpecialitiesMen,	Family	MedicineWomen,	Family	MedicineWomen,	Other	specialties	 16	rely	on	an	overly	simplistic	calculus	(physician	and	service	counts,	or	physician	to	population	ratios)	to	determine	both	current	and	future	supply	(Lavis	&	Birch,	1997;	O’Brien-Pallas	et	al.,	2001;	Tomblin	Murphy,	2005).		These	measures,	particularly	when	taken	in	isolation,	are	insufficient	in	that	they	mask	the	lack	of	uniformity	in	the	distribution	of	physicians,	and	are	unable	to	account	for	other	important	factors	such	as	changes	in	activity	and	practice	patterns.	Despite	these	broad	limitations,	these	measures	continue	to	be	broadly	used,	frequently	without	mention	of	other	contextual	factors.	Health	human	resources	planning	should	take	into	account	variables	other	than	raw	physician	headcounts	and	physician	to	population	ratios	(Esmail,	2007;	The	BCMA	Council	on	Health	Economics	and	Policy,	2011;	Watson	et	al.,	2006),	including	variables	on	both	the	requirements	(population	need)	and	supply	(service	availability)	side	(Stephen	Birch	&	Kephart,	2007;	Dreesch	et	al.,	2005;	O’Brien-Pallas	et	al.,	2001).	On	the	supply	side,	forecasts	should	account	for	the	combined	effects	of	changes	in	physician	activity	and	service	delivery	patterns,	the	potential	impact	of	retirement	and	parental	leaves,	feminization,	and	the	new	work	patterns	of	the	younger	cohort	of	physicians	(Chan,	1999;	Singh	&	Ontario	Ministry	of	Health	and	Long-Term	Care,	2010;	The	BCMA	Council	on	Health	Economics	and	Policy,	2011).			 17	1.4.1	Gender	and	health	human	resources	policy	and	planning		Health	human	resource	policy	has	been	criticized	as	suffering	from	“gender	blindness”	(Hilary	Standing,	2000).	In	her	paper,	Health-manpower	Planning	or	Gender	Relations?	The	Obvious	and	the	Oblique,	Kazanjian	writes:	All	early	governance	policies	and	most	current	public	policies	have	been	designed	by	policy-makers	who	have	defined	each	situation	from	a	male	perspective,	resulting	in	a	delivery	system	that	favours	male	patterns	of	labour	market	participation,	and	rewards	male	life-cycle	activity	patterns,	and	perpetrates	the	gender	gap	in	the	hierarchy	of	professionals.	….	The	description,	analysis	and	interpretation	of	this	market	situation	have	not,	traditionally,	taken	into	consideration	the	female	experience	(p.168)	(Kazanjian,	1993).			Expectations	around	working	hours	and	other	conditions	of	service	are	predicated	on	male	patterns	of	employment	(Hilary	Standing,	2000).	Women	who	cannot	accommodate	these	patterns	are	seen	as	a	“problem”	or	“not	productive”	(Dacre	&	McKinstry,	2008;	Hilary	Standing,	2000).		However,	including	gender	as	a	key	consideration	in	health	human	resource	planning	is	not	based	on	accommodating	the	unique	needs	and	wants	of	female	care	providers:		“The	arguments	for	taking	gender	seriously	in	human	resources	policy	and	planning	are	not,	therefore,	based	on	special	pleading	for	women	(or	on	a	unilateral	view	that	women	employees	always	have	different	needs	or	interests	than	men),	but	on	the	need	to	develop	a	much	more	effective	way	of	using	the	health	human	resources	that	exist	to	meet	the	considerable	challenges	of	providing	competent	health	care…		This	means	enabling	qualified	women	to	operate	effectively	as	workers	while	avoiding	the	pitfalls	of	stereotyping	women	as	a	“problem”.	This	entails	incorporating	gender	into	health	services	policy	and	planning	frameworks.”	-	(Hilary	Standing,	2000)		All	things	considered,	HHR	policy	and	planning	has	entirely	failed	to	account	for	the	effect	of	changing	gender	roles	in	society	on	the	health	labour	market		 18	(Kazanjian,	1993).	Existing	policies	governing	HHR	planning,	and	widely-adopted	models	of	physicians	supply	implicitly	assume	that	the	gender	of	the	physician	has	no	impact	on	what	care	is	provided,	to	which	patients,	how	much,	or	how	often	(Boulis	&	Jacobs,	2008).	Those	policy	and	planning	exercises	that	do	acknowledge	a	gender	effect	either	at	the	individual	or	population	level	apply	it	as	if	it	was	static	descriptor	that	has	a	single,	consistent	and	entirely	predictable	impact	on	supply	issues	(Schofield,	2009)	.		The	most	common	of	these	static	measures	is	full-time	equivalents	(FTEs),	where	an	average	female	physician	is	simulated	as	contributing	a	static	proportion	(always	less	than	1.0)	of	the	activity	of	a	male	physician	(whose	work	is	typically	valued	at	1.0	FTE).	The	Health	Human	Resources	Supply	Model,	which	was	developed	by	Health	Canada,	is	an	example	of	a	planning	tool	that	relies	on	an	FTE	calculation.	In	it,	the	activity	of	an	average	female	physician	is	weighted	at	0.75	FTEs	compared	to	1.0	FTEs	for	male	physicians	(Basu	&	Rajbhandary,	2004).	FTEs	do	not	take	into	account	differences	in	practice	patterns,	changes	in	relative	activity	over	careers,	or	other	relevant	concepts	that	impact	service	supply	and	may	be	differential	by	physician	gender.	Gender	is	relevant	for	health	human	resources	policy	and	planning	for	myriad	related	reasons	(Schofield,	2009).	Chief	among	them,	and	most	relevant	for	this	work,	gender	affects	occupational	choice	(in	this	case,	physician	specialty)	(Tyrrell	et	al.,	1999),	and	career	activity	(e.g.	(Aasland	&	Rosta,	2011;	Crossley	et	al.,	2009;	Kazanjian,	1993))	and	practice	patterns	(e.g.(Boerma	&	van	den	Brink-	 19	Muinen,	2000).1	Second,	it	affects	physician	training	through	the	provision	of	gender-sensitive	health	services.	Third,	it	affects	the	“sensitivity”	of	the	doctor-patient	relationship	(Standing,	2000).		1.4.2	Conceptual	model	There	is	no	existing	(conceptual)	HHR	planning	model	that	explicitly	includes	gender	as	a	relevant	planning	consideration.	I	have	chosen	to	expand	an	existing	model	to	include	a	gender	module,	which	will	conceptualize	the	ways	in	which	physician	gender	should	be	considered	in	an	HHR	policy	and	planning	process.	I	will	describe	the	model	in	its	current	form	here,	and	follow	that	with	a	description	of	the	added	gender	module	below.		1.4.2.1	Needs-based	health	human	resources	planning	model	The	HHR	planning	model	created	by	O’Brien-Pallas	and	colleagues	(2001)	and	enhanced	by	Bourgeault	and	colleagues	(2015)	“identifies	the	constructs	that	influence	the	requirements	for	and	supply	of	human	resources	(population	health	needs,	education	and	training,	supply	of	providers,	organization	of	work	and	production	and	the	prevailing	contexts	in	which	all	these	constructs	are	experienced)	and	the	pathways,	both	direct	(independent)	and	indirect	(interactions	between	influences),	through	which	these	influences	operate”	(Stephen	Birch	&	Kephart,	2007)	(Figure	1.6).	It	has	been	empirically	tested	(Tomblin	Murphy,	2005),	and	improves	upon	earlier	HHR	planning	models	by																																																									1	The	ways	in	which	gender	has	an	impact	on	activity	and	practice	patterns	are	explored	in	the	systematic	review	presented	in	Chapter	2.			 20	capturing	the	interplay	between	supply	elements	that	were	typically	considered	only	in	isolation	(O’Brien-Pallas	&	Birch,	2001;	Tomblin	Murphy,	2005).			Figure	1.6:	Health	Human	Resources	Conceptual	Model	(reproduced	from	(Bourgeault,	Demers,	James,	&	Bray,	2015))		 While	considering	carefully	elements	affecting	the	supply	of	health	services,	the	model	is	“needs-based”	in	that	it	recognizes	that	the	supply	must	be	matched	as	closely	as	possible	with	the	health	needs	of	the	population	(O’Brien-Pallas,	2002).	There	is	therefore	no	assumption	of	correlation	between	need	and	measures	of	supply,	such	as	utilization	and	expenditures	(Birch,	Eyles,	&	Newbold,	1996;	Lomas,	Stoddart,	&	Barer,	1985).	Rather,	supply	requirements	for	HHR	are	derived	from	the	need	for	health	services,	not	from	(current)	baseline	supply	and	mix	of	services	available	(Stephen	Birch	&	Kephart,	2007).	The	framework	emphasizes	that	health		 21	human	resources	planning	occurs	within	health	care	planning	more	broadly,	and	also	within	(as	opposed	to	independent	of)	other	public	policy	planning	(Stephen	Birch	&	Kephart,	2007).	This	thesis	is	concerned	primarily	with	the	supply-side	considerations,	and	generating	HHR	planning	model	inputs	and	predictions	that	are	as	accurate	as	possible.	It	represents	an	extension	of	the	supply,	and	planning	and	forecasting	sections	of	the	O’Brien-Pallas	model.	These	components	of	the	model,	and	others	that	are	directly	relevant,	are	described	below.	Complete	descriptions	of	all	model	elements	can	be	found	elsewhere	(Federal/Provincial/Territorial	Advisory	Committee	on	Health	Delivery	and	Human	Resources,	2007).	The	“supply”	box	on	the	left	hand	side	of	the	model	includes	the	number,	type	and	geographic	distribution	of	providers	–	in	this	case,	primary	care	physicians	(Canadian	Institute	for	Health	Information,	2001a).	The	demographic	and	educational	characteristics	of	providers,	as	well	as	deaths,	retirements,	leaves	of	absence,	immigration,	emigration	and	education,	all	have	an	impact	on	supply.		“Production”	includes	education	and	training	for	future	health	care	providers,	and	estimates	future	capacity	(O’Brien-Pallas,	2002).	Included	in	this	element	is	the	geographic	distribution	of	medical	school	and	residency	slots,	which	should	be	linked	to	population	health	needs.	Broadly,	“financial	resources”	focuses	on	the	allocation	of	dollars	to	health	care,	and	specifically	to	HHR,	recognizing	that	the	decision	about	the	size	of	the	budgetary	share	should	be	based	on	meeting	the	specific	health	needs	of	the	population.	An	appropriate	balance	between	human	and	physical	capital,	purchased		 22	with	those	public	and	private	financial	resources,	helps	to	ensure	that	population	heath	needs	are	met	effectively	and	efficiently	(O’Brien-Pallas,	2002).		Careful	choices	about	the	appropriate	quantity,	mix	and	distribution	of	both	human	and	physical	capital	should	be	made	within	the	context	of	the	pool	of	available	financial	resources	(Federal/Provincial/Territorial	Advisory	Committee	on	Health	Delivery	and	Human	Resources,	2007).		“Management,	organization	and	delivery	of	services”	refers	to	the	underlying	characteristics	of	the	health	system	in	which	care	is	delivered.	This	includes,	for	example,	the	degree	of	centralization	or	decentralization,	structural	arrangements,	and	organizational	culture	(O’Brien-Pallas,	2002).	Organization	characteristics	influence,	among	other	things,	provider	job	satisfaction,	which	impacts	absenteeism	and	retention	rates,	which	have	a	dampening	impact	on	downstream	supply	(O’Brien-Pallas,	2002).	The	supply	side	of	the	model	is	iterative:	Supply,	financial	resources,	and	management,	organization	and	delivery	of	services	are	used	to	make	decisions	about	supply	and	production,	which	are	then	monitored	to	adjust	forecasts.		“Planning	and	forecasting”	represents	the	specific	methods	and	tools	used	to	predict	HHR	and	other	resource	requirements	to	meet	the	needs	of	the	population.	Thus	this	element	reflects	the	data	requirements,	assumptions	and	limitations	of	current	HHR	planning	practices	(Federal/Provincial/Territorial	Advisory	Committee	on	Health	Delivery	and	Human	Resources,	2007).	Bourgeault	and	colleagues,	embellished	this	section	to	include	an	explicit	focus	on	the	variation	of	productivity		 23	and	activity	rates	between	health	professionals	(Bourgeault,	Demers,	James,	&	Bray,	2015).		“Resource	deployment	and	utilization”	reflects	what	resources,	both	human	and	physical	capital,	are	deployed	to	provide	health	services	to	the	population,	and	the	quantity	and	nature	of	services	that	are	used	by	that	population	(O’Brien-Pallas,	2002).	Bourgeault	and	colleagues	added	an	expanded	“deployment	and	distribution	module”	that	includes	micro,	meso,	and	macro	level	factors	that	influence	both	planning	and	forecasting	and	deployment	and	distribution.	This	includes	for	example	influences	of	different	models	of	care,	supporting	health	care	infrastructure,	and	regulatory	or	legal	influences.		It’s	notable	that	although	demographic	characteristics	are	mentioned	as	impacting	supply,	there	is	no	explicit	mention	of	gender,	nor	an	explanation	of	how	particular	demographic	changes	would	be	expected	to	impact	supply.	Neither	gender,	nor	demographics	more	broadly,	are	mentioned	in	any	other	model	elements	of	the	model.		1.4.2.2	The	gender	module		Although	gender	issues	impact	both	the	supply	and	need	components	of	this	model,	here	I	will	focus	solely	on	the	impact	of	physician	gender	on	supply	and	deployment	considerations.		Existing	evidence,	which	will	be	discussed	in	Chapter	2,	suggests	that	physician	gender	independently	impacts	everything	from	occupational	choice	and	career	path,	to	activity	patterns	and	practice	style,	all	of	which	have	a	tangible	impact	on	HHR	planning.			 24	The	“supply”	box	should	be	expanded	to	include	current	staffing	levels	for	each	specialty,	disaggregated	by	gender	and	location,	as	well	as	gender-related	differences	in	skill	mix	and	practice	patterns.	Historically,	women	were	likely	to	be	concentrated	in	certain	“ancillary”	health	occupations	(Armstrong	&	Armstrong,	2010),	and	were	actively	excluded	from	established	professions	(Crompton	&	Sanderson,	1990).	Although	this	trend	is	slowly	changing,	women	remain	relatively	poorly	represented	in	more	senior	management	and	academic	health	positions,	as	well	as	within	certain	physician	specialties	(Canadian	Resident	Matching	Service,	2012;	Reed	&	Buddeberg-Fischer,	2001).		Tracking	supply	should	also	incorporate	any	gender	differences	in	career	activity	patterns	and	work	practices,	including	scopes	of	practice,	how	tasks	are	distributed	and	shared	between	workers.	Female	physicians	have	different	life	cycle	activity	patterns,	more	frequently	leaving	the	workforce	for	family	reasons	and	then	returning	later,	or	taking	advantage	of	part-time	work	options	to	better	balance	work	and	home-life	responsibilities.	These	issues	have	a	clear	effect	on	both	supply	and	deployment,	which	is	very	much	an	iterative	process.	They	are	discussed	at	length	in	Chapter	2.		Lastly,	it	is	important	to	at	least	acknowledge	the	implicit	linkages	between	the	formal	and	informal	care	domains.	Because	women	are	more	likely	to	carry	the	“lion’s	share”	of	any	informal	care	burden	(such	as	childcare),	policies	and	practices	that	impact	supply	in	within	the	formal	care	domain	(either	increasing	or	decreasing	staffing	levels,	for	example)	will	necessarily	differentially	impact	women	within	the	informal	care	domain	(Standing,	1997).		 25	The	”production”	box	should	expanded	to	include	enrolment	in	and	completion	of	education	and	training,	as	well	as	rates	of	specialization,	and	uptake	of	rural	practice	and	other	incentives	by	gender.	There	is	a	clear	need	to	critically	examine	the	relationship	between	the	current	array	of	recruitment	and	retention	strategies	and	the	gender	composition	of	the	workforce.	Specific	training	and	career	pathways,	and	any	changes	therein	should	be	examined	with	an	eye	to	whether	they	differentially	impact	women.	For	example,	lengthening	of	training	that	occurs	primarily	during	childbearing	years	could	be	differentially	difficult	for	female	trainees	who	are	more	likely	to	be	balancing	family	responsibilities	(Standing,	2000).	Drop-out	and	completion	rates	should	be	monitored,	and	reasons	for	lower	attainment	among	women,	when	present,	should	be	studied	and	resolved.			 Female	physicians	are	under-represented	in	rural	areas	(Doescher,	Ellsbury,	&	Hart,	2000).	Thus,	the	geographic	distribution	of	residency	slots	poses	a	particular	challenge	when	a	gender-focused	lens	is	applied.	Women	face	different	constraints	with	respect	to	living	and	working	in	these	areas,	compared	to	those	working	in	metropolitan	centers.	Issues	of	role-strain	–	resulting	from	attempts	to	balance	and	manage	competing	role	obligations	(Waters,	1993)	–	may	be	aggravated	in	rural	areas	(Rourke,	Rourke,	&	Brown,	1996).	For	example,	rural	areas	are	less	likely	to	have	large	pools	of	physicians	for	job	sharing,	making	it	more	difficult	to	regulate	office	hours,	or	find	locum	coverage	for	maternity	leave	(Rourke	et	al.,	1996).	Additionally,	a	desire	for	a	more	selective	scope	of	practice	can	make	rural	locations	less	appealing	for	female	physicians	(Rourke	et	al.,	1996).		 26		 Current	attempts	to	increase	the	attractiveness	of	rural	practice	are	primarily	financial,	or	focused	on	getting	physicians	to	spend	time	in	rural	communities	early	in	their	careers.	Making	both	entry	to	and	maintenance	of	rural	residency	and	practice	more	appealing	to	female	physicians	may	take	a	different	set	of	incentives	than	those	currently	implemented.	Women	early	in	career	are	more	likely	to	be	constrained	by	family	or	marital	demands	compared	to	male	physicians,	which	makes	the	current	slate	of	incentives	less	useful	for	them.			 The	“management,	organization,	and	delivery”	of	the	health	system	also	has	gender	elements	that	should	be	considered	in	any	health	human	resources	planning	exercise.	Certain	organizational	cultures,	for	example,	can	act	as	barrier	to	career	advancement	for	women,	such	that	although	they	are	not	explicitly	excluded	from	work,	they	are	instead	often	relegated	to	“ill-defined	support	roles”	(Davies,	1996).		Additionally,	existing	evidence	suggests	that	female	physicians	in	academic	roles	report	a	lower	sense	of	belong	and	fewer	positive	relationships	in	the	workplace	compared	with	male	physicians,	resulting	in	lower	feelings	of	self-efficacy	in	career	advancement	(Pololi,	Civian,	Brennan,	Dottolo,	&	Krupat,	2012).	The	desire	to	balance	family	and	professional	demands	may	lead	physicians	–	more	often	female	physicians	–	to	circumvent	existing	organizational	culture	by	establishing	their	own	organizations	(practices),	where	they	have	the	freedom	to	establish	their	own	schedules,	volumes	of	work,	patterns	of	practice,	and	culture	either	in	groups	or	on	their	own	(DeLaat,	2007).	Management,	organization	and	delivery	should	be	expanded	to	include	a	specific	focus	on	how	organizational	culture	may	differentially	impact	female	employees.					 27		 The	“resource	deployment	and	utilization”	portion	of	the	model	should	be	expanded	to	include	the	efficiency	and	effectiveness	of	service	delivery	disaggregated	by	provider	gender.		Quantitative	and	qualitative	assessments	of	patient	outcomes	and	satisfaction,	according	to	the	gender	of	their	physician	should	be	conducted.	Provider	health	status	and	job	satisfaction	should	also	be	studied	with	an	explicit	focus	on	gender	differences.				 Lastly,	the	use	of	tools	for	“planning	and	forecasting”	should	reflect	the	expanded	considerations	discussed	in	the	supply	and	forecasting	sections.	They	should	incorporate	qualitative	and	quantitate	data	requirements	to	track	gender	differences	in	current	and	predicted	supply	levels.		1.5	Research	objectives	and	hypotheses			The	specific	research	objectives	and	hypothesis	for	this	thesis	were	generated	from	the	needs-based	HHR	planning	conceptual	framework	and	the	added	gender	module	described	in	Section	1.4.2,	and	were	refined	based	on	an	extensive	review	of	the	literature	that	is	presented	in	Chapter	2.			Objective	1:	To	examine	differences	in	income,	as	well	as	contact	and	service	patterns	(activity),	for	male	and	female	physicians	(Chapter	4).	• Hypothesis	1.1:	Female	PCPs	will	have	lower	age-adjusted	activity	levels	(contacts,	visits,	dollars	billed)	per	unit	time	compared	with	their	male	counterparts.			 28	• Hypothesis	1.2:	The	difference	in	age-adjusted	activity	levels	per	unit	time	for	male	vs.	female	physicians	will	decrease	with	time	(period/cohort	effect)	• Hypothesis	1.3:	The	difference	in	activity	levels	will	be	greatest	during	childbearing	years,	and	smallest	amongst	primary	care	physicians	aged	65	and	over.	Objective	2:	To	investigate	gender	differences	in	income	relating	to	differences	in	payments	for	clinical	care	and	in	uptake	of	clinical	and	non-clinical	incentive	payments	(Chapter	5).	• Hypothesis	2.1:	Male	physicians	will	be	more	likely	to	take	advantage	of	incentive	(and	non-clinical)	payments,	and	these	payments	will	make	up	a	larger	proportion	of	their	total	incomes.			• Hypothesis	2.2:	Incentives	and	non-clinical	payments	will	represent	a	larger	proportion	of	physician	income	over-time,	regardless	of	gender.	At	the	same	time,	absolute	per-capita	payments	for	clinical	care	will	decline.				Objective	3:	To	determine	differences	in	the	characteristics	of	patient	populations	seen	by	male	as	compared	to	female	PCPs	(Chapter	6).	• Hypothesis	3.1:	Female	physicians	will	see	more	female	patients,	and	fewer	elderly	ones.	Male	and	female	physicians	will	be	equally	likely	to	treat	sick	patients	with	multiple	chronic	diseases	or	disabilities.		 29	Objective	4:	To	examine	how	patterns	of	service	delivery	differ	for	male	and	female	physicians,	once	the	characteristics	of	the	patient	population,	and	fee-for-service	activity	levels	have	been	accounted	for	(Chapter	7).		• Hypothesis	4.1:	Female	PCPs	will	be	more	likely	to	refer	patients	to	other	forms	of	care	(including	specialists,	and	diagnostic	and	laboratory	testing).		• Hypothesis	4.2:	Female	PCPs	will	be	less	likely	to	bill	for	out-of-office	care	provision,	and	a	greater	proportion	will	be	characterized	as	“office	only”	providers.		• Hypothesis	4.3:	More	female	than	male	PCPs	will	provide	obstetrical	services;	however,	the	proportion	of	physicians	who	provide	obstetrical	care	will	decline	over	time	for	physicians	of	both	genders.	• Hypothesis	4.4:	Female	physicians	will	have	more	frequent	(consistent)	contact	with	their	patients.	1.6	Thesis	roadmap		The	goal	of	this	research	is	to	examine	the	impact	of	the	feminization	of	BC’s	primary	care	physician	workforce	on	the	supply	of	primary	health	services,	as	an	empiric	exploration	of	several	elements	of	the	gender-module	I	have	applied	to	the	supply	side	of	O’Brien-Pallas	and	colleagues’	needs-based	health	human	resources	planning	framework.	In	particular,	I	focus	on	the	impact	that	a	demographic	shift	in	the	physician	population	has	on	supply	forecasting	using	both	direct	work	measures	(billings),	and	also	through	practice	pattern	differences	that	may	result	in		 30	downstream	changes	in	the	required	supply	of	specialty	services,	or	diagnostic,	imaging	or	laboratory	services.		In	Chapter	2,	I	apply	the	supply	side	gender	module	through	a	systematic	review	of	the	extant	literature	that	examines	the	impact	of	primary	care	workforce	feminization.	I	focus	on	five	themes	that	are	represented	in	the	supply,	resource	deployment	and	utilization,	provider	outcomes,	and	system	outcomes	elements	of	the	model:	time	spent	working,	intensity	of	work,	scope	of	work,	and	practice	characteristics.	I	reflect	on	how	these	themes	may	be	affecting	current	supply	as	well	as	the	planning	for	and	forecasting	of	future	resource	deployment	and	utilization.		Chapter	3	describes	the	data	resources	used	for	my	empiric	work,	the	methodology	for	cohort	creation,	and	key	concepts	and	variables	that	appear	throughout	all	subsequent	chapters.			 The	remaining	chapters	present	the	bulk	of	the	new	empiric	work,	which	seeks	to	address	some	of	the	knowledge	gaps	identified	in	the	literature	outlined	in	Chapter	2	and	more	comprehensively	examines	the	potential	impact	of	primary	care	workforce	feminization	on	service	delivery.	Specifically,	Chapter	4	examines	differences	in	activity	levels	over	career	trajectories	between	male	and	female	PCPs,	generating	estimates	of	total	years	of	productive	clinical	practice.		Chapter	5	delves	more	deeply	into	activity	differences,	examines	the	change	in	billing	patterns	between	clinical	and	non-clinical	payments	over	time,	and	by	gender.	Chapters	6	and	7	look	more	closely	at	the	nature	of	differences	in	practice	between	male	and	female	PCPs	by	comparing	the	characteristics	of	patient	populations	and	patterns	of	service	delivery.				 31		 Finally,	in	Chapter	8	I	will	summarize	the	key	results	from	each	chapter,	revisit	the	conceptual	model	and	discuss	whether	and	how	the	results	from	the	preceding	analytic	chapters	fit	therein.	I	will	also	draw	some	overall	conclusions,	present	limitations,	and	suggest	directions	for	future	work	and	policy	action.			 	1.7	Publication	of	thesis	chapters	Some	chapters	in	this	thesis	have	been	published	(or	are	being	prepared)	as	stand-alone	manuscripts,	which	leads	to	some	repetition	within	the	thesis.	This	has	been	minimized	as	much	as	possible.	Additionally,	the	reader	will	note	the	use	of	both	the	active	singular	(Chapters	3-7)	and	active	plural	voices	(Chapter	2),	which	reflects	the	contributions	my	committee	and	coauthors	made	during	the	preparation	and	publication	of	manuscripts	that	are	based	on	thesis	chapters.			 		 32	CHAPTER	2:	Systematic	literature	review2	As	described	in	Chapter	1,	the	demographics	of	the	primary	care	physician	(PCP)	workforce	in	Canada,	and	most	industrialized	nations,	are	shifting.	In	several	countries,	the	proportion	of	PCPs	who	are	women	has	doubled	or	nearly	doubled	over	the	last	30	years	(Dumontet,	Le	Vaillant,	&	Franc,	2012;	Harrison,	Britt,	&	Charles,	2011).	Globally,	32%	of	all	physician	graduates	worldwide	are	female,	and	that	percentage	is	higher,	on	average,	in	family	medicine	(American	Medical	Association,	2012).	This	shift	in	workforce	demographic	has	the	potential	to	impact	future	service	supply,	both	within	the	primary	care	context,	and	through	increased	derived	demand	for	specialist	physician	services,	laboratory	technicians,	imaging	technicians	or	other	health	professionals,	outside	of	primary	health	care.		There	has	been	no	comprehensive	synthesis	of	the	existing	literature	on	this	topic.	Our	objective	in	undertaking	a	systematic	review	is	to	examine	the	existing	evidence	related	to	the	impact	of	this	demographic	shift	effect	on	the	supply	of	physician	services.	Specifically,	we	reviewed	studies	that	compared	male	and	female	PCPs	in	terms	of	the	amount	of	time	they	spent	working,	how	intensely	they	worked	(i.e.	the	number	of	services	or	patient	encounters	per	unit	time),	and	whether	their	practice	and	service	characteristics	differed.																																																										2	A	version	of	this	chapter	has	been	published:	Hedden,	L,	Barer,	M.L.,	Cardiff,	K.,	McGrail,	K.M.,	Law,	M.R.,	Bourgeault,	I.L.	The	implications	of	the	feminization	of	the	primary	care	physician	workforce	on	service	supply:	a	systematic	review.	Human	Resources	for	Health	2014	12:32.			 33	2.1	Methodology		2.1.1	Search	strategy	and	inclusion	criteria	In	an	effort	to	ensure	comprehensiveness,	we	used	multiple	search	strategies	to	locate	both	peer-reviewed	and	grey	literature	sources.	Peer	reviewed	literature	was	selected	from	Medline	(OVID),	Embase,	and	Web	of	Science.	We	limited	the	search	to	English	language	articles	published	between	January	1990	and	January	2013.		Database-specific	search	terms	included	variations	on	“physician”,	“women,”	and	“workforce”	(see	Appendix	1	for	the	full	search	strategies).	We	identified	relevant	grey	literature	using	the	Canadian	Health	Research	Library,	ProQuest	Dissertations	and	Theses,	and	the	Canadian	Health	Human	Resource	Network	Library	(http://www.hhr-rhs.ca/index.php?option=com_content&view=article&id=168&Itemid=78&lang=en).	We	also	conducted	searches	of	the	websites	of	organizations,	groups,	governments,	associations,	and	professional	bodies	identified	using	the	Canadian	Agency	for	Drugs	and	Technologies	in	Health’s	“Grey	Matters”	guide	to	grey	literature	(Canadian	Agency	for	Drugs	and	Technologies	in	Health,	2013).	Additionally,	we	completed	forward	and	reverse	citation	searches	(snowballing)	of	included	peer-reviewed	articles	using	Google	Scholar.			We	imported	search	results	into	a	reference	manager	and	removed	any	duplicates.	We	screened	all	abstracts	for	relevance	to	the	research	topic	and	pulled	relevant	articles.	A	second	reviewer	(Karen	Cardiff)	independently	reviewed	all	full-text	articles	using	the	inclusion	and	exclusion	criteria	in	Table	2.1	and	thematic		 34	typology	in	Table	2.2,	and	disagreements	were	resolved	by	discussion.	We	computed	a	Kappa	statistic	for	inter-rater	reliability.	Studies	were	not	excluded	due	to	quality	issues;	however,	methodological	concerns	are	presented	as	part	of	both	the	results	and	discussion	sections.		2.1.2	Data	abstraction	and	article	typology		We	abstracted	and	summarized	the	following	data	from	all	included	articles:	citation;	country;	objectives;	study	sample,	response	and	drop-out	rates	(where	applicable);	study	design	(cross-sectional	or	longitudinal);	data	collection	(administrative,	survey,	or	other	primary	data);	analytic	methodology;	outcome	measure(s);	and	results.				 		 35	Table	2.1	Inclusion	and	exclusion	criteria	Inclusion	Criteria	 Exclusion	Criteria	Publication	Details	Published	between	January	1990	and	January	2013;	published	in	English.	 Published	before	January	1990	or	after	January	2013;	published	in	a	language	other	than	English.	Participants/Population	Primary	care	physicians	(studies	focusing	on	all	physicians	were	included	only	if	results	pertaining	to	primary	care	physicians	were	presented	separately)	Other	physician	specialties;	all	physicians,	where	separate	analysis	for	primary	care	physicians	is	not	presented.		Comparison	Male	to	female	primary	care	physicians.3		 Does	not	compare	male	and	female	physicians.		Outcome	Measures	A	measure	of	one	or	more	of	the	following:	time	spent	working,	intensity	of	work,	scope	of	work,	or	practice	characteristics.4		None	of	time	spent	working,	intensity	of	work,	scope	of	work,	or	practice	characteristics.	Design	Original	research	 Editorials,	comments	or	commentaries,	letters;	reviews	articles;	reports	with	no	primary	data	analysis.																																																											3	Specialist	physicians	(such	as	paediatricians,	or	general	internists)	who	may	practice	like	primary	care	physicians	on	occasion	(i.e.	acting	as	a	point	of	entry	to	the	health	care	system,	providing	person-focused	care	over	time,	and	acting	as	a	coordinator	for	care	provided	elsewhere)	were	not	included.		4	Raw	or	adjusted	results	for	one	or	more	of	these	measures	must	be	presented.	If	these	measures	were	included	as	covariates	in	a	multivariate	modeling	exercise	(e.g.	for	income,	for	example),	the	study	was	excluded	unless	raw	comparisons	on	one	of	these	outcomes	are	also	presented. 	 36	We	coded	articles	using	a	typology	designed	with	the	intention	of	capturing	any	practice	differences	between	male	and	female	physicians	that	could,	either	directly	or	indirectly,	affect	the	availability	of	primary	health	care	services.	It	was	developed	based	on	the	gender	module	and	O’Brien-Pallas	framework	as	described	in	Chapter	1,	sections	1.4.2	(Federal/Provincial/Territorial	Advisory	Committee	on	Health	Delivery	and	Human	Resources,	2007),	and	specifically	with	focus	on	themes	that	are	represented	in	the	supply	and	resource	deployment	elements	of	the	model:	time	spent	working,	intensity	of	work,	scope	of	work,	and	practice	characteristics.		It	includes	variations	in	what	care	is	delivered,	to	whom,	and	how	much.	The	typology	consists	of	five	themes	and	11	sub-themes.	Table	2.2	includes	examples	of	how	each	thematic	area	may	be	linked	to	changes	in	service	availability.	We	conducted	a	qualitative	examination	of	study	quality	by	assessing	the	following	items:	clarity	of	research	questions	and	objectives;	appropriateness	of	study	design;	sample	size	and	representativeness;	validity	of	measures;	addressing	possible	confounders;	and	generalizability.				 		 37	Table	2.2:	Article	Typology	Theme	 Sub-Theme	 Potential	effect	on	supply	-	Direct/Indirect		Years	of	practice	 • Retirement	• Leaves	of	absence		 Direct	–	e.g.	shortening	of	career	or	more	lengthy	absences	from	practice	Hours	of	work	 • Full-	vs.	part-time	work	• Time	spent	on	patient	care		• Time	spent	on	administrative	responsibilities,	professional	development	Direct	–	e.g.	less	time	spent	working	overall,	or	less	time	spent	on	direct	patient	care	in	favour	of	other	responsibilities	Intensity	of	work	 • Number	of	services/time	• Number	of	patients/time	 Direct	(lower	service	or	patient	volumes)	Scope	of	work	 • Patient	characteristics	• Service	provision	 Indirect	(restrictions	in	scope	of	practice,	or	basket	of	services	delivered;	restricted	patient	population;	reduced	availability	of	out-of-office	or	off-hours	care)	Practice	characteristics	 • Location	• Group	practice	vs.	solo	practice	Indirect	(imbalance	between	urban-	vs.	rural-based	practices	leading	to	shortages	in	some	areas,	oversupply	in	others)		2.2	Results		2.2.1	Search	results		The	initial	search	of	Medline,	Embase	and	Web	of	Science	located	1476	citations,	of	which	205	were	duplicates.	We	screened	the	abstracts	from	the	remaining	1271	for	relevance	to	the	topic,	and	excluded	1224,	leaving	47	peer-reviewed	articles.	We	identified	an	additional	27	studies	from	grey	sources	and		 38	through	snowballing	of	references	in	selected	articles.		These	74	sources	were	retained	for	full-text	review	(Figure	1).		Of	these,	34	studies	met	the	inclusion	criteria;	they	are	summarized	in	Appendix	2.	The	K-coefficient	for	inter-rater	agreement	beyond	change	was	0.84.	Thirty	of	the	34	included	studies	(88%)	had	been	published	in	peer-	reviewed	fora.	Fifteen	of	the	34	(44%)	were	conducted	in	Canada,	four	(12%)	in	the	US,	and	five	(15%)	in	the	United	Kingdom.	Twenty-seven	studies	(79%)	used	a	cross-sectional	methodology.	Of	these	21	(78%)	used	retrospective	survey	data,	five	(19%)	used	administrative	data,	and	one	employed	prospective	primary	data	collection.		Of	the	seven	(21%)	studies	that	used	longitudinal	methods,	four	(57%)	used	administrative	data,	one	combined	administrative	and	survey	data,	and	two	(29%)	used	surveys	alone.								 39		Potentially	relevant	citations	identified	through	Medline,	Embase,	and	Web	of	Science:	N	=	1476	Abstract	screening:	N	=	1271	Duplicates	removed:	N	=	205	Abstracts	excluded	for	lack	of	relevance:	N	=	1224	Records	included	in	full-text	review:	N	=	47	Citations	identified	through	snowballing:	N	=	5	Citations	identified	in	grey	literature:	N	=	22	Full-text	articles	excluded:	N	=	40	• Not	original	research:	N	=	8	• Not	primary	care:	N	=	5	• No	male/female	comparison:	N	=	11	• Did	not	include	outcome	of	interest:	N	=	15	• Out	of	date	range:	N	=	1	Full-text	articles	included:	N	=	34	Figure	2.1:	Search	Results		 40	2.2.2	Thematic	results	Hours	of	work,	intensity	of	work	(defined	here	as	number	of	services	or	patient	encounters	per	unit	time),	and	scope	of	work	featured	in	18	(53%),	14	(42%)	and,	17	(50%)	studies	respectively	(Figure	2.1).	Practice	characteristics	were	examined	in	seven	(21%)	studies,	and	years	of	practice	was	a	focus	in	only	four	(12%).		Themes	with	a	direct	impact	on	service	availability	(years	of	practice,	hours	and	intensity	of	work)	were	more	commonly	featured	(26	articles,	76%)	than	those	that	affect	supply	or	availability	of	services	indirectly	(practice	characteristics,	scope	of	practice)	(18	articles,	53%).	Slightly	more	than	sixty	percent	of	the	included	studies	focused	on	a	single	thematic	area.				 41	Figure	2.2:	Thematic	Results 		 42	2.2.2.1	Hours	of	work	All	18	studies	that	examined	hours	of	work	found	that	female	PCPs	tended	to	self-report	working	fewer	hours	than	their	male	counterparts.	Few	of	these	studies,	however,	presented	results	that	adjusted	for	physician	age,	practice	characteristics	or	other	factors	that	may	confound	the	relationship	between	physician	sex	and	work	hours	(e.g.	(Boerma	&	van	den	Brink-Muinen,	2000;	Gravelle	&	Hole,	2007)).	In	their	survey	of	English	general	practitioners,	Gravelle	et	al.	found	that	the	average	difference	in	hours	per	week	worked	between	males	and	females	was	11.8	hours	(Gravelle	&	Hole,	2007).	Forty-five	percent	(5.3	hours)	of	this	difference	was	due	to	the	greater	proportion	of	male	PCPs	at	each	age	working	full	time,	and	46%	(5.4	hours)	was	due	to	female	PCPs	reducing	their	hours	more	than	male	PCPs	who	have	the	same	family	circumstances.	The	final	9%	(1.1	hours)	of	the	difference	was	due	to	differences	in	physician	demographics	(e.g.	age)	and	practice	characteristics	(e.g.	size	of	practice)	(Gravelle	&	Hole,	2007).		In	their	European	study,	Boerma	et	al.	found	that,	on	average,	male	PCPs	worked	more	hours	per	week,	excluding	on-call	time	(45.1	vs.	36.2)	(Boerma	&	van	den	Brink-Muinen,	2000).	In	countries	where	the	difference	in	hours	was	statistically	significant	(12	of	32	study	countries),	male	PCPs	worked	more	in	ten,	and	female	PCPs	worked	more	in	two	(Boerma	&	van	den	Brink-Muinen,	2000).	Results	from	North	America	are	similar,	with	female	PCPs	working	between	four	and	14.5	fewer	patient-care	hours	per	week	(Atkin,	2000;	Carek,	King,	Hunter,	&	Gilbert,	2003;	McMurray	et	al.,	2002;	Norton,	Dunn,	&	Soberman,	1994;	Slade	&	Busing,	2002;	Watson	et	al.,	2006).		 43		Female	PCPs	were	more	likely	to	report	working	part-time	(31.6%	vs.	11.1%)	(Gravelle	&	Hole,	2007;	Keane,	Woodward,	Ferrier,	Cohen,	&	Goldsmith,	1991),	and	billed	Canadian	provincial	health	insurance	plans	for	fewer	months	of	the	year	(Cohen,	Ferrier,	Woodward,	&	Goldsmith,	1991).	Having	children	under	the	age	of	18	increased	the	probability	that	female	PCPs	worked	part-time,	but	had	no	effect	on	male	PCPs	(Gravelle	&	Hole,	2007).		Despite	consistent	differences	found	in	hours	worked	overall,	and	specifically	in	hours	spent	on	patient	care,	male	and	female	PCPs	tended	to	spend	a	similar	amount	of	time	on-call	(Atkin,	2000;	Carek	et	al.,	2003;	Raymont,	Lay-Yee,	Pearson,	&	Davis,	2005).		Three	of	the	included	studies	examined	longitudinal	trends	in	work	hours	for	male	and	female	physicians	(Aasland	&	Rosta,	2011;	Crossley	et	al.,	2009;	Watson	et	al.,	2006).	In	their	study	on	PCP	labour	supply	in	Canada,	Crossley	et	al.	found	a	secular	decline	in	hours	of	patient	care	between	1982	and	2003	(Crossley	et	al.,	2009).	Although	female	physicians	were	found	to	have	worked	fewer	hours	than	male	physicians,	a	change	in	the	behavior	of	male	PCPs	accounted	for	a	greater	proportion	of	the	decline	in	hours	of	patient	care	than	did	the	growing	proportion	of	females	in	the	workforce.		The	gap	in	hours	worked	between	male	and	female	PCPs	diminished	over	the	study	period	(Crossley	et	al.,	2009).		They	also	reported	that,	for	female	physicians	only,	there	was	a	significant	age	effect	on	hours	of	patient	care:	hours	declined	up	to	approximately	age	38,	and	then	gradually	increased	with	age	(Crossley	et	al.,	2009).		This	would	be	consistent	with	a	“childbearing	years”	effect.	Aasland	et	al.	found	that	the	gap	between	male	and	female	PCPs’	hours	of		 44	work	is	also	narrowing	in	Norway,	with	female	PCPs	having	worked	significantly	fewer	hours	than	male	PCPs	between	2000	and	2006,	but	not	in	2008	(Aasland	&	Rosta,	2011).		In	that	country,	however,	physicians’	hours	have,	on	the	whole,	increased	rather	than	declined,	with	the	increase	in	hours	obviously	being	more	marked	amongst	female	physicians	(Aasland	&	Rosta,	2011).	2.2.2.2	Intensity	of	work		Eleven	studies	compared	the	number	of	services	per	unit	of	time	delivered	or	number	of	patients	seen	for	male	and	female	PCPs.	Of	these,	five	presented	multivariate	results,	controlling	for	the	effect	of	physician	and	patient	characteristics,	or	other	confounders.		Cohen	et	al.,	Woodward	and	Hurley,	and	the	Canadian	Institute	for	Health	Information	all	found	that	Canadian	male	PCPs’	bill	for	more	services	compared	with	their	female	colleagues,	and	that	physician	gender	contributed	significantly	to	explaining	variation	in	service	activity		(Canadian	Institute	for	Health	Information,	2001b;	Cohen	et	al.,	1991;	C.	A.	Woodward	&	Hurley,	1995).	Boerma	and	colleagues	similarly	found	that	European	female	PCPs	have	on	average	4.1	(or	14%)	fewer	office	contacts	per	day.	This	difference	in	office	contacts	was	only	significant	in	12	of	the	32	study	countries,	and	in	half	of	these,	female	physicians	had	significantly	more	daily	contacts	than	male	physicians	(Boerma	&	van	den	Brink-Muinen,	2000).		Additionally,	when	results	were	restricted	to	only	include	physicians	who	worked	full	time,	the	sex-related	difference	in	contacts	dropped	to	2.3	fewer	contacts	per	day	for	female	physicians,	and	a	significant	difference	was	found	in	only	six	of	32	countries.	Of	these,	women	had	significantly	more	contacts	per	day	in	three		 45	(Boerma	&	van	den	Brink-Muinen,	2000).		Consistent	with	the	age-stratified	results	reported	for	hours	worked,	Constant	and	Legere	reported	that	the	difference	between	male	and	female	PCPs	peaks	between	the	ages	of	36	and	40,	and	declines	thereafter	(Constant	&	Leger,	2008).		Unadjusted	results	from	the	remaining	studies	were	relatively	consistent:	male	PCPs	were	reported	to	deliver	more	services	than	female	PCPs	(700	vs.	399/month)	(Keane	et	al.,	1991),	and	to	have	more	patient	encounters	(between	32	and	72/week)	(for	example:	(Atkin,	2000;	Norton	et	al.,	1994;	Raymont	et	al.,	2005)).	Female	PCPs,	however,	were	found	to	manage	more	problems	per	patient	encounter	(157.8	vs.	145.4	per	100	encounters)	and	spend	40%	more	time	with	each	patient	(20.5	vs.	14.4	minutes)	(Britt,	Bhasale,	Miles,	Meza,	&	Sayer,	1996;	Chaytors,	Szafran,	&	Crutcher,	2001).		In	their	longitudinal	examination	of	intergenerational	differences	in	workloads	of	physicians	from	six	Canadian	provinces,	Watson	et	al.	found	that	between	1992	and	2001,	female	PCPs	reduced	their	workloads	(defined	as	number	of	visits	per	year)	by	6.1%,	while	male	workloads	remained	stable.		The	result	was	an	accentuated	difference	in	workload	over	time:	female	physicians’	workloads	were,	on	average,	74%	of	the	workloads	of	their	male	counterparts	in	1992,	and	68%	in	2001	(Watson	et	al.,	2006).		These	results	run	somewhat	counter	to	those	reported	by	Crossley	et	al.	who	found	that	the	gap	in	self-reported	hours	worked	between	male	and	female	physicians	was	narrowing	(Crossley	et	al.,	2009).	It	is	possible	that	these	conflicting		 46	results	could	be	caused	by	some	combination	of	differences	in	time	periods	used	for	analysis	(1982-2003	vs.	1992-2001),	outcome	measure	(hours	vs.	billed	consultations)	or	other	differences	in	methodology	(Crossley	et	al.,	2009).	If	one	takes	both	sets	of	results	at	face	value	and	attempts	to	reconcile	them,	a	possible	conclusion	would	be	that	male	PCPs	are	reducing	their	hours	while	maintaining	visit	counts,	while	female	PCPs	are	maintaining	their	hours,	but	are	decreasing	their	visits.	Taking	account	of	other	results	cited	here,	it	may	be	that	female	PCPs	are	simply	changing	their	style	of	practice,	taking	more	time	with	each	patient	and	dealing	with	more	problems	per	visit.	The	other	conclusion	that	can	be	drawn	from	these	results	is	that	measuring	physician	productivity	is	difficult,	and	that	the	numerator	(outputs	or	outcomes	per	unit	of	activity)	matters	(Evans,	Schnieder,	&	Barer,	2010).	2.2.2.3	Scope	of	work		Patient	Characteristics:	Compared	with	male	PCPs,	female	PCPs	saw	a	higher	proportion	of	female	patients	(Bensing,	van	den	Brink-Muinen,	&	de	Bakker,	1993;	Britt	et	al.,	1996;	Cohen	et	al.,	1991;	Keane	et	al.,	1991),	in	all	age	groups,	but	especially	in	the	15-49	age	category	(Cohen	et	al.,	1991;	Keane	et	al.,	1991).	They	also	saw	fewer	elderly	patients	than	their	male	counterparts	(Carek	et	al.,	2003;	Harrison	et	al.,	2011).	These	results	survived	multivariate	analyses	that	accounted	for	the	age	of	physician,	practice	location,	and	graduation	period	(Cohen	et	al.,	1991).		Care	Delivered:	Controlling	for	patient	and	physician	demographics,	female	PCPs	were	significantly	more	likely	to	manage	issues	related	to	the	reproductive	or		 47	female	genital	system	(Bensing	et	al.,	1993;	Britt	et	al.,	1996;	Harrison	et	al.,	2011),	as	well	as	psychological	and	social	problems	(Bensing	et	al.,	1993;	Britt	et	al.,	1996;	Harrison	et	al.,	2011).		Female	physicians	were	less	likely	to	manage	issues	of	the	musculoskeletal,	or	male	genitourinary	systems	(Britt	et	al.,	1996;	Harrison	et	al.,	2011).			With	respect	to	obstetrical	and	prenatal	care,	results	from	U.S.-based	literature	were	inconsistent	with	those	from	Canada.	In	the	U.S.,	male	and	female	PCPs	were	equally	likely	to	provide	prenatal	care,	with	or	without	delivery	(Carek	et	al.,	2003).	In	contrast,	in	Canada,	female	physicians	were	more	likely	than	their	male	counterparts	to	provide	prenatal	care,	but	were	less	likely	to	provide	intrapartum	care	(Keane	et	al.,	1991).		After	adjusting	for	problems	per	encounter,	as	well	as	physician,	practice	and	patient	characteristics,	Australian	male	PCPs	had	a	higher	rate	of	prescribing	(4.3%	more	medications	per	100	patients)	(Harrison	et	al.,	2011).	Female	PCPs	recorded	19.5%	more	clinical	treatments	(e.g.	education	and	counseling),	18.5%	more	referrals,	8.1%	more	imaging	ordered,	and	9.6%	more	pathology	tests	ordered	(Harrison	et	al.,	2011).		In	their	1993	study	on	service	delivery	trends	for	male	and	female	PCPs	in	the	Netherlands,	Bensing	and	colleagues	found	that	female	physicians	wrote	fewer	prescriptions	and	performed	fewer	technical	interventions	compared	with	male	physicians;	however,	they	ordered	more	laboratory	tests	(Bensing	et	al.,	1993).	They	found	no	difference	in	the	rate	of	referrals	to	specialists	(Bensing	et	al.,	1993).		Chan	and	colleagues	examined	the	referral	rates	for	Canadian	male	and		 48	female	PCPs.	Like	Harrison	et	al.,	they	found	that	female	physicians	referred	to	specialists	about	10%	more	frequently	than	their	male	colleagues	after	making	adjustments	for	patient	age	and	gender	(Chan	&	Austin,	2003).	Boerma	and	van	den	Brink-Muinen	found	that	male	European	PCPs	were	more	involved	in	technical	procedures;	however	the	difference	was	smaller	in	countries	with	a	gate-keeping	system	(Boerma	&	van	den	Brink-Muinen,	2000).		Out-of-Office	and	Off-Hours	Care:	Five	studies	examined	the	provision	of	out-of-office	and/or	off-hours	care	(Bergeron	et	al.,	1999;	Boerma	&	van	den	Brink-Muinen,	2000;	Carek	et	al.,	2003;	Keane	et	al.,	1991;	Raymont	et	al.,	2005).	In	1991,	Keane	et	al.	reported	that	a	smaller	proportion	of	Canadian	female	than	male	PCPs	billed	for	home	visits	(1.5	vs.	3.7	per	100	patients)	and	after-hours	care	(7.0	vs.	9.6	per	100	patients,	after	controlling	for	the	effects	of	place	and	date	of	MD	graduation,	practice	location,	certification	status,	and	work	status	(Keane	et	al.,	1991).			Adjusted	for	patient,	physician	and	practice	characteristics,	male	PCPs	also	more	routinely	made	long-term	care	facility	visits	(50.6%	vs.	35.5%	for	females),	and	home	visits	(49.0%	vs.	33.8%	for	females)	(Boerma	&	van	den	Brink-Muinen,	2000).	Male	PCPs	were	also	more	likely	than	their	female	counterparts	to	bill	for	time	in	the	hospital	(14.8.%	vs.	13.1%,	emergency	room	(37.0%	vs.	14.2%),	or	for	surgical	assists	(64.8%	vs.	47.2%)	(Boerma	&	van	den	Brink-Muinen,	2000).	Consistent	with	the	multivariate	results	from	Keane	et	al.	and	Boerma	et	al.,	the	two	studies	that	report	only	bivariate	results	found	that	female	PCPs	were	less	likely	to	provide	after-hours	services	(Carek	et	al.,	2003;	Raymont	et	al.,	2005),	make	house	calls	(e.g	12.7%	vs.	15.2%	for	men),	and	spend	significantly	more	of		 49	their	work	time	in	office	or	clinic	practice	(87.9%	vs.	80.9%	for	men	(Carek	et	al.,	2003).	This	is	in	contrast	to	findings	reported	by	Bergeron	et	al.	who	report	that	although	male	physicians	make	more	home	visits	compared	with	female	physicians,	they	spend	an	almost	equal	amount	of	time	on	this	activity	(5.7	vs.	5.2	hours/week)	(Bergeron	et	al.,	1999).	2.2.2.4	Years	of	practice	Patterns	of	retirement	(or	practice	leave)	were	examined	in	four	of	the	included	studies	(Brett,	Arnold-Reed,	Hince,	Wood,	&	Moorhead,	2009;	French	et	al.,	2006;	Leese,	Young,	&	Sibbald,	2002;	McKinstry,	Colthart,	Elliott,	&	Hunter,	2006),	and	results	are	mixed.	French	and	colleagues	found	that	a	similar	proportion	of	male	and	female	PCPs	in	Scotland	intend	to	retire	at	age	59	(French	et	al.,	2006).	In	their	study	of	Australian	physicians,	Brett	et	al.	report	that	male	PCPs	were	more	likely	to	intend	to	retire	before	age	65:	seventy-five	percent	of	women	compared	with	59%	of	men	reported	that	they	intended	to	work	to	normal	retirement	age	(rather	than	retiring	early)	(Brett	et	al.,	2009).		In	their	survey	of	physicians	who	had	recently	left	practice,	however,	Leese	et	al.	found	that	female	leavers	tended	to	be	younger,	and	to	have	children	under	the	age	of	18	(Leese	et	al.,	2002).		This	suggests	that	child-rearing	responsibilities	play	a	key	role	in	decisions	to	leave	practice,	and	that	female	PCPs	are	more	likely	to	leave	practice	for	reasons	other	than	full	retirement,	compared	with	their	male	counterparts.	Leaves	of	absence,	for	reasons	of	childbearing	or	otherwise,	were	not	a	focus	in	any	of	the	articles	included	in	this	review.				 50	2.2.2.5	Practice	characteristics	Female	PCPs	practicing	across	Europe	and	in	Australia	were	less	likely	than	men	to	work	in	solo	practice	rather	than	in	small	or	large	groups	(Europe:	27%	of	women	found	to	work	in	solo	practice	vs.	45.2%	of	men	(Boerma	&	van	den	Brink-Muinen,	2000);	Australia:	4.6%	of	women	work	in	solo	practice,	vs.	13.2%	of	men	(Harrison	et	al.,	2011).		In	the	United	States,	male	and	female	PCPs	are	about	equally	likely	to	practice	within	a	small	group	(32.7%	vs.	38.3%)	(Carek	et	al.,	2003).	Female	PCPs	practicing	in	Europe	were	significantly	less	likely	to	practice	in	rural	areas	compared	with	their	male	counterparts	(14.9%	vs.	27.2%	rural).		In	contrast,	in	the	U.S.	and	Australia,	women	and	men	were	equally	likely	to	choose	rural	practice	(Carek	et	al.,	2003;	Harrison	et	al.,	2011).		Female	PCPs	in	Europe	were	more	likely	to	work	in	inner	city	locations	(33.7%	vs.	18.0%)	(Boerma	&	van	den	Brink-Muinen,	2000).			2.3	Discussion	The	intent	of	this	systematic	review	was	to	examine	the	impact	of	the	increasing	proportion	of	women	in	the	PCP	workforce	on	service	delivery	in	five	areas	that	could	affect	such	projections	of	service	supply:	years	of	practice,	hours	of	work,	intensity	of	work,	scope	of	work,	and	practice	characteristics.	Compared	with	their	male	colleagues,	female	PCPs:		• Self-report	fewer	hours	of	work	(excluding	on-call	time).	• Have	fewer	patient	encounters,	and	deliver	fewer	services	(perhaps	as	an	artifact	of	working	fewer	hours),	but	spend	longer	with	their	patients	during		 51	a	contact,	and	deal	with	more	separate	presenting	problems	during	each	visit.		• Write	fewer	prescriptions,	but	order	more	laboratory	tests,	and	refer	patients	on	to	specialists	more	frequently.	• See	more	female	patients	and	fewer	geriatric	patients.		• Provide	less	out-of-office	(including	home,	nursing	home	and	hospital	visits)	and	off-hours	care.		The	scale	of	the	impact	of	these	findings	on	future	effective	physician	supply	is	difficult	to	determine	with	currently	available	evidence,	given	that	very	few	studies	looked	at	time	trends	or	years	of	practice,	and	results	from	those	that	did	are	inconsistent.		Also,	the	full	impact	will	depend	critically	on	future	trends	in	the	feminization	of	the	work	force.		In	Canada,	and	in	the	UK	and	other	parts	of	Europe,	the	proportion	of	medical	students	who	are	female	ensures	that	the	overall	supply	of	physicians	will	continue	to	become	increasingly	female	in	the	near	term.				Given	that	fact,	the	differences	in	practice	patterns	between	male	and	female	PCPs	could	result	in	increased	derived	demand	for	specialist	physician	services,	laboratory	technicians,	imaging	technicians	or	other	health	professionals,	outside	of	primary	health	care.		The	fact	that	female	PCPs	spend	less	time	in	off-hours	care,	and	are	less	likely	to	serve	patients	at	home	and	in	nursing	homes,	could	increase	the	reliance	on	already-stretched	emergency	departments	and	walk-in	clinics	as	a	source	of	primary	health	care,	and	force	a	rethinking	of	how	medical	care	is	delivered	to	patients	outside	standard	office	hours	and	locations.		 52	It	is	important	to	consider	the	effects	of	childbearing	and	childrearing,	which	were	mentioned	in	several	studies,	but	were	seldom	explicitly	investigated,	and	were	not	the	primary	focus	of	any	of	the	research	documents	reviewed	here.	Female	PCPs	who	had	children	under	age	18	worked	fewer	hours	per	week	and	were	more	likely	to	have	self-reported	part-time	status	compared	with	women	who	did	not.	The	dampening	effect	of	children	on	work	hours	was	twice	as	large	for	women	as	it	was	for	men.	And,	one	study	found	that	once	family	circumstances	were	accounted	for,	the	gender	of	the	physician	had	no	significant	effect	on	hours	worked	(Gravelle	&	Hole,	2007).	An	important	issue	that	was	not	covered	in	any	of	the	literature	reviewed	here	is	the	balance	between	work	and	household	responsibilities	among	physicians.	One	study	found	that	female	physicians	spent	more	time	on	unwaged	childcare	and	other	household	than	male	physicians	(Woodward,	Williams,	Ferrier,	&	Cohen,	1996).	Once	unwaged	household	responsibilities	were	accounted	for,	female	PCPs	who	have	children	worked	an	average	of	90.5	hours	a	week,	compared	with	68.6	hours	per	week	for	males	with	children	(Woodward	et	al.,	1996).	2.3.1	Consistency	of	results	Results	were	strongly	consistent	across	some	of	the	thematic	areas,	and	relatively	less	so	in	others.	In	particular,	results	relating	to	the	hours	and	intensity	of	work	were	consistent	across	studies.	In	other	areas,	such	as	practice	characteristics,	results	were	highly	variable.		 53	The	results	of	this	review	demonstrate	that	the	drivers	of	observed	differences	between	male	and	female	PCPs	are	complex	and	nuanced.	The	size	of	an	observed	gender	difference	varied	based	on	the	characteristics	of	the	health	care	system	under	study	and	on	whether	the	possible	confounding	effects	of	physician	age,	practice	characteristics,	and	in	particular,	family	characteristics	and	part-time	status	were	adequately	controlled.	There	were	at	least	36	different	health	care	systems	represented	by	the	studies	included	in	this	review.	Inconsistent	results	across	studies	may	be	caused	by	health	care	system	differences	including,	but	not	limited	to,	physician	remuneration	mechanisms	and	policies,	the	gatekeeping	role	of	general	practitioners,	and	general	employment	policies.	An	exploration	of	the	role	of	such	system	differences	was	well	beyond	the	scope	of	this	review,	but	is	an	important	area	for	future	research.	Inconsistent	results	could	also	be	a	function	of	methodological	and	measurement	differences	across	studies,	and	whether	the	confounding	effects	of	other	physician,	patient,	and	practice	characteristics	have	been	accounted	for.	For	example,	gender	differences	in	the	number	of	patient	contacts	per-day	disappeared	once	full-	vs.	part-time	status	had	been	accounted	for	in	work	by	Boerma	et	al	(Boerma	&	van	den	Brink-Muinen,	2000).	Differences	in	hours	worked	depended	on	whether	auxiliary	activities	such	as	on-call	time	were	included	as	part	of	“hour	worked”	(Carek	et	al.,	2003).	Similarly,	differences	in	care	provision	were	attenuated	once	patient	characteristics	and	practice	location	was	accounted	for	(e.g.	(Britt	et	al.,	1996;	Harrison	et	al.,	2011)).			 54	2.3.2	Quality	of	included	studies	As	part	of	our	qualitative	assessment	of	study	quality,	we	identified	some	significant	methodological	concerns	with	the	studies	included	in	this	review.	For	the	most	part,	they	relied	on	cross-sectional	retrospective	surveys.		Such	surveys	are	always	subject	to	recall	bias,	though	unless	there	were	systematic	male	vs.	female	differences	in	accuracy	of	recall,	this	may	not	be	an	issue	in	this	particular	circumstance.	But	surveys	do	tend	to	produce	inflated	estimates	of	hours	worked	for	those	who	report	high	hours	(more	often	male	physicians)	and	deflated	estimates	for	those	reporting	low	hours	(more	often	female	physicians),	which	may	exaggerate	any	true	gender	difference	(Williams,	2004).	Many	studies	relied	on	small,	often	unbalanced	samples,	raising	concerns	about	selection	bias.	All	but	one	study	failed	to	adjust	statistically	for	multiple	comparisons,	despite	conducting	as	many	as	155	separate	statistical	significance	tests	(Chaytors	et	al.,	2001).		Perhaps	even	more	concerning,	however,	is	that	12	(35%)	studies	presented	only	unadjusted,	bivariate	results,	failing	to	control	for	the	potential	confounding	effects	of	other	physician,	patient	or	practice	characteristics	(for	example,	(Bensing	et	al.,	1993;	Carek	et	al.,	2003;	Raymont	et	al.,	2005)).	Additionally	6	(18%)	undertook	only	rudimentary	stratification	(for	patient	age	and	gender,	for	example)	(for	example,	(Bensing	et	al.,	1993;	Cohen	et	al.,	1991;	Keane	et	al.,	1991;	Weyrauch,	Boiko,	&	Feeny,	1995)).	Statistical	methods	controlling	for	confounders	may	not	yet	have	been	accepted	practice	in	this	field	when	some	of	these	earlier	papers	were	published,	which	may	explain	their	limited	use.	Comparisons	between	adjusted	and	unadjusted	results	suggest	that	physician	age,	family	characteristics	and	practice		 55	location,	at	a	minimum,	can	have	important	influences	on	apparent	male-female	differences	in	key	practice	and	productivity	indicators.	For	example,	older	physicians	–	who	are	more	likely	to	be	male	–	tend	to	see	more	older	patients	(Boerma	&	van	den	Brink-Muinen,	2000),	and	physicians	who	work	in	rural-based	clinics	practice	differently	from	physicians	who	practice	in	urban	centres	(Chaytors	et	al.,	2001).	Thus	the	impacts	of	physician	age	and	practice	location	may	be	conflated	with	a	gender	effect	in	unadjusted	analyses,	since	female	PCPs	tend	to	be	younger	(Britt	et	al.,	1996)	and	more	likely	to	work	in	urban	centres	in	some	countries	(Boerma	&	van	den	Brink-Muinen,	2000).		2.3.3	Remaining	knowledge	gaps	and	future	research	Given	the	reliance	on	cross-sectional	and	survey	data,	and	the	relative	underutilization	of	longitudinal	or	administrative	datasets	in	this	area,	there	remains	a	need	to	critically	examine	activity	levels,	over	time	and	at	a	population	level,	adjusting	for	the	potentially	confounding	effects	of	age	and	cohort.	The	issue	of	retirement	patterns	has	also	not	been	adequately	examined	with	reference	to	the	effects	on	working.	It	is	possible,	for	example,	that	although	female	PCPs	work	less,	especially	around	childbearing	years,	they	may	retire	later	than	their	male	counterparts,	reducing	or	even	eliminating	a	career	difference	in	time	spent	working.	While	historically	this	may	not	have	been	true,	trends	over	time	suggest	that	it	might	become	so	in	future.	The	key	point	is	that	differences	in	retirement	patterns	between	male	and	female	physicians	may	partially	or	wholly	offset	other	trends	in	service	provision,	when	viewed	over	an	entire	life	cycle.		Leaves	of	absence		 56	taken	for	parental	or	other	reasons	should	also	be	examined	for	their	effects	on	both	time	and	intensity	of	working.	No	studies	included	in	this	review	examined	absences	from	practice.		To	date,	the	literature	examining	other	practice	differences	between	male	and	female	physicians	that	could	have	an	important	impact	on	health	human	resources	planning	has	been	limited.	More	studies	comparing	the	patient	populations	of	male	and	female	PCPs	–	beyond	simple	gender	concordance	and	patient	age	–	are	certainly	warranted.	Specifically,	very	little	work	has	been	done	examining	differences	in	patient	morbidity	levels,	or	chronic	disease	burdens.	Additionally,	more	nuanced	investigations	of	service	mix,	problems	seen,	and	care	delivered	would	address	currently	unanswered,	but	important,	questions	bearing	on	the	future	provision	of	physician	services.	For	example,	differences	in	practice	style	between	male	and	female	physicians	have	currently	received	little	attention	beyond	comparisons	of	time	taken	for	each	appointment.		Issues	of	work-life	balance	and	child-rearing	and	household	responsibilities	are	also	under-researched,	especially	given	their	observed	impact	on	full-	vs.	part-time	job	status	and	working	hours	(Gravelle	&	Hole,	2007;	C.	A.	Woodward	et	al.,	1996).	In	the	2007	and	2010	Canadian	National	Physician	Surveys,	the	majority	of	respondents	identified	attaining	balance	between	personal	and	professional	life	as	the	most	important	factor	for	a	satisfying	practice	(College	of	Family	Physicians	of	Canada,	The	Royal	College	of	Physicians	and	Surgeons	of	Canada,	&	The	Canadian	Medical	Association,	2011).	Physicians,	regardless	of	gender,	are	increasingly	(and	not	unreasonably)	seeking	a	work	environment	that	provides	this	balance,	without		 57	compromising	the	quality	of	care	they	provide	to	their	patients	(College	of	Family	Physicians	of	Canada,	2004).		Secular	trends	in	time	made	available	for	clinical	practice	obviously	have	direct	implications	for	projections	of	physician	service	provision.	2.3.4	Limitations		This	systematic	review	used	comprehensive	search	strategies	encompassing	multiple	peer-reviewed	and	grey-literature	sources	to	maximize	capture	of	relevant	articles	and	minimize	publication	bias.	The	restriction	of	articles	to	those	published	in	English	and	within	the	last	23	years	may	have	eliminated	some	potentially	relevant	studies.	Additionally,	because	the	area	of	research	is	not	yet	well-indexed,	and	the	specific	topic	area	is	broad,	some	studies	that	would	be	relevant,	but	whose	main	comparison	was	not	male	versus	female	PCPs	may	have	been	missed.		Our	decision	to	include	only	those	studies	that	focused	on	primary	care	physicians,	defined	here	as	general	practitioners	or	family	medicine	specialists,	(rather	than	also	including	other	specialists	–	such	as	general	internists	or	pediatricians	–	who	may	practice	like	primary	care	physicians	under	certain	circumstances)	may	limit	the	generalizability	of	our	results,	particularly	with	respect	to	research	from	the	United	States.		An	additional	limitation	is	the	decision	not	to	eliminate	studies	that	were	deemed	of	poor	quality.	The	methodologies	employed	in	many	of	the	studies	is	certainly	far	from	ideal,	with	many	relying	on	small,	unbalanced	samples,	retrospective	surveys,	and	incomplete	(or	no)	control	for	the	effect	of	confounding		 58	factors.	These	studies	were,	however,	retained	in	the	review	since	none	of	the	30	included	would	have	achieved	the	level	of	guidance	required	for	formal	guidelines	(for	example,	those	issued	by	the	Cochrane	Collaboration)	and	thus	there	was	no	straightforward	way	to	gauge	methodological	quality.		Meta-analytic	techniques	could	have	been	a	useful	way	to	summarize	the	research	within	individual	thematic	and	sub-thematic	areas;	however,	small	numbers	and	the	variance	in	outcome	measures	even	within	individual	sub-themes	was	too	great	to	allow	for	the	use	of	those	tools.	2.3.5	Implications	for	health	human	resource	planners		Projections	of	physician	supply	must	take	into	account	variables	other	than	estimated	future	physician	headcounts.	At	a	minimum,	more	robust	measures	that	account	for	gender	differences	in	service	volumes,	but	that	also	address	the	implications	of	the	differences	in	patient	mix,	service	mix,	and	practice	style	between	male	and	female	physicians	need	to	be	developed	and	used	as	evidence	in	these	areas	becomes	available.		Other	demographic	and	workforce	factors,	such	as	the	impact	of	physician	age	and	cohort	–	should	also	be	considered.		2.4	Conclusions	Compared	with	their	male	counterparts,	female	PCPs	spend	less	time	working,	and	deliver	less	care.	Evidence	as	to	whether	this	gap	is	narrowing	is	mixed.	The	effect	of	child	rearing	is	critically	important,	affecting	female	PCPs	far	more	than	their	male	counterparts,	in	terms	of	impact	on	participation	in	clinical	practice.	Once	the	effect	of	family	characteristics	has	been	accounted	for,	sex	has	no		 59	effect	on	time	spent	working.	Issues	of	work-life	balance,	caregiving	and	child-rearing	responsibilities	warrant	attention	in	future	research.	The	literature	focuses	heavily	on	differences	in	the	amount	of	work	done	by	female	compared	with	male	physicians,	and	is	almost	exclusively	based	on	retrospective	surveys	with	some	significant	methodological	limitations.	These	studies	tell	us	nothing	about	differences	in	the	appropriateness	or	quality	of	care.	Also,	more	research	examining	differences	in	practice	characteristics,	and	patient/service	mix,	is	warranted	in	order	to	support	the	development	of	robust	forecasts	of	physician	supply.		Such	forecasts	would	ideally	take	into	account	sex-related	differences	in	volume,	but	also	the	implications	of	the	differences	in	patient/service	mix	and	practice	style,	and	temporal	trends	in	each	of	these.		The	extant	literature	suggests	that	secular	trends	in	hours	of	work	may	dominate	sex-related	differences	in	service	provision.	This	thesis	addresses	some	of	the	gaps	in	the	literature	identified	here	using	British	Columbia’s	rich	administrative	data	resources.	In	particular,	I	conduct	a	longitudinal	analysis	examining	differences	in	activity	levels	between	male	and	female	physicians,	adjusting	for	secular	trends	related	to	physician	age	and	cohort	(Chapters	4	and	5).	I	also	conduct	a	nuanced	examination	of	differences	in	patient	and	service	mix	(Chapters	6	and	7),	and	discuss	resulting	service	supply	trends.			 		 60	CHAPTER	3:	Data	sources,	file	preparation,	and	study	variables	3.1	Outline	of	data	sources	and	linkage	procedures		The	majority	of	the	data	resources	I	used	for	this	project	fall	under	the	authority	of	the	BC	Ministry	of	Health,	and	are	held	by	Population	Data	BC	(http://www.popdata.bc.ca)	for	approved	research	purposes.	Access	to	Population	Data	BC’s	holdings	is	governed	by	a	Data	Access	Framework	that	ensures	that	research	projects	are	in	compliance	with	BC’s	Freedom	of	Information	and	Protection	of	Privacy	Act	(Freedom	of	Information	and	Protection	of	Privacy	Act,	1996).	In	order	to	access	these	data,	applicants’	proposed	work	must	be	reviewed	by	an	ethics	committee,	as	well	as	undergo	traditional	academic	peer	review	to	demonstrate	scientific	merit.		I	obtained	ethics	approval	from	the	University	of	British	Columbia’s	Behavioral	Research	Ethics	Board,	and	was	granted	permission	to	use	de-identified,	individual-level	data	from	the	data	stewards,	the	BC	Ministry	of	Health	and	the	College	of	Physicians	and	Surgeons	of	British	Columbia	(CPSBC).	My	work	underwent	peer	review	as	part	of	my	applications	for	doctoral	research	funding5,	and	as	part	of	the	School	of	Population	and	Public	Heath’s	Thesis	Screening	Panel.		Data	from	Population	Data	BC	can	be	linked	to	a	selection	of	external	datasets.	I	requested	access	to	two	external	files:	the	Physician	Registry,	which	falls	under	the	authority	of	the	CPSBC,	and	Alternative	Payment	Plan	(APP)	data,	which																																																									5	Applications	were	submitted	to	the	Canadian	Institutes	of	Health	Research,	the	Western	Regional	Training	Centre	for	Health	Services	Research,	and	the	Transdisciplinary	Understanding	and	Training	on	Research	–	Primary	Health	Care	programs	for	peer-reviewed	research	support.			 61	are	held	by	the	BC	Ministry	of	Health.	Analysts	at	Population	Data	BC	were	responsible	for	conducting	the	linkage	between	the	CPSBC’s	physician	registry,	the	alternative	payments	data,	and	its	own	data	holdings.	They	also	de-identified	the	data,	adding	unique	study	ID	codes	for	each	individual	in	the	resulting	dataset.		These	unique	identifiers	were	used	to	link	records	for	individuals	(both	physicians	and	patients)	over	time	and	to	link	physicians	with	specific	patients.	I	requested	data	for	the	years	2004/5	through	to	2011/12,	which	was	the	most	recent	year	available	at	the	time	of	data	request.		The	specific	datasets	to	which	I	requested	access	for	this	dissertation	include	the	following:	• Medical	Services	Plan	(MSP)	Practitioner	File	• CPSBC	Physician	Registry	• MSP	Consolidation	File	• MSP	Payment	Information	File	• Discharge	Abstracts	Database	• Vital	Statistics	–	Deaths		• APP	Database	The	data	extracted	from	these	databases	were	used	to	generate	the	outputs	listed	in	Table	3.1.	Complete	descriptions	of	each	database,	and	how	I	used	the	data	therein,	are	included	in	sections	3.2	to	3.7	below.						 62	Table	3.1:	Data	Sources	and	Relevant	Outputs	Data	Outputs	 Relevant	Data	Set	Physician	roster	 MSP	Practitioner	File	CPSBC	Physician	Registry	APP	Dataset	Physician	demographic	characteristics	 MSP	Practitioner	File	CPSBC	Physician	Registry	Patient	demographic	characteristics	 MSP	Consolidation	File	Patient	health	status	 MSP	Payment	Information	File	Discharge	Abstract	Database	(DAD)	Vital	Statistics	-	Deaths	Physician	activity/salary	 MSP	Payment	Information	File	APP	Dataset	Patient	care	approaches	 MSP	Payment	Information	File		3.2	Datasets	3.2.1	MSP	Practitioner	File	and	CPSBC	Registry	The	CPSBC	Registry	(referred	to	henceforth	as	“the	Registry”)	is	a	database	that	contains	demographic	and	geographic	and	information	on	all	registered	and	practicing	physicians	in	BC.	Only	data	pertaining	to	physicians	with	a	specialty	code	indicating	that	they	practice	primary	medicine	was	used	in	this	thesis.		The	database	also	includes	a	status	field	(indicating	whether	a	physician	is	actively	practicing,	or	retired),	age,	sex,	year	of	graduation,	place	of	training,	practice	location,	and	date	of	death	where	applicable.		 63	The	Registry	is	used	is	to	generate	the	MSP	Practitioner	file,	which	is	stewarded	by	Population	Data	BC.	Both	databases	therefore	have	the	same	set	of	fields.	The	Practitioner	File,	however,	contains	data	only	for	those	physicians	who	have	fee-for-service	(FFS)	billings.	Physicians	who	were	remunerated	entirely	under	an	alternative	payment	model	would	not	appear	in	the	Practitioner	File.	In	the	case	of	any	disagreement	between	the	CPSBC	Registry	and	the	MSP	Practitioner	file,	I	relied	on	the	data	from	the	CPSBC	registry	(since	it	is	used	in	the	creation	of	the	Practitioner	File	by/for	the	Ministry	of	Health).			 I	used	the	MSP	Practitioner	File	(in	combination	with	the	CPSBC	Registry	and	APP	database)	to	create	a	complete	roster	of	BC’s	active	primary	care	physicians.	Included	in	the	roster	are	all	physicians	with	a	specialty	code	of	00	(general	practice	or	family	medicine)	and	who	billed/were	paid	a	minimum	of	$1	(in	either	MSP	or	FFS	remuneration)	in	at	least	one	year	between	2005/06	and	2011/12.		3.2.2	MSP	Consolidation	File	The	MSP	Consolidation	File	is	the	central	demographics	database	maintained	by	Population	Data	BC;	it	contains	demographic	and	regional	data	for	every	resident	of	BC,	updated	annually,	and	excluding	those	whose	health	care	services	coverage	is	provided	by	the	Federal	Government6,	from	April	1985	onwards.	The	content	of	the	File	is	regularly	cleaned	and	validated	by	Population	Data	BC.		In	particular,	care	is	taken	to	ensure	consistency	in	demographic	information	over	time.																																																										6Exclusions	include	Aboriginal	peoples	living	on	reserve,	the	Canadian	Forces,	veterans,	and	inmates	in	federal	penitentiaries.	This	represents	a	very	small	segment	of	the	BC	population,	but	one	whose	health	status	tends	to	be	much	poorer	(Adelson,	2005)				 64	For	this	thesis,	I	developed	the	following	demographic	variables	using	the	MSP	Consolidation	File:	patient	year	of	birth,	sex,	geographic	location	(health	authority	and	health	services	delivery	area,	and	neighbourhood	income	quintile,	(used	as	a	proxy	measure	for	socioeconomic	status).		3.2.3	MSP	Payment	Information	File	The	MSP	Payment	Information	File	is	a	comprehensive	record	of	fee-for-service	physician	payments	including	for	laboratory	and	diagnostic	procedures,	referable	back	to	the	individual	billing	physician	(meaning	that	the	data	is	at	the	physician	level)	(“Population	Data	BC	Website	-	Medical	Services	Plan	Payment	Information	File,”	2013).	Each	claim	record	contains	the	fee	item	code	and	amount	paid	as	well	as	the	date	of	service	delivery,	and	a	most	responsible	diagnostic	code	(i.e.	the	diagnosis	that	led	to	the	delivery	of	a	particular	service).	Records	are	linked	to	the	individual	patient	who	received	the	service.	Diagnoses	are	summarized	using	the	International	Classification	of	Diseases	(ICD)-9	codes,	which	have	been	validated	for	specificity	and	completeness	(Williams	&	Young,	1996).	The	majority	of	the	records	in	the	MSP	Payment	Information	File	have	been	submitted	directly	to	MSP	by	practitioners’	offices,	using	an	electronic	system.	MSP	also	conducts	regular	audits	and	quality	checks,	to	ensure	the	accuracy	of	the	database	(“Population	Data	BC	Website	-	Medical	Services	Plan	Payment	Information	File,”	2013).		For	the	purposes	of	this	thesis,	the	Payment	Information	File	was	used	to	measure	the	activity	of	individual	physicians	over	time.	See	Section	3.5.2	for	an		 65	explanation	of	activity	measures.	It	was	also	used	to	characterize	physicians’	visit	patterns	and	patient	care	approaches,	and	to	create	measures	of	patient	morbidity.		Additional	details	are	provided	in	Sections	3.5.3	and	3.5.4	below.		3.2.4	APP	Database	An	important	limitation	of	the	MSP	payment	information	file	is	that	it	does	not	capture	services	that	are	paid	by	non-fee-for-service	methods.	These	alternative	payments	include,	for	example,	salaried	and	sessional	payment	arrangements,	service	agreements,	and	on-call	payments,	and	are	maintained	in	a	separate	APP	database.	Each	record	in	this	database	indicates	the	type	of	payment	(e.g.	on-call),	the	dollar	value,	the	date,	and	the	claiming	physician.	The	complete	list	of	payment	types	is	described	in	Appendix	3.	There	are	two	key	limitations	associated	with	the	use	of	the	APP	database.	First,	unlike	the	MSP	payment	records,	records	in	the	APP	dataset	do	not	include	information	on	the	specific	service(s)	delivered,	and	are	not	linked	to	the	individual	patient	who	received	the	service.		Second,	this	dataset	covers	only	2005/6	to	2011/12,	which	is	why	I	selected	these	years	as	my	study	period.	I	used	the	APP	database	to	identify	physicians	in	the	MSP/CPSBC	practitioner	files	who	were	receiving	all	or	a	portion	of	their	total	2005/6-2011/2	income	from	non-fee-for-service	sources,	and	to	generate	accurate	overall	physician-level	payment	figures	(which	are	a	combination	of	fee-for-service	MSP	and	non-fee-for-service	APP	payments).			 66	3.2.5	Discharge	Abstracts	Database	and	Vital	Statistics	The	DAD	contains	information	on	all	inpatient	(acute,	chronic	or	rehabilitation)	or	same-day	surgical	admissions	to	acute	care	hospitals	from	April	1985,	recorded	at	the	time	of	patient	discharge	or	death	(“Population	Data	BC	Website	-	Discharge	Abstracts	Database,”	2013).	Data	on	out-of-province	hospital	stays	for	BC	residents	are	also	included.	For	each	patient	stay,	the	database	includes	patient	ID	and	diagnoses,	level	of	care	(e.g.	Acute	Care,	Day	Surgery	etc.),	and	resource	intensity	weight,	which	indicates	the	intensity	of	service	delivery	and	is	used	primarily	for	hospital	service	costing.	Diagnostic	reasons	for	admission	are	summarized	using	ICD	codes.	ICD-9	was	used	in	BC	hospitals	until	2002	when	ICD-10	was	introduced.			 The	BC	Vital	Statistics	death	registry	includes	all	deaths	registered	in	the	province,	regardless	of	where	those	deaths	occurred.		Deaths	that	occur	in	hospital	are	also	registered	in	the	DAD.	Each	record	contains	the	underlying	cause	of	death	–	coded	using	ICD-9	codes	until	1999	and	then	ICD-10	thereafter	–	and	the	date	and	place	of	death.			 For	the	purposes	of	this	thesis,	I	used	the	DAD	and	the	cause	of	death	recorded	in	either	the	DAD	or	Vital	Statistics	Death	records	to	supplement	the	MSP	Payment	Information	File	as	a	source	of	information	on	patient	diagnoses	with	which	to	generate	morbidity	variables.			 67	3.3	Analytic	file	creation	All	data	management	and	analyses	were	completed	using	Statistical	Analysis	System	(SAS).		The	initial	data	provided	by	Population	Data	BC	included	individual	datasets	for	each	fiscal	year	of	MSP	data7	and	each	fiscal	year	of	APP	data.	The	MSP	files	contained	a	single	record	for	each	physician-patient	year,	describing	the	number	and	nature	of	their	patient	contacts,	as	well	as	their	respective	demographic	characteristics.		The	data	in	this	file	were	aggregated	by	creating	yearly	summary	records	for	each	physician,	describing	that	physician’s	activity,	as	well	as	the	characteristics	of	the	patients	the	physician	interacted	with	in	that	year.	These	annual	files	allowed	for	the	tracking	of	changes	in	physician	demographic	data,	for	example,	practice	location	over	time.		The	annual	APP	datasets,	which	were	prepared	and	maintained	separately,	similarly	contained	a	single	record	for	each	physician,	and	described	that	physician’s	APP	activity	for	that	year.		The	datasets	were	prepared	for	analysis	using	the	following	steps:		1) Collapse	the	physician-patient	pair	data	in	the	MSP	annual	files	by	physician,	creating	one	annual	record	for	each	physician,	summarizing	activity	across	all	patients;	2) Merge	the	set	of	MSP	annual	files	into	a	single	database,	adding	a	variable	to	indicate	the	source	fiscal	year;	3) Merge	the	set	of	APP	annual	files	into	a	single	database,	adding	a	variable	to	indicate	the	source	fiscal	year8;	4) Merge	APP	fields	into	the	MSP	database,	such	that	each	physician	has	MSP	and	APP	income	sources	for	each	year	of	activity.																																																										7	This	file	included	data	from	the	MSP	Practitioner	File,	CPSBC	Physician	Registry,	MSP	Consolidation	File,	MSP	Payment	Information	File,	the	DAD,	and	Vital	Statistics.	8	Since	the	APP	data	files	do	not	contain	information	at	the	physician-patient	pair	level,	no	collapsing	was	necessary	before	the	annual	files	could	be	merged.				 68	5) Transpose	the	dataset	on	practitioner	id,	creating	a	secondary	file	with	one	record	per	physician,	covering	all	years	of	that	physician’s	activity.	I	conducted	my	analyses	on	two	datasets:	one	person-time	database,	where	each	physician	has	as	many	records	as	years	they	are	active	in	the	dataset;	and	a	second	where	each	physician	has	a	single	record	summarizing	all	years	of	activity.		3.4	Creation	of	physician	cohort		The	process	for	selecting	my	physician	cohort	is	described	in	Figure	3.1	below.	I	identified	all	those	physicians	who	had	a	CPSBC	specialty	of	“general	practice”	(code:	00),	which	includes	general	practitioners	and	doctors	with	a	“family	practice”	specialty.	This	initial	cohort	contained	6735	primary	care	physicians,	4205	(62%)	males	and	2530	(38%)	females.		I	removed	physicians	who,	at	any	point	during	the	study	period,	had	an	ineligible	practice	status.	I	deemed	a	status	“ineligible”	if	it	meant	that	the	record	of	a	physician’s	activity	was	incomplete,	or	that	the	physician	was	inactive	for	a	reason	other	than	retirement	or	leave	of	absence.	This	screen	resulted	in	the	removal	of	156	physicians	(61	women	(39%)	and	95	men	(61%)),	leaving	a	cohort	of	6579	(4110	males	(64%)	and	2469	females	(38%).	Ninety	eight	percent	of	the	cohort	was	retained.			 69		Figure	3.1:	Selection	of	physician	cohort		3.5	Variables	and	indicators	A	list	of	the	variables	and	indicators	constructed	for	this	study	is	included	in	a	data	dictionary	in	Appendix	4.	Brief	descriptions	of	key	variables	are	presented	here.	Additional	details	about	how	each	variable	was	constructed	and	used	can	be	found	in	subsequent	chapters.		Table	3.2	outlines	how	each	set	of	variables	is	used,	in	each	hypothesis.		N	=	6735 4205	(62%)	males 2530	(38%)	females	 N	=	6579 4110	(64%)	males 2469	(38%)	females	 N	=	156	Excluded	 			Out-of-province:	 N=143 			Opted-out:	 	 N=3	 			Suspended:	 	 N=9 			De-enrolled:	 N=1 	 70	Table	3.2:	Variable	Matrix	Variables		 Hypotheses*	1.1	 1.2	 1.3	 2.1	 2.2	 3.1	 3.2	 4.1	 4.2	 4.3	Physician	Gender	 		 		 		 		 		 		 	 		 		 		Physician	Demographics:	Age,	Location,	Cohort,	Place	of	Training	 		 		 		 		 		 		 	 		 		 		Activity:	Billings,	Contacts,	Services	 		 		 		 		 		 		 	 		 		 		Activity:	Clinical,	Incentives,	Non-Clinical	 		 		 		 		 		 		 	 		 		 		Patient	Characteristics:	Age,	Gender,	Morbidity	 		 		 		 		 		 		 	 		 		 		Practice	Patterns:	Referrals	 		 		 		 		 		 		 	 		 		 		Practice	Patterns:	Practice	Patterns	 		 		 		 		 		 		 	 		 		 		Key:	 Primary	independent	variable	 Explanatory	Variables	 Outcome	Variable	*Chapter	4:	1.1:	Female	PCPs	will	have	lower	age-adjusted	activity	levels	(contacts,	visits,	dollars	billed)	per	unit	time	compared	with	their	male	counterparts.	1.2:	The	difference	in	age-adjusted	activity	levels	per	unit	time	for	male	vs.	female	physicians	will	decrease	with	time	(period/cohort	effect)	1.3	The	difference	in	activity	levels	will	be	greatest	during	childbearing	years,	and	smallest	amongst	primary	care	physicians	aged	65	and	over.	Chapter	5:	2.1:	Male	physicians	will	be	more	likely	to	take	advantage	of	clinical	and	non-clinical	incentive	payments,	and	these	payments	will	make	up	a	larger	proportion	of	their	total	incomes.			2.2:	Clinical	incentives	and	non-clinical	incentives	will	represent	a	larger	proportion	of	physician	income	over-time,	regardless	of	gender.	At	the	same	time,	absolute	per-capita	payments	for	clinical	care	will	decline.				Chapter	6:	3.1:	Female	physicians	will	see	more	female	patients,	and	fewer	elderly	ones.	Male	and	female	physicians	will	be	equally	likely	to	treat	sick	patients	with	multiple	chronic	diseases	or	disabilities.	3.2:	If	differences	in	patient	population	characteristics	are	observed,	these	differences	will	affect	any	gender	difference	observed	in	2.1	and	2.2.		Chapter	7:	4.1:	Female	PCPs	will	be	more	likely	to	refer	patients	to	other	forms	of	care	(including	specialists,	and	diagnostic	and	laboratory	testing).		4.2:	Female	PCPs	will	be	less	likely	to	bill	for	out-of-office	care	provision,	and	a	greater	proportion	will	be	characterized	as	“office	only”	providers.	4.3	More	female	than	male	PCPs	will	provide	obstetrical	services;	however,	the	proportion	of	physicians	who	provide	obstetrical	care	will	decline	over	time	for	physicians	of	both	genders		4.4:	Female	physicians	will	have	more	frequent	(consistent)	contact	with	their	patients.		 71	3.5.1	Physician	characteristics		(Usage:	Explanatory	variables	in	Chapter	4	-	hypotheses	1.1,	1.2,	1.3,	Chapter	5	–	hypotheses	2.1,	2.2,	Chapter	6	–	hypothesis	3.1,3.2,	and	Chapter	7	–	hypotheses	4.1,	4.2,	4.3,	4.4)		 Physician	gender	(male	or	female)	was	used	as	the	primary	independent	variable	for	all	thesis	objectives.	Other	demographic	factors	–	practice	location,	and	age	(in	ten-year	intervals),	time	period,	cohort	(year	of	medical	school	graduation	in	ten-year	intervals),	and	location	of	training	(Canada	or	international)	–	were	included	as	important	intervening	and	confounding	factors.	These	data	were	generated	annually	from	the	CPSBC	Registry	and	MSP	Practitioner	File,	capturing	any	change	in	a	physician’s	practice	location	over	time.	I	derived	practice	rurality	using	existing	health	service	delivery	areas	(Table	3.3).	Table	3.3:	Practice	rurality	Practice	Rurality	 Health	Service	Delivery	Areas	Metropolitan	 Fraser	Valley	Simon	Fraser	South	Fraser	Richmond	Vancouver	South	Vancouver	Island	Urban-Dominated	 Okanagan	North	Shore/Garibaldi	Central	Vancouver	Island	Northern	Interior	Rural-Dominated	 East	Kootenay	Kootenay/Boundary	Thompson/Cariboo	North	Vancouver	Island	North	West	North	East			 72	3.5.2	Physician	activity		3.5.2.1	Services,	contacts	and	dollars	billed		(Usage:	Outcome	variables	in	Chapter	4	–	hypotheses	1.1,	1.2,	1.3	and	Chapter	7	–	hypothesis	4.4;	Explanatory	variables	in	Chapter	5	–	hypotheses	2.1,	2.2)		 I	measured	physician	activity	using	three	complementary	metrics:	service	counts	per	unit	time,	patient	contacts	per	unit	time,	and	dollars	billed	per	unit	time.	I	conducted	most	analyses	using	activity/year.	Service	counts	are	billings	for	individual	fee	items.	A	physician	may	bill	for	multiple	services	in	a	single	patient	encounter	(referred	to	henceforth	as	a	“patient	contact”).	Dollars	paid	is	simply	the	total	of	the	fee	items	paid	per	physician,	per	contact	or	per	unit	time.		To	conduct	a	longitudinal	analyses	of	dollars	paid	(as	a	proxy	for	physician	activity	level),	I	needed	to	account	for	artificial	changes	in	activity	levels	attributable	to	changes	in	fee	level	over	time.	Thus,	I	valued	all	items	at	the	fee	level	effective	April	1,	2010	(Pascali,	1995).	Additionally,	fee	codes	are	frequently	retired,	and	new	ones	are	added	to	replace	them.	To	ensure	consistency	over	time,	I	used	a	crosswalk	that	accounted	for	any	changes	in	individual	fee	codes.	The	crosswalk	was	developed	for	earlier	work	with	these	datasets	and	has	been	successfully	used	in	several	other	projects	(e.g.	(Barer	et	al.,	2004;	McGrail	et	al.,	2011)).	The	crosswalk	and	fee	level	adjustments	together	are	used	to	create	a	measure	of	output	in	which	individual	services	provided	and	paid	for	at	any	time	during	the	study	period	are	weighted	by	their	relative	billed	value	at	a	single	point	in	time	(April	2010).	The	relationship	between	service	counts,	patient	contacts	and	billings	in	the	FFS	context	is	illustrated	here	(Barer	et	al.,	2004;	McGrail	et	al.,	2011):			 73	"##	$%&'()#%	*+,#-%.)'/#, 1ℎ3(.45-	6 = 48-)54)( ∗ (#/:.4#(48-)54) ∗ ;##	5%&'()#%	#+,#-%.)'/#(#/:.4# 	The	use	of	these	three	measures	collectively	to	investigate	physician	activity	allows	for	the	examination	of	some	interesting	nuances	in	physician	practice	patterns,	ones	that	may	vary	for	male	compared	to	female	physicians.		For	example,	over	a	specified	time	period,	two	physicians	may	bill	for	the	same	service	count,	but	may	see	a	dramatically	different	number	of	patients,	based	on	how	often	they	provide	multiple	services	per	patient	encounter.	Similarly,	physicians	who	bill	a	certain	dollar	value	may	bill	for	a	high	service	count	with	an	over-representation	of	low-cost	fee	items,	or	they	may	bill	for	a	low	service	count	with	an	over-representation	of	high-cost	fee	items.		In	addition	to	these	activity	measures,	I	compared	the	total	number	of	unique	patients	seen	per	year	for	male	and	female	physicians,	and	also	the	percentage	of	unique	patients	with	whom	physicians	had	three	or	more	contacts	in	a	single	year.		3.5.2.2	Payment	categories:	Clinical	care,	clinical	incentives,	non-clinical	incentives,	other	(Usage:	Outcome	variables	in	Chapter	5	-	hypotheses	2.1,	2.2;	Outcome	variables	in	Chapter	6	–	hypothesis	3.2)		 I	split	physician	payments	into	four	basic	categories:	payments	for	clinical	services,	clinical	incentives,	non-clinical	incentives	and	other	(Figure	3.2).		I	have	defined	payments	for	clinical	services	as	remuneration	specifically	for	the	delivery	of	health	services.	This	category	includes	MSP	billings	for	clinical	services,	and	APP	service,	salary	and	sessional	contracts.	Clinical	incentives	are	remuneration	provided	for	specific	types	of	clinical	care	(or	for	caring	for	particular	types	of		 74	patients)	provided	over	and	above	(on	top	of)	direct	care	delivery.	This	category	includes	payments	that	are	part	of	the	General	Practice	Services	Committee’s	(GPSC)	Full	Service	Family	Practice	Program.	Non-clinical	incentives,	in	contrast,	are	incentives	and	bonuses	for	other,	non-clinical	activities.	This	category	includes	on-call	payments	(Medical	On-Call	Availability	Program	(MOCAP)),	and	rural	and	remote	program	payments.	The	specific	fees	categories	from	both	MSP	and	APP	sources,	and	their	alignment	within	clinical	and	non-clinical	categories	are	listed	in	Table	3.4.	More	detail	can	be	found	in	Appendix	3.		Figure	3.2:	Physician	compensation,	divided	into	clinical	and	non-clinical	payments											Physician	CompensationMSP	Sources	(FFS)Clinical		Payments Clinical	IncentivesGPSC	IncentivesNon-Clinical	IncentivesAlternative	Payments	(non-FFS)Clinical	Payments Non-Clinical	IncentivesRural	and	RemoteOn-Call	PaymentsOther	 75	Table	3.4:	Fee	items	in	clinical	and	non-clinical	payment	categories	Payment	 Data	Source	Clinical	Payments			Billings	for	clinical	services	 MSP	billings			APP	Service	Contract	 APP	billings			APP	Sessional	Contract	 APP	billings			APP	Salaried	Contract	 APP	billings			HA	Service	Contract	 APP	billings			HA	Sessional	Contract	 APP	billings			HA	Salaried	Contract	 APP	billings			Clinical	Academic	Service	Contract			 APP	billings			Assigned	Fee	for	Service	 APP	billings			Primary	Care	Organizations:	ACG	Payments	 APP	billings			Pathology	 APP	billings			Diagnostic	 APP	billings	Clinical	Incentive	Payments			Chronic	and	complex	diseases	incentive	 MSP	billings	(GPSC	incentive)			Telephone	and	patient	conferences	 MSP	billings	(GPSC	incentive)			Mental	health	incentive	 MSP	billings	(GPSC	incentive)			Cardiovascular	risk	assessment	 MSP	billings	(GPSC	incentive)			Personal	health	risk	assessment	 MSP	billings	(GPSC	incentive)			Palliative	care	incentives	 MSP	billings	(GPSC	incentive)			Maternity	care	incentives		 MSP	billings	(GPSC	incentive)	Non-Clinical	Incentive	Payments	On-Call	Payments			MOCAP	-	On	Call	Payments	 APP	billings			Doctor	of	the	Day	 APP	billings			Other	On-Call	 APP	billings	Rural	and	Remote	Incentives			Isolation	Allowance	 APP	billings			Rural	Retention	Program	Flat	Fee	Sum	 APP	billings			Recruitment	incentive	(one-off	payment)	 APP	billings			Recruitment	Contingency	Fund	 APP	billings			Education	Expenses	 APP	billings			Rural	Continuing	Medical	Education	 APP	billings			Locum	Accommodation	 APP	billings			Vehicle	/	Transportation	Allowance	 APP	billings	Other	Non-Clinical	Incentives		 			WBC	forms	 MSP	billings		 76	Payment	 Data	Source			Tray	fees	 MSP	billings			Form	fees	 MSP	billings			After-	hours	premiums	 MSP	billings	Other	Payments	Other	 APP	billings			Accidental	death	and	dismemberment	 APP	billings			Overhead	-	Office	Support	 APP	billings			Administrative	stipend	 APP	billings			Academic	stipend	 APP	billings			Mortgage	Relief/housing	allowance	 APP	billings			Relocation	allowance	 APP	billings			Professional	Fees/membership	Allowance	 APP	billings			Continuing	Medical	Education	 APP	billings		 I	compared	the	proportion	of	income	accounted	for	by	MSP	vs.	APP	sources,	as	well	as	clinical,	clinical	incentives,	and	non-clinical	incentives	for	male	and	female	physicians	over	time.		I	also	examined	the	uptake	of	incentive	payments,	and	MOCAP	payments,	by	gender,	assessing	whether	these	payments	make	up	a	significantly	larger	portion	of	overall	income	for	physicians	of	a	particular	demographic	group.	I	also	used	both	a	dichotomous	definition	(i.e.	received	payments	for	time	on-call	or	did	not),	as	well	as	average	on-call	dollars	per	year	per	physician.			3.5.3	Patient	characteristics		For	all	analyses	of	the	characteristics	of	physicians’	patient	populations,	I	rely	solely	on	FFS	data	from	the	MSP	payments.	The	APP	database,	as	stated	in	section	3.2.4,	does	not	contain	data	on	the	specific	service(s)	delivered	to	individual	patients,	and	is	therefore	not	useful	here.	In	all	multivariate	exercises,	I	include	a	variable	indicating	the	percentage	of	a	physician’s	income	that	comes	from	APP		 77	sources	to	adjust	for	the	fact	that	I	am	only	looking	at	the	proportion	of	work	paid	under	a	FFS	model.			3.5.3.1	Demographic	Variables	(Usage:	Outcome	variables	in	Chapter	6	–	hypotheses	3.1;	Explanatory	variables	in	Chapter	6	-	hypothesis	3.2	and	Chapter	7	–	hypotheses	4.1,	4.2,	4.3,	4.4)		 Patient	characteristics	variables	were	generated	using	data	from	the	MSP	Consolidation	File.	Patient	age,	gender,	and	income	quintile	were	used	to	generate	summary	statistics	characterizing	the	visit	patterns	of	individual	physicians.		I	was	particularly	interested	in	examining	the	proportion	of	visits	with	male	vs.	female	patients;	the	proportion	of	visits	with	patients	over	age	65	or	age	75;	and	the	proportion	of	visits	with	patients	in	the	lowest	income	quintile.	The	latter	two	groups	are	important	high-needs	populations	who	are	therefore	more	likely	to	be	high	users	of	health	services.			3.5.3.2	Morbidity	(Usage:	Outcome	variables	in	Chapter	6	–	hypotheses	3.1;	Explanatory	variables	in	Chapter	6	-	hypothesis	3.2	and	Chapter	7	–	hypotheses	4.1,	4.2,	4.3,	4.4)		 Patient	morbidity	was	measured	using	Johns	Hopkins’	Aggregated	Diagnostic	Groupings	(ADGs),	which	were	generated	using	the	ICD-9	and	ICD-10	diagnostic	codes	in	the	MSP	Payment	Information	File	and	the	DAD.		ADGs	are	a	method	for	quantifying	a	predicted	morbidity	burden	experienced	by	a	particular	individual	or	population	based	on	the	specifics	of	their	previous	resource	utilization	patterns	(Health	Services	Research	and	Devlopment	Centre	at	Johns	Hopkins	University,	2011;	Johns	Hopkins	University,	2012).	Eight	of	the	thirty-two	possible	ADGs	are	considered	“major	conditions”.	The	computation	of	the	ADG	groupings	is	described		 78	in	Appendix	5.	Analyses	were	undertaken	yearly,	and	only	patients	who	resided	in	the	province	of	BC	for	275	days	in	a	given	year	(as	per	BC	Medical	Insurance	Plan	stipulations)	were	included.		This	approach	has	been	validated	in	earlier	work,	in	(Reid,	MacWilliam,	Verhulst,	Roos,	&	Atkinson,	2001)and	outside	of	(Reid	et	al.,	2001)	B.C.	I	examined	the	proportion	of	male	and	female	physicians’	annual	contacts	that	were	with	patients	who	represented	a	high	morbidity	burden	(3	or	more	ADGs,	or	1	or	more	of	the	8	major	ADGs).		3.5.4	Patient	care	approaches			As	with	the	analyses	on	patient	characteristics,	the	examination	of	patient	care	approaches	is	restricted	to	the	FFS	data	from	MSP	payments.	The	APP	database,	as	stated	in	section	3.2.4,	does	not	contain	data	on	the	specific	service(s),	nor	their	locations.	As	previously	noted,	I	include	a	variable	indicating	the	percentage	of	a	physician’s	income	that	comes	from	APP	sources	to	adjust	for	the	fact	that	I	am	only	looking	at	the	proportion	of	work	paid	under	a	FFS	model.			3.5.4.1	Referral	patterns		(Usage:	Dependent	variables	in	Chapter	7	–	hypotheses	4.1)		 I	examined	how	frequently	primary	care	physicians	refer	their	patients	on	to	four	other	categories	of	health	care	services:	medical	specialists,	surgical	specialists,	laboratory	testing,	or	diagnostic	imaging.	Referral	data	were	obtained	using	the	MSP	Payment	File.	I	modeled	rates	of	referrals	(and	proportion	of	visits	resulting	in	a	referral)	across	these	four	categories	controlling	for	the	impact	of	the	physician’s		 79	activity	level	and	characteristics	(average	age,	proportion	female,	and	morbidity	level)	of	their	patient	population.		3.5.4.2	Scope	of	Practice		(Usage:	Dependent	variables	in	Chapter	7	–	hypotheses	4.2,	4.3)		 Out-of-Office	Care:	Billings	for	services	delivered	outside	of	a	physician’s	office	are	classified	as	out-of-office	care.	I	examined	whether	male	and	female	physicians	provided	any	out-of-office	care	at	four	locations:	homes,	long-term	care	facilities,	emergency	rooms,	or	hospitals	(not	including	emergency)	using	location	codes	and	specific	fee-items.	Physicians	who	have	no	billings	for	care	at	any	of	these	locations	these	items	are	referred	to	as	practicing	“in-office-only”.	I	also	calculated	the	proportion	of	services	billed	for	out-of-office	care,	comparing	male	and	female	physicians.	Lastly	I	compared	the	proportion	of	male	and	female	physicians	who	billed	for	service	delivery	occurring	outside	of	office	hours.		Obstetrical	Care:	The	proportion	of	male	versus	female	physicians	who	provide	obstetrical	care	(deliveries	in	particular,	and	obstetrical	care	overall)	were	similarly	examined	on	a	yearly	basis.	These	yes/no	values	were	examined	for	all	years	an	individual	physician	had	fee-for-service	payments.		I	also	examined	how	much	of	a	physician’s	total	activity	was	related	to	obstetrical	care	provision.	My	interest	was	in	differences,	and	patterns	of	change,	in	obstetrical	practice	involvement	by	gender,	age,	location,	and	practice	characteristics	(e.g.	proportion	of	practice	made	up	of	female	patients).			.		Mental	Health:	I	examined	the	proportion	of	total	activity	for	male	and	female	physicians	that	was	related	to	the	provision	of	mental	health	care	(as		 80	identified	by	the	proportion	of	contacts	with	mental	health	related	ICD-9	codes).	I	also	examined	the	uptake	of	the	community-based	mental	health	management	fee,	and	the	provision	of	counselling	visits	on	a	yearly	basis.			 	 81	CHAPTER	4:	Remuneration	and	Activity		4.1	Introduction	This	chapter	is	an	in-depth,	longitudinal,	examination	of	related	but	distinct	measures	of	primary	care	physician	activity	–	total	compensation,	patient	contacts,	and	service	volumes	–	with	a	focus	on	gender	differences.	It	builds	upon	existing	international	and	Canadian	literature	on	gender-related	differences	in	activity	over	the	life	course,	reviewed	in	Chapter	2.			Much	of	the	extant	international	literature	in	this	area	has	used	cross-sectional	surveys	that	asked	questions	about	time	spent	working	(see	for	example	Boerma	&	van	den	Brink-Muinen,	2000;	Gravelle	&	Hole,	2007b;	Keane,	Woodward,	Ferrier,	Cohen,	&	Goldsmith,	1991)).		The	results	of	these	studies	suggest	that	female	primary	care	physicians	(PCPs)	work	fewer	hours	and	are	more	likely	to	report	working	part-time	compared	with	their	male	counterparts	(Gravelle	&	Hole,	2007;	Keane	et	al.,	1991).			But	these	studies	all	suffer	from	well-understood	survey	methods	biases	(Hedden	et	al.,	2014).		In	addition,	time	spent	working	can	only	provide	an	at-best	incomplete	view	of	physician	activity.		International	studies	that	have	attempted	to	measure	activity	directly	have	also	tended	to	rely	on	survey	data.	They	report	that	female	physicians	tend	to	have	fewer	patient	encounters	and	lower	services	volumes	(which	may	be	an	artifact	of	working	fewer	hours),	but	that	they	also	spend	more	time	with	each	patient	and	deliver	more	services	per	encounter	(Boerma	&	van	den	Brink-Muinen,	2000;		 82	McKinstry	et	al.,	2006;	Weeks	&	Wallace,	2006).	In	almost	all	cases,	inadequate	adjustments	were	made	for	the	potentially	confounding	effects	of	age	and	cohort.		There	is	very	little	Canadian	literature	that	examines	activity	patterns	for	male	and	female	PCPs,	and	like	the	international	literature,	most	of	the	papers	are	cross-sectional	and	rely	on	survey	data.	Additionally,	the	two	existing	Canadian	studies	that	used	a	longitudinal	methodology	produced	inconsistent	results:	Watson	and	colleagues	(2006)	reported	that	the	gender-related	activity	gap	widened	between	1992	and	2001,	while	Crossley	and	colleagues	find	that	the	gap	was	narrowing	(2009).	Crossley	et	al.	(2009)	measured	activity	indirectly,	focusing	on	hours	worked.	Watson	et	al.	(2006)	used	a	population-based	approach	and	administrative	data	to	measure	activity	directly;	however,	they	were	unable	to	account	for	payments	to	physicians	occurring	outside	of	the	traditional	fee-for-service	(FFS)	remuneration	scheme.	This	is	an	important	limitation.	The	inability	to	account	for	alternative	payment	programs	(APP)	may	obscure	any	true	gender-related	activity	differences	if	female	physicians	are	more	likely	to	be	receiving	mixed	or	full	APP	remuneration.	In	that	case,	the	remaining	FFS	data	would	be	an	underestimate	of	female	physicians’	activity	levels	relative	to	male	physicians.	To	my	knowledge,	no	studies	have	to	this	point	examined	gender-related	activity	differences	in	BC	using	a	complete	cohort	of	physicians	and	physicians’	payment	records,	including	APP	programs.			In	this	chapter,	I	examine	differences	in	overall	activity	level	using	a	longitudinal,	population-based	approach.	I	tease	apart	the	extent	to	which	any		 83	observed	gap	in	activity	is	driven	by	patient	contacts	and/or	service	volumes,	or	differential	uptake	of	APP	programs.		Specific	research	questions	are	as	follows:		• Question	1.1:	On	average,	does	average	annual	activity	(number	of	patient	contacts,	number	of	services,	dollars	billed,	cost	per	service,	and	percentage	APP)	vary	for	male	versus	female	PCPs,	controlling	for	age,	cohort,	and	practice	location?	How	has	activity	changed	over	time?		• Question	1.2:	Is	there	a	gender-related	difference	in	activity	level	across	the	complete	career	trajectory,	and	is	the	magnitude	of	that	difference	age	dependent?			I	hypothesize	that	female	PCPs	will	have	lower	age-adjusted	activity	levels	(contacts,	services,	dollars	billed)	per	unit	time	compared	with	their	male	counterparts,	but	that	this	difference	will	decrease	over	the	study	period.	Further,	male	physicians	will	have	a	higher	payment	per	patient	contact,	but	female	physicians	will	deliver	more	services	per	contact,	reflecting	key	differences	in	practice	‘style’.	A	larger	proportion	of	the	income	of	female	physicians	will	come	from	APP	remuneration.			4.2	Methods,	variables,	and	data	sources	The	datasets	and	study	cohort	used	for	this	analysis	were	described	in	Chapter	3,	Sections	3.2	and	3.4.	This	chapter	focuses	only	on	those	variables	used	to	measure	physician	activity	and	demographics,	and	therefore	draws	on	data	from	the	Medical	Services	Plan	(MSP)	Practitioner	File,	College	of	Physicians	and	Surgeons	(CPSBC)	Physician	Registry,	MSP	Payment	Information	File,	and	the	Alternative	Payments	Program	(APP).	No	patient	level	data	are	used	in	the	analyses	described		 84	in	this	chapter,	and	therefore	the	MSP	consolidation	file,	discharge	abstract	database,	and	Vital	Statistics	databases	were	not	used.		4.2.1	Dependent	variables	The	dependent	variables	of	interest	for	this	chapter	are	all	either	payment-related	or	activity-related.	All	variables	were	described	in	depth	in	Chapter	3,	Section	3.5.2.	In	brief,	I	measured	total	annual	financial	compensation,	which	includes	payments	through	both	fee-for-service	(FFS)	(from	the	MSP	payment	information	file)	and	non-FFS	(from	the	APP	database)	payment	mechanisms.			I	also	used	annual	service	and	contact	counts,	as	well	as	the	total	number	of	unique	patients,	all	of	which	were	generated	from	fee-for-service	data	only.	Thus,	when	results	refer	to	payments,	they	reflect	all	payments	from	all	sources,	while	references	to	“activity”	reflect	only	what	can	be	seen	in	the	FFS	data.		4.2.2	Explanatory	variables	I	was	primarily	interested	in	the	independent	impact	of	physician	gender	on	all	activity	variables	described	above.	I	used	physician	demographic	characteristics,	including	age	(in	10-year	intervals),	year	of	graduation	(in	10-year	cohorts),	location	of	training	(Canada	or	international),	and	practice	rurality	as	explanatory	covariates.	Due	to	significant	collinearity	between	age	and	graduation	cohort9,	undertaking	the	statistical	analyses	required	that	I	include	only	one	of	the	two.	I	chose	to	include	the	age	categories	rather	than	the	graduation	cohorts.	However,																																																									9	Age	and	year	of	graduation	had	a	Pearson	correlation	coefficient	of	0.90,	p<0.0001.	Including	both	variables	in	the	models	produced	significant	issues	caused	by	collinearly.		 85	because	these	two	variables	are	so	closely	aligned,	the	age	group	variable	–	which	was	designed	capture	change	in	activity	and	practice	pattern	as	physicians	age	–	will	likely	also	capture	any	effects	related	to	changes	in	physician	training	over	time	(cohort	effect).	A	limitation	of	this	approach	is	that	although	the	models	adjust	for	the	combined	effect	of	age	and	cohort	(allowing	for	the	independent	examination	of	the	effect	of	gender),	they	are	unable	to	differentiate	the	independent	effects	of	each.			I	included	the	study	year,	from	2005-06	to	2011-12,	to	examine	activity	trends	over	time,	and	valued	all	fee	items	at	the	fee	levels	effective	April	1,	2010	(Pascali,	1995).	The	fee-adjustment	process	was	described	in	Chapter	3,	Section	3.5.2.		 In	those	models	where	the	dependent	variable	depended	solely	on	fee-for-service	activity	measures	(patient	contacts,	services	delivered,	and	number	of	unique	patients),	percentage	of	total	payments	from	APP	sources	was	also	included	as	an	explanatory	covariate.	This	was	done	to	account	for	the	fact	that	information	on	individual	patient	contacts	and	services	provided	are	only	available	in	the	MSP	Payment	Information	File.		As	a	result,	these	measures	will	reflect	only	a	proportion	of	a	physician’s	total	activity	if	that	physician	is	paid	under	mixed	remuneration.		4.2.3	Statistical	analyses	4.2.3.1	Descriptive	statistics,	univariate	and	bivariate	analyses			 I	computed	measures	of	central	tendency	and	dispersion	(mean,	range,	standard	deviation	and	variance	where	relevant)	for	both	demographic	and	activity	variables,	and	produced	counts	of	all	missing	values.		I	also	constructed	frequency		 86	tables	or	line	graphs	for	each	measure	to	assess	distributions,	examining	normality,	skew	and	kurtosis.			I	used	chi-square	tests	to	examine	whether	(categorical)	demographic	characteristics	differed	for	male	and	female	physicians.	I	computed	preliminary,	unadjusted	measures	of	association	between	the	dependent	variables	and	all	explanatory	variables,	and	between	physician	gender	and	the	explanatory	demographic	characteristics.	I	used	chi-square	tests	to	examine	the	association	between	gender	and	categorical	demographic	variables,	and	a	Wilcox-Mann-Whitney	two-sample	test	to	compute	an	unadjusted	association	between	gender	and	count	or	continuous	demographic	or	activity	variables	(e.g.	age,	salary,	contacts,	etc.).	I	also	performed	some	bivariate	analyses	between	explanatory	variables	in	order	to	identify	potential	collinearity	that	may	have	required	consideration	in	subsequent	analyses.	Of	particular	concern	was	the	potential	for	collinearity	between	physician	age	and	year	of	graduation,	which	I	assessed	using	a	simple	R2.	4.2.3.2	Multivariate	modeling		Question	1.1:		The	determinants	of	variation	in	annual	contacts,	services,	unique	patient	counts	and	compensation	were	all	modeled	using	mixed-effects	multivariate	linear	models.	Preliminary	analysis	of	the	three	activity	variables	suggested	that	their	distribution	was	not	normal,	as	all	three	had	a	pronounced	right	skew,	with	several	outliers.	Because	of	concerns	with	potentially	violating	the	assumptions	within	a	standard	linear	regression	model,	I	log-transformed	total	remuneration,	contact	and	service	counts,	and	modeled	these	transformed	dependent	variables	under	a	normal	distribution	with	an	identity	link	function;		 87	however,	when	I	compared	the	model	fit	between	the	log-transformed	and	non-transformed	variables,	there	was	almost	no	difference.		As	a	result,	I	elected	to	model	all	four	variables,	non-transformed,	under	a	normal	distribution,	since	the	estimation	results	are	more	easily	interpretable10.		The	probability	density	function	for	a	general	normal	distribution,	6	~	= >, ?@ ,	is	given	by,			;A +; >, ? = 	 1? 2E	 #F AFG H@IH 	where	µ	and	?	are	the	location	and	scale	parameters	respectively.	Total	compensation	is	modeled	as	a	linear	function	of	time,	gender,	and	other	physician	demographic	covariates:		3JK = LM +	LOPK +	L@QJ +	LR PK ∗ QJ +	LS$JK + LTUJ + LVWJK + LX UJ ∗ WJK + LY1JK+ ZMJ + ZOJPK + [JK 	ZMJZOJ ~= 00 , ?]M@ ?]M]O?]M]O ?]O@ 		[JK~= 0, ?@ 	Where:	y	=	total	compensation	(in	constant	dollars)	($)	for	physician	i	in	year	j	T=	time	at	year	j	G	=	gender	for	physician	i		A	=	physician	age	group	for	physician	i	in	year	j	L	=	location	of	training	for	physician	i		R	=	practice	rurality	for	physician	i	in	year	j																																																									10	For	the	contacts,	services	and	unique	patient	models,	I	also	attempted	to	fit	a	Poisson	model,	which	tends	to	be	better	suited	to	count	data;	however,	in	all	three	cases,	I	was	unable	to	achieve	convergence	without	resorting	to	a	fixed-effects	model.	Fixed-effects	models	would	be	methodologically	inferior	in	this	case	because	they	assume	that	the	repeated	observations	for	each	physician	are	independent	from	each	other,	which	they	obviously	are	not.	Using	a	random	effect	(specifically	for	physician-level	residuals)	accounts	for	the	correlation	of	multiple	measures	on	the	same	physician	without	having	to	include	practitioner	id	as	a	model	covariate.			 88	P	=	percent	APP	for	physician	i	in	year	j		The	majority	of	the	physicians	in	the	dataset	contribute	more	than	one	year	of	activity	data	and	therefore	I	have	multiple	annual	measures	for	the	same	physician,	which	cannot	be	regarded	as	independent	from	each	other.		I	included	random	effects	for	subjects	to	resolve	the	issue	of	correlation	amongst	measures	from	the	same	subject,	allowing	each	subject	to	have	a	different	baseline	activity	level.	I	also	included	a	random	effect	for	slope,	allowing	each	individual	physician’s	activity	level	to	change	at	a	different	rate	over	time,	and	a	random	effect	for	subject-level	residuals.	The	residual	random	effect	was	assumed	to	have	a	first	order	auto-regressive	correlation	structure,	which	assumes	that	measures	from	the	same	subject	will	be	correlated,	and	that	measures	closer	in	time	will	be	more	highly	correlated	than	those	further	apart				The	models	for	contacts,	services,	and	unique	patient	counts	take	the	same	form;	however,	a	variable	for	percent	of	income	from	APP	sources	is	added	to	adjust	for	the	fact	that	both	contact	and	service	count	measures	can	reflect	only	the	fee-for-service	components	of	a	physician’s	total	compensation.	For	annual	patient	contacts:	3JK = LM +	LOPK +	L@QJ +	LR PK ∗ QJ +	LS$JK + LTUJ + LVWJK + LX UJ ∗ WJK + LY1JK+ ZMJ + ZOJPK + [JK 	ZMJZOJ ~= 00 , ?]M@ ?]M]O?]M]O ?]O@ 		[JK~= 0, ?@ 	Where:	y	=	annual	contact	count	for	physician	i	in	year	j	T=	time	at	year	j	G	=	gender	for	physician	i			 89	A	=	physician	age	group	for	physician	i	in	year	j	L	=	location	of	training	for	physician	i		R	=	practice	rurality	for	physician	i	in	year	j	P=	percent	APP	for	physician	i	in	year	j		I	used	the	same	model	for	two	additional	dependent	variables:	services	delivered	and	unique	patient	counts.	As	with	the	model	for	overall	compensation,	I	included	random	effects	for	subject	intercept,	slope	and	residuals	to	allow	for	individual-level	variation	in	baseline	activity	and	change	in	activity	level	over	time.		APP	uptake:	I	examined	the	importance	of	APP	programs	in	overall	compensation	using	both	a	dichotomous	and	three-category	characterization.	A	preliminary	examination	of	the	distribution	of	percentage	of	income	from	APP	sources	suggests	bimodality,	with	one	mode	at	0%	and	a	second	at	100%.	Eighty	three	percent	of	the	cohort	has	values	within	25%	of	either	zero	or	100%	APP.	Thus	the	variable	lends	itself	well	to	a	dichotomous	characterization.	I	defined	high-adopters	as	anyone	whose	total	compensation	was	at	least	50%	from	APP	sources,	and	low-adopters	as	anyone	whose	total	compensation	was	less	than	50%	APP.		I	performed	some	sensitivity	analyses	by	changing	the	low	vs.	high	cut-off	value	to	45%	and	then	55%.	I	modeled	the	dichotomous	variable	(with	all	three	cut-offs)	using	a	cross-sectional	fixed	effects	model	under	a	binary	(logistic)	distribution	with	a	logit	link	function,	using	data	from	2011-12	only.	I	tested	a	longitudinal	approach	(using	all	study	years)	and	found	that	neither	the	time	variable	nor	time-by-gender	interaction	terms	were	significant,	and	thus	elected	to	use	a	cross-sectional	model	for	ease	of	interpretation.	I	ran	the	model	on	the	most	recent	year	of	data	(2011-12)	because	a	physician’s	practice	location	(rurality	and	health	authority)	may	change		 90	on	a	yearly	basis,	precluding	the	use	of	running	the	model	on	an	overall	average	across	all	years	of	data11.	The	logistic	function	is	given	by	; ^; , = 11 + #F(`ab	`cA)	The	odds	of	being	a	high-	vs.	low-adopter	of	APP	in	2011-12	was	modeled	as	a	function	of	physician	characteristics:	log	(3J) = LM +	LOQJ +	L@$J + LRUJ + LSWJ + LT UJ ∗ WJK + [J 		 Where:		y	=	odds	of	being	a	high-adopter	of	APP	for	physician	i	G	=	gender	for	physician	i	A	=	physician	age	group	for	physician	i	L	=	location	of	training	for	physician	i	R	=	practice	rurality	for	physician	i		Because	preliminary	analyses	of	the	APP	data	suggested	that	female	physicians	were	more	likely	than	male	physicians	to	be	clustered	near	0	and	100%	(male	physicians	were	more	likely	to	be	receiving	mixed	remuneration),	I	also	built	an	ordinal	logistic	model	–	an	extension	of	the	binary	logistic	model	–	where	I	broke	percentage	APP	into	three	categories:	low	uptake	(less	than	25%	of	payments);	moderate	uptake	(25-75%	of	payments);	and	high	uptake	(greater	than	75%	of	payments):	log( 3R + 3@1 − 3R − 3@) = LM +	LOQJ +	L@$J + LRUJ + LSWJ + LT UJ ∗ WJK + [J 																																																											11	I	tested	the	model	on	each	of	the	other	years	of	data	and	found	no	significant	change	in	either	the	direction	or	magnitude	of	the	effect	estimates.			 91	Where:		Y3	=	odds	of	being	a	high-adopter	of	APP	for	physician	i	Y2	=	odds	of	being	a	moderate-adopter	of	APP	for	physician	i	G	=	gender	for	physician	i	A	=	physician	age	group	for	physician	i	L	=	location	of	training	for	physician	i	R	=	practice	rurality	for	physician	i		As	an	additional	form	of	sensitivity	analysis,	I	re-ran	both	the	binary	and	ordinal	APP	models	using	only	clinical	delivery	data,	with	percentage	APP	defined	as	percent	of	clinical	payments	coming	from	APP	sources	(rather	than	percentage	of	total	payments)12.	My	intention	was	to	examine	whether	gender	differences	in	APP	uptake	resulted	from	differences	in	clinical	care	income	(uptake	of	salaried	and	sessional	arrangements,	for	example)	or	from	differential	uptake	of	non-clinical	incentives	within	the	APP	data.	The	results	from	these	models	were	comparable	and	consistent	with	the	original	binary	and	ordinal	APP	models,	suggesting	either	that	the	gender	differences	observed	are	dominated	by	differences	in	clinical	care	uptake,	or	that	they	are	of	approximately	the	same	magnitude	for	clinical	and	non-clinical	APP.	These	explanations	are	explored	further	in	Chapter	5.		Question	1.2:	I	modeled	activity	levels	for	each	of	ten	five-year	age	brackets	(<30,	30-<35…,	65-<70,	70+)	as	a	function	of	physician	gender,	rurality,	and	training	location	using	fixed	effects	generalized	linear	models	(GLMs)	under	a	normal	distribution:	3J = LM +	LOQJ +	L@$J + LRUJ + LSWJ + LT UJ ∗ WJK + LV1JK + [J																																																										12	See	Chapter	3:	Section	3.5.2,	Table	3.3	for	definitions.		 92	Where:		y	=	annual	total	compensation,	contacts,	services,	or	unique	patients	for	each	physician	i	in	a	particular	five	year-age	bracket	G	=	gender	for	physician	i	L	=	location	of	training	for	physician	i	R	=	practice	rurality	for	physician	i	P=	percent	APP	for	physician	i		I	used	the	results	from	the	models	to	plot	age-	and	gender-based	career	activity	profiles	as	least	squares	mean	percentages.		I	built	all	models	using	the	“top-down”	strategy	as	outlined	by	Verbeke	and	Molenberghs	(2000):	1. Specify	a	mean	structure	for	the	model,	adding	all	fixed-effects	to	create	a	“loaded	mean	structure”;	2. Select	a	covariance	structure	for	random	effects;	and	3. Select	a	residual	covariance	structure.	I	tested	model	assumptions	using	residual	diagnostics	and	goodness	of	fit	parameters.	I	examined	the	log	likelihood	of	the	parameter	estimates,	and	deviance	(negative	twice	the	log-likelihood)	to	determine	which	parameters	best	characterize	the	data.	I	compared	the	fit	of	candidate	models	using	Akaike’s	Information	Criterion	(AIC)	and	Schwarz’s	Bayesian	Information	Criterion	(BIC).		4.2.4	Some	notes	on	model	interpretation		 The	multivariate	models	in	this	chapter	contain	two	interaction	terms:	one	(gender*year)	includes	one	categorical	and	one	continuous	variable;	the	other	(rurality*training	location)	includes	two	categorical	variables.	The	inclusion	of	the		 93	gender*year	interaction	term	is	intended	to	test	the	relationship	between	physician	gender,	time,	and	the	dependent	variable–	i.e.	to	test	whether	the	relationship	between	gender	and	the	dependent	variable	of	interest	is	changing	over	time.	In	models	where	this	term	is	significant,	rather	than	interpreting	the	gender	effect	estimate	in	insolation,	I	consider	the	combined	effect	of	the	independent	gender	term	and	the	gender*time	interaction.	Both	the	effect	estimate,	and	standard	errors	(and	therefore	95%	confidence	intervals)	are	combined,	and	I	can	produce	a	gender	effect	for	each	study	year.	I	combined	the	standard	errors	for	these	terms	using	the	delta	method	(Efron	&	Tibshirani,	1986).		 The	interpretation	of	the	rurality*training	location	interaction	term	is	slightly	more	complex.			The	models	include	a	series	of	dummy	variables	for	each	possible	combination	of	rurality	(three	categories)	and	training	location	(two	categories).	Two	of	these	combinations	–	practice	in	an	urban-dominated	area	and	international	training,	and	practice	in	a	rural	area	and	international	training	–	produce	effect	estimates	that	indicate	whether	international	training	modifies	the	relationship	between	rural	practice	and	the	dependent	variable.	Thus,	in	models	where	one	or	both	of	these	terms	are	significant,	I	can	comment	on	the	independent	effects	of	practice	location	and	international	training,	as	well	as	the	combined	effect	of	the	two	together.		As	with	the	gender	and	time	interaction,	the	effect	estimates	and	the	standard	errors	are	combined,	producing	an	overall	effect	of	location,	training,	and	the	interaction	together.			 The	proportion	of	physicians’	total	payments	that	are	generated	through	APP	sources	is	included	in	all	models	in	this	thesis.	As	explained	in	Chapter	3,	section		 94	3.2.4,	the	APP	database	does	not	contain	data	on	the	specific	services	delivered,	nor	the	patient	who	received	the	services.	Thus,	this	control	variable	is	included	solely	in	order	to	adjust	for	the	fact	that	there	may	be	differential	uptake	of	APP	programs	by	gender,	which	would	produce	a	bias	if	the	MSP	data	were	examined	in	isolation.	It	is	expected	that	individuals	who	earn	a	larger	proportion	of	their	income	through	APP	sources	would	have	smaller	contact	and	service	counts.	Because	it	is	included	only	as	a	control	term,	I	do	not	discuss	its	influence	on	the	dependent	variables,	except	in	the	case	of	the	modelling	of	total	remuneration.	In	that	model	(Section	4.3.3.1),	I	am	looking	at	both	MSP	and	APP	services	with	the	same	granularity	(total	payments	per	year),	and	can	address	questions	about	whether	physicians	whose	income	is	largely	APP	tend	to	have	higher	or	lower	total	remuneration	compared	to	those	whose	income	is	largely	from	MSP	sources.		4.3	Results	4.3.1	Cohort	description	and	demographics	The	final	study	cohort	included	6579	primary	care	physicians,	of	whom	2469	(38%)	were	female	and	4110	(62%)	were	male.		Two	hundred	and	fifty	eight	(4%,	113	females	and	145	males)	physicians	received	100%	of	their	total	payments	from	APP	sources	for	all	their	years	of	activity.13		On	average,	physicians	were	active	in	the	cohort	for	5.50	of	the	possible	7	years	(SD	=	2.2).		Male	physicians	were	active																																																									13	These	individuals	were	excluded	from	the	contacts	and	services	analyses	but	remained	in	the	study	cohort	for	the	examinations	of	total	compensation	and	percentage	APP.			 95	significantly	longer	than	female	physicians	(5.6	years,	compared	to	5.3	years,	t-value:	-5.80,	p<0.0001).		Male	and	female	physicians	differed	across	all	demographic	characteristics	(Table	4.1).	Male	PCPs	were	older,	with	a	mean	age	(in	2012)	of	53.57	compared	to	46.64	for	female	PCPs	(Z	=		-21.1,	P<0.0001).	They	also,	not-surprisingly,	were	more	likely	to	have	graduated	medical	school	earlier	(chi-square	=	461.0,	p<0.0001).		Male	physicians	were	also	more	likely	to	have	received	their	medical	training	outside	Canada	(34.0%	vs.	24.0%,	chi-square	=	71.7,	p<0.0001).				 		 96	Table	4.1:	Cohort	demographics	by	physician	gender			 Males	 Females	 Total			 N=4110	(62%)	 N=2469	(38%)	 N=	6579	Mean	age	in	2012	(sd)1	 53.57	(12.65)	 46.64	(10.98)	 50.97	(12.51)	Age	group	in	2012	(%)2	<35	 293	(7.13)	 383	(15.51)	 676	(10.28)	35-<45	 761	(18.52)	 712	(28.84)	 1473	(22.39)	45-<55	 1115	(27.13)	 769	(31.15)	 1884	(28.64)	55-<65	 1102	(26.81)	 463	(18.75)	 1565	(23.79)	65+	 839	(20.41)	 142	(5.75)	 981	(14.91)	Health	Authority	in	2012	(%)3	Interior	Health	 631	(15.35)	 343	(13.89)	 974	(14.80)	Fraser	Health	 882	(21.46)	 474	(19.20)	 1356	(20.61)	Vancouver	Coastal	Health	 845	(20.56)	 759	(30.74)	 1604	(24.38)	Vancouver	Island	Health	 734	(17.86)	 392	(15.88)	 1126	(17.12)	Northern	Health	 227	(5.52)	 106	(4.29)	 333	(5.06)	Missing	 32	(0.78)	 30	(1.22)	 62	(0.94)	Not	active	in	2011/12	 759	(18.47)	 365	(14.78)	 1124	(17.08)	Practice	rurality	in	2012	(%)4	Metropolitan	 1885	(45.86)	 1269	(30.88)	 3154	(47.94)	Urban	dominated	 877	(21.34)	 496	(12.07)	 1373	(20.87)	Rural	dominated	 557	(13.55)	 309	(7.52)	 866	(13.16)	Missing	 32	(0.78)	 30	(0.73)	 62	(0.94)	Not	active	in	2011/12	 759	(18.47)	 365	(8.88)	 1124	(17.08)	Graduation	year	(%)5	<1970	 515	(12.53)	 78	(1.90)	 593	(9.01)	1970-<1980	 990	(24.09)	 298	(7.25)	 1288	(19.58)	1980-<1990	 1050	(25.55)	 628	(15.28)	 1678	(25.51)	1990-<2000	 924	(22.48)	 716	(17.42)	 1640	(24.93)	2000+	 631	(15.35)	 749	(18.22)	 1380	(20.98)	Trained	internationally	(%)6*	 1370	(34.00)	 580	(24.00)	 1950	(30.25)	1Wilcoxon-Mann-Whitney,	Z	=-21.1,	p<0.0001	 4chi-square	=	12.7,	p=0.0053	2chi-square	=	452.09,	p<0.0001			 	 5chi-square	=	455.7,	p<0.0001	3chi-square	=	79.5,	p<0.0001	 	 6chi-square	=	71.7,	p<0.0001		I	also	found	significant	differences	between	male	and	female	PCPs	in	terms	of	practice	location	(in	the	2011-12	fiscal	year).	Male	physicians	were	more	likely	to	practice	in	areas	that	are	rural	dominated	(16.6%	of	male	physicians,	compared		 97	with	14.7%	of	female	physicians),	while	female	physicians	were	more	likely	to	practice	in	metropolitan	areas	(51.2	%	of	female	physicians,	compared	with	45.8%	of	male	physicians,	chi-square	for	practice	rurality	=	12.7,	p=0.0053).		Female	physicians	were	more	likely	to	be	practicing	within	Vancouver	Coastal	Health	(36.0%	of	female	PCPs	practiced	in	Vancouver	Coastal	Health,	compared	with	25.2%	of	male	PCPs).		Male	physicians	were	more	likely	to	practice	within	Interior	Health,	Fraser	Health,	Vancouver	Island	Health,	and	Northern	Health	(chi-square	for	Health	Authority	=	79.5,	p<0.0001).			4.3.2	Unadjusted	payments	and	activity		Payments	to	physicians	across	all	study	years,	including	both	FFS	and	APP	sources,	averaged	$200,715	per	capita	per	year	(SD=$137,780)	(Table	4.2).		Female	PCPs	earned	less	than	male	PCPs	($148,433	vs.	$232,121	per	year,	Z=-23.68,	p<0.0001).	Female	physicians	also	had	fewer	patient	contacts	per	year	(2906	patient	contacts	vs.	4639	patient	contacts,	Z=-19.63,	p<0.0001),	which	equates	to	a	difference	of	approximately	5	patients	per	active	day14.		Male	physicians	delivered	more	services	total	(4920	vs.	3217	per	year,	Z=-17.99,	p<0.0001),	and	more	services	per	patient	per	year	compared	to	female	physicians	(1.06	and	1.11	services	per	patient	contact	respectively	(Z=25.19,	p<0.0001).	They	also	billed	a	larger	amount	for	each	contact,	but	the	difference	in	terms	of	real	dollars	is	small	($57.34	for	male	physicians	and	$56.01	for	female	physicians,	Z=-3.50,	p=0.0002).																																																													14	Patients	per	day	is	defined	as	total	number	of	patient	contacts	per	year	divided	by	the	number	days	for	which	the	physician	had	any	active	billings.		 98	Table	4.2:	Physician	payments	and	activity	averaged	across	all	study	years		Variable	(SD)	Males	 Females	 Total	N=4110	(62%)	 N=2469	(38%)	 N=	6579	Total	compensation1	 $232,122	(146,994)	 $148,434	(101,222)	 $200,715	(137,780)	APP	payments2	 $33,486	(59,059)	 $28,106	(50,099)	 $31,467	(55,922)	Percent	APP3	 18.61	(29.58)	 21.32	(31.86)	 19.63	(30.48)	Services/year	†4	 4920	(3875)	 3217	(2737)	 4285	(3590)	Contacts/year	†5	 4639	(3518)	 2906	(2443)	 3993	(3269)	Contacts/day	†6	 21.76	(12.61)	 17.26	(9.23)	 20.06	(11.67)	Cost/contact	†7	 $57.34	(72.24)	 $56.01	(84.09)	 $56.84	(76.87)	Services/contact	†8	 1.06	(0.14)	 1.11	(0.10)	 1.08	(0.13)	Unique	patients/year	†9	 1781	(1397)	 1277	(1014)	 1593	(1291)	Contacts/unique	patient/year	†10	 2.76	(2.22)	 2.31	(1.97)	 2.60	(2.14)	Services/unique	patient/year	†11	 2.90	(2.27)	 2.54	(2.03)	 2.77	(2.19)	1Wilcoxon-Mann-Whitney,	Z=-23.68,	p<0.0001	 7Wilcoxon-Mann-Whitney,	Z=-3.50,	p=0.0002	2Wilcoxon-Mann-Whitney,	Z=-3.26,	p=0.0006	 8Wilcoxon-Mann-Whitney,	Z=25.19,	p<0.0001	3Wilcoxon-Mann-Whitney,	Z=0.74,	p=0.2273	 9Wilcoxon-Mann-Whitney,	Z=-14.98,	p<0.0001	4Wilcoxon-Mann-Whitney,	Z=-17.99,	p<0.0001	 10Wilcoxon-Mann-Whitney,	Z=-9.24,	p<0.0001	5Wilcoxon-Mann-Whitney,	Z=-41.98,	p<0.0001	 11Wilcoxon-Mann-Whitney,	Z=-6.48,	p<0.0001	6Wilcoxon-Mann-Whitney,	Z=-17.27,	p<0.0001	 	†	Excludes	individuals	whose	salaries	are	from	100%	APP	sources	(N=	258)		 Male	physicians	had	larger	practices,	seeing	significantly	more	unique	patients	per	year	(1781	vs.	1277,	Z=-14.98,	p<0.0001).	They	also	had	more	visits	per	year	with	each	of	those	unique	patients	(2.76	vs.	2.31	visits	per	unique	patient	per	year,	Z=-9.25,	p<0.0001),	and	delivered	more	services	2.90	vs.	2.54	services	per	unique	patient	per	year,	Z=-6.48,	p<0.0001).	Female	PCPs	generated	a	larger	proportion	of	their	income	from	APP	sources	(21.3%	vs.	18.6%);	however	this	difference	is	not	significant.	Female	physicians	were,	however,	more	likely	than	male	physicians	to	be	clustered	at	the	tails	of	this	distribution,	having	either	none	or		 99	all	of	their	income	from	APP	sources	(44.50%	vs.	41.58%	at	zero,	and	5.59%	vs.	3.05%	percent	at	100%	APP).	Payment	and	activity	levels	varied	across	other	demographic	characteristics	as	well	(Table	4.3,	for	2011/12).	Physicians	trained	in	Canada	had	lower	total	remuneration	than	their	internationally	trained	counterparts.		Uptake	of	APP	programs	also	varied	substantially	across	the	demographic	groupings	with	younger	and	Canadian-trained	physicians	receiving	a	larger	proportion	of	their	total	payments	through	APP	sources.	Physicians	in	the	middle	age	groups	(between	35	and	55)	and	graduation	cohorts	(1970-2000)	were	generally	more	active	than	the	youngest	and	oldest	physicians,	having	higher	all-clinical-source	incomes	and	higher	contact	and	service	counts			 		 100	Table	4.3:	Mean	physician	activity,	by	demographic	characteristics	(for	2011/12)			 N	(%)	 Total	Compensation	 Contacts	 Services	 Percent	APP	Total	active	physicians	 5455	 $220,041		 4187	 4481	 18.23	Age	group	 	<35	 567	 $189,903	 3266	 3537	 23.34	35-<45	 1159	 $236,868	 4270	 4590	 20.28	45-<55	 1705	 $245,299	 4810	 5147	 15.70	55-<65	 1389	 $204,734	 4346	 4587	 14.80	65+	 635	 $127,643	 2836	 2980	 11.22	Training1	Within	Canada	 3790	 $202,360	 3734	 4020	 19.85	International	 1551	 $262,249	 5256	 5571	 14.34	Health	Authority2	Interior	 974	 $224,397	 3964	 4219	 14.63	Fraser	 1356	 $266,383	 5487	 5848	 13.04	Vancouver	Coastal	 1604	 $189,247	 3657	 4081	 24.51	Vancouver	Island	 1126	 $190,456	 3657	 3883	 14.27	Northern		 333	 $291,545	 3822	 4059	 32.34	Practice	rurality2	Metropolitan	 3154	 $216,935	 4452	 4493	 14.64	Urban	dominated	 1373	 $223,573	 4064	 4323	 16.71	Rural	dominated	 866	 $235,210	 3601	 3830	 22.52	Graduation	year	<1970	 335	 $150,703	 3293	 3473	 10.58	1970-<1980	 1122	 $224,018	 4724	 5006	 14.38	1980-<1990	 1511	 $246,324	 4721	 5051	 15.91	1990-<2000	 1405	 $235,009	 4234	 4562	 20.48	2000+	 1082	 $181,247	 3088	 3336	 24.92	1.	N=114	training	location	is	unknown																2.	N=62	health	authority/rurality	are	unknown 	 Activity	was	also	differentiated	by	practice	location,	with	physicians	in	rural-dominated	areas	earning	more	overall	but	seeing	fewer	patients	and	delivering	fewer	services	within	the	FFS	remuneration	scheme.	Similarly,	physicians	practicing	in	Northern	Health	had	the	highest	total	payments,	but	seemingly	saw	the	fewest	patients;	this	latter	finding	is	an	artifact	caused	by	the	absence	of	patient	count/interaction	information	in	the	APP	data,	and	the	fact	that	physicians	in	that		 101	region	received	more	payments	through	APP	programs	(and	the	rural	and	remote	incentive	payment	program	in	particular).		4.3.3	Question	1.1:	Multivariate	results		On	average,	does	average	annual	activity	(number	of	patient	contacts,	number	of	services,	dollars	billed,	cost	per	service,	and	percentage	APP)	vary	for	male	versus	female	PCPs,	controlling	for	age,	cohort,	and	practice	location?	How	has	activity	changed	over	time?			4.3.3.1Total	compensation	Descriptive	and	preliminary	bivariate	statistical	work	suggested	that	all	of	the	physician	demographic	characteristics	had	significant	associations	with	compensation;	all	of	these	characteristics	were	retained	in	the	compensation	model.	After	adjusting	for	other	demographic	factors,	total	compensation	for	female	physicians	was	significantly	lower	than	that	for	their	male	counterparts	across	the	entire	study	period	(Table	4.4).	In	2005-06,	this	difference	was	approximately	$89,000	(combined	effect	estimate	for	gender	and	gender*year	interaction:		-$89,356,	95%	CI:	-$88,666-	-$98,095).	By	2011-12,	the	difference	was	reduced	to	approximately	$79,900	(combined	effect	estimate	for	gender	and	gender*year	interaction:	-$79,928,	95%	CI:	-	$71,189-	-$80,618).		This	represents	a	10%	reduction	in	the	gender-drive	income	difference	over	the	seven-year	period.	There	was	also	a	small	but	statistically	significant	gender-independent	decline	in	total	compensation	across	each	study	year	(effect	estimate:	-$1104,	95%	CI:	-$420--$1,789).		Physician	gender	was	by	far	the	most	important	predictor	of	physician	income,	even	accounting	for	the	narrowing	of	the	gender	gap	by	the	end	of	the	study	period.			 102	Table	4.4:	Multivariate	modeling	results:	total	compensation,	patient	contacts,	services	and	unique	patients	Variables	Model	1:	Total	Compensation	 Model	2:	Patient	Contacts		 Model	3:	Services	 Model	4:	Unique	Patients	Estimate	(Est)	 95%	Confidence	Interval	 Est	 95%	Confidence	Interval	 Est	 95%	Confidence	Interval	 Est	 95%	Confidence	Interval	Intercept	 $187,225	 $180,065-$194,385*	 4785	 4,628-4,942*	 5119	 4,927-5,311*	 2006	 1,936-2,075*	Sex	(female)	 -$90,928	 -$98,557-	-$83,300*	 -1992	 -2,177-	-1,808*	 -2026	 -2,248-	-1,803*	 -589	 -669-	-509*	Year	(continuous)	 -$1,104	 -$1,789-	-$420‡	 -115	 -130-	-100*	 -125	 -145-	-105*	 -38	 -45-	-31*	Sex*year	interaction	(female)	 $1,571	 $462-	$2,681‡	 83	 59-108*	 97	 64-130*	 21	 8-33†	Age:	35-<45	 $26,242	 $21,417-$31,067*	 419	 326-512*	 510	 388-632*	 45	 1-90‡	Age:	45-<55	 $40,087	 $34,377-$45,797*	 576	 464-688*	 719	 574-863*	 28	 -25-81	Age:	55-<65	 $31,327	 $24,755-$37,899*	 505	 375-635*	 632	 467-798*	 -46	 -107-14	Age:	65+	 -$4,589	 -$12,822-$3,643	 65	 -96-226	 95	 -110-300	 -177	 -252-	-102*	Rurality:	urban-dominated	 $9,225	 $2,545-$15,906	 -124	 -258-11	 -195	 -363-	-28	 -5	 -66-56	Rurality:	rural-dominated	 $11,041	 $2,556-$19,525‡	 -270	 -442-	-99	 -397	 -609-	-184‡	 -126	 -203-	-49	International	training	 $33,908	 $25,336-$42,481*	 1033	 842-1,223*	 1099	 878-1,320*	 429	 350-508*	Rurality*training	interaction	(urban*international)	 -$9,152	 -$21,445-$3,141		 -258	 -507-	-9‡	 -294	 -604-16		 -250	 -$362-	-$138*	Rurality*training	interaction	(rural*international)	 $2,335	 -$12,137-$16,806		 -367	 -662-	-72‡	 -323	 -687-41		 -194	 -$326-	-$63‡	Percentage	APP	 $72,625	 $67,612-$77,639*	 -1716	 -1,814	-	-1,619*	 -2099	 -2,227	-	-1,970*	 -843	 -891-	-796*	*p<0.0001	†	p<0.001		‡p<0.05  	 	 	 	 			 		 103	Physician	incomes	showed	an	inverse	U	pattern	with	age,	increasing	to	ages	45-55,	then	declining	(effect	estimate	for	45-<55:	$40,087,	95%	CI:	$34,377-$45,797).		Physicians	working	in	rural-dominated	areas,	and	those	who	received	their	training	internationally	also	earned	more	(effect	estimate	for	rural-dominated	practice	location:	$11,041,	95%	CI:	$2,556-$19,525;	effect	estimate	for	international	training:	$33,098,	95%	CI:	$25,336-$42,481).		These	effects	are	entirely	independent	(i.e.	the	increased	earnings	for	physicians	in	rural-dominated	areas	is	not	explained	by	the	fact	that	more	internationally-trained	physicians	practice	in	those	locations).	They	are	also	additive.	Thus,	internationally-trained	physicians	located	in	rural	areas	would	be	expected	to	earn	an	additional	$44,139	($11,041+$33,098)	compared	to	Canadian-trained	physicians	in	rural	areas.	The	proportion	of	income	from	APP	sources	also	affected	physician	payments,	with	physicians	paid	entirely	through	APP	earning	$72,625	more	than	those	working	exclusively	under	FFS	remuneration	arrangements	(95%	CI:	$67,612-$77,639).		4.3.3.2	Contacts	and	services	The	results	of	the	contacts	and	services	models	are	very	similar	(Table	4.4).	In	2005-06,	after	adjusting	for	other	demographic	characteristics,	female	physicians	had	an	average	of	1909	fewer	patient	contacts	(95%	CI:	for	gender	and	gender	by	year	interaction	term	1701-2118	fewer)	and	delivered	1928	fewer	services	(95%	CI:	for	gender	and	gender	by	year	interaction	term	1672-2184	fewer)	over	the	course	of	an	average	year.		In	both	cases,	the	gender	gap	narrowed	over	the	study	period	such	that	by	2011-12,	female	physicians	had	1409	fewer	contacts	(95%	CI:	for	gender	and	gender	by	year	interaction	term	1201-1619	fewer)	and	delivered		 104	1344	fewer	services	(95%	CI:	for	gender	and	gender	by	year	interaction	term	1089-1600	fewer).	This	represents	a	26%	reduction	in	the	gender	difference	for	patient	contacts,	and	a	30%	reduction	in	the	difference	in	services.		As	with	total	compensation,	I	found	a	small	but	statistically	significant	reduction	in	the	number	of	contacts	and	services	over	the	study	period.	Physicians	had	115	fewer	fee-for-service	patient	contacts	per	year,	and	delivered	125	fewer	fee-paid	services	(95%	CI:	for	contacts	100-130	fewer;	95%	CI:	for	services:	105-145	fewer).	Taken	together,	the	time	trend	and	interaction	between	time	and	gender	suggest	that	while	the	number	of	fee-reimbursed	contacts	and	services	delivered	by	male	physicians	was	declining,	the	same	was	not	true	for	female	physicians.	Their	contact	and	service	counts	were	stable	throughout	the	study	period.		Contact	and	service	count	patterns	followed	a	parabolic	curve	with	respect	to	physician	age,	peaking	between	45	and	55	years,	and	mirroring	the	trend	I	observed	for	total	compensation	(effect	estimate	for	age	45-55	for	contacts:	576,	95%	CI:	464-688;	for	services:	719,	95%	CI:	574-863).	Physicians	who	trained	internationally	saw	1032	more	patients	per	year	on	average	(95%	CI:	878-1320),	and	delivered	1099	more	services	than	their	Canadian-trained	counterparts	(95%	CI:	842-1223).		I	found	no	difference	in	patient	contacts	based	on	location;	however	the	variables	for	location	interacting	with	international	training	were	significant	and	negative,	indicating	that	internationally-trained	physicians	working	outside	of	metropolitan	areas	had	fewer	patient	contacts	than	Canadian-trained	physicians	in	these	same	areas.	In	contrast,	I	did	find	some	location	based	differences	in	the	service	model:		 105	physicians	working	in	rural-dominated	areas	delivered	397	fewer	services	annually	than	those	in	metropolitan	areas.		The	interaction	terms	for	location	and	international	training	were	not	significant.			4.3.3.3	Unique	patients		In	2005-06,	female	physicians	saw	an	average	of	568	fewer	unique	patients	than	their	male	counterparts	(95%	CI	for	combined	gender	and	gender	by	year	interaction:	477-660	fewer)	(Table	4.4).	As	I	observed	in	the	compensation,	contact	and	service	models,	the	gender	difference	decreased	over	the	study	period:	by	2011-12,	female	physicians	were	seeing	an	average	of	445	fewer	unique	patients	(95%	CI:	for	combined	gender	and	gender	by	year	interaction:	354-537	fewer).	Also	consistent	with	earlier	models,	physicians	were	seeing	fewer	unique	patients	year	on	year	(effect	estimate:	-38,	95%	CI:	-31-	-45).			 Like	contact	and	service	counts,	the	count	of	unique	patients	also	follows	a	parabolic	curve;	however,	it	peaks	between	ages	35	and	45,	with	45	additional	unique	patients	compared	to	physicians	under	age	35	(95%	CI:	1-90).	Physicians	over	65	saw	by	far	the	smallest	number	of	unique	patients	per	year	(effect	estimate	for	65+:	-177,	95%	CI:	-102-	-252).	I	found	no	difference	in	the	number	of	unique	patients	in	a	physician’s	practice	based	on	practice	location.	Physicians	who	trained	internationally	saw	428	more	unique	patients	per	year	than	Canadian-trained	physicians.	The	interaction	terms	for	urban	location	and	international	training,	and	rural	location	and	international	training	were	both	negative	and	significant;	however	the	combined	odds	ratios	were	not.	This	suggests	that	location	of	practice	is	dampening	the	effect	of	international	training:	internationally-trained	physicians		 106	who	worked	in	urban-	or	rural-dominated	locations	had	no	more	unique	patients	than	Canadian-trained	physicians	in	these	same	locations.	Internationally-trained	physicians	working	in	metropolitan	areas,	however,	did	still	have	more	unique	patients	than	their	Canadian-trained	counterparts.	This	may	be	suggestive	of	differences	in	practice	style	by	training	location,	but	ones	that	are	mitigated	by	increasing	rurality	of	practice.			 In	all	four	activity	models,	physician	gender	had	by	far	the	largest	effect	size	of	any	of	the	covariates,	even	by	the	end	of	the	study	period	when	the	gender	differences	were	less	pronounced.15	Male	physicians	had	significantly	higher	total	compensation	and	delivered	more	services	to	more	unique	patients;	however,	those	differences	narrowed	over	the	study	period	for	all	activity	outcomes,	with	female	physicians’	earnings	increasing	over	the	study	period,	while	contact,	service	and	unique	patient	counts	remained	stable.	4.3.3.3	APP	remuneration	Compared	to	male	physicians,	female	physicians	had	significantly	higher	odds	of	being	high-adopters	of	APP	remuneration,	at	all	three	cut-off	levels	(Table	4.5)	(45%	OR:	1.29,	95%	CI:	1.10-1.50;	50%	OR:	1.27,	95%	CI:	1.09-1.49;	55%	OR:	1.23,	95%	CI:	1.16-1.32).		The	odds	of	being	a	high	adopter	at	any	of	the	three	cut-offs	decreased	with	age,	with	those	over	age	65	being	the	least	likely	to	be	high-adopters	(45%	OR:	0.41,	95%	CI:	0.29-0.59;	50%	OR:	0.41,	95%	CI:	0.29-0.59;	55%																																																									15	The	proportion	of	income	from	APP	services	frequently	had	a	larger	effect	size	than	physician	gender;	however,	I	am	choosing	not	to	count	this	since	the	reason	for	including	it	in	the	models	(with	the	exception	of	the	total	compensation	model)	was	to	account	for	the	fact	that	the	outcome	variables	reflect	activity	within	the	FFS	realm	only.		As	such,	the	APP	term	is	included	as	a	proxy	for	the	remaining	missing	activity.		 107	OR:	0.52,	95%	CI:	0.45-0.60).	Internationally-trained	physicians	had	lower	odds	of	being	high-adopters	of	APP	programs	at	all	three	cut-off	points	(45%	OR:	0.79,	95%	CI:	0.63-0.99;	50%	OR:	0.78,	95%	CI:	0.62-0.99;	55%	OR:	0.70,	95%	CI:	0.64-0.77).		Table	4.5:	Multivariate	modeling	results:	binary	APP	uptake	Variables	Model	1:		45%	Cut-off	 Model	2:		50%	Cut-off	 Model	3:		55%	Cut-off	Odds	Ratio	 95%	Confidence	Interval	 Odds	Ratio	 95%	Confidence	Interval	 Odds	Ratio	 95%	Confidence	Interval	Sex	(female)	 1.29	 1.1	-	1.5‡	 1.27	 1.09	-	1.49‡	 1.23	 1.16	-	1.32*	Age:	35-<45	 1.08	 0.84	-	1.40	 1.08	 0.83	-	1.40	 0.99	 0.90	-	1.10	Age:	45-<55	 0.68	 0.53	-	0.88‡	 0.67	 0.52	-	0.87‡	 0.70	 0.63	-	0.78*	Age:	55-<65	 0.62	 0.48	-	0.81†	 0.61	 0.46	-	0.8†	 0.66	 0.59	-	0.73*	Age:	65+	 0.41	 0.29	-	0.59*	 0.41	 0.29	-	0.59*	 0.52	 0.45	-	0.60*	Rurality:	urban-dominated	 0.86	 0.70	-	1.06	 0.84	 0.68	-	1.03	 0.69	 0.64	-	0.76*	Rurality:	rural-dominated	 0.81	 0.63	-	1.04	 0.78	 0.60	-	1.01	 0.69	 0.62	-	0.77*	International	training	 0.79	 0.63	-	0.99‡	 0.78	 0.62	-	0.99‡	 0.70	 0.64	-	0.77*	Rurality*training	interaction	(urban*international)	 0.59	 0.38	-	0.92‡	 0.59	 0.38	-	0.94‡	 0.70	 0.58	-	0.85†	Rurality*training	interaction	(rural*international)	 0.59	 0.35	-	0.99‡	 0.53	 0.31	-	0.91‡	 0.84	 0.68	-	1.04	*p<0.0001	†	p<0.001		‡p<0.05	 	 	 	 	 	 	Additionally,	physicians	in	urban-	and	rural-dominated	areas	were	less	likely	to	be	high	adopters,	but	only	at	the	highest	cut-off	value	(OR	for	urban-dominated:	0.69,	95%	CI:	0.64-0.76;	OR	for	rural-dominated:	0.69,	95%	CI:	0.62-0.77).	The	interaction	terms	for	both	urban-dominated	location	and	international	training,	and	rural-dominated	location	and	international	training	were	significant	and	negative	at	all	three	cut-off	values	(e.g.	combined	OR	for	rural	location,	international	training	and	interaction	term	at	45%	cutoff:	0.19,	95%	CI:	0.01-0.75).	This	suggests	that	internationally	trained	physicians	practicing	in	urban-	or	rural-dominated	locations	had	lower	odds	of	being	high	APP	adopters	compared	to	Canadian-trained		 108	physicians	in	the	same	locations;	however	the	confidence	intervals	were	substantial,	suggesting	significant	variability	in	the	data.		Results	for	the	ordinal	logistic	regression	were	similar	(Table	4.6),	with	female	physicians	having	1.3.1	times	the	odds	of	being	in	the	high	uptake	category	(75%	of	more	income	from	APP)–	compared	to	the	combined	effect	of	low-	or	moderate-APP	uptake	(95%	CI:	1.14-1.50).16		As	with	the	binary	APP	models,	physicians	in	successive	age	categories	had	lower	odds	of	being	high	adopters,	as	did	individuals	who	trained	in	international	locations.	Unlike	the	binary	models,	however,	physicians	practicing	in	rural-dominated	locations	had	higher	odds	of	being	in	the	high	uptake	category	(OR:	1.25,	95%	CI:	1.02-1.54).	This	may	be	a	statistical	artifact	of	the	bimodal	distribution	of	the	APP	variable,	whereby	individuals	practicing	in	rural-dominated	are	more	likely	to	be	in	the	highest	and	lowest	uptake	categories,	but	less	likely	to	be	moderate	APP	adopters.																																																										16	This	could	also	be	interpreted	as	females	having	a	1.30	times	higher	odds	of	being	in	either	moderate-	or	high-uptake	categories.		 109	Table	4.6:	Multivariate	results:	ordinal	logistic	APP	uptake	Variables		 Odds	Ratio	 95%	Confidence	Interval	Sex	(female)	 1.31	 1.14	-	1.50†	Age:	35-<45	 1.05	 0.84	-	1.32	Age:	45-<55	 0.68	 0.54	-	0.85†	Age:	55-<65	 0.67	 0.53	-	0.84†	Age:	65+	 0.46	 0.34	-	0.63*	Rurality:	urban-dominated	 0.98	 0.82	-	1.18	Rurality:	rural-dominated	 1.25	 1.02	-	1.54‡	International	training	 0.74	 0.60	-	0.92‡	Rurality*training	interaction	(urban*international)	 0.69	 0.47	-	1.01	Rurality*training	interaction	(rural*international)	 0.68	 0.45	-	1.02	*p<0.0001	†	p<0.001		‡p<0.05	 	 	4.3.4	Question	1.2:	Multivariate	results	Is	there	a	gender-related	difference	in	activity	level	across	the	complete	career	trajectory,	and	is	the	magnitude	of	that	difference	age	dependent?				 Gender	was	a	significant	predictor	of	total	compensation,	contacts,	and	services	in	all	of	the	five-year	multivariate	GLMs	(Figure	4.1,	4.2,	4.3	and	4.4).		For	total	compensation	(Figure	4.1),	percent	differences	ranged	from	23%	for	physicians	under	age	30	to	55%	for	physicians	between	ages	35	and	40	(meaning	female	physicians	under	age	30	earned	79%	of	what	their	male	counterparts	earned),	and	57%	between	the	ages	of	35	and	40.	The	maximum	percent	difference	was	in	two	age	brackets:	between	ages	35	and	40	and	between	ages	65	and	70.	The	difference	was	smallest	between	50	and	60.			 110		Figure	4.1:	Five-year	multivariate	GLMs,	total	compensation	The	results	for	patient	contacts	and	service	counts	are	similar	to	those	presented	for	total	compensation	(Figures	4.3	and	4.3).	Percent	differences	ranged	from	22%	to	52%	for	contacts,	19%	to	51%	for	services	and	12%	to	60%	for	unique	patients,	and	peaks	in	differences	occurred	at	age	groups	35-<40	and	65-<70	for	all	variables.	The	percent	difference	was	generally	smallest	in	the	youngest	and	oldest	age	brackets	across	all	dependent	variables;	however,	this	may	be	related	to	small	N’s	and	thus	substantially	wider	confidence	intervals	for	female	physicians	in	the	upper	age	categories	(N=271	for	age	65-70,	and	N=130	for	70+).	0102030405060050000100000150000200000250000300000350000400000<30 30-<35 35-<40 40-<45 45-<50 50-<55 55-<60 60-<65 65-<70 70+Annual	Total	CompensationAge	GroupMen Women Percent	Difference	 111		Figure	4.2:	Five-year	multivariate	GLMs,	patient	contacts	The	differences	in	activity	were	significant	across	all	age	categories,	and	at	no	point	did	mean	female	physicians’	compensation,	contact,	service	or	unique	patient	counts	meet	or	exceed	those	of	male	physicians.	Importantly	however,	the	percent	difference	in	activity	was	certainly	not	constant	across	the	work	life-span,	ranging	from	12%	in	the	mid-career	years,	up	to	60%	(in	the	case	of	unique	patients,	for	example)	from	age	35	to	<40.				0102030405060010002000300040005000600070008000<30 30-<35 35-<40 40-<45 45-<50 50-<55 55-<60 60-<65 65-<70 70+Annual	ContactsAge	GroupMen Women Percent	Difference	 112		Figure	4.3:	Five-year	multivariate	GLMs,	services		Figure	4.4:	Five-year	multivariate	GLMs,	unique	patient	counts	0102030405060010002000300040005000600070008000<30 30-<35 35-<40 40-<45 45-<50 50-<55 55-<60 60-<65 65-<70 70+Annual	ServicesAge	GroupMen Women Percent	Difference010203040506070050010001500200025003000<30 30-<35 35-<40 40-<45 45-<50 50-<55 55-<60 60-<65 65-<70 70+Unique	PatientsAge	GroupMen Women Percent	Difference	 113	4.4	Discussion	Consistent	with	existing	international	and	Canadian	literature,	and	with	my	initial	hypotheses,	this	analysis	suggests	that	female	primary	care	physicians	in	BC	earn	less	than	their	male	colleagues,	see	fewer	patients,	and	deliver	fewer	services	(Boerma	&	van	den	Brink-Muinen,	2000;	Canadian	Institute	for	Health	Information,	2001b;	Cohen	et	al.,	1991;	C.	A.	Woodward	&	Hurley,	1995).	They	also	see	fewer	unique	patients	in	a	year.	Most	past	research	examining	gender	differences	in	activity	suffered	from	significant	methodological	weaknesses,	however,	rendering	this	work	a	significant	contribution	to	the	literature	in	this	area.		The	differences	in	activity	levels	between	male	and	female	physicians	narrowed	significantly	over	the	study	period	(reductions	of	30%	for	services,	26%	for	patient	contact,	and	10%	for	total	remuneration);	however,	even	in	2011-12,	physician	gender	remained	the	most	important	explanation	for	differences	in	total	remuneration,	patient	contacts	and	services.			Between	2005-06	and	2011-12,	I	found	significant	declines	in	all	measures	of	activity,	suggesting	reduced	physician	workloads	over	time.	I	also	found	a	parabolic	activity	curve	related	to	physician	age:	across	all	measures	of	activity,	the	youngest	and	oldest	age	groups	had	the	lowest	contact	and	service	counts,	and	total	remuneration,	and	the	smallest	practice	sizes	(fewest	unique	patients).	Differences	in	activity	based	on	practice	or	training	location	were	small,	and	not	always	significant.			The	results	reported	in	this	chapter	demonstrate	that	female	physicians	were,	at	least	during	this	period	of	time,	more	likely	than	their	male	colleagues	to	be		 114	high-adopters	of	APP.		To	my	knowledge,	this	is	the	first	study	to	include	non-fee-for-service	payments	in	an	examination	of	activity	measures	for	primary	care	physicians.	The	exclusion	of	non-FFS	data	is	a	substantial	limitation	in	the	existing	literature,	introducing	a	bias	and	spuriously	inflating	the	activity	differences	between	male	and	female	PCPs,	as	demonstrated	here	by	the	results	showing	considerably	higher	APP	participation	amongst	female	PCPs.	This	also	represents	a	significant	contribution	to	the	existing	literature,	given	that	more	physicians,	and	female	physicians	in	particular,	are	moving	to	mixed	or	to	100%	APP-based	remuneration.			Work	by	Watson	and	colleagues	suggested	that	a	gender-related	activity	difference	widened	between	1992	and	2001	(Watson	et	al.,	2006).	Watson	et	al.	also	reported	the	existence	of	a	cohort	effect	whereby	younger	and	middle-aged	physicians	carried	smaller	workloads	in	2003	than	did	their	same-aged	colleagues	in	1993.	This	intergenerational	effect	did	not	interact	with	gender	(Watson	et	al.,	2006).		In	contrast,	I	found	that	while	male	physicians	are	continuing	to	earn	more,	and	to	have	higher	patient	and	service	counts,	this	difference	is	beginning	to	close,	with	female	physicians’	earnings	increasing	over	time,	and	their	contact	and	service	counts	remaining	stable,	relative	to	their	male	counterparts.	The	lack	of	agreement	between	my	findings	and	those	of	Watson	et	al.	could	be	interpreted	in	two	ways:	first,	it	is	possible	that	the	direction	of	the	interaction	effect	between	gender	and	time	flipped	between	2001,	the	last	year	examined	by	Watson	et	al.,	and	2005,	the	first	year	I	analyzed.	What	might	be	more	likely,	however,	is	that	the	inclusion	of	the	APP	data	made	it	possible	for	the	analysis	here	to	provide	a	more	accurate		 115	assessment	of	activity,	particularly	among	female	physicians	who	are	more	likely	to	be	high	adopters	of	APP	and	who	have	a	larger	proportion	of	their	total	remuneration	coming	from	APP	sources.	Examining	the	MSP	(FFS)	data	in	isolation	would	produce	the	results	observed	by	Watson	and	colleagues	if	a)	female	physicians	were	more	likely	to	receive	APP	remuneration;	and	b)	the	proportion	of	their	activity	related	to	APP	was	increasing	over	their	period	of	study.	Because	I	did	not	have	access	to	data	prior	to	2005,	I	cannot	confirm	with	certainty	that	this	is	the	case;	however,	it	makes	intuitive	sense	given	that	provincial	spending	on	APP	programs	and	the	proportion	of	physicians	who	receive	some	remuneration	from	those	programs	has	been	increasing	consistently	since	the	program’s	inception	in	1968,	both	in	BC	and	in	Manitoba	where	Watson	and	colleagues	performed	their	analysis	(Canadian	Institute	for	Health	Information,	2005a;	Office	of	the	Auditor	General	of	British	Columbia,	2003).	I	found,	however,	that	the	proportion	of	payments	from	APP	sources	did	not	change	between	2005	and	2012,	suggesting	a	leveling	off	of	this	trend.		As	noted	above,	I	found	that	the	number	of	patient	contacts,	services	delivered,	and	unique	patients	declined	significantly	for	both	male	and	female	primary	care	physicians	(though	less	so	for	females)	over	the	study	period.	Although	total	remuneration	also	declined,	the	percentage	change	was	much	smaller.	These	seemingly	conflicting	results	could	be	explained	by	increases	in	the	proportion	of	physician	income	(and	therefore	activity)	from	APP	sources;	however,	I	found	no	significant	time	trend	in	APP	uptake.	The	results	could	also	be	related	to	increases	in	the	availability	of	clinical	and	non-clinical	incentives	during	the	study		 116	period;	this	is	the	subject	of	Chapter	5.		Regardless	of	reasons,	however,	the	decline	in	contacts,	services,	and	patient	loads	over	time	could	be	taken	as	evidence	of	a	“backward	bending	supply	curve”,	which	is	an	economic	concept	suggesting	that	beyond	a	certain	income	level,	people	will	substitute	more	leisure	time	for	higher	workloads	(McGuire	&	Pauly,	1991).	If	income	per	unit	of	time	can	be	increased,	PCPs	may	be	disinclined	to	increase	patient	practice	sizes,	or	work	hours.			4.4.1	Limitations	Strengths	and	limitations	that	apply	to	the	thesis	broadly	are	discussed	in	Chapter	8.		The	analysis	in	this	chapter	had	some	specific	limitations.	Chief	among	them	is	the	fact	that	I	have	no	way	of	analyzing	the	specific	activities	that	occur	within	clinical	APP	programs,	nor	the	specific	patients	those	activities	are	linked	to.	This	means	that	while	I	was	able	to	look	at	overall	compensation	as	a	measure	of	activity,	I	could	not	look	at	contact	or	service	counts	for	the	APP	portion	of	that	compensation.	To	take	account	of	the	fact	that	there	would	be	lower	contact	and	service	counts	observed	among	individuals	whose	remuneration	sources	were	mixed,	I	included	percentage	APP	as	a	covariate	in	all	multivariate	models.	Additionally,	I	excluded	individuals	whose	income	was	100%	APP	from	the	multivariate	contact	and	service	models.	This	exclusion	represented	approximately	4%	of	my	overall	cohort	and	slightly	more	female	physicians.		A	second	important	limitation	is	the	lens	through	which	I	have	looked	at	“activity”.	Using	payments	in	conjunction	with	contact	and	service	counts,	and	unique	patient	counts,	produces	an	incomplete	picture	of	physician	activity.	For		 117	example,	it	is	not	possible	to	measure	overall	hours	worked,	nor	time	spent	with	each	patient,	using	administrative	data	resources.	Under	a	fee-for-service	model,	a	physician	who	works	two	hours	and	sees	eight	patients	may	earn	the	same	income	as	another	physician	who	sees	those	same	eight	patients	in	a	single	hour.		There	is	some	existing	evidence	to	suggest	that	female	physicians	do	spend	significantly	more	time	with	each	patient	and	manage	more	independent	problems	per	patient	(Britt	et	al.,	1996;	Chaytors	et	al.,	2001).	Further,	longer	visit	length	has	been	linked	to	improvements	in	patient	and	physician	satisfaction	(Dugdale,	Epstein,	&	Pantilat,	1999),	and	increases	in	the	provision	of	preventative	medicine	and	screening	(Dugdale	et	al.,	1999;	Goldbloom,	1978;	Morrell,	Evans,	Morris,	&	Roland,	1986;	Roland,	Bartholomew,	Courtenay,	Morris,	&	Morrell,	1986).			We	have	no	measure	of	work	related	to	non-clinical	activities,	such	as	management	of	practice,	administration	or	education.	Thus,	although	I	have	attempted	to	be	rigorous	in	my	approach	to	activity	measurement,	looking	beyond	compensation	at	contact	and	service	patterns,	and	including	APP	data,	my	examination	of	activity	is	necessarily	incomplete.		An	additional	limitation	is	the	overall	number	of	years	under	study.	I	would	have	liked	to	be	able	to	further	investigate,	and	control	for,	any	period	effects	on	activity	levels.	Because	the	APP	data	are	not	available	for	years	prior	to	2005-06,	I	was	unable	to	do	this.	Also	I	was	unable	to	include	both	physician	age	and	cohort	in	the	model	separately.	Bivariate	analyses	suggested	that	age	and	graduation	cohort	were	highly	collinear,	which	precluded	including	both	in	the	same	model.	However,	because	of	this	high	degree	of	correlation,	the	inclusion	of	physician	age	in	my		 118	model	actually	accounts	for	the	combined	effects	of	both	age	and	graduation	cohort.	Though	this	is	a	methodological	strength	–	allowing	me	assess	gender	effects	independent	of	age	and	graduation	cohort	–	it	does	mean	that	I	was	unable	to	differentiate	activity	differences	caused	by	aging,	and	those	caused	by	changes	in	training	over	time.	4.4.2	Implications	The	research	and	policy	implications	of	this,	and	the	remaining	analytic	chapters,	are	described	as	part	of	Chapter	8.	This	analysis	addresses	a	substantive	and	important	gap	in	knowledge	related	to	the	potential	impact	of	the	feminization	of	the	workforce	and	changes	in	activity	patterns	with	time.		It	suggests	that	although	there	is	a	gender-related	gap	in	overall	activity,	this	gap	is	narrowing.	It	also	suggests	that	physicians	are	reducing	their	patient	contact	and	service	counts	over	time,	while	maintaining	their	total	remuneration	levels.	This	finding	will	be	explored	in	more	depth	in	subsequent	chapters.						 		 119	CHAPTER	5:	Clinical	activities	and	incentives	5.1	Introduction		Physicians	in	BC	are	increasingly	being	compensated	under	non-fee-for-service	(FFS)	remuneration	arrangements	(alternative	payment	plans	(APP)),	and	for	activities	outside	of	direct	patient	care.		Examples	of	non-fee-for-service	income	include	bonuses,	incentives,	and	on-call	payments	(British	Columbia	Ministry	of	Health,	2015).	APP	payments	to	physicians	now	account	for	approximately	20%	of	overall	payments	for	physician	services	(British	Columbia	Ministry	of	Health,	2015).			In	Chapter	4,	I	established	that	there	is	a	gender	difference	in	overall	(fee-for-service-based)	activity	levels,	with	male	primary	care	physicians	(PCP)	earning	more	overall	and	seeing	more	patients	and	delivering	more	services	annually	compared	to	female	physicians.		I	also	established	that	women	are	more	likely	to	be	paid	using	alternative	arrangements.	This	chapter	examines	the	extent	to	which	payments	for	activities	or	functions	other	than	direct	clinical	care	delivery	(including	clinical	incentives,	non-clinical	incentives	and	others)	within	both	the	FFS	and	APP	remuneration	schemes	may	contribute	to	the	observed	gender	differences	in	income	and	activity	over	time.			 Throughout	this	chapter	I	refer	to	“compensation	for	direct	clinical	care	delivery”.	I	am	defining	this	as	payment	specifically	for	the	provision	of	health	care	services.	Clinical	incentive	payments,	in	contrast,	are	treated	as	bonuses	for	the	delivery	of	health	care	services	to	a	specific	segment	of	the	population,	provided	in	addition	to	a	physician’s	regular	clinical	fees.	Non-clinical	payments	include		 120	remuneration	for	non-clinical	activity	(e.g.	academic	services	contracts),	and	bonuses	and	incentives	for	care	provision	at	a	certain	time	or	in	a	particular	location	(e.g.	incentives	for	time	spent	on-call,	or	bonuses	for	practicing	in	rural	or	isolated	settings).	The	complete	list	of	payment	types	and	their	division	into	direct	clinical	services,	clinical	incentives,	and	non-clinical	payments	was	presented	in	Chapter	3,	section	3.5.2.			 In	their	report	on	the	uptake	of	the	General	Practices	Services	Committee’s	Full	Service	Family	Practice	Program	incentives	(which	represent	the	majority	of	clinical	incentive	payments	to	BC’s	primary	care	physicians),	Hollander	and	Tessaro	(2009)	find	that	by	2007/08,	71.7%	of	PCPs	received	at	least	one	clinical	incentive	payment	through	the	program;	however,	they	did	not	examine	uptake	by	gender,	nor	did	they	measure	the	proportion	of	physicians’	total	income	accounted	for	by	these	incentives	(Hollander	&	Tessaro,	2009).	Neither	the	Medical	On-Call	Availability	Program	(MOCAP),	nor	the	rural	recruitment	and	retention	programs	–	which	together	make	up	the	majority	of	non-clinical	payments	–	have	been	evaluated	to	estimate	uptake	as	a	proportion	of	total	compensation,	or	differential	uptake	by	gender.			 There	is	limited	existing	literature	on	whether	gender	affects	the	likelihood	of	participation	in	on-call	programs.	Previous	studies	found	that	in	spite	of	gender	differences	in	hours	worked	overall,	and	in	hours	spent	on	patient	care,	male	and	female	physicians	tend	to	spend	a	similar	amount	of	time	on-call	(Atkin,	2000;	Carek	et	al.,	2003;	Raymont	et	al.,	2005).		All	of	these	studies	rely	on	cross	sectional	surveys.			 121	To	my	knowledge	there	are	no	prior	studies	that	have	used	payment	data	to	examine	the	relationships	among	gender,	time,	and	on-call	activity.	Similarly,	there	is	no	existing	literature	that	examines	trends	in	the	distribution	of	physician	payments	between	clinical-	and	non-clinical	compensation	in	BC,	or	in	Canada	more	broadly,	or	the	relationship	between	any	such	trends	and	physician	gender.	I	was	particularly	interested	in	the	extent	to	which	gender-based	differences	in	uptake	of	non-clinical	incentives	were	becoming	explanatory	of	gender	differences	in	overall	activity	or	compensation.		The	objective	of	the	work	found	in	the	remainder	of	this	chapter	is	to	investigate	gender	differences	in	remuneration	for	clinical	vs.	non-clinical	care	and	specifically	in	uptake	of	clinical	and	non-clinical	incentive	payments.	The	specific	research	questions	I	am	addressing	are	as	follows:	• Question	2.1:	Do	clinical	payments	make	up	a	significantly	larger	proportion	of	total	income	for	male	or	female	physicians?	• Question	2.2:	Are	male	or	female	physicians	more	likely	to	bill	for	clinical	and	non-clinical	incentive	payments,	and	has	this	changed	over	time?	• Question	2.3:	How	has	the	makeup	of	physicians’	income	changed	with	time	and	are	changes	differential	by	gender?	• Question	2.4:	Does	uptake	of	on-call	and	rural	and	remote	incentive	payments	vary	for	male	versus	female	PCPs,	controlling	for	age,	period,	and	practice	location?		I	hypothesize	that	male	physicians	will	be	more	likely	to	bill	for	clinical	and	non-clinical	incentive	payments,	and	that	these	payments	will	make	up	a	larger	proportion	of	their	total	income	than	of	the	total	income	of	their	female		 122	counterparts.		Also,	I	hypothesize	that	clinical	and	non-clinical	incentives	will	represent	a	growing	proportion	of	physician	income	over	time,	regardless	of	gender.		5.2	Methods,	variables,	and	data	sources	Information	on	the	datasets	and	study	cohort	used	for	this	analysis	can	be	found	in	Chapter	3,	Sections	3.2	and	3.4.	As	with	Chapter	4,	this	study	deals	only	with	variables	related	to	physician	payments	and	demographics	and	therefore	draws	data	from	the	Medical	Services	Plan	(MSP)	Practitioner	File,	College	of	Physicians	and	Surgeons	(CPSBC)	Physician	Registry,	MSP	Payment	Information	File,	and	APP	databases	only.	Since	no	patient-level	data	are	used	in	this	analysis,	the	MSP	consolidation	file,	discharge	abstract	database,	and	Vital	Statistics	databases	were	not	used.		5.2.1	Dependent	variables	This	analysis	examines	payments	to	physicians	in	finer	detail	than	the	analyses	in	the	previous	chapter.		I	built	multivariate	models	with	the	following	dependent	variables:	• Proportion	of	total	remuneration	for	direct	clinical	care	delivery;	• Remuneration	for	clinical	incentives	(measured	as	both	yes	or	no,	and	proportion	of	total	payments);	• Remuneration	for	non-clinical	incentives	(measured	as	both	yes	or	no,	and	proportion	of	total	payments);	and		 123	• Uptake	of	on-call	payments	and	rural	and	remote	incentives	(yes	or	no	only)17.			Full	definitions	and	descriptions	for	these	variables	are	provided	in	Chapter	3,	Section	3.5.2.	5.2.2	Explanatory	variables		As	with	the	analyses	in	Chapter	4,	I	looked	at	the	independent	impact	of	physician	gender	on	all	the	remuneration	dependent	variables	described	above.	I	used	the	same	set	of	physician	demographic	characteristics	as	explanatory	covariates:	age,	training	location	(Canada	or	international),	and	practice	rurality.	I	also	included	percentage	of	total	payments	from	APP	sources	as	a	covariate	in	the	models	to	account	for	the	different	clinical,	clinical	incentive,	and	non-clinical	payment	types	and	amounts	that	occur	within	the	FFS	and	APP	remuneration	schemes.	Opportunities	for	billing	for	incentive	payments	–	which	make	up	a	large	proportion	of	non-clinical	payments	–	are	greater	within	traditional	fee-for-service	remuneration	arrangements.	Therefore,	I	expected	that	physicians	whose	income	is	largely	or	entirely	fee-for-service	would	be	likely	to	receive	more	of	these	incentives.	Lastly,	I	included	the	study	year,	from	2005-06	to	2011-12,	to	examine	trends	in	payments	for	clinical	care	delivery	and	incentives	over	time.	Definitions	for	each	of	these	variables	are	provided	in	Chapter	3,	Section	3.5.1.																																																										17	I	elected	not	to	model	the	proportion	of	physicians’	income	from	MOCAP	payments	or	rural	and	remote	incentives	because	the	descriptive	data	suggested	that	these	payments	accounts	for	a	very	small	percentage	of	total	payments	for	the	vast	majority	of	physicians,	and	because	it	is	subsumed	into	non-clinical	incentive	payments	more	generally.			 124	5.2.3	Statistical	analyses	5.2.3.1	Descriptive	statistics,	univariate	and	bivariate	analyses			 This	analysis	uses	the	same	cohort	as	was	used	in	Chapter	4.		Cohort	demographics	will	not	be	repeated	here;	detail	can	be	found	in	Table	4.1.	I	computed	measures	of	central	tendency	and	dispersion	for	clinical	and	non-clinical	activities,	on-call	payments,	and	incentives	and	bonuses,	and	for	the	proportion	of	total	payments	for	each	category.		I	constructed	a	line	graph	for	each	continuous	variable	to	assess	skew.		I	computed	unadjusted	measures	of	association	between	the	proportional	payment	outcomes	listed	above,	physician	gender	and	other	physician	demographics	using	Wilcoxon-Man-Whitney	two	sample	statistics.	I	constructed	frequency	tables	and	conducted	chi-square	tests	to	examine	the	difference	between	male	and	female	physicians	and	across	physician	demographic	categories.	These	analyses	were	used	to	assess	potential	for	confounding	and	collinearity	in	multivariate	model	estimations	to	come.	5.2.3.2	Multivariate	modeling	and	longitudinal	analyses	Question	2.1:		The	proportion	of	a	physician’s	annual	payments	related	to	clinical	care	delivery	is	highly	left-skewed,	and	has	natural	boundaries	at	zero	and	one.	These	properties	violate	assumptions	within	a	standard	linear	regression	model,	making	this	modeling	technique	inappropriate.	Instead,	I	modeled	the	logit	of	the	proportion	of	physician	payments	related	to	clinical	care	delivery	using	a		 125	linear	mixed	effects	model	under	a	normal	distribution	with	an	identity	link	function1819.	The	probability	distribution	for	a	logit-normal	distribution	for	a	variable,	!,		is	given	by:	"# !; %, ' = 1'√2, 1!(1 − !) 01(23456(#)17)89:8 		When	computing	the	logit	of	the	proportion	of	payments	for	direct	clinical	care,	any	values	that	are	zeros	or	ones	become	negative	infinity	and	positive	infinity	respectively.	Thus,	these	values	are	necessarily	excluded	from	the	model.	In	order	to	avoid	the	exclusion,	I	rescaled	the	zero	and	one	values	using	this	formula	(Smithson	&	Verkuilen,	2006):			;< = [; > − 1 + 0.5]> 	Where:		y’	=	rescaled	proportion	of	payments	for	clinical	care	delivery		y	=	original	value	for	proportion	of	payments	for	clinical	care	delivery	N	=	number	of	records	in	the	dataset,	where	a	record	is	a	single	year	of	activity	for	an	individual	physician	 		Thus,	with	the	35,973	records	(person	years)	in	my	dataset,	the	185	zeros	would	become	0.00001	and	the	1987	ones	would	become	0.99999.	I	modeled	the	logit	of	the	proportion	of	a	physician’s	annual	income	related	to	direct	clinical	care																																																									18	The	link	function	is	the	statistical	relationship	between	the	linear	predictors	and	the	mean	of	the	distribution	function.	19	I	also	modeled	the	proportion	of	a	physician’s	income	for	direct	clinical	care	under	a	beta	distribution	and	logit	link	function;	however,	I	found	that	the	model	fit	was	inferior	to	that	of	the	logit	normal	model.		 126	delivery	as	a	function	of	physician	demographics,	time,	and	the	percentage	of	payments	from	APP	sources:		DEFGH(;5I) = JK +	JMNI +	J9O5 +	JP NI ∗ O5 +	JRS5I + JTU5 + JVW5I + JX U5 ∗ W5I+ JYZ5I + [K5 + [M5NI + \5I 	]K5]M5 ~> 00 , '_K9 '_K_M'_K_M '_M9 			\5I~> 0, '9 	Where:		y	=	proportion	of	payments	for	clinical	care	delivery	for	physician	i	in	year	j	T=	time	at	year	j	G	=	gender	for	physician	i		A	=	age	group	for	physician	i	in	year	j	L	=	location	of	training	for	physician	i		R	=	practice	rurality	for	physician	i	in	year	j	P=	percent	APP	for	physician	i	in	year	j		This	model	includes	a	variable	for	the	percent	of	a	physician’s	income	from	APP	sources	to	account	for	the	fact	that	opportunities	for	billing	for	incentive	payments	are	greater	within	traditional	FFS	payment	arrangements.		The	random	effects	structure	mirrors	the	one	I	used	for	the	models	presented	in	Chapter	4.	I	included	random	effects	at	the	subject	level	to	account	for	correlation	in	repeated	measures,	and	I	included	random	effects	for	slope	and	for	subject-level	residuals.	I	assumed	the	residual	random	effect	would	follow	a	first-order	autoregressive	correlation	structure.		Question	2.2:		Clinical	Incentives:	The	distribution	of	the	proportion	of	physicians’	income	from	clinical	incentive	payments	has	considerable	right	skew	and	heteroscedasticity,	making	an	estimation	approach	based	on	assuming	an	underlying	normal	distribution	inappropriate.	Additionally,	more	than	30%	of	the		 127	total	physician-year	records	(11,484	of	35,013)	have	zero	values,	which	would	be	excluded	under	a	beta	regression	model,	rendering	it	unusable	as	well.	Rescaling	the	zero	and	one	values	using	the	strategy	developed	by	Smithson	and	Verkuilen	(2006),	and	then	using	a	logit	transformation,	would	also	produce	misleading	results.	Zero	values	become	negative	infinity,	and	therefore	become	highly	influential	observations,	pulling	the	predicted	probabilities	towards	zero.	This	could	also	affect	the	effect	estimate	for	gender.	If,	for	example,	more	males	than	females	have	zero	values,	the	odds	for	males	will	be	pulled	more	strongly	towards	zero,	compared	to	females.	This	does	turn	out	to	be	the	case.	Among	those	with	zero	values,	64%	are	male	and	36%	are	female	(chi-square	=	883,	p<0.0001).		To	solve	this	problem,	I	used	a	combined	modeling	strategy	similar	to	one	recommended	by	Fletcher	et	al.	(Fletcher,	MacKenzie,	&	Villouta,	2005).	This	approach	consists	of	separately	modeling	the	occurrence	of	a	zero	value	(meaning	a	physician	did	not	bill	for	any	clinical	incentives	in	a	particular	year)	as	a	binary	variable	using	a	logistic	modeling	approach,	and	then	modeling	the	proportion	without	the	zero	values	using	an	appropriate	linear	mixed	model.	Thus,	first	I	modeled	whether	or	not	a	physician	received	any	form	of	clinical	incentive	payment	during	each	study	year	with	a	longitudinal	log	linear	model	under	the	assumption	of	a	binary	distribution,	and	logit	link	function.			The	logistic	function	is	given	by	" `; a = 11 + 01(bcd	be#)		 128	Where	k	and	p	are	the	location	and	scale	parameters	respectively20,	and	e	is	Euler’s	number	(e=2.71828…).		I	modeled	the	dichotomous	dependent	variable	as	a	function	of	physician	gender,	demographics,	and	the	proportion	of	income	coming	from	APP	sources:		DEF(;5I) = JK +	JMNI +	J9O5 +	JP NI ∗ O5 +	JRS5I + JTU5 + JVW5I + JX U5 ∗ W5I+ JYZ5I + \5I 	 	\5I~> 0, '9 	Where:		y	=	Y/N	for	billing	clinical	incentive	payments	for	physician	i	in	year	j	T=	time	at	year	j	G	=	gender	for	physician	i		A	=	age	group	for	physician	i	in	year	j	L	=	location	of	training	for	physician	i		R	=	practice	rurality	for	physician	i	in	year	j	P=	percent	APP	for	physician	i	in	year	j		I	included	one	random	effect	for	subject-level	residuals	under	a	first-order	auto-regressive	correlation	matrix.21			Second,	I	modeled	the	proportion	of	a	physician’s	income	from	clinical	incentives	only	for	those	records	that	had	a	non-zero	value.	I	rescaled	the	values	of	one	(N=	2	person	years)	according	to	the	formula	from	Smithson	and	Verkuilen																																																									20	The	location	parameter,	k,	is	the	location	on	a	horizontal	x-axis	where	a	distribution	is	centered.	The	scale	parameter	dictates	the	statistical	dispersion	of	a	distribution.	A	large	value	for	p	would	indicate	data	that	are	very	spread	out.	In	a	standard	normal	distribution,	the	scale	parameter	is	the	standard	deviation.	For	the	standard	normal	distribution,	the	location	parameter	would	a	mean	of	zero,	and	the	scale	parameter	would	be	the	standard	deviation.		21	My	preference	would	have	been	to	include	additional	random	effects	for	individual	physician	intercept	and	slope;	however,	issues	with	model	convergence	prevented	this.	I	did	run	the	model	with	each	random	effect	individually	and	found	no	difference	in	results.		 129	(2006).22		I	modeled	the	logit	of	the	dependent	variable	using	a	linear	mixed	effects	model	under	a	normal	distribution	with	a	logit	link	function.	As	with	the	binary	log	linear	model,	I	was	only	able	to	include	a	random	effect	for	subject-level	residuals	due	to	convergence	issues23;	however,	I	did	test	whether	including	random	effects	for	subject	intercept	and	slope	instead	of	subject	residuals	would	affect	the	effect	estimates	or	variance	therein,	and	they	did	not.	I	also	tested	the	use	of	beta	regression	but	found	that	the	model	fit	was	inferior	and	convergence	issues	were	worse.	I	modeled	the	proportion	of	a	physician’s	payments	from	clinical	incentives	as	a	function	of	physician	gender,	demographics,	and	the	proportion	of	their	income	coming	from	APP	sources:		DEFGH(;5I) = JK +	JMNI +	J9O5 +	JP NI ∗ O5 +	JRS5I + JTU5 + JVW5I + JX U5 ∗ W5I+ JYZ5I + \5I 	 		\5I~> 0, '9 		Where:		y	=	proportion,	clinical	incentive	payments	for	physician	i	in	year	j	T=	time	at	year	j	G	=	gender	for	physician	i		A	=	age	group	for	physician	i	in	year	j	L	=	location	of	training	for	physician	i		R	=	practice	rurality	for	physician	i	in	year	j	P=	percent	APP	for	physician	i	in	year	j																																																									22	There	were	only	two	values	of	one	for	this	variable,	representing	one	male	physician	and	one	female	physician.		23	The	models	would	not	converge	in	SAS.	Lack	of	convergence	of	the	numerical	algorithms	for	maximum	likelihood	estimation	in	logistic	models	may	be	caused	by	a	variety	of	issues	including	limits	to	computational	power,	or	issues	estimating	the	maximum	likelihood	function	when	the	maximum	likelihood	estimate	does	not	exist	(Allison,	2003).			 130		Non-Clinical	Incentive	Payments:	I	used	the	same	modeling	approach	to	examine	non-clinical	incentive	payments	as	for	clinical	incentives:	I	modeled	a	dichotomous	dependent	variable	using	a	longitudinal	log	linear	model	under	a	binary	distribution	and	logit	link,	and	the	logit	of	the	proportional	dependent	variable	using	a	linear	mixed	effects	model	under	a	normal	distribution	with	an	identity	link	function.	The	list	of	covariates,	and	random	effects	structure	is	also	the	same24.	For	the	binary	model:		DEF(;5I) = JK +	JMNI +	J9O5 +	JP NI ∗ O5 +	JRS5I + JTU5 + JVW5I + JX U5 ∗ W5I+ JYZ5I + \5I 	 	\5I~> 0, '9 	Where:		y	=	Y/N	for	billing	non-clinical	incentive	payments	for	physician	i	in	year	j	T=	time	at	year	j	G	=	gender	for	physician	i		A	=	age	group	for	physician	i	in	year	j	L	=	location	of	training	for	physician	i		R	=	practice	rurality	for	physician	i	in	year	j	P=	percent	APP	for	physician	i	in	year	j		For	the	logit	normal	model:	DEFGH(;5I) = JK +	JMNI +	J9O5 +	JP NI ∗ O5 +	JRS5I + JTU5 + JVW5I + JX U5 ∗ W5I+ JYZ5I + \5I 	 		\5I~> 0, '9 	Where:																																																										24	As	with	the	models	for	clinical	incentives,	I	was	unable	to	include	more	than	a	single	random	effect	(for	subject-level	residuals);	however,	I	did	test	whether	swapping	in	the	random	effect	for	subject	slope	or	intercept	affected	my	results.	It	did	not	in	either	case.		 131	y	=	proportion,	non-clinical	incentive	payments	for	physician	i	in	year	j	T=	time	at	year	j	G	=	gender	for	physician	i		A	=	age	group	for	physician	i	in	year	j	L	=	location	of	training	for	physician	i		R	=	practice	rurality	for	physician	i	in	year	j	P=	percent	APP	for	physician	i	in	year	j		Both	models	include	a	random	effect	for	individual-level	residuals	with	a	first	order	auto-regressive	correlation	matrix.	Consistent	with	the	other	logit-proportion	models,	I	rescaled	the	values	of	one	(N=	120	person	years)	down	to	0.99999	in	the	proportion	model.	Question	2.3:	To	further	investigate	the	extent	to	which	the	proportions	of	physicians’	income	from	payments	for	direct	clinical	care,	clinical	incentives	and	non-clinical	incentives	changed	year-on-year	over	the	course	of	the	study	period,	and	whether	any	such	changes	were	differential	by	gender,	I	used	fixed	effects	generalized	linear	models	(GLMs)	for	each	study	year,	under	a	normal	distribution	and	identity	link	function:		;5 = JK +	J9O5 +	JRS5I + JTU5 + JVW5I + JX U5 ∗ W5I + JYZ5I + \5 		Where:		y	=	proportion	of	income	related	to	direct	clinical	care,	clinical	incentives	or	non-clinical	incentives	for	physician	i	in	a	particular	study	year		G	=	gender	for	physician	i	A	=	age	group	for	physician	i	in	year	j	L	=	location	of	training	for	physician	i	R	=	practice	rurality	for	physician	i	P=	percent	APP	for	physician	i	Because	I	did	not	transform	the	dependent	variable	for	the	GLMs,	no	rescaling	of	values	was	necessary.		I	plotted	the	least	squares	mean	percentage		 132	associated	with	the	independent	effect	of	gender,	to	track	the	change	in	the	trend	over	time,	year	on	year.		This	approach	supplements	the	longitudinal	analyses	presented	as	part	of	Questions	2.1	and	2.2	in	two	ways:	1)	the	use	of	a	linear,	non-transformed	approach	allows	me	to	include	the	zero	and	one	values	(without	rescaling),	as	well	as	those	in	between,	modeling	all	physicians	in	the	cohort	within	the	same	model;	and	2)	use	of	a	series	of	cross	sectional	models	allows	for	a	closer	examination	of	change	over	time,	allowing	for	the	change	of	slope	year	by	year,	while	still	isolating	the	gender	effect	from	the	impact	of	other	model	covariates.25			 Question	2.4:	On-Call	Payments:	I	modeled	on-call	program	uptake	as	dichotomous	(billed	for	time	on-call	or	did	not,	for	each	study	year)	using	a	longitudinal	logistic	log	linear	model	under	a	binary	(logistic)	distribution	and	a	logit	link	function,	with	physician	gender,	demographic	characteristics,	and	percentage	APP	as	covariates:		DEF(;5I) = JK +	JMNI +	J9O5 +	JP NI ∗ O5 +	JRS5I + JTU5 + JVW5I + JX U5 ∗ W5I+ JYZ5I + \5I 	 	\5I~> 0, '9 	Where:		y	=	Y/N	for	billing	time	on-call	for	physician	i	in	year	j	T	=	time	at	year	j	G	=	gender	for	physician	i		A	=	age	group	for	physician	i	in	year	j	L	=	location	of	training	for	physician	i																																																										25	A	limitation	of	this	approach	is	that	because	the	GLMs	are	cross	sectional,	I	cannot	assess	the	statistical	significance	of	any	yearly	changes;	however,	combined	with	longitutinal	model,	I	can	determine	the	statistical	significance	of	any	change	in	the	observed	gender	difference	over	the	course	of	the	study	period,	but	can	also	visualize	whether	the	rate	of	change	is	likely	constant	year	on	year.		 133	R	=	practice	rurality	for	physician	i	in	year	j	P=	percent	APP	for	physician	i	in	year	j		As	with	the	other	logistic	modeling	exercises,	I	used	a	single	random	effect	for	subject-level	residuals	under	a	first	order	auto-regressive	correlation	matrix,	and	tested	random	effects	for	subject	intercept	and	slope.	Rural	and	Remote	Incentive	Programs:	I	modeled	uptake	of	rural	and	remote	incentive	programs	using	the	same	approach,	using	a	binary	dependent	variable	and	log	linear	mixed	effects	model	under	a	binary	distribution	and	logit	link	function,	and	a	single	random	effect	for	subject-level	residuals:	DEF(;5I) = JK +	JMNI +	J9O5 +	JP NI ∗ O5 +	JRS5I + JTU5 + JVW5I + JX U5 ∗ W5I+ JYZ5I + \5I 	 	\5I~> 0, '9 	Where:		y	=	Y/N	for	billing	rural/remote	incentives	for	physician	i	in	year	j	T	=	time	at	year	j	G	=	gender	for	physician	i		A	=	age	group	for	physician	i	in	year	j	L	=	location	of	training	for	physician	i		R	=	practice	rurality	for	physician	i	in	year	j	P	=	percent	APP	for	physician	i	in	year	j		 The	models	in	this	chapter	all	contain	interaction	terms,	as	well	as	proportion	explanatory	covariates.	For	a	discussion	on	how	these	terms	should	be	interpreted,	please	see	Chapter	4	section	4.2.4.	For	details	on	my	model-building	strategy,	please	refer	back	to	Chapter	4,	section	4.2.3.2	(Verbeke	&	Molenberghs,	2000).				 134	5.3	Results	The	cohort	for	this	analysis	includes	N=6579	physicians:	2469	females	(38%)	and	4110	males	(62%).	Demographics	as	well	as	overall	measures	of	activity	are	described	in	section	4.3.1.	Since	all	measures	here	are	constructed	solely	from	payment	data	(rather	than	from	contacts	or	service	counts),	physicians	whose	income	was	generated	entirely	from	APP	sources	were	not	excluded	from	this	analysis.		5.3.1	Descriptive	statistics	and	bivariate	results		Physicians	earned	85.86%	(approximately	$172,00026)	of	their	total	remuneration	through	direct	clinical	care	delivery,	6.58%	($13,000)	through	clinical	incentives	and	6.98%	($14,002)	through	non-clinical	incentives	and	0.58%	through	other	sources	(e.g.	continuing	medical	education	stipend,	administrative	fees),	averaged	over	the	entire	study	period.	Female	physicians	were	significantly	more	likely	to	bill	for	clinical	incentives	(75.98%	vs.	73.77%,	p=0.041)	and	significantly	less	likely	to	bill	for	non-clinical	incentives	(94.05%	vs.	95.5%,	p=0.0092)	(Table	5.1).		Table	5.1:	Proportion	of	physicians	who	billed	or	any	clinical	or	non-clinical	incentives,	by	gender,	2005-2012			Males	 Females	 Total	N=4110	(62%)	 N=2469	(38%)	 N=	6579	Clinical	Incentives	(%)1	 3032	(73.77%)	 1876	(75.98%)	 4098	(74.60%)	Non-clinical	incentives	(%)2	 3925	(95.50%)	 2322	(94.05%)	 6247	(94.95%)	1chi-square	=3.97,	p=0.0461		2chi-square	=6.79,	p=0.0092 																																																									26	In	2010	constant	dollars.		 135	Payments	for	clinical	care	delivery	accounted	for	a	significantly	larger	proportion	of	female	(compared	to	male)	physicians’	total	payments	(87.8%	vs.	85.5%,	Z=	7.56,	p<0.0001)	(Table	5.2).	Each	of	the	other	categories	of	payment	[clinical	incentives	(7.0%	for	males,	5.4%	for	females,	Z=	-7.56,	p<0.0001),	non-clinical	incentives	(7.0%	for	males	and	6.8%	for	females,	Z=	-8.38,	p<0.0001)	and	other	activities	(1.0%	for	males	and	0.7%	for	females,	Z=-12.75,	p<0.0001)]	accounted	for	a	larger	proportion	of	male	physicians’	incomes.	Male	physicians	also	billed	more,	in	2010	constant	dollars,	across	all	three	non-clinical-service	categories.		Female	physicians	were	less	likely	to	bill	for	time	on	call	(chi-square	26.59,	p<0.0001)	or	for	rural	and	remove	incentives	(chi-square	5.61,	p=0.0179)	(Table	5.3).	Table	5.2:	Payment	types	as	a	percentage	($)	of	total	compensation,	by	gender,	averaged	across	2005-2012			Males	 Females	 Total	N=4110	(62%)	 N=2469	(38%)	 N=	6579	Clinical	billings1	 85.31%	($198,016)		 87.30%	($129,588)		 85.86%	($172,336)	Clinical	incentives2	 7.02%	($16,304)		 5.43%	($8,054)		 6.58%	($13,208)		Non-clinical	incentives3	 7.03%	($16,315)	 6.84%	($10,150)		 6.98%	($14,002)		Other	payments4	 0.64%	($1,486)		 0.43%	($641)	 0.58%	($1,168)	Total	billings		 100%	($232,122)	 100%	($148,434)	 100%	($200,715)	1Wilcoxon-Mann-Whitney,	Z=7.56,	p<0.0001	 3Wilcoxon-Mann-Whitney,	Z=-12.75,	p<0.0001			2Wilcoxon-Mann-Whitney,	Z=-8.38,	p<0.0001	 	 		Table	5.3:	Number	(%)	of	male	and	female	physicians	who	billed	for	time	on	call	or	rural	and	remote	incentives,	2005-2012		 Males	 Females	 Total		 N=4110	(62%)	 N=2469	(38%)	 N=	6579	On-call	payments1	 1894	(46.08%)	 977	(39.57%)	 2871	(43.68%)	Rural	&	remote2	 1231	(29.95%)	 672	(27.22%)	 1903	(28.93%)	1chi-square	=26.59	p<0.0001	 2chi-square	=5.61	p=0.0179		 136	Uptake	of	clinical	and	non-clinical	incentive	payments	also	varied	across	several	specific	demographic	groupings.	Clinical	incentives	accounted	for	between	6.0%	and	11.0%,	and	non-clinical	incentives	accounted	for	between	2.8%	and	19.9%,	of	total	income,	across	different	physician	age	groups,	graduation	cohorts,	and	health	authorities	(Table	5.4).				 137	Table	5.4:		Primary	care	physician	payments,	by	demographic	characteristics	(for	2011/12)	27			 N	 Total	Compensation	 Clinical	Billings	%	($)	 Clinical	Incentives	%	($)	 Non-Clinical	Incentives	%	($)	 Other	Payments	%	($)	Total	active	physicians	 5455	 $220,041		 83.7%	($184,156)	 9.5%	($21,002)	 6.7%	($14,800)	 0.0%	($84)	Age	group	 	<35	 567	 $189,903	 83.8%	($159,168)	 7.0%	($13,295)	 9.1%	($17,248)	 0.0%	($191)	35-<45	 1159	 $236,868	 83.3%	($197,295)	 9.5%	($22,451)	 7.2%	($17,025)		 0.0%	($96)		45-<55	 1705	 $245,299	 83.3%	($204,373)	 10.7%	($26,359)		 5.9%	($14,519)		 0.0%	($46)		55-<65	 1389	 $204,734	 84.3%	($172,580)		 10.4%	($21,323)		 5.3%	($10,805)		 0.0%	($26)		65+	 635	 $127,643	 89.4%	($114,148)		 6.0%	($7,705)		 4.5%	($5,790)		 0.0%	($0)		Training1	Within	Canada	 3790	 $202,360	 84.4%	($170,742)		 9.3%	($18,732)		 6.3%	($12,847)		 0.0%	($37)		International	 1551	 $262,249	 82.3%	($215,852)		 10.2%	($26,690)		 7.4%	($19,504)		 0.1%	($203)		Health	Authority2	Interior	Health	 974	 $224,397	 78.8%	($176,853)		 10.9%	($24,537)		 10.2%	($22,902)		 0.0%	($104)		Fraser	Health	 1356	 $266,383	 88.0%	($234,430)		 9.2%	($24,515)		 2.8%	($7,438)		 0.0%	($0)		Vancouver	Coastal	 1604	 $189,247	 86.5%	($163,760)		 9.0%	($17,071)		 4.4%	($8,412)		 0.0%	($3)		Vancouver	Island	 1126	 $190,456	 82.0%	($156,203)		 11.0%	($20,996)		 7.0%	($13,252)		 0.0%	($6)		Northern	Health	 333	 $291,545	 73.5%	($214,201)		 6.3%	($18,259)		 19.9%	($58,041)		 0.4%	($1,045)		Practice	rurality2	Metropolitan	 3154	 $216,935	 87.4%	($189,637)		 9.6%	($20,728)		 3.0%	($6,568)		 0.0%	($2)		Urban	dominated	 1373	 $223,573	 81.4%	($181,978)		 10.4%	($23,205)		 8.2%	($18,236)		 0.1%	($154)		Rural	dominated	 866	 $235,210	 74.7%	($175,586)		 8.4%	($19,641)		 16.7%	($39,301)		 0.1%	($289)																																																										27	N=5455	physicians	who	had	at	least	one	clinical	billing	in	2011/12.	1124	physicians	did	not	bill	during	that	fiscal	year	and	are	not	included	in	this	table,	or	in	Table	5.5		 138			 N	 Total	Compensation	 Clinical	Billings	%	($)	 Clinical	Incentives	%	($)	 Non-Clinical	Incentives	%	($)	 Other	Payments	%	($)	Graduation	year	<1970	 335	 $150,703	 85.7%	($129,097)		 9.1%	($13,710)		 5.2%	($7,896)		 0.0%	($0)		1970-<1980	 1122	 $224,018	 83.9%	($187,849)		 11.0%	($24,570)		 5.2%	($11,583)		 0.0%	($16)		1980-<1990	 1511	 $246,324	 83.0%	($204,480)		 10.9%	($26,938)		 6.0%	($14,853)		 0.0%	($52)		1990-<2000	 1405	 $235,009	 83.9%	($197,185)		 8.7%	($20,532)		 7.3%	($17,232)		 0.0%	($60)		2000+	 1082	 $181,247	 83.9%	($152,071)		 6.6%	($11,877)		 9.4%	($17,041)		 0.1%	($258)		1.	N=114	whose	training	location	is	unknown	2.	N=62	whose	health	authority	and	rurality	are	unknown			 139	Physicians	practicing	in	rural-dominated	areas	not	surprisingly	earned	much	more	from	non-clinical	incentive	payments	than	physicians	practicing	in	urban	or	metropolitan	areas	(16.7%	of	income	from	non-clinical	incentive	payments,	compared	with	only	3.0%	in	metropolitan	areas).		This	difference	is	driven	primarily	by	the	uptake	of	incentive	payments	through	the	Rural	Practice	Subsidiary	Agreement	for	rural	and	remote	physicians	(Joint	Standing	Committee	on	Rural	Issues,	2012)	(Table	5.5).		A	larger	proportion	of	rural	physicians’	income	also	came	from	on-call	payments	(6.0%	of	total	payments	vs.	0.9%	for	physicians	in	metropolitan	areas).	Physicians	in	the	younger	age	categories	received	a	larger	proportion	of	their	income	from	non-clinical	incentives,	and	a	smaller	proportion	from	clinical	incentives	compared	with	physicians	in	older	age	categories	(with	the	exception	of	those	above	age	65).		 		 140	Table	5.5:	Physician	payments,	by	demographic	characteristics	(for	2011/12)			 N	 On-Call	Payments		%	($)	 Rural	and	Remote	Incentives	%	($)	Total	active	physicians	 5455	 2.2%	($4,932)	 1.6%	($3,535)	Age	group	 	  <35	 567	 3.0%	($5,633)	 2.6%	($5,073)	35-<45	 1159	 2.4%	($5,616)		 2.0%	($4,678)		45-<55	 1705	 2.0%	($4,816)		 1.5%	($3,715)		55-<65	 1389	 1.6%	($3,271)		 1.4%	($2,936)		65+	 635	 0.9%	($1,202)		 1.6%	($1,981)		Training1	 		 		Within	Canada	 3790	 2.1%	($4,220)		 1.7%	($3,512)		International	 1551	 2.5%	($6,449)		 2.1%	($5,404)		Health	Authority2	 	  Interior	Health	 974	 3.5%	($7,876)		 3.2%	($7,179)		Fraser	Health	 1356	 0.6%	($1,569)		 0.1%	($214)		Vancouver	Coastal	Health	 1604	 1.8%	($3,455)		 0.7%	($1,246)		Vancouver	Island	Health	 1126	 2.2%	($4,193)		 1.8%	($3,404)		Northern	Health	 333	 6.6%	($19,314)		 9.1%	($26,453)		Practice	rurality2	 	  Metropolitan	 3154	 0.9%	($1,886)		 0.1%	($296)		Urban	dominated	 1373	 2.7%	($5,994)		 2.4%	($5,375)		Rural	dominated	 866	 6.0%	($14,221)		 6.7%	($15,716)		Graduation	year	 	  <1970	 335	 1.5%	($2,292)		 1.5%	($2,286)		1970-<1980	 1122	 1.6%	($3,540)		 1.3%	($2,974)		1980-<1990	 1511	 2.1%	($5,116)		 1.5%	($3,682)		1990-<2000	 1405	 2.4%	($5,701)		 2.1%	($4,842)		2000+	 1082	 3.2%	($5,711)		 2.9%	($5,261)		1.	N=114	whose	training	location	is	unknown	2.	N=62	whose	health	authority	and	rurality	are	unknown	5.3.2	Question	2.1:	Multivariate	results		Do	clinical	payments	make	up	a	significantly	larger	proportion	of	total	income	for	male	or	female	physicians?			 In	2005-06,	after	adjusting	for	the	impact	of	demographic	characteristics,	and	percentage	of	payments	from	APP	sources,	I	found	no	difference	in	the		 141	proportion	of	income	related	to	clinical	care	delivery	for	male	compared	to	female	physicians	(Table	5.6);	however,	by	2011-12,	female	physicians	earned	a	significantly	higher	proportion	of	their	income	from	direct	clinical	care	delivery	(combined	OR	for	gender	and	gender	by	year	interaction:	1.73,	95%	CI:	1.52-1.93).	The	proportion	of	income	from	direct	clinical	care	delivery	declined	significantly	over	the	study	period	(OR:	0.91,	95%	CI:	0.89-0.92),	widening	the	gender	gap	over	time,	with	this	source	of	income	becoming	relatively	more	important	for	female	PCPs	than	for	male	over	the	course	of	the	study	period.		Table	5.6:	Multivariate	logit-normal	results:	Percent	clinical	income	Variables		Model	1:	Percent	Clinical	Income	Odds	Ratio	 95%	Confidence	Interval	Sex	(female)	 1.10	 0.95-1.27	Year	(continuous)	 0.91	 0.89-0.92*	Sex*year	interaction	(female)	 1.07	 1.04-1.10*	Age:	35-<45	 0.87	 0.78-0.97‡	Age:	45-<55	 0.81	 0.71-0.91†	Age:	55-<65	 0.93	 0.81-1.07	Age:	65+	 1.29	 1.08-1.54‡	Rurality:	urban-dominated	 0.49	 0.43-0.57	Rurality:	rural-dominated	 0.21	 0.17-0.24*	International	training	 0.89	 0.76-1.04		Rurality*training	interaction	(urban*international)	 0.73	 0.57-0.94‡	Rurality*training	interaction	(rural*international)	 0.76	 0.57-1.02		Percentage	APP	 1.08	 0.97-1.22		*p<0.0001	†	p<0.001		‡p<0.05	 	  	Physicians	working	in	rural-dominated	areas	had	a	significantly	lower	proportion	of	their	income	from	direct	clinical	care	(OR:	0.21,	95%	CI:	0.17-0.24),	as	did	physicians	in	the	35-<45-	and	45-<55-year	age	groups	(OR:	35-<45:	0.87,	95%		 142	CI:	0.78-0.97;	OR:	45-<55:	0.81,	95%	CI:	0.71-0.91).	The	interaction	between	rural	location	and	international	training	was	significant,	suggesting	that	internationally	trained	physicians	in	rural	areas	earned	a	significantly	lower	proportion	of	their	income	from	direct	clinical	care	compared	to	metropolitan	physicians,	but	a	greater	proportion	compared	to	Canadian-trained,	rurally-located	physicians	(OR	for	rural	location,	international	training,	and	rural	location	by	international	training	interaction:	0.32,	95%	CI:	0.14-0.50).	5.3.3	Question	2.2:	Multivariate	results	Are	male	or	female	physicians	more	likely	to	bill	for	clinical	and	non-clinical	incentive	payments?		 In	2005-06,	I	observed	no	difference	in	the	odds	of	billing	any	clinical	incentives	by	gender;	however,	by	2011-12,	female	physicians	showed	a	higher	likelihood	of	billing	any	clinical	incentives	compared	to	male	physicians,	when	demographic	variables	and	percentage	APP	were	held	constant	(combined	OR	for	gender	and	gender	by	year	interaction:	1.49,	95%	CI:	1.28-1.69)	(Table	5.7).	In	contrast,	female	physicians	were	consistently	less	likely	to	bill	for	any	non-clinical	incentives	(OR:	0.73,	95%	CI:	0.54-1.00).			 		 143	Table	5.7:	Multivariate	binary	results:	clinical	and	non-clinical	incentives		 Model	1:	Clinical	Incentives	 Model	2:	Non-Clinical	Incentives	Variables		 Odds	Ratio	 95%	Confidence	Interval	 Odds	Ratio	 95%	Confidence	Interval	Sex	(female)	 0.77	 0.63-0.96‡	 0.73	 0.54-1.00	Year	(continuous)	 1.26	 1.23-1.28*	 0.95	 0.91-0.98‡	Sex*year	interaction	(female)	 1.10	 1.06-1.13*	 1.00	 0.94-1.05	Age:	35-<45	 1.48	 1.27-1.73*	 0.99	 0.80-1.23	Age:	45-<55	 1.94	 1.62-2.32*	 1.02	 0.80-1.29	Age:	55-<65	 1.56	 1.27-1.91*	 0.72	 0.57-0.93‡	Age:	65+	 0.46	 0.36-0.59*	 0.35	 0.26-0.45*	Rurality:	urban-dominated	 1.83	 1.51-2.22*	 2.08	 1.66-2.61*	Rurality:	rural-dominated	 2.89	 2.29-3.66*	 4.26	 2.91-6.25*	International	training	 1.38	 1.10-1.74‡	 1.52	 1.21-1.91†	Rurality*training	interaction	(urban*international)	 1.33	 0.93-1.89		 1.00	 0.61-1.63		Rurality*training	interaction	(rural*international)	 1.02	 0.68-1.55		 0.83	 0.40-1.71		Percentage	APP	 0.02	 0.01-0.02*	 0.30	 0.25-0.36*	*p<0.0001	†	p<0.001		‡p<0.05	 	 	 	 	Physicians	whose	practices	were	in	rural-dominated	locations	had	almost	three	times	the	odds	of	billing	clinical	incentives	(OR:	2.89,	95%CI	2.29-3.66),	and	more	than	four	times	the	odds	of	billing	non-clinical	incentives	compared	to	physicians	whose	practices	were	in	metropolitan	areas	(OR:	4.26,	95%CI	2.91-6.25).	Additionally,	physicians	who	trained	outside	Canada	were	more	likely	to	bill	for	both	clinical	and	non-clinical	incentives	compared	to	their	Canadian-trained	counterparts	(OR	for	clinical	incentives:	1.38,	95%	CI:	1.10-1.74;	OR	for	non-clinical	incentives:	2.08,	95%CI	1.66-2.61).		Physicians	aged	35-<45,	45-<55,	and	55-<65	were	all	significantly	more	likely	to	bill	for	clinical	incentives	compared	to	those	age	<35	(OR	35-<45:	1.48,	95%	CI:	1.27-1.73;	OR	45-<55:	1.94,	95%	CI:	1.62-2.32;	and	OR	55-<65:	1.56,	95%		 144	CI:	1.27-1.91).	Physicians	over	age	65,	in	contrast	were	the	least	likely	to	bill	for	the	incentives	(OR:	0.46	95%	CI:	0.36-0.59).		The	odds	of	billing	for	clinical	incentives	increased	significantly	over	the	study	period		(OR:	1.26,	95%	CI:	1.23-1.28).	The	increase	over	the	course	of	the	study	period	is	likely	to	be	a	function	of	availability,	since	the	payments	were	first	introduced	in	2003,	and	then	ramped	up	significantly	in	2007.	Adjusted	for	age,	and	other	physician	characteristics,	physicians	in	2011-12	were	more	than	four	times	as	likely	as	those	in	2005-06	to	bill	for	any	clinical	incentives.			At	the	same	time,	the	odds	of	billing	for	non-clinical	incentives	declined	(OR:	0.95,	95%	CI:	0.91-0.98).	Unlike	the	pattern	observed	for	clinical	incentives,	physicians	aged	55-<65	and	65+	were	significantly	less	likely	to	bill	for	non-clinical	incentives	compared	to	physicians	under	age	45	(OR	for	55-<65:	0.72,	95%	CI:	0.57-0.93;	OR	65+:	0.35,	95%	CI:	0.26-0.45).			 In	2005-06,	I	observed	no	gender	difference	in	the	proportion	of	income	generated	through	clinical	incentive	payments	(Table	5.8).	By	2011-12,	however,	female	physicians	earned	a	significantly	smaller	proportion	of	their	income	from	clinical	incentives	compared	with	male	physicians,	all	other	variables	held	constant		(combined	OR	for	gender	and	gender	by	year	interaction	term:	0.81,	95%	CI:	0.66-0.95).	Thus,	by	2011-12,	female	physicians	were	more	likely	to	bill	for	at	least	one	clinical	incentive;	however,	the	incentives	accounted	for	a	smaller	proportion	of	the	income	of	female	physicians	among	all	those	who	billed	for	any.				 145	Table	5.8:	Multivariate	logit-normal	results:	Clinical	and	non-clinical	incentives	Variables		Model	1:	Percent	Clinical	Incentives	 Model	2:	Percent	Non-Clinical	Incentives	Odds	Ratio	 95%	Confidence	Interval	 Odds	Ratio	 95%	Confidence	Interval	Sex	(female)	 1.15	 1.03-1.28‡	 0.98	 0.90-1.07	Year	(continuous)	 1.31	 1.29-1.33*	 1.01	 1.00-1.02	Sex*year	interaction	(female)	 0.95	 0.93-0.97*	 0.98	 0.96-1.00‡	Age:	35-<45	 1.29	 1.19-1.40*	 0.88	 0.82-0.94†	Age:	45-<55	 1.52	 1.39-1.66*	 0.84	 0.78-0.91*	Age:	55-<65	 1.47	 1.33-1.62*	 0.77	 0.71-0.84*	Age:	65+	 1.01	 0.89-1.15	 0.80	 0.71-0.89*	Rurality:	urban-dominated	 0.97	 0.88-1.06	 1.92	 1.76-2.08*	Rurality:	rural-dominated	 0.95	 0.85-1.06	 4.73	 4.27-5.24*	International	training	 0.88	 0.80-0.98‡	 1.07	 0.97-1.18		Rurality*training	interaction	(urban*international)	 1.17	 0.99-1.38		 1.28	 1.10-1.49‡	Rurality*training	interaction	(rural*international)	 1.13	 0.94-1.36		 1.16	 0.98-1.39		Percentage	APP	 0.09	 0.08-0.10*	 2.52	 2.33-2.73*	*p<0.0001	†	p<0.001		‡p<0.05	 	    	The	proportion	of	income	generated	through	clinical	incentives	increased	over	time	(OR:	1.31,	95%CI	1.29-1.33).	Physicians	in	the	middle	age	groups	(35-<65)	received	a	larger	proportion	of	their	income	from	clinical	incentives	compared	to	those	in	the	youngest	age	group.	I	found	no	difference	based	on	practice	location;	however,	physicians	who	trained	internationally	received	a	smaller	proportion	of	their	income	from	clinical	incentive	payments.	In	2005-06,	I	found	no	gender	difference	in	the	proportion	of	income	from	non-clinical	incentives	among	those	who	did	bill	for	at	least	one.	By	2011-12,	however,	female	physicians	received	a	significantly	smaller	proportion	of	their	incomes	from	non-clinical	incentives	(combined	OR	for	gender	and	gender	by	year		 146	interaction:	0.86,	95%	CI:	0.80-0.96).		Physicians	in	older	age	groups	(35-<45,	45-<55,	55-<65,	65+)	received	a	significantly	lower	proportion	of	their	income	from	non-clinical	incentives	compared	to	physicians	under	age	35.	Not	surprisingly,	physicians	located	in	rural-dominated	areas	received	a	larger	proportion	of	their	income	from	non-clinical	incentive	payments,	compared	to	those	located	in	a	metropolitan	area	(OR:	4.73,	95%	CI:	4.27-5.24).	I	found	no	difference	in	the	proportion	of	income	from	non-clinical	incentives	based	on	training	location;	however,	the	interaction	between	training	location	and	urban	practice	location	was	significant.	Internationally-trained	physicians	working	in	urban-dominated	areas	received	a	significantly	higher	proportion	of	their	income	in	the	form	of	non-clinical	incentives,	compared	with	Canadian-trained	physicians	in	the	same	practice	areas	(combined	OR	for	urban-dominated	location,	international	training,	and	the	interaction	term	between	them:	2.63,	95%	CI:	1.75-3.52).	5.3.4	Question	2.3:	Multivariate	results		How	has	the	makeup	of	physicians’	income	changed	with	time	and	are	changes	differential	by	gender?		 The	make-up	of	physicians’	income	changed	significantly	over	the	course	of	the	study	period,	and	changed	differently	for	male	versus	female	physicians	(Figures	5.1,	5.2,	and	5.3).	There	was	a	substantial	decrease	in	the	proportion	of	income	from	direct	clinical	care	delivery	for	both	males	and	females	(Figure	5.1).			 147		Figure	5.1:	Annual	fixed	effects	GLMs	for	percent	clinical	care	delivery28		 There	was	no	apparent	difference	between	males	and	females	in	either	2005-06	or	2006-07;	however,	the	gender	gap	widened	dramatically	in	2007/08	and	peaked	in	2008/09.		The	change	in	the	proportion	of	income	from	direct	clinical	care	delivery	equates	to	a	drop	of	approximately	7%	for	male	physicians	(from	87.9%	of	income	in	2005-06	to	81.08%	of	income	in	2011-12)	and	5%	for	female	physicians	(from	88.30%	to	83.57%).																																																									28	A	note	on	interpretation:	This	graph	represents	the	least	squares	mean	percent	for	gender	generated	by	seven	cross-sectional	fixed	effects	GLMs.	The	least	squares	mean	percent	is	the	expected	percentage	income	accounted	for	by	direct	clinical	care	for	male	and	female	physicians	in	each	study	year,	calculated	by	using	weighted	averages	of	the	effect	estimates	for	all	categorical	study	covariates,	an	mean	values	for	continuous	covariates.	Figures	5.2	and	5.3	can	be	interpreted	in	the	same	fashion,	with	outcomes	for	percent	clinical	incentives	and	percent	non-clinical	incentives.		0.00%0.50%1.00%1.50%2.00%2.50%3.00%3.50%4.00%4.50%76.00%78.00%80.00%82.00%84.00%86.00%88.00%90.00%2005-06 2006-07 2007-08 2008-09 2009-10 2010-11 2011-12Percent	DifferenceLeast	Squares	Mean	Percent	(of	Total	Payments)Fiscal	YearMen Women Percent	Difference	 148	Over	the	same	period,	the	proportion	of	income	from	clinical	incentive	payments	increased	significantly	(Figure	5.2).	As	with	payments	for	direct	clinical	care	delivery,	there	was	no	gender	difference	in	either	2005-06	or	2006-07,	but	a	substantial	gender	gap	appeared	in	2007-08	(the	year	the	roll-out	of	the	General	Practice	Services	Committee’s	substantial	chronic	disease	management	incentive	was	complete)	and	was	sustained	for	the	remaining	study	years.	Clinical	incentive	payments	accounted	for	an	estimated	0.77%	of	income	for	males	in	2005-06	and	increased	to	7.85%	by	2011-12.	They	accounted	for	0.98%	and	6.94%	of	income,	respectively,	for	the	same	two	years	for	female	physicians.			Figure	5.2:	Annual	fixed	effects	GLMs	for	percent	clinical	incentives	0.00%1.00%2.00%3.00%4.00%5.00%6.00%7.00%8.00%0.00%1.00%2.00%3.00%4.00%5.00%6.00%7.00%8.00%9.00%2005-06 2006-07 2007-08 2008-09 2009-10 2010-11 2011-12Percent	DifferenceLeast	Squares	Mean	Percent	(of	Total	Payments)Fiscal	YearMen Women Percent	Difference	 149	Results	from	the	binary	and	logit-normal	models	presented	in	section	5.3.3	suggest	that	the	observed	increase	in	proportion	is	due	to	a	combined	increase	in	the	odds	of	physicians	billing	for	any	clinical	incentives,	and	an	increase	in	the	proportion	of	income	accounted	for	by	these	incentives	among	those	physicians	who	received	them.		The	percentage	of	income	from	non-clinical	incentive	payments	did	not	change	as	dramatically	over	the	course	of	the	study	period,	increasing	from	9.12%	to	9.82%	for	women	and	from	9.42%	to	10.48%	for	men	(Figure	5.3).	This	is	consistent	with	the	fact	that	there	were	no	major	changes	in	the	types	or	values	of	incentives	offered	in	this	category	over	the	study	period.		Figure	5.3:	Annual	fixed	effects	GLMs	for	percent	non-clinical	incentives	0.00%0.50%1.00%1.50%2.00%2.50%3.00%3.50%4.00%4.50%0.00%2.00%4.00%6.00%8.00%10.00%12.00%14.00%2005-06 2006-07 2007-08 2008-09 2009-10 2010-11 2011-12Percent	DifferenceLeast	Squares	Mean	Percent	(of	Total	Payments)Fiscal	YearMen Women Percent	Difference	 150	5.3.5	Question	2.4:	Multivariate	results		Does	uptake	of	on-call	and	rural	and	remote	incentive	payments	vary	for	male	versus	female	PCPs,	controlling	for	age,	period,	and	practice	location?		 In	2005-06,	female	physicians	had	significantly	lower	odds	of	billing	for	time	on-call	compared	to	male	physicians	(combined	OR	for	gender	and	gender	by	year	interaction:	0.58,	95%	CI:	0.48-0.68).	By	2011-12,	the	gender	difference	decreased	but	was	still	significant	(combined	OR	0.72,	95%	CI:	0.62-0.81)	(Table	5.9).	The	odds	of	billing	for	time	on-call	decreased	over	the	study	period	(OR:	0.94,	95%CI:	0.92-0.96)	for	both	male	and	female	physicians.	Physicians	in	successively	older	age	groups	were	increasingly	less	likely	to	bill	for	time	on	call	compared	to	their	younger	counterparts	(for	e.g.	OR:	age	65+:	0.31,	95%	CI:	0.27-0.36).	Physicians	in	urban-	or	rural-dominated	locations	had	much	higher	odds	of	billing	for	time	on	call	compared	to	their	metropolitan	counterparts	(OR:	urban-dominated:	2.22,	95%	CI:	2.00-2.47;	OR:	rural-dominated:	4.95,	95%	CI:	4.36-5.62).			 		 151	Table	5.9:	Multivariate	binary	results:	on-call	payments	and	rural	and	remote	incentives	 Model	1:	On-Call	Payments	 Model	2:	Rural	and	Remote	Incentives	Variables		 Odds	Ratio	 95%	Confidence	Interval	 Odds	Ratio	 95%	Confidence	Interval	Sex	(female)	 0.56	 0.49-0.65*	 0.80	 0.68-0.94‡	Year	(continuous)	 0.95	 0.93-0.96*	 1.00	 0.98-1.02	Sex*year	interaction	(female)	 1.03	 1.01-1.06‡	 1.00	 0.97-1.04	Age:	35-<45	 0.81	 0.74-0.89*	 0.78	 0.69-0.87*	Age:	45-<55	 0.74	 0.67-0.81*	 0.83	 0.74-0.93‡	Age:	55-<65	 0.52	 0.47-0.59*	 0.71	 0.62-0.80*	Age:	65+	 0.31	 0.27-0.36*	 0.40	 0.34-0.47*	Rurality:	urban-dominated	 2.22	 2.00-2.47*	 19.52	 17.41-21.88*	Rurality:	rural-dominated	 4.95	 4.36-5.62*	 129.49	 114.13-146.92*	International	training	 1.07	 0.93-1.22		 2.01	 1.71-2.37*	Rurality*training	interaction	(urban*international)	 1.05	 0.86-1.27		 1.05	 0.87-1.27		Rurality*training	interaction	(rural*international)	 1.20	 0.96-1.50		 0.74	 0.60-0.92‡	Percentage	APP	 1.89	 1.73-2.06*	 6.51	 5.81-7.29*	*p<0.0001	†	p<0.001		‡p<0.05	 	    	 Female	physicians	had	0.80	times	lower	odds	of	billing	for	rural	and	remote	incentives	(OR:	0.80,	95%CI:	0.68-0.94),	even	after	adjusting	for	the	impact	of	the	rurality	of	the	physician’s	practice	location.	The	odds	of	billing	for	a	rural	and	remote	incentive	did	not	change	over	time,	nor	was	there	any	significant	gender*time	interaction.	Not	surprisingly,	physicians	in	urban-	or	rural-dominated	locations	were	vastly	more	likely	to	bill	for	rural	and	remote	incentives	(OR:	urban-dominated:	19.52,	9%%	CI	17.41-21.88;	OR:	rural-dominated:	129.49,	95%	CI:	114.3-146.92),	as	were	physicians	who	trained	internationally,	independent	of	location	(OR:	2.01,	95%	CI:	1.71-2.37).	Internationally	trained	physicians	located	in	rural-dominated	areas	had	higher	odds	of	billing	for	the	incentives	compared	to		 152	Canadian-trained,	rurally-located	physicians	(combined	OR	for	rural	location,	international	training	and	the	interaction	term:	193.19,	95%	CI:	98.61-287.77).	5.4	Discussion	In	Chapter	4	I	found	that	physicians’	total	compensation	(in	constant	dollars)	fell	a	total	of	4%	between	2005	and	2012	for	men,	and	increased	by	3%	for	women.	Over	the	same	period,	the	number	of	annual	patient	contacts	and	services	delivered	(for	male	physicians;	less	so	for	female	physicians)	dropped	by	17%,	and	practice	sizes	dropped	by	13%	despite	no	corresponding	increases	in	the	uptake	of	alternative	payment	programs	over	the	same	period.	In	this	Chapter,	my	main	objective	was	to	examine	whether	uptake	of	clinical	and	non-clinical	incentive	programs	was	differential	by	gender.	However,	reflecting	on	my	findings	from	Chapter	4,	I	also	chose	to	disaggregate	physician	remuneration	to	determine	whether	shifts	in	the	proportion	of	payments	for	direct	clinical	care,	clinical	incentives,	and	non-clinical	incentives,	could	be	responsible	for	the	incongruous	trends	between	contacts	and	services	counts	on	the	one	hand,	and	total	remuneration	on	the	other.			 I	found	that	in	2005-06,	there	was	no	gender	difference	in	the	proportion	of	total	compensation	from	direct	clinical	care.	By	2011-12,	however,	female	physicians	earned	a	greater	proportion	of	their	total	compensation	from	direct	clinical	care	compared	to	male	physicians	(i.e.	from	either	fee-for-service	billings	or	from	sessional,	salary,	or	service	agreements).	While	the	proportion	of	income	from	direct	clinical	care	declined	significantly	over	the	course	of	the	study	period	for	both		 153	males	and	females,	the	decline	was	more	dramatic	for	male	physicians.	Both	of	these	trends	are	consistent	with	the	decrease	in	annual	contacts	and	services	observed	in	Chapter	4.			 At	the	same	time,	the	uptake	of	clinical	incentives,	in	terms	of	both	the	number	(proportion)	of	physicians	billing	for	them,	and	the	proportion	of	total	remuneration,	increased	significantly.	Thus,	as	incentives	increased,	patient	contact	counts	declined.	This	might	not	be	surprising,	since	the	GPSC	incentives	were	at	least	partially	intended	to	buy	longer	patient	visits,	and	longer	visits	would	result	in	fewer	patients	seen	in	a	day,	smaller	annual	contact	counts,	and	presumably	less	need	for	repeat	visits.	The	reduction	in	the	number	of	contacts	might	not	in	and	of	itself,	be	problematic	if	the	incentive	payments	improved	patient	health	outcomes	and	satisfaction,	or	if	this	change	in	practice	patterns/style	reduced	downstream	health	services	utilization;	however,	there	is	evidence	that	suggests	that	this	is	probably	not	the	case.	A	review	of	the	incentive	program	found	that	measures	of	access	to,	and	coordination	of,	healthcare	actually	declined	over	the	period	of	time	since	the	program	was	introduced	(Lavergne	et	al.,	2014).	In	addition	to	these	troubling	changes	in	primary	care	practice,	reductions	in	the	number	of	patients	seen	per	physician	per	year	(both	total	contacts	and	unique	patients)	are	also	potentially	problematic	in	light	of	the	fact	that	approximately	200,000	British	Columbians	do	not	have	a	regular	source	of	primary	care,	but	wish	they	did	(Doctors	of	BC,	General	Practice	Services	Committee,	&	Government	of	British	Columbia,	2015).		 154		 I	also	found	a	parabolic	curve	in	the	uptake	of	both	clinical	and	non-clinical	incentives	with	respect	to	physician	age,	with	the	eldest	and	youngest	physicians	being	less	likely	to	bill	for	either	them.	Combined	with	the	findings	from	Chapter	4	indicating	reduced	activity	levels	(total	compensation,	contacts,	services	and	unique	patient	counts),	this	finding	suggests	that	physicians	in	these	groups	are	practicing	less,	and	in	a	more	restrictive	fashion,	relative	to	physicians	aged	35-65.	This	finding	will	be	explored	in	further	detail	through	the	examination	of	patient	characteristics	and	practice	patterns	in	Chapters	6	and	7	respectively.			 This	chapter	also	examined	the	uptake	of	two	specific	non-clinical	incentives:	on-call	payments	and	rural	and	remote	incentives.	Female	physicians	were,	over	the	study	period,	only	half	as	likely	as	their	male	counterparts	to	bill	for	time	on-call.	Also,	the	proportion	of	physicians	who	billed	for	time	on-call	declined	significantly	over	the	study	period	and	this	effect	was	more	substantial	for	female	physicians,	widening	the	observed	gender	gap	with	time.	I	am	unaware	of	any	changes	to	the	program	or	associated	fees	that	would	provide	an	exogenous	explanation	for	the	decrease	in	uptake.		On-call	programs	provide	compensation	to	physicians	who	provide	emergency	care	to	patients	as	part	of	a	rotation	(physician	group)	(Kornelsen	&	Grzybowski,	2010).	Compensation	is	provided	according	to	physician	response	times.	On-call	programs	are	of	particular	importance	in	rural	and	remote	communities,	where	call	rotas	rely	heavily	on	primary	care	physicians	who	have	experience	in	more	than	one	practice	area	(Kornelsen	&	Grzybowski,	2010).	Reductions	in	the	number/proportion	of	physicians	participating	in	on-call		 155	programs	could	exacerbate	a	long-standing	access	issue	for	those	individuals	living	in	rural	areas	in	particular.		 Female	physicians	were	also	less	likely	to	bill	for	rural	and	remote	incentives	even	after	adjusting	for	the	impact	of	practice	rurality.	This	is	an	interesting	finding	since	the	eligibility	for	billing	incentives	within	BC’s	Rural	Subsidiary	Agreement	is	based	solely	on	practice	location.	The	observed	gender	difference	may	be	a	result	of	residual	confounding,	or	a	lack	of	precision	in	the	rurality	measure	that	is	differential	by	gender.	Areas	that	are	labeled	“rural-dominated”,	for	example,	still	have	a	degree	of	internal	rural	variation	that	may	affect	eligibility	for	rural	incentives.	It	is	possible	that	within	those	regions,	female	physicians	were	more	likely	to	work	in	areas	not	eligible	for	the	incentives	compared	to	their	male	counterparts.		5.4.1	Limitations	As	with	Chapter	4,	an	important	strength	of	this	work	is	the	inclusion	of	data	from	the	APP	dataset,	the	use	of	which	completes	the	hitherto	incomplete	picture	of	physician	compensation.	The	inclusion	of	this	data	source	allows	for	the	examination	of	several	important	payment	types	that	are	not	available	in	the	MSP	fee-for-service	dataset	including,	in	particular,	non-clinical	incentive	payments	(such	as	those	for	time	on-call	or	for	rural	and	remote	practice).			 An	important	strength	of	the	work	described	in	this	chapter	is	the	cohort	included	in	this	analysis.	Because	the	variables	used	here	involve	the	classification	of	income	sources,	rather	than	details	of	specific	patient	encounters,	I	did	not	have		 156	to	exclude	physicians	paid	entirely	under	alternative	remuneration	models.	Rather,	I	used	a	complete	cohort	of	primary	care	physicians	who	billed	during	at	least	one	study	year,	ameliorating	risk	of	selection	bias	in	my	results.			 The	modeling	strategy	is	an	additional	strength.	Breaking	down	clinical	and	non-clinical	incentive	models	into	binary,	and	then	proportional	dependent	variables	allowed	me	to	look	at	whether	the	independent	impact	of	gender	was	different	for	“any	incentives”	versus	“how	much	of	that	incentive”.	The	fact	that	the	gender	effect	was	different	across	these	two	model	types	supports	the	fact	that	this	was	the	appropriate	strategy,	suggesting	that	the	factors	that	affect	whether	a	physician	bills	for	an	incentive	payment	are	different,	or	carry	different	weights,	than	those	that	affect	the	participation	rate	in	activities	eligible	for	the	incentive..	Also,	the	inclusion	of	the	GLMs	allowed	me	to	examine	the	whole	cohort,	and	to	examine	annual	changes	in	incentive	uptake	over	the	course	of	the	study	period.			 A	limitation	of	this	work	is	the	range	of	data	I	was	able	to	access	(2005-2012).	The	GPSC’s	Full-Service	Family	practice	program	was	established	in	2002,	and	the	roll-out	of	clinical	incentive	payments	occurred	between	2002	and	2008	(Tregillus	&	Cavers,	2011).	It	would	have	been	preferable	to	be	able	to	compare	the	proportion	of	income	for	clinical	care	delivery,	and	for	clinical	incentives	in	particular,	prior	to	as	well	as	following	implementation.		Unfortunately	the	APP	data	database	not	made	available	by	the	B.C.	Ministry	of	Health	for	years	prior	to	2005-06.				 Another	possible	limitation	of	this	analysis	is	the	lack	of	information	about	patient	encounters.	In	order	to	include	a	complete	cohort	of	physicians,	I	did	not	use		 157	data	found	in	the	fee-for-service	encounters	that	would	help	to	elucidate	whether	differences	in	uptake	of	clinical	incentives	in	particular	are	related	to	the	underlying	demographics	and	morbidity	characteristics	of	the	patient	population.	I	do	address	this	limitation	in	Chapter	6	by	building	multivariate	models	that	incorporate	patient	encounter	data,	but	these	exclude,	by	necessity,	physicians	receiving	100%	of	their	income	through	alternative	payment	programs.	Additional	strengths	and	limitations	that	apply	to	the	thesis	generally	are	included	in	Chapter	8.	5.4.2	Implications	The	proportion	of	a	physician’s	income	from	direct	care	delivery	declined	over	the	course	of	the	study	period,	consistent	with	the	reduction	in	service	and	contact	counts	observed	in	Chapter	4,	and	particularly	for	male	physicians;	this	decline	shows	no	signs	of	slowing.	Clinical	incentives	have	made	up	the	difference	in	incomes,	with	no	evidence	so	far	of	improved	access	to,	or	quality	of,	care	for	British	Columbians.	With	200,000	residents	not	having	a	regular	family	doctor	but	ostensibly	seeking	one,	this	trend	is	particularly	troubling.			 		 158	CHAPTER	6:	Patient	characteristics	6.1	Introduction	As	a	first	step	in	the	examination	of	how	physician	gender	influences	practice	patterns,	this	chapter	examines	the	demographic	and	morbidity	characteristics	of	the	patient	populations	for	male	and	female	primary	care	physicians	(PCPs).	This	analysis	supplies	some	additional	context	for	the	findings	from	Chapters	4	and	5	and	feeds	into	the	examination	of	gender-driven	differences	in	practice	patterns	that	will	be	presented	in	Chapter	7		 Two	(now	quite	dated)	studies	conducted	in	Ontario	found	that	female	graduates	of	McMaster	university’s	medical	program	saw	a	higher	proportion	of	female	patients	in	all	age	groups	compared	with	their	male	counterparts	(Cohen	et	al.,	1991;	Keane	et	al.,	1991).	These	results	are	consistent	with	what	has	been	found	internationally	(Aasland	&	Rosta,	2011;	Bensing	et	al.,	1993;	Britt	et	al.,	1996).	International	studies	have	also	found	that	female	physicians	tend	to	see	fewer	older	patients	compared	to	their	male	counterparts	(Carek	et	al.,	2003;	Harrison	et	al.,	2011).		To	my	knowledge,	no	studies	examining	differences	in	the	patient	populations	for	male	and	female	PCPs	have	ever	been	conducted	in	BC,	and	none	have	been	conducted	in	Canada	since	1991.		Additionally,	none	of	the	existing	studies	I	reviewed	addressed	the	question	of	possible	physician-gender	differences	in	patient	morbidity	burden.		 159	In	Chapter	5,	I	found	that	a	higher	percentage	of	female	physicians’	total	payments	came	from	direct	clinical	care	delivery,	and	that	they	were	less	likely	to	bill	for	clinical	incentives.	In	this	chapter,	I	examine	the	extent	to	which	differential	uptake	of	these	incentive	payments	and	compensation	for	non-clinical	activities	may	be	related	to	differences	in	the	patient	populations	seen	by	male	versus	female	PCPs	(and	therefore	differences	in	underlying	rates	of	eligibility	for	incentive	payments).	The	specific	research	questions	I	am	addressing	are	as	follows:	• Question	3.1:	What	are	the	differences	in	service	population	demographics	for	male	versus	female	PCPs?		o 3.1.1:	What	is	the	age	distribution	(median,	mean,	etc.)	of	patients	seen	by	female	versus	male	PCPs?		What	is	the	proportion	of	patients	over	the	age	65?	Over	age	75?	o 3.1.2:	What	is	the	gender	distribution	of	patients	seen	by	female	versus	male	PCPs?			• Question	3.2:	Are	there	PCP	gender-related	differences	in	the	mix	of	their	patients’	medical	needs?		o 3.2.1:	What	was	the	PCP	gender-specific	distribution	of	patients	across	the	eight	major	aggregated	diagnostic	groups	(ADGs)?		o 3.2.2:	What	percentage	of	the	patient	populations	of	each	PCP	gender	had	three	or	more	ADGs?		• Question	3.3:	Do	clinical	payments	continue	to	make	up	a	larger	proportion	of	the	incomes	of	female	physicians	than	male	once	the	characteristics	of	their	respective	patient	populations	are	accounted	for?	Does	uptake	of	clinical	and		 160	non-clinical	incentives	vary	by	gender	once	patient	population	characteristics	are	held	constant?	I	hypothesize	that	female	physicians	will	see	more	female	patients,	and	fewer	seniors,	and	also	that	male	and	female	physicians	will	be	equally	likely	to	treat	patients	who	experience	an	elevated	morbidity	burden.	If	I	observe	no	difference	in	morbidity	of	the	patient	populations	of	male	and	female	PCPs,	I	further	hypothesize	that	the	inclusion	of	patient	characteristics	will	not	affect	the	observed	gender	differences	in	uptake	of	incentive	payments.	If,	however,	I	do	find	a	morbidity	difference,	I	hypothesize	that	the	inclusion	of	patient	characteristics	will	ameliorate	the	gender	differences	in	incentive	uptake	reported	on	in	Chapter	5.	6.2	Methods		 This	analysis	relies	on	the	same	study	cohort	and	datasets	used	in	Chapter	4		(which	was	described	in	Chapter	3,	sections	3.2	and	3.4).	It	focuses	on	the	characteristics	of	the	patient	population	seen	by	male	versus	female	PCPs,	and	therefore	also	uses	data	from	the	Medical	Services	Plan	(MSP)	Consolidation	File	for	patient	demographic	data.	Additionally,	the	MSP	payment	information	file,	acute	hospital	discharge	abstract	and	Vital	Statistics	Death	databases	were	used	to	describe	patient	morbidity.		The	characteristics	of	the	patient	population	can	be	described	only	using	MSP	(fee-for-service)	data.	Thus,	the	Alternative	Payment	Program	(APP)	database	is	only	used	in	this	analysis	to	generate	the	percentage	for	a	physician’s	income	from	fee-for-service	(FFS)	vs.	non-FFS	activity.		 161	6.2.1	Dependent	variables			 The	dependent	variab