Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Physician and patient preferences for stroke prophylaxis in atrial fibrillation Kuo, I fan 2014

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

Item Metadata


24-ubc_2014_spring_kuo_ifan.pdf [ 41.75MB ]
JSON: 24-1.0167221.json
JSON-LD: 24-1.0167221-ld.json
RDF/XML (Pretty): 24-1.0167221-rdf.xml
RDF/JSON: 24-1.0167221-rdf.json
Turtle: 24-1.0167221-turtle.txt
N-Triples: 24-1.0167221-rdf-ntriples.txt
Original Record: 24-1.0167221-source.json
Full Text

Full Text

PHYSICIAN	  AND	  PATIENT	  PREFERENCES	  FOR	  STROKE	  PROPHYLAXIS	  	  IN	  ATRIAL	  FIBRILLATION	  	  by	  I	  fan	  Kuo	  	  PharmD,	  The	  University	  of	  British	  Columbia,	  2009	  	  A	  THESIS	  SUBMITTED	  IN	  PARTIAL	  FULFILLMENT	  OF	  THE	  REQUIREMENTS	  FOR	  THE	  DEGREE	  OF	  	  MASTER	  OF	  SCIENCE	  in	  THE	  FACULTY	  OF	  GRADUATE	  AND	  POSTDOCTORAL	  STUDIES	  (Pharmaceutical	  Sciences)	  	  THE	  UNIVERSITY	  OF	  BRITISH	  COLUMBIA	  (Vancouver)	  	  April	  2014	  	  ©	  I	  fan	  Kuo,	  2014	  ii	  	  Abstract	  Purpose.	  	  To	  derive	  and	  compare	  relative	  preferences	  of	  physicians	  and	  patients	  for	  selecting	  oral	  antithrombotics	  in	  atrial	  fibrillation	  (AF).	  	  	  Methods.	  Elicitation	  task:	  Time	  trade-­‐off	  (TTO)	  and	  best	  worst	  scaling	  (BWS)	  choice	  experiments	  were	  constructed	  from	  literature	  review	  and	  expert	  opinion,	  reflecting	  four	  attributes	  relevant	  to	  oral	  antithrombotic	  selection	  in	  the	  setting	  of	  stroke	  prevention	  in	  AF	  –	  frequency	  of	  laboratory	  monitoring,	  annual	  risk	  of	  stroke,	  annual	  risk	  of	  major	  bleed,	  availability	  of	  reversal	  agent.	  	  	  Analysis.	  Utilities	  based	  on	  the	  patient	  TTO	  data	  were	  derived	  and	  analyzed	  for	  subgroup	  differences.	  Relative	  utilities	  based	  on	  the	  BWS	  choice	  data	  were	  derived	  using	  the	  conditional	  logit	  model	  and	  latent	  class	  analysis.	  The	  Wilcoxon	  signed-­‐rank	  test	  was	  performed	  to	  assess	  the	  difference	  in	  preference	  for	  each	  attribute	  level	  between	  the	  Best-­‐Worst	  score	  for	  the	  two	  groups.	  	  	  Results.	  33	  physicians	  and	  58	  patients	  completed	  the	  choice	  experiment.	  Both	  groups	  favoured	  “annual	  stroke	  risk	  of	  0%”	  as	  the	  most	  valued	  attribute-­‐level,	  and	  “annual	  stroke	  risk	  of	  10%”	  was	  the	  least	  favourable	  attribute-­‐level.	  Significant	  preference	  differences	  between	  the	  two	  perspectives	  for	  several	  of	  the	  attribute	  levels	  were	  also	  found.	  The	  results	  points	  out	  that	  while	  both	  groups	  had	  strong	  positive	  preferences	  for	  “annual	  stroke	  0%”,	  the	  physicians	  iii	  	  valued	  this	  attribute	  level	  to	  be	  much	  more	  desirable	  than	  the	  patients.	  Similar	  observation	  applies	  to	  “annual	  stroke	  10%”,	  where	  physicians	  had	  much	  stronger	  negative	  preferences	  for	  the	  attribute	  level	  compared	  to	  the	  patients	  as	  well	  as	  the	  range	  of	  annual	  bleeding	  rates,	  suggesting	  that	  physicians	  in	  general,	  were	  more	  stroke	  and	  bleed	  averse	  than	  the	  patients.	  	  	  Conclusions.	  There	  is	  a	  general	  congruence	  in	  physician	  and	  patient	  preferences	  for	  stroke	  prophylaxis,	  however,	  the	  strength	  of	  preferences	  differ	  for	  several	  attributes	  differ	  between	  the	  two	  groups.	  Using	  a	  BWS	  choice	  experiment,	  a	  preference	  elicitation	  method	  based	  on	  the	  random	  utility	  theory,	  this	  is	  the	  first	  study	  that	  quantitatively	  evaluated	  the	  preferences	  of	  physicians	  and	  patients	  for	  stroke	  prophylaxis	  in	  atrial	  fibrillation	  and	  provides	  important	  insights	  into	  clinical	  decision-­‐making	  in	  a	  patient-­‐centered	  care	  model.	  	  iv	  	  Preface	  All	  work	  presented	  in	  this	  thesis,	  including	  design,	  study	  participant	  recruitment,	  analysis	  and	  write-­‐up	  were	  conducted	  by	  the	  Master’s	  candidate.	  	  The	  following	  certificates	  of	  approval	  were	  received	  to	  conduct	  the	  research	  at	  different	  institutions:	  1. Valuation	  of	  patient	  and	  physician	  preferences	  for	  stroke	  prophylaxis	  in	  atrial	  fibrillation:	  the	  INTENT	  study,	  Providence	  Health	  Care	  Research	  Ethics	  Board,	  the	  University	  of	  British	  Columbia	  (H12-­‐01983)	  2. Valuation	  of	  patient	  and	  physician	  preferences	  for	  stroke	  prophylaxis	  in	  atrial	  fibrillation:	  the	  INTENT	  study,	  Vancouver	  Coastal	  Health	  Authority	  Research	  Study	  (#V12-­‐01983)	  	  v	  	  Table	  of	  Contents	  Abstract	  ...................................................................................................................................................	  ii	  Preface	  ....................................................................................................................................................	  iv	  Table	  of	  Contents	  ..................................................................................................................................	  v	  List	  of	  Tables	  .......................................................................................................................................	  viii	  List	  of	  Figures	  .......................................................................................................................................	  ix	  List	  of	  Abbreviations	  ............................................................................................................................	  x	  Acknowledgements	  ............................................................................................................................	  xi	  Chapter	  1:	  Introduction	  ...............................................................................................................................	  1	  1.1	   Patient-­‐centered	  decision-­‐making:	  current	  state	  and	  challenges	  ..............................................	  1	  1.2	   Patient	  preferences:	  definition	  and	  importance	  in	  decision-­‐making	  ..........................................	  3	  1.3	   Preference	  elicitation	  methods	  ...................................................................................................	  5	  1.4	   Oral	  antithrombotics	  for	  stroke	  prevention	  in	  patients	  with	  atrial	  fibrillation	  .........................	  15	  1.5	   Patient-­‐centered	  decision-­‐making:	  current	  knowledge	  gaps	  ...................................................	  17	  1.6	   Thesis	  objectives	  and	  organization	  ...........................................................................................	  23	  Chapter	  2:	  Comparison	  of	  patient	  preferences	  and	  physician	  judgments	  in	  clinical	  decision-­‐making:	  a	  literature	  review	  ........................................................................................................................................	  27	  2.1	   Introduction	  ..............................................................................................................................	  27	  2.2	   Methods	  ...................................................................................................................................	  28	  2.3	   Findings	  ....................................................................................................................................	  29	  2.4	   Discussion	  .................................................................................................................................	  34	  vi	  	  Chapter	  3:	  Patients’	  preferences	  for	  stroke	  prophylaxis	  in	  atrial	  fibrillation:	  time	  trade-­‐off	  (TTO)	  and	  best-­‐worst	  scaling	  (BWS)	  experiment	  .......................................................................................................	  51	  3.1	   Introduction	  ..............................................................................................................................	  51	  3.2	   Methods	  ...................................................................................................................................	  53	  3.3	   Results	  ......................................................................................................................................	  60	  3.4	   Discussion	  .................................................................................................................................	  65	  Chapter	  4:	  Physicians’	  preferences	  for	  stroke	  prophylaxis	  in	  atrial	  fibrillation	  .........................................	  77	  4.1	   Introduction	  ..............................................................................................................................	  77	  4.2	   Methods	  ...................................................................................................................................	  79	  4.3	   Results	  ......................................................................................................................................	  83	  4.4	   Discussion	  .................................................................................................................................	  86	  Chapter	  5:	  Comparison	  of	  patients’	  and	  physicians’	  preferences	  for	  stroke	  prophylaxis	  in	  atrial	  fibrillation	  ...................................................................................................................................................................	  95	  5.1	   Introduction	  ..............................................................................................................................	  95	  5.2	   Methods	  ...................................................................................................................................	  97	  5.3	   Results	  ......................................................................................................................................	  99	  5.4	   Discussion	  ...............................................................................................................................	  101	  Chapter	  6:	  Discussion	  ..............................................................................................................................	  109	  6.1	   Summary	  of	  major	  findings	  and	  study	  implications	  ...............................................................	  109	  6.2	   Limitations	  ..............................................................................................................................	  115	  6.3	   Directions	  of	  future	  research	  ..................................................................................................	  118	  6.4	   Knowledge	  translation	  ...........................................................................................................	  119	  6.5	   Conclusion	  ..............................................................................................................................	  119	  vii	  	  Bibliography	  .....................................................................................................................................	  121	  Appendix	  A:	  Patient	  Questionnaire	  ..........................................................................................	  136	  Appendix	  B:	  Physician	  Questionnaire	  .....................................................................................	  179	  	  viii	  	  List	  of	  Tables	  Table	  1.1	  Axioms	  of	  von	  Neumann-­‐Morgenstern	  utility	  theory	  	  .................................................................................	  25	  Table	  1.2	  CHADS2	  	  risk	  criteria	  ......................................................................................................................................	  26	  Table	  1.3	  CHA2DS2-­‐VASc	  risk	  criteria	  ...........................................................................................................................	  26	  Table	  2.1:	  Summary	  of	  characteristics	  of	  included	  studies	  comparing	  physician	  and	  patient	  preferences	  .................	  39	  Table	  2.2:	  Summary	  of	  results	  of	  included	  studies	  ......................................................................................................	  45	  Table	  3.1:	  Attributes	  and	  levels	  in	  BWS	  choice	  experiment	  .........................................................................................	  71	  Table	  3.2:	  Characteristics	  of	  patient	  participants	  ........................................................................................................	  72	  Table	  3.3:	  TTO	  utilities	  for	  each	  health	  state	  ...............................................................................................................	  73	  Table	  3.4:	  TTO	  utilities	  by	  subgroups	  ...........................................................................................................................	  73	  Table	  3.5:	  Relative	  preferences	  estimates	  from	  conditional	  logit	  and	  latent	  class	  models	  (BWS	  choice	  experiment)	  74	  Table	  4.1:	  Attributes	  and	  levels	  in	  BWS	  choice	  experiment	  .........................................................................................	  90	  Table	  4.2:	  Characteristics	  of	  physician	  participants	  ....................................................................................................	  91	  Table	  4.3:	  Relative	  preference	  estimates	  from	  conditional	  logit	  model	  ......................................................................	  92	  Table	  5.1:	  Best-­‐worst	  score	  for	  patients	  and	  physicians	  ...........................................................................................	  106	  	  ix	  	  List	  of	  Figures	  Figure	  3.1:	  Part-­‐worth	  utility	  estimates	  for	  attribute	  levels	  ........................................................................................	  75	  Figure	  3.2:	  Relative	  importance	  of	  attributes	  ..............................................................................................................	  76	  Figure	  4.1:	  Part-­‐worth	  utility	  estimates	  for	  attribute	  levels	  ........................................................................................	  93	  Figure	  4.2:	  Relative	  importance	  of	  attributes	  ..............................................................................................................	  94	  Figure	  5.1:	  Patient	  choices	  of	  oral	  antithrombotics	  relative	  to	  warfarin	  based	  on	  stated	  preferences	  according	  to	  baseline	  stroke	  risk	  ....................................................................................................................................................	  107	  Figure	  5.2:	  Physician	  choices	  of	  oral	  antithrombotics	  relative	  to	  warfarin	  based	  on	  stated	  preferences	  according	  to	  baseline	  stroke	  risk	  ....................................................................................................................................................	  108	  	  	  	  	  	  x	  	  List	  of	  Abbreviations	  	  Abbreviations	   Definition	  AF	   Atrial	  fibrillation	  ASA	   Acetylsalicylic	  acid	  BWS	   Best-­‐worst	  scaling	  CA	   Conjoint	  analysis	  CNS	   Central	  nervous	  system	  CPG	   Clinical	  practice	  guideline	  DCE	   Discrete	  choice	  experiment	  EBM	   Evidence-­‐based	  medicine	  LCA	   Latent	  class	  analysis	  LCM	   Latent	  class	  model	  GP	   General	  practitioner	  PCC	   Patient-­‐centered	  care	  PONV	   Postoperative	  nausea	  and	  vomiting	  TTO	   Time	  trade-­‐off	  SG	   Standard	  gamble	  VKA	   Vitamin	  K	  antagonist	  	  xi	  	  Acknowledgements	  I	  would	  like	  to	  thank	  my	  supervisor	  and	  co-­‐supervisor,	  Dr.	  Larry	  Lynd	  and	  Dr.	  Carlo	  Marra	  for	  their	  guidance	  and	  support	  in	  this	  research	  project.	  I	  would	  also	  like	  to	  thank	  the	  rest	  of	  my	  supervisory	  committee	  members,	  Dr.	  Karin	  Humphries,	  Dr.	  Ross	  Tsuyuki,	  Dr.	  Robert	  Boone	  and	  Dr.	  Frank	  Abbott	  for	  their	  input	  and	  feedback	  in	  conducting	  this	  research.	  I	  would	  like	  to	  express	  my	  gratitude	  to	  Dr.	  Nick	  Bansback	  for	  generously	  providing	  consultation	  and	  advice	  in	  the	  area	  of	  preference	  elicitation.	  I	  am	  also	  sincerely	  grateful	  for	  the	  help	  from	  Ms.	  Salma	  Lalji,	  our	  IT	  specialist	  who	  assisted	  me	  in	  setting	  up	  the	  web-­‐based	  surveys.	  I	  would	  also	  like	  to	  thank	  Ms.	  Kathy	  Hui-­‐Qing	  Li	  in	  her	  significant	  contributions	  to	  help	  me	  understand	  and	  perform	  the	  statistical	  analysis	  needed	  for	  this	  project.	  Much	  gratitude	  also	  goes	  out	  to	  Ms.	  Michelle	  Lee,	  who	  provided	  assistance	  in	  patient	  recruitment	  at	  the	  AF	  clinic	  during	  a	  time	  of	  medical	  health	  impairment.	  And	  I	  owe	  particular	  thanks	  to	  Ms.	  Lilla	  Roy	  for	  her	  constant	  support	  and	  encouragement	  throughout	  the	  entire	  process.	  	  	  I	  would	  also	  like	  to	  thank	  the	  Heart	  and	  Stroke	  Foundation	  of	  Canada	  and	  the	  Canadian	  Institutes	  of	  Health	  Research	  for	  providing	  a	  three-­‐year	  salary	  stipend	  for	  me	  to	  pursue	  training	  in	  health	  research	  and	  complete	  this	  research	  project.	  	  1	  	  Chapter	  1: Introduction	  1.1 Patient-­‐centered	  decision-­‐making:	  current	  state	  and	  challenges	  1.1.1 From	  evidence-­‐based	  medicine	  to	  patient-­‐centered	  medicine:	  a	  paradigm	  shift	  Since	  the	  early	  1990s,	  evidence-­‐based	  medicine	  (EBM)	  has	  been	  viewed	  as	  an	  important	  milestone	  in	  the	  evolution	  of	  modern	  health	  care	  delivery	  (1).	  Evidence-­‐based	  medicine	  epitomizes	  the	  best	  available	  scientific	  evidence	  on	  treatment	  alternatives,	  combining	  with	  clinical	  expertise	  to	  provide	  the	  most	  appropriate	  treatment	  for	  the	  patients	  (1–5).	  It	  has	  become	  increasingly	  paramount	  in	  informing	  decision-­‐making	  at	  the	  regulatory	  and	  individual	  patient	  levels.	  A	  notable	  derivative	  of	  EBM	  is	  the	  development	  of	  clinical	  practice	  guidelines	  (CPGs)	  and	  protocols,	  which	  are	  systematically	  developed	  in	  consensus	  groups,	  with	  grading,	  interpretation	  and	  translation	  of	  evidence	  into	  recommendations	  to	  inform	  the	  clinicians	  on	  making	  diagnostic	  and	  therapeutic	  decisions	  for	  their	  patients	  (3,6,7).	  While	  EBM	  holds	  the	  patients’	  best	  interest	  at	  its	  core	  and	  practice	  guidelines	  are	  an	  established	  tool	  for	  quality	  improvement	  in	  clinical	  practice,	  it	  has	  been	  criticized	  that	  EBM	  is	  essentially	  a	  physician-­‐centered	  approach	  to	  patient	  care	  as	  it	  focuses	  on	  the	  clinician’s	  (or	  a	  group	  of	  clinicians’)	  interpretation	  of	  the	  evidence	  and	  neglects	  the	  values	  or	  preferences	  of	  the	  patients	  –	  the	  other	  key	  stakeholder	  in	  the	  clinical	  decision-­‐making	  process	  (2,6,8).	  It	  has	  also	  been	  pointed	  out	  that	  even	  the	  scientific	  evidence	  used	  to	  formulate	  guidelines,	  for	  example	  randomized	  controlled	  trials	  (RCTs),	  are	  not	  patient-­‐centered	  whereby	  the	  study	  population,	  intervention	  and	  outcome	  of	  interest	  are	  researcher-­‐defined	  where	  treatment	  benefit	  is	  largely	  focused	  on	  improvement	  in	  clinical	  endpoints	  (1,2).	  2	  	  	  During	  the	  last	  decade,	  there	  has	  been	  a	  movement	  to	  shift	  the	  focus	  of	  modern	  health	  care	  delivery	  from	  EBM	  to	  patient-­‐centered	  medicine	  or	  patient-­‐centered	  care	  (PCC)	  (1,9–11).	  Patient-­‐centered	  care	  is	  defined	  by	  the	  Institute	  of	  Medicine	  as	  care	  that	  is	  respectful	  of	  patient	  preferences	  and	  in	  which	  patient	  values	  guide	  clinical	  decisions	  (12).	  First	  introduced	  by	  Balint	  in	  the	  1950s,	  the	  concept	  of	  patient	  centeredness	  is	  not	  entirely	  new	  (13,14).	  Reasons	  behind	  the	  renewed	  attention	  in	  a	  patient	  centered	  approach	  are	  manifold.	  It	  is	  ideologically	  attractive	  as	  it	  resonates	  with	  the	  basic	  humanistic,	  biopsychosocial	  and	  ethical	  aspects	  of	  patient	  autonomy,	  where	  the	  recommendations	  being	  made	  should	  reflect	  the	  values	  of	  those	  affected	  (1,15–19).	  It	  is	  also	  a	  natural	  progression	  to	  focus	  on	  patient	  outcomes	  with	  the	  growing	  demands	  for	  quality	  and	  safety	  in	  health	  care	  (10).	  From	  the	  capitalistic	  standpoint,	  the	  patient	  is	  increasingly	  perceived	  as	  a	  consumer	  of	  the	  health	  care	  system,	  in	  which	  case	  customer	  satisfaction	  should	  take	  priority	  (10).	  Last	  but	  not	  least,	  there	  is	  also	  the	  belief	  that	  PCC	  serves	  the	  practical	  purpose	  of	  improving	  patient	  adherence	  to	  treatment	  recommendations	  (20).	  	  	  Since	  the	  mid-­‐1990s,	  there	  has	  been	  a	  call	  for	  the	  integration	  of	  patient	  values	  in	  the	  decision-­‐making	  process	  (5,18,21–25).	  However,	  it	  has	  proven	  to	  be	  a	  difficult	  task	  to	  structurally	  incorporate	  patient	  preferences	  in	  the	  decision-­‐making	  process	  as	  patient	  centeredness	  remains	  an	  elusive	  concept	  (18,26–28).	  One	  reason	  is	  because	  of	  the	  continuously	  evolving	  decision-­‐making	  models.	  Under	  the	  PCC	  model,	  decision-­‐making	  can	  be	  broadly	  categorized	  into	  three	  types:	  1)	  health	  provider-­‐as-­‐agent,	  2)	  informed	  decision-­‐making,	  and	  3)	  shared	  3	  	  decision-­‐making	  (29).	  Depending	  on	  the	  decision-­‐making	  model	  adopted,	  there	  are	  several	  places	  where	  patient	  preferences	  can	  be	  integrated.	  For	  instance,	  patient	  preferences	  can	  be	  incorporated	  at	  the	  practice	  development	  stage	  (6,9,26,30),	  or	  at	  the	  point-­‐of-­‐care	  between	  a	  physician	  and	  a	  patient	  (14,19,21,29).	  One	  other	  critical	  obstacle	  to	  integrating	  patient	  preferences	  is	  the	  varied	  interpretation	  of	  the	  term	  patient	  ‘preference’.	  The	  lack	  of	  uniform	  consensus	  on	  the	  definition	  of	  patient	  preference	  has	  in	  part	  hindered	  research	  progress	  in	  this	  area	  and	  consequently	  an	  approach	  to	  integrating	  patient	  preference	  in	  health	  care	  (19).	  	  1.2 Patient	  preferences:	  definition	  and	  importance	  in	  decision-­‐making	  In	  the	  setting	  of	  clinical	  decision-­‐making,	  the	  term	  preference	  is	  often	  applied	  in	  the	  most	  literal	  sense	  where	  a	  patient’s	  ‘preferences	  represent	  his	  or	  her	  final	  choices	  from	  many	  possible	  treatment	  options	  (31).	  Specifically	  in	  the	  shared	  decision-­‐making	  model,	  it	  is	  often	  presumed	  that	  patient	  preference	  is	  elicited	  at	  the	  patient-­‐physician	  encounter	  where	  the	  physician	  informs	  the	  patient	  of	  the	  risk	  and	  benefit	  probability	  associated	  with	  various	  drug	  treatments	  and	  asks	  what	  the	  patient’s	  inclination	  is	  in	  choosing	  a	  therapeutic	  alternative.	  Preferences	  in	  this	  context	  is	  largely	  subjective	  and	  loosely	  defined	  across	  the	  clinical	  settings	  as	  it	  is	  not	  known	  what	  facets	  of	  the	  health	  outcomes	  actually	  constitute	  the	  patient’s	  preference.	  Preferences	  in	  this	  definition	  also	  rest	  on	  the	  assumption	  that	  patient	  preference	  is	  exclusively	  composed	  of	  health-­‐related	  attributes.	  	  4	  	  The	  more	  formal,	  and	  robust	  definition	  of	  ‘preference’,	  which	  is	  also	  used	  interchangeably	  with	  the	  terms	  ‘value’	  and	  ‘utility’,	  is	  deeply	  rooted	  in	  the	  application	  of	  utility	  theories	  in	  health	  care	  (32,33).	  Subjective	  preferences	  became	  a	  quantifiable	  term	  when	  Von	  Neumann	  and	  Morgenstern	  expressed	  preference	  as	  a	  utility	  function	  where	  rationality	  can	  be	  modeled	  to	  maximize	  an	  expected	  value	  through	  making	  a	  trade-­‐off	  between	  two	  states	  or	  features	  (e.g.	  shorter	  time	  for	  better	  quality	  of	  life,	  benefit	  for	  risk	  of	  side	  effects)	  (32,33).	  In	  this	  context,	  preference	  is	  the	  relative	  desirability	  an	  individual	  exhibits	  for	  a	  particular	  health	  state	  or	  attribute	  measured	  on	  a	  scale,	  whereby	  a	  more	  desirable	  state	  (e.g.	  good	  health)	  is	  assigned	  a	  higher	  value	  than	  a	  less	  desirable	  state	  (e.g.	  impaired	  health)	  (18).	  While	  the	  utility	  of	  each	  state	  or	  feature	  is	  not	  observed	  by	  researchers,	  the	  choice	  decisions,	  or	  trade-­‐off	  made	  by	  the	  individual	  can	  reveal	  the	  underlying	  utility	  or	  value	  associated	  with	  that	  health	  state	  or	  feature	  (34).	  It	  is	  in	  making	  a	  trade-­‐off	  that	  the	  preference	  expressed	  by	  the	  individual	  becomes	  an	  explicit	  entity.	  In	  this	  thesis,	  the	  term	  preference	  is	  herein	  referred	  to	  this	  latter	  concept.	  	  	  In	  the	  era	  of	  patient	  centered	  medicine,	  it	  is	  becoming	  accepted	  that	  good	  clinical	  decision-­‐making	  integrates	  not	  only	  the	  best	  evidence,	  clinical	  expertise	  and	  clinical	  uncertainty	  (the	  probability	  that	  various	  health	  outcomes	  will	  occur),	  but	  also	  patients’	  preferences	  for	  the	  health	  outcomes	  (35–37).	  Recommendations	  have	  been	  made	  to	  integrate	  patient	  preferences	  into	  clinical	  practice	  guidelines	  (CPG)	  by	  many	  CPG	  organizations	  as	  well	  as	  the	  World	  Health	  Organization	  (6,26,30,38,39).	  It	  has	  been	  recognized	  that	  the	  importance	  of	  incorporating	  evidence	  of	  patient	  preferences	  is	  greater	  when	  the	  decision	  is	  ‘preference-­‐sensitive’,	  i.e.	  when	  5	  	  the	  evidence	  for	  therapy	  is	  weak	  or	  conflicting,	  when	  several	  effective	  options	  co-­‐exist,	  when	  there	  is	  a	  clear	  trade-­‐off	  between	  benefit	  and	  risk,	  or	  when	  patients	  and	  clinicians	  may	  have	  disparate	  views	  (6,30,40,41).	  	  1.3 Preference	  elicitation	  methods	  An	  overview	  of	  preference	  elicitation	  techniques	  Several	  formal	  methods	  exist	  to	  measure	  preferences	  in	  health	  care	  (18,34,42).	  Qualitative	  methods	  such	  as	  phone	  or	  in-­‐person	  interviews,	  focus	  groups,	  questionnaires	  and	  surveys	  impart	  descriptive	  information	  about	  patients’	  preferences	  for	  specific	  treatment	  features	  that	  are	  important	  to	  them	  (43).	  While	  qualitative	  preference	  elicitation	  methods	  are	  generally	  less	  cognitively	  burdensome	  to	  respondents,	  they	  do	  not	  allow	  the	  quantitative	  valuation	  of	  the	  patients’	  preferences	  and	  thus	  have	  limited	  application	  for	  use	  in	  terms	  of	  both	  health	  economic	  evaluations	  and	  the	  integration	  of	  patient	  preferences	  on	  a	  systematic	  level.	  	  	  Of	  the	  quantitative	  preference	  elicitation	  methods,	  patient	  preferences	  can	  be	  sought	  through	  evaluating	  actual	  trade-­‐off	  choices	  made	  by	  the	  individuals	  in	  real-­‐life	  setting	  as	  seen	  in	  ‘revealed	  preference	  methods’,	  or	  through	  asking	  individuals	  to	  state	  their	  preferences	  by	  making	  choices	  from	  a	  set	  of	  two	  or	  more	  hypothetical	  scenarios	  as	  seen	  in	  ‘stated	  preference	  methods’	  (34,44–47).	  In	  stating	  a	  preference	  that	  involves	  trade-­‐offs,	  the	  value	  (or	  utility)	  that	  an	  individual	  associates	  with	  the	  health	  states	  or	  attributes	  is	  revealed	  (34).	  	  	  6	  	  Stated	  preference	  methods	  can	  be	  broadly	  categorized	  into	  cardinal	  and	  ordinal	  preference	  techniques.	  Cardinal	  methods	  derive	  preference	  estimates	  that	  reflect	  the	  strength	  of	  preference	  for	  the	  outcome	  or	  health	  states	  relative	  to	  the	  others	  on	  an	  interval	  scale.	  In	  contrast,	  ordinal	  preference	  instruments	  produce	  relative	  preference	  weights	  of	  two	  or	  more	  alternatives	  or	  attributes	  through	  establishing	  the	  ordering	  or	  ranking	  of	  preferences	  (34).	  	  	  Cardinal	  preference	  elicitation	  methods	  The	  most	  commonly	  used	  cardinal	  preference	  methods	  include	  the	  time	  trade-­‐off	  (TTO),	  and	  standard	  gamble	  (SG).	  Central	  to	  the	  cardinal	  preference	  elicitation	  methods	  is	  the	  expected	  utility	  theory,	  also	  called	  the	  von	  Neumann-­‐Morgenstern	  (NM)	  utility	  theory,	  that	  has	  formed	  the	  framework	  for	  modern	  health-­‐related	  decision-­‐making	  (32,48).	  The	  normative	  model	  developed	  by	  von	  Neumann	  and	  Morgenstern	  is	  built	  on	  the	  assertion	  of	  how	  a	  rational	  individual	  ‘ought’	  to	  make	  decisions	  when	  faced	  with	  uncertain	  outcomes.	  The	  theory	  can	  be	  expressed	  in	  the	  following	  terms	  (48):	  	  	  𝑈 𝑥? , 𝑝? ,… , 𝑥?, 𝑝? = 𝑝?𝑈 𝑥? +⋯+ 𝑝?𝑈(𝑥?)	   	   	   	   (Eq.	  1.2.1)	  	  𝑈 𝑤 + 𝑥? , 𝑝? … ,𝑤 + 𝑥?, 𝑝? > 𝑈 𝑤  𝑈 𝑥? , 𝑝? … , 𝑥?, 𝑝? > 0	   	   (Eq.	  1.2.2)	  	  𝑈"(. ) > 0	   	   	   	   	   	   	   	   	   (Eq.	  1.2.3)	  7	  	  where	  𝑥? 	  is	  the	  outcome	  of	  event	  𝑖,	  𝑝? 	  is	  the	  probability	  of	  event	  𝑖	  happening,	  and	  𝑤	  is	  individual’s	  asset	  at	  the	  beginning	  before	  making	  the	  choice.	  Equation	  1.2.1	  sets	  the	  expectation	  rule,	  where	  the	  utility	  is	  equivalent	  to	  the	  expected	  value	  of	  its	  component.	  Equation	  1.2.2	  outlines	  the	  asset	  integration	  principle,	  whereby	  a	  rational	  individual	  will	  participate	  in	  making	  the	  trade-­‐off	  if	  the	  utility	  of	  the	  choice	  outcomes	  in	  combination	  with	  his	  or	  her	  current	  asset	  𝑤	  is	  higher	  than	  the	  utility	  of	  the	  current	  asset.	  The	  last	  equation	  1.2.3	  implies	  that	  the	  rational	  individual	  is	  risk	  averse	  and	  therefore	  prefers	  certain	  outcomes	  to	  equivalent	  risky	  options.	  The	  above	  equations	  describe	  the	  axioms	  of	  von	  Neumann-­‐Morgenstern	  utility	  theory:	  completeness,	  transitivity,	  continuity	  and	  independence.	  Detailed	  description	  of	  these	  axioms	  can	  be	  found	  in	  Table	  1	  (32).	  Proponents	  of	  the	  expected	  utility	  theory	  claim	  that	  any	  rational	  person	  would	  comply	  with	  these	  axioms	  when	  faced	  with	  risks,	  such	  that	  their	  preferred	  course	  of	  action	  would	  correspond	  to	  the	  outcome	  with	  the	  highest	  expected	  utility.	  	  In	  cardinal	  methods	  such	  as	  the	  TTO	  and	  SG,	  individuals	  are	  asked	  to	  make	  trade-­‐offs	  between	  probabilities,	  uncertainties	  or	  risks	  associated	  with	  a	  health	  states	  or	  features	  associated	  with	  a	  health	  delivery	  or	  therapeutic	  option.	  In	  the	  SG	  method,	  respondents	  are	  asked	  to	  choose	  between	  two	  choices,	  Choice	  B	  with	  the	  probability	  of	  a	  sure	  outcome,	  or	  Choice	  A	  with	  the	  varying	  probability	  p	  of	  a	  specified	  health	  benefit	  and	  probability	  1− 𝑝	  of	  a	  specified	  adverse	  outcome	  (e.g.	  death)	  (32,47).	  The	  probability	  𝑝	  at	  which	  the	  respondent	  becomes	  indifferent	  8	  	  between	  taking	  the	  gamble	  (Choice	  A)	  or	  remaining	  in	  current	  health	  state	  (Choice	  B)	  is	  then	  the	  individual’s	  utility	  for	  the	  specified	  health	  state	  i	  (32).	  	  	  In	  the	  TTO	  method,	  preference	  estimates	  are	  generated	  by	  asking	  the	  individuals	  to	  trade-­‐off	  years	  of	  perfect	  health	  against	  a	  remaining	  lifetime	  in	  a	  particular	  health	  state	  (32,34).	  Respondents	  are	  shown	  two	  alternatives,	  A	  and	  B.	  In	  Alternative	  B,	  respondent	  would	  remain	  in	  a	  hypothetically	  described	  health	  state	  i	  with	  a	  life	  expectancy	  of	  𝑡	  years	  and	  in	  Alternative	  A,	  respondent	  would	  live	  in	  optimal	  health	  but	  with	  a	  shorter	  life	  expectancy	  of	  𝑥	  years.	  Respondents	  are	  prompted	  to	  trade	  the	  number	  of	  years	  in	  Alternative	  A	  for	  a	  state	  of	  optimal	  health	  in	  Alternative	  B.	  The	  value	  of	  t	  at	  which	  the	  respondent	  becomes	  indifferent	  generates	  the	  utility	  for	  the	  health	  state	  𝑖	  through	  the	  calculation	  of	  𝑥/𝑡	  (32,43,49).	  	  While	  cardinal	  preference	  weights	  are	  widely	  used	  in	  health	  economics	  for	  cost-­‐utility	  analysis,	  these	  elicitation	  methods	  have	  come	  under	  heavy	  criticism	  (50–52).	  Many	  decision	  theorists	  and	  modern	  health	  economists	  challenge	  the	  fundamental	  axiom	  of	  independence,	  arguing	  that	  actual	  choice	  behavior	  are	  not	  always	  independent	  and	  that	  especially	  in	  medical	  decisions,	  one’s	  specified	  preferences	  are	  not	  necessarily	  constant	  given	  another	  stated	  preference.	  In	  other	  words,	  while	  it	  may	  be	  expected	  that	  an	  individual	  will	  likely	  make	  decisions	  based	  on	  the	  highest	  expected	  utility	  in	  a	  one-­‐time	  lottery	  scenario,	  the	  rule	  does	  not	  apply	  to	  situations	  of	  chronic	  illnesses	  involving	  significant	  risks	  such	  as	  death,	  where	  expectations	  are	  known	  to	  be	  influenced	  by	  random	  variables,	  and	  thus	  are	  not	  normative	  (53).	  	  9	  	  From	  the	  survey	  administration	  perspective,	  the	  hypothetical	  alternatives	  provided	  in	  cardinal	  elicitation	  methods	  are	  complex	  and	  require	  a	  fair	  amount	  of	  cognitive	  ability	  and	  concentration	  in	  order	  to	  make	  a	  choice	  that	  is	  fully	  reflective	  of	  the	  axioms	  of	  the	  expected	  utility	  theory	  (34).	  Others	  have	  also	  pointed	  out	  that	  utilities	  generated	  by	  SG	  could	  be	  contaminated	  by	  risk	  aversion	  of	  the	  individuals	  (34,50).	  In	  addition,	  preference	  estimates	  from	  the	  TTO	  methods	  are	  	  predisposed	  to	  bias	  of	  time	  preference	  (i.e.	  individuals	  having	  different	  preference	  for	  an	  immediate	  health	  state	  versus	  one	  that	  is	  20	  years	  ahead),	  duration	  effects	  and	  the	  likelihood	  of	  some	  respondents	  unwilling	  to	  trade-­‐off	  any	  life	  years	  (50).	  Another	  limitation	  of	  the	  TTO	  and	  SG	  methods	  is	  the	  lack	  of	  their	  validity	  in	  measuring	  preferences	  for	  temporary	  health	  states	  (54).	  Traditionally,	  TTOs	  and	  SGs	  have	  been	  used	  in	  measuring	  preference	  estimates	  for	  chronic	  health	  states	  such	  as	  chronic	  pain,	  or	  post	  strokes	  where	  the	  time	  the	  respondents	  are	  willing	  to	  give	  up	  reveals	  meaningful	  trade-­‐offs.	  Perhaps	  the	  most	  apparent	  shortcoming	  of	  the	  cardinal	  preference	  elicitation	  methods	  is	  that	  these	  techniques	  only	  allow	  the	  trade-­‐off	  of	  one	  health	  state	  or	  feature	  at	  a	  time	  and	  thus	  poorly	  simulate	  the	  decision-­‐making	  in	  the	  real	  world	  where	  respondents	  often	  have	  to	  make	  simultaneous	  trade-­‐offs	  amongst	  multiple	  attributes	  related	  to	  a	  health	  delivery	  or	  therapy.	  	  	  Ordinal	  preference	  elicitation	  methods	  Ordinal	  techniques	  have	  gained	  popularity	  in	  health	  economics	  in	  the	  recent	  years	  and	  are	  thought	  to	  overcome	  some	  of	  the	  biases	  inherent	  in	  the	  cardinal	  methods	  (55–57).	  The	  basis	  of	  the	  ordinal	  preference	  elicitation	  methods	  is	  centered	  on	  the	  classic	  consumer	  theory	  and	  the	  10	  	  random	  utility	  theory	  (56,58).	  The	  former	  assumes	  that	  the	  respondents	  are	  rational	  decision	  makers	  and	  would	  make	  choices	  to	  maximize	  utilities.	  The	  random	  utility	  theory	  (RUT)	  is	  where	  ordinal	  preference	  instruments	  are	  distinguished	  from	  the	  von	  Neumann-­‐Morgenstern	  expected	  utility	  theory.	  Under	  the	  random	  utility	  theory,	  individuals	  have	  a	  construct	  of	  utilities	  for	  choice	  alternatives	  for	  which	  they	  can	  perfectly	  discriminate.	  In	  addition,	  there	  is	  also	  a	  component	  of	  individual	  choice	  behavior	  that	  is	  intrinsically	  probabilistic.	  This	  random	  variation	  may	  be	  due	  to	  unobserved	  attributes	  or	  factors	  affecting	  the	  individual	  choice	  or	  inter-­‐individual	  differences,	  measurement	  errors	  and/or	  function	  specification.	  The	  sum	  of	  this	  random,	  unobserved	  utility	  and	  the	  systematic,	  explainable	  utility	  makes	  up	  the	  latent	  utility	  and	  can	  thus	  be	  described	  by	  the	  following	  term	  (56):	  	  𝑈™ = 𝑉™ + 𝜀™ 	   	   	   	   	   	   	   	   (Equation	  1.2.4)	  	  where	  𝑈™ 	  is	  the	  latent	  utility,	  or	  choices	  as	  a	  result	  of	  choosing	  alternative	  𝑗.	  𝑈™ ,	  consists	  of	  a	  systematic	  part	  𝑉™ 	  and	  a	  random,	  probabilistic	  component	  𝜀™ .	  An	  important	  distinction	  to	  note	  here	  is	  that	  the	  utilities	  generated	  by	  ordinal	  elicitation	  methods	  are	  relative	  estimates,	  whereas	  the	  vNM	  utilities	  in	  cardinal	  methods	  are	  absolute	  numbers	  on	  a	  scale	  between	  0	  and	  1.	  In	  other	  words,	  utility	  estimates	  in	  ordinal	  methods	  are	  meaningful	  only	  amongst	  the	  attributes	  and	  levels	  within	  the	  preference	  experiment,	  and	  cannot	  be	  extrapolated	  and	  compared	  directly	  with	  utilities	  of	  other	  preference	  tasks.	  	  	  11	  	  Two	  common	  ordinal	  preference	  elicitation	  methods	  are	  the	  discrete	  choice	  experiments	  (DCEs)	  and	  ranking	  techniques	  such	  as	  the	  best-­‐worst	  scaling	  (BWS)	  choice	  experiment	  (56,58).	  Both	  methods	  are	  typically	  administered	  as	  a	  survey	  and	  allow	  the	  evaluation	  of	  relative	  importance	  of	  health	  states	  or	  attributes	  of	  the	  intervention	  (or	  services,	  practices,	  products)	  to	  establish	  the	  simultaneous	  trade-­‐offs	  individuals	  are	  willing	  to	  make	  between	  different	  levels	  of	  attributes	  associated	  with	  the	  intervention	  (58).	  In	  DCEs,	  sets	  of	  health	  profiles	  based	  on	  a	  descriptive	  system	  composed	  of	  levels	  of	  a	  finite	  number	  of	  important	  attributes	  are	  constructed.	  Preference	  estimates	  are	  then	  obtained	  by	  presenting	  respondents	  with	  two	  or	  more	  sets	  of	  profiles	  at	  a	  time	  and	  asking	  them	  to	  choose	  the	  one	  they	  believe	  to	  be	  the	  most	  desirable	  option.	  This	  process	  is	  repeated	  with	  different	  profile	  sets	  each	  with	  varying	  levels	  of	  the	  attributes,	  so	  that	  each	  respondent	  will	  have	  made	  choices	  between	  ten	  and	  twenty	  hypothetical	  treatment	  options.	  The	  utility	  derived	  represents	  the	  relative	  value	  that	  the	  respondent	  associates	  with	  the	  particular	  level	  within	  that	  attribute	  of	  interest.	  Comparison	  of	  the	  utilities	  helps	  determine	  which	  characteristics	  of	  a	  drug	  therapy	  or	  services	  most	  likely	  drive	  the	  respondents’	  treatment	  choices	  and	  facilitates	  the	  determination	  of	  the	  “marginal	  rates	  of	  substitution”	  or	  the	  amount	  of	  one	  attribute	  that	  an	  individual	  is	  potentially	  willing	  to	  forego	  in	  order	  to	  gain	  more	  of	  another	  attribute	  (i.e.	  the	  magnitude	  of	  risk	  a	  patient	  is	  willing	  to	  accept	  in	  order	  to	  realize	  a	  greater	  likelihood	  of	  receiving	  a	  benefit).	  	  	  The	  chief	  drawback	  of	  DCE	  is	  that	  a	  large	  number	  of	  choice	  questions	  are	  needed	  to	  construct	  a	  choice	  model,	  which	  invariably	  leads	  to	  either	  the	  need	  for	  larger	  sample	  size	  or	  more	  12	  	  respondent	  burden	  (56,59).	  Best-­‐worst	  scaling	  choice	  experiments	  have	  recently	  acquired	  the	  reputation	  as	  an	  ideal	  preference	  elicitation	  method	  as	  they	  present	  as	  a	  way	  of	  obtaining	  more	  information	  than	  DCEs	  without	  having	  to	  burden	  the	  respondents	  with	  full	  ranking	  of	  all	  the	  choice	  options.	  Currently,	  there	  are	  three	  types	  of	  BWS	  being	  used	  in	  health	  research,	  known	  as	  the	  object-­‐based,	  profile-­‐based	  and	  multi-­‐profile-­‐based	  BWS,	  or	  Case	  1,	  2,	  and	  3	  BWS,	  respectively	  (59,60).	  	  In	  the	  Object-­‐based	  BWS	  (Case	  1	  BWS)	  choice	  task,	  respondents	  are	  asked	  to	  choose	  the	  best	  and	  worst	  options	  amongst	  a	  list	  of	  objects,	  or	  attributes	  without	  the	  associated	  levels	  (59,60).	  Case	  1	  BWS	  are	  simple	  to	  design	  and	  administer,	  and	  useful	  for	  eliciting	  values	  for	  broad	  issues	  or	  concepts	  in	  health.	  Examples	  include	  researches	  done	  by	  Finn	  and	  Louviere	  to	  investigate	  public’s	  opinion	  about	  food	  supply	  policy	  (61),	  and	  more	  recently	  in	  Australia	  to	  examine	  the	  citizens’	  views	  on	  what	  were	  the	  most	  and	  least	  important	  principles	  in	  the	  health	  care	  system	  (62).	  The	  main	  disadvantage	  of	  Case	  1	  BWS	  is	  that	  it	  only	  allows	  the	  valuation	  of	  the	  attributes	  and	  not	  individual	  levels	  associated	  with	  the	  attributes.	  	  	  In	  profile-­‐based	  BWS	  (Case	  2	  BWS),	  respondents	  are	  shown	  a	  single	  profile	  of	  varying	  attribute	  levels	  and	  asked	  to	  choose	  the	  best	  and	  worst	  levels	  from	  each	  choice	  task.	  With	  each	  profile	  set,	  the	  respondents	  are	  choosing	  a	  pair	  of	  attribute	  levels	  that	  are	  furthest	  apart	  on	  their	  utility	  scale	  (56).	  Utilities	  for	  each	  attribute	  level	  are	  then	  revealed	  through	  the	  number	  of	  times	  they	  were	  chosen	  as	  best	  and	  worst	  in	  relation	  to	  the	  other	  attribute	  levels.	  	  This	  allows	  13	  	  the	  direct	  comparison	  of	  all	  attribute	  levels	  on	  a	  common	  scale,	  a	  main	  advantage	  of	  Case	  2	  BWS	  over	  traditional	  DCEs,	  where	  relative	  preferences	  are	  estimated	  within	  each	  attribute	  (i.e.	  a	  reference	  level	  is	  needed	  for	  each	  attribute	  in	  DCEs)	  (56,60,63).	  Case	  2	  BWS	  is	  also	  regarded	  highly	  over	  DCEs	  by	  some	  researchers	  as	  it	  obtains	  more	  information	  and	  is	  less	  cognitively	  burdensome	  compared	  to	  traditional	  DCEs.	  However,	  Case	  2	  BWS	  does	  not	  allow	  the	  estimation	  of	  welfare	  measures	  such	  as	  marginal	  rate	  of	  substitution	  or	  willingness	  to	  pay	  because	  such	  analysis	  requires	  the	  respondents	  to	  make	  discrete	  choices	  between	  alternatives,	  not	  within	  alternatives	  as	  seen	  in	  Case	  2	  BWS	  (59).	  	  Case	  3	  BWS,	  or	  multi-­‐profile-­‐based	  BWS	  is	  similar	  to	  traditional	  DCEs	  in	  that	  respondents	  are	  presented	  with	  choice	  task	  of	  multiple	  alternatives	  (59,60).	  Instead	  of	  having	  to	  pick	  the	  best	  alternative	  as	  in	  traditional	  DCEs,	  they	  are	  also	  required	  to	  choose	  the	  worst.	  Compared	  to	  traditional	  DCEs,	  Case	  3	  BWS	  yields	  more	  preference	  information	  as	  it	  asks	  respondents	  to	  reveal	  best-­‐worst	  pairs	  of	  alternatives.	  However,	  like	  traditional	  DCEs,	  a	  different	  utility	  scale	  is	  modeled	  for	  each	  attribute;	  thus,	  direct	  comparison	  of	  all	  attribute	  level	  like	  that	  seen	  in	  Case	  2	  BWS	  is	  not	  possible.	  	  	   	  Best-­‐worst	  scaling	  choice	  experiments	  are	  similar	  to	  DCEs	  in	  that	  both	  are	  deeply	  rooted	  in	  the	  random	  utility	  theory,	  where	  the	  utility	  of	  an	  attribute	  is	  proportional	  to	  the	  frequency	  it	  was	  chosen	  by	  the	  respondent	  in	  preference	  to	  other	  attribute	  (or	  attribute	  levels).	  Both	  DCEs	  and	  BWS	  have	  been	  used	  to	  value	  health	  states	  but	  also	  healthcare	  services,	  protocols,	  practices,	  14	  	  interventions	  and	  policies	  (64–69).	  Compared	  to	  the	  cardinal	  preference	  elicitation	  methods,	  ordinal	  techniques	  are	  relatively	  easier	  for	  respondents	  to	  comprehend	  and	  simulate	  a	  more	  real	  world	  decision-­‐making	  scenario	  where	  multiple	  attribute	  levels	  are	  being	  considered	  in	  making	  a	  decision.	  One	  area	  where	  more	  research	  is	  still	  needed	  in	  ordinal	  methods	  is	  how	  to	  incorporate	  ordinal	  preference	  weights	  for	  use	  in	  economic	  evaluation	  even	  though	  there	  has	  been	  some	  proposed	  methodology	  for	  transforming	  ordinal	  measures	  into	  cardinal	  estimates	  (70).	  	  	  One	  type	  of	  stated	  preference	  method	  that	  had	  not	  been	  discussed	  so	  far	  is	  conjoint	  analysis	  (CA),	  which	  is	  considered	  by	  many	  health	  researchers	  as	  a	  type	  of	  ordinal	  elicitation	  method.	  Similar	  to	  DCEs,	  CAs	  are	  also	  administered	  in	  the	  format	  of	  a	  survey	  consisting	  of	  a	  combination	  of	  varying	  attributes	  and	  levels,	  based	  on	  a	  full	  factorial	  or	  partial	  factorial	  design.	  The	  resemblance	  between	  the	  two	  methods	  has	  led	  to	  some	  misleading	  terminologies	  where	  DCEs	  are	  sometimes	  called	  “choice-­‐based	  conjoint	  analyses.”	  It	  is	  important	  to	  note,	  however,	  that	  the	  two	  elicitation	  methods	  are	  far	  from	  similar	  and	  a	  clear	  distinction	  should	  be	  made	  when	  making	  reference	  to	  DCEs	  as	  CAs.	  In	  contrast	  to	  DCEs,	  which	  are	  deeply	  rooted	  in	  random	  utility	  theory,	  CAs	  developed	  out	  of	  the	  theory	  of	  conjoint	  measurement	  (CM),	  a	  theory	  that	  is	  primarily	  concerned	  about	  the	  behaviour	  of	  algebraic	  processes	  in	  the	  ranking	  exercise,	  and	  not	  the	  behaviour	  of	  how	  choices	  are	  made.	  Because	  the	  axioms	  underpinning	  CM	  differ	  from	  classical	  economic	  theory,	  the	  preference	  data	  cannot	  be,	  and	  should	  not	  be	  readily	  translated	  into	  welfare	  measurements	  (i.e.	  willingness	  to	  pay).	  Another	  main	  criticism	  of	  CM	  theory	  is	  that	  15	  	  there	  is	  no	  consideration	  of	  error	  in	  the	  framework	  of	  CM,	  whereas	  in	  RUT,	  error	  was	  inherent	  in	  the	  utility	  component	  such	  that	  a	  large	  error	  would	  be	  expected	  to	  change	  the	  utility	  estimates.	  Recent	  decades	  have	  seen	  “enhancements”	  of	  CAs,	  including	  assigning	  ratings	  to	  choice	  profiles	  instead	  of	  rankings,	  or	  the	  use	  of	  different	  statistical	  models	  to	  analyze	  choice	  data.	  As	  Louviere	  et	  al.	  have	  pointed	  out,	  these	  modifications	  are	  all	  analytical	  enhancement	  and	  do	  no	  concern	  the	  decision	  making	  process.	  Hence,	  despite	  recent	  developments	  in	  CAs,	  there	  exist	  fundamental	  differences	  between	  CAs	  and	  DCEs	  and	  a	  distinction	  must	  be	  made	  between	  the	  two	  methods	  in	  the	  literature.	  	  	  1.4 Oral	  antithrombotics	  for	  stroke	  prevention	  in	  patients	  with	  atrial	  fibrillation	  Atrial	  fibrillation	  (AF)	  is	  the	  most	  common	  cardiac	  arrhythmia,	  affecting	  approximately	  200,000	  to	  250,000	  Canadians	  (71).	  The	  global	  prevalence	  of	  AF	  is	  estimated	  to	  be	  approximately	  1%,	  which	  is	  projected	  to	  increase	  by	  two-­‐	  to	  three-­‐fold	  by	  2050	  with	  the	  aging	  population	  (72–74).	  Atrial	  fibrillation	  accounts	  for	  approximately	  15-­‐25%	  of	  all	  ischemic	  strokes	  and	  is	  known	  to	  carry	  a	  two-­‐fold	  increase	  in	  mortality	  rate	  and	  a	  higher	  risk	  of	  vascular	  cognitive	  impairment	  than	  strokes	  unrelated	  to	  AF	  (75–81).	  The	  risk	  of	  stroke	  is	  different	  for	  each	  AF	  patient	  and	  may	  vary	  between	  0.2%	  per	  year	  to	  more	  than	  10%	  depending	  on	  risk	  factors	  (81–83).	  The	  CHADS2	  and	  CHA2DS2VASc	  are	  two	  of	  several	  schemes	  that	  were	  developed	  and	  validated	  to	  stratify	  the	  stroke	  risk	  in	  AF	  patients.	  These	  risk	  prediction	  calculators	  consider	  risk	  factors	  such	  as	  age,	  hypertension,	  diabetes,	  previous	  strokes,	  and	  other	  vascular	  diseases	  as	  important	  contributors	  to	  increased	  stroke	  risk	  (see	  Table	  1.2	  and	  1.3)	  	  16	  	  	  Currently,	  oral	  antithrombotics	  remain	  the	  cornerstone	  of	  stroke	  prevention	  in	  patients	  with	  AF.	  Acetylsalicylic	  acid	  (ASA)	  for	  patients	  at	  low	  risk	  of	  stroke	  and	  Vitamin	  K	  antagonists	  (VKA)	  such	  as	  warfarin	  for	  patients	  at	  moderate-­‐to-­‐high	  risk	  have	  been	  considered	  the	  treatments	  of	  choice	  for	  the	  past	  decades	  (84).	  While	  evidence	  suggests	  that	  both	  agents	  can	  reduce	  the	  risk	  of	  stroke,	  use	  of	  either	  agent	  is	  associated	  with	  an	  increased	  risk	  of	  a	  major	  bleed	  (e.g.	  upper	  GI	  bleed,	  hemorrhagic	  stroke),	  especially	  with	  warfarin	  (85–87).	  In	  the	  last	  three	  years,	  novel	  oral	  antithrombotics	  such	  as	  direct	  thrombin	  inhibitors	  (e.g.	  dabigatran),	  direct	  Xa	  inhibitors	  (e.g.	  rivaroxaban,	  apixaban)	  and	  combination	  therapy	  of	  clopidogrel	  and	  ASA	  have	  also	  been	  studied	  in	  AF	  patients	  as	  stroke	  prophylaxis	  (88–92).	  Some	  of	  these	  therapeutic	  agents	  have	  been	  found	  to	  be	  superior,	  or	  non-­‐inferior,	  to	  warfarin,	  the	  gold	  standard	  of	  therapy,	  in	  terms	  of	  efficacy;	  the	  most	  recent	  Canadian	  clinical	  practice	  guidelines	  for	  AF	  management	  recommend	  dabigatran	  as	  first-­‐line	  therapy	  (93).	  While	  these	  new	  therapies	  provide	  more	  options	  to	  clinicians	  and	  patients	  in	  choosing	  an	  antithrombotic	  for	  stroke	  prophylaxis,	  the	  rapid	  emergence	  of	  new	  drugs	  poses	  a	  challenging	  dilemma	  for	  clinical	  decision	  making	  as	  there	  is	  currently	  no	  head-­‐to-­‐head	  comparison	  of	  the	  new	  oral	  antithrombotics.	  With	  at	  least	  four	  more	  new	  oral	  antithrombotics	  in	  the	  clinical	  development	  process	  (94),	  clinicians	  and	  regulatory	  agencies	  will	  face	  even	  more	  challenges	  in	  determining	  the	  relative	  benefits	  and	  risks	  of	  these	  new	  drugs	  to	  determine	  the	  most	  appropriate	  treatment	  strategy.	  While	  conducting	  head-­‐to-­‐head	  comparison	  trials	  seems	  to	  be	  the	  obvious	  solution	  to	  this	  problem,	  this	  would	  be	  costly,	  labor	  intensive	  and	  time	  consuming	  such	  that	  a	  long	  period	  of	  time	  would	  have	  lapsed	  before	  17	  	  findings	  of	  such	  trials	  could	  be	  translated	  into	  clinical	  practice.	  Thus,	  a	  new	  method	  that	  is	  rapid,	  accurate,	  and	  cost-­‐effective	  in	  comparing	  oral	  antithrombotics	  is	  needed	  to	  solve	  this	  clinical	  predicament.	  	  1.5 Patient-­‐centered	  decision-­‐making:	  current	  knowledge	  gaps	  1.5.1 Lack	  of	  explicit	  integration	  of	  patient	  preference	  in	  clinical	  decision-­‐making	  Decision-­‐making	  at	  the	  physician-­‐patient	  level	  Currently,	  most	  clinicians	  view	  patient-­‐centered	  decision-­‐making	  as	  discussions	  that	  take	  place	  at	  the	  point-­‐of-­‐care	  between	  the	  patient	  and	  physician	  with	  or	  without	  prediction	  rules,	  risk	  calculators	  and	  decision	  aids	  (95–97).	  While	  this	  may	  sound	  straight	  forward,	  preferences	  elicited	  at	  this	  level	  are	  not	  entirely	  explicit	  for	  several	  reasons.	  One	  apparent	  shortcoming	  concerning	  preference	  elicitation	  at	  the	  physician-­‐patient	  level	  is	  inconsistency.	  Patient	  preferences	  elicited	  by	  different	  physicians	  can	  be	  highly	  variable	  as	  there	  is	  no	  standardized	  approach	  to	  eliciting	  preferences	  from	  patients.	  Such	  standardized	  approaches	  would	  also	  be	  difficult	  to	  establish	  as	  clinicians	  have	  their	  own	  communication	  styles	  and	  preconceived	  preferences	  for	  therapeutic	  alternatives.	  Also,	  the	  approach	  undertaken	  by	  the	  same	  clinician	  may	  also	  be	  highly	  inconsistent	  depending	  on	  the	  individual	  patients	  and	  circumstantial	  factors	  (i.e.	  logistics,	  physician	  workload,	  etc.).	  	  	  Apart	  from	  clinician	  factors,	  patient	  input	  on	  decision-­‐making	  has	  been	  shown	  to	  be	  highly	  variable	  depending	  on	  the	  individual’s	  background	  and	  psychosocial	  factors,	  which	  18	  	  consequently	  lead	  to	  inconsistent	  preference	  elicitation.	  Several	  studies	  have	  shown	  that	  patients	  who	  tend	  to	  be	  more	  anxious	  are	  less	  likely	  to	  participate	  in	  informed	  decision-­‐making	  (41,98,99).	  It	  is	  also	  typically	  assumed	  that	  patients	  want	  to	  participate	  in	  shared	  decision-­‐making	  but	  this	  is	  not	  necessarily	  the	  case.	  In	  a	  study	  by	  Swenson	  et	  al.,	  69%	  of	  the	  patients	  preferred	  patient-­‐centered	  approaches	  to	  decision-­‐making	  while	  the	  rest	  preferred	  a	  doctor-­‐centered	  style	  (100).	  In	  another	  4-­‐year	  observational	  study	  by	  Arora	  and	  McHorney,	  again	  one-­‐third	  of	  the	  patients	  preferred	  to	  leave	  their	  medical	  decisions	  to	  their	  physicians	  (101).	  This	  evidence	  does	  not	  imply	  the	  lack	  of	  preference	  or	  preference	  neutrality	  on	  the	  patient’s	  part,	  but	  on	  the	  contrary,	  the	  importance	  of	  using	  established	  instruments	  to	  formally	  elicit	  patients’	  valuation	  implicitly	  and	  explicitly.	  	  	  The	  use	  of	  decision	  aids	  or	  decision-­‐making	  support	  tools	  appears	  to	  be	  an	  attractive	  solution	  to	  standardizing	  patient	  preference	  elicitation	  at	  the	  physician-­‐patient	  level,	  however,	  the	  same	  problem	  of	  inconsistency	  applies	  with	  the	  use	  of	  these	  tools.	  Even	  though	  decision	  aids	  invite	  patient’s	  input	  into	  the	  decision,	  they	  are	  limited	  to	  situations	  when	  the	  clinicians	  decide	  to	  administer	  the	  tools	  and	  to	  patients	  who	  are	  able	  to	  use	  them.	  There	  are	  also	  issues	  concerning	  the	  evidence	  for	  such	  decision	  support	  tools.	  Firstly,	  there	  is	  no	  agreed	  method	  on	  identifying	  the	  appropriate	  evidence	  for	  decision	  support	  tools.	  Also,	  most	  evidence	  for	  decision	  support	  tools	  does	  not	  uphold	  the	  same	  rigor	  as	  clinical	  trials.	  This	  research	  is	  not	  registered	  in	  advance,	  and	  the	  outcomes	  can	  be	  easily	  modified	  and	  thus	  prone	  to	  selective	  reporting	  (19).	  It	  has	  also	  been	  pointed	  out	  that	  there	  is	  evidence	  of	  infiltration	  of	  panels	  producing	  decision-­‐support	  19	  	  tools	  by	  influential	  experts	  with	  strong	  conflict	  of	  interests	  (19,102–105).	  Overall,	  it	  is	  assumed	  that	  decision	  support	  tools	  would	  improve	  patient	  outcomes	  but	  there	  is	  a	  lack	  of	  evidence	  to	  support	  them.	  A	  review	  demonstrated	  that	  while	  decision	  aids	  may	  be	  helpful	  in	  improving	  patient	  knowledge,	  expectations,	  and	  reducing	  decisional	  conflict,	  they	  do	  not	  affect	  patient	  outcomes	  such	  as	  satisfaction,	  or	  health	  outcomes	  (106).	  Another	  problem	  with	  relying	  on	  decision	  support	  tools	  to	  elicit	  patient	  preferences	  is	  that	  the	  tools	  need	  to	  be	  updated	  with	  emerging	  evidence,	  which	  can	  be	  labour	  intensive	  and	  inefficient.	  	  	  The	  major	  issue	  with	  preference	  elicitation	  at	  the	  physician-­‐patient	  level,	  with	  or	  without	  decision	  support	  tools	  is	  that	  only	  preferences	  for	  a	  select	  number	  of	  risk	  and	  benefit	  parameters	  can	  be	  elicited	  before	  incurring	  huge	  burden	  on	  the	  patients.	  Most	  importantly,	  preferences	  elicited	  through	  this	  approach	  are	  not	  entirely	  explicit	  as	  patients	  are	  not	  asked	  to	  make	  simultaneous	  trade-­‐offs	  between	  risks	  and	  benefits	  –	  to	  make	  this	  trade-­‐off	  would	  require	  cognitive	  exercises	  from	  the	  patients’	  using	  validated	  instruments,	  which	  is	  typically	  beyond	  the	  capacity	  at	  the	  physician-­‐patient	  level.	  	  	  Decision-­‐making	  at	  the	  practice	  guideline	  level	  As	  discussed	  above,	  patient-­‐centered	  decision-­‐making	  at	  the	  physician-­‐patient	  level	  is	  subject	  to	  inconsistent	  application	  and	  is	  unlikely	  to	  fully	  represent	  the	  patient	  preference.	  Therefore,	  recommendations	  have	  been	  made	  to	  integrate	  patient	  preferences	  into	  clinical	  practice	  guidelines	  (CPGs)	  by	  many	  CPG	  organizations	  as	  well	  as	  the	  World	  Health	  Organization	  (30,39).	  20	  	  To	  date,	  some	  effort	  has	  been	  made	  to	  integrate	  patient	  input	  by	  involving	  consumers	  and	  patients	  on	  CPG	  development	  committees;	  however,	  this	  practice	  has	  had	  limited	  application	  to	  date	  (18,26,27,107,108).	  A	  recent	  study	  found	  that	  only	  20%	  of	  the	  730	  Canadian	  guideline	  committees	  included	  patients	  or	  consumers	  in	  the	  development	  process	  (108).	  In	  a	  study	  by	  Chong	  et	  al.	  that	  reviewed	  70	  CPGs	  endorsed	  by	  the	  Ontario	  Medical	  Association,	  only	  4%	  of	  the	  guideline	  text	  addressed	  preferences,	  compared	  to	  24%	  for	  effectiveness	  (27).	  The	  lack	  of	  evidence	  on	  patient	  preference	  in	  CPGs	  can	  be	  attributed	  to	  the	  limited	  access	  to	  formal	  analytic	  decision	  models	  by	  development	  committee	  as	  well	  as	  the	  infeasibility	  of	  conducting	  preference	  surveys	  at	  the	  time	  of	  guideline	  development	  (18,30).	  This	  calls	  for	  a	  new	  method	  of	  integrating	  patient	  preferences	  beyond	  the	  point-­‐of-­‐care	  encounter	  that	  uses	  objective	  preference	  measures	  and	  is	  easily	  accessible	  and	  utilizable	  by	  the	  guideline	  development	  committees.	  	  1.5.2 Limited	  patient	  preference	  data	  in	  atrial	  fibrillation	  While	  some	  progress	  has	  been	  made	  to	  investigate	  the	  benefit-­‐risk	  balance	  using	  patient	  preferences	  in	  atrial	  fibrillation,	  there	  are	  methodological	  shortcomings	  in	  those	  analyses	  (25,109–111).	  Gage	  et	  al.	  used	  a	  decision	  analysis	  stratified	  by	  the	  number	  of	  stroke	  risk	  factors	  to	  compare	  the	  potential	  benefits	  of	  preference-­‐based	  therapy	  with	  ‘warfarin-­‐for-­‐all	  therapy’	  (109).	  While	  patients’	  preferences	  elicited	  using	  a	  time	  trade-­‐off	  (TTO)	  method	  were	  incorporated	  into	  estimating	  quality-­‐adjusted	  life	  years	  (QALYs)	  gained	  and	  the	  cost	  of	  the	  two	  treatment	  strategies,	  the	  study	  only	  considered	  3	  different	  health	  states	  focusing	  on	  strokes	  21	  	  (i.e.	  mild,	  moderate	  and	  severe	  strokes)	  (112).	  Because	  of	  the	  lack	  of	  integration	  of	  potential	  harm	  associated	  with	  antithrombotic	  treatment	  in	  calculating	  the	  QALYs,	  this	  analysis	  did	  not	  completely	  quantify	  potential	  benefits	  and	  risks	  of	  all	  treatment	  options.	  In	  another	  study	  by	  Man-­‐Son-­‐Hing	  and	  colleagues,	  a	  decision-­‐analytic	  model	  using	  Markov	  subtrees	  was	  constructed	  to	  assess	  how	  the	  risk	  of	  upper	  gastrointestinal	  (GI)	  bleeding	  influences	  the	  choice	  of	  antithrombotic	  therapy	  in	  patients	  with	  AF	  (25).	  While	  the	  analysis	  considered	  patients’	  preferences	  for	  six	  health	  states	  including	  both	  harm	  (i.e.	  upper	  GI	  bleed)	  and	  benefit	  (i.e.	  ischemic	  stroke),	  other	  relevant	  harm-­‐related	  outcomes	  such	  as	  cerebral	  hemorrhage	  or	  lower	  GI	  bleed	  was	  not	  taken	  into	  account.	  In	  a	  systematic	  review	  by	  Maclean	  et	  al.,	  16	  studies	  were	  identified	  that	  looked	  at	  patient	  preferences	  in	  the	  AF	  population.	  Excluding	  the	  studies	  that	  utilized	  decision	  aids	  to	  elicit	  patient	  preferences,	  the	  rest	  of	  the	  studies	  measured	  preferences	  using	  either	  the	  TTO	  or	  SG	  methods	  (111).	  Preferences	  were	  found	  to	  be	  highly	  variable	  depending	  on	  the	  patient	  population	  and	  how	  the	  health	  states	  were	  defined.	  All	  of	  these	  studies	  elicited	  patient	  preferences	  for	  stroke.	  A	  few	  also	  measured	  minor	  stroke	  (or	  non-­‐debilitating	  stroke)	  and	  major	  bleed	  (most	  commonly	  described	  as	  a	  GI	  bleed).	  Given	  the	  valuation	  methods	  used,	  only	  preferences	  of	  clinical	  health	  states	  were	  elicited	  in	  those	  studies.	  Other	  non-­‐clinical	  features	  associated	  with	  antithrombotics	  that	  may	  influence	  patient	  decision-­‐making,	  for	  example	  frequency	  of	  monitoring,	  and	  cost,	  were	  also	  not	  examined	  in	  those	  studies.	  It	  is	  also	  important	  to	  point	  out	  that	  while	  TTO	  and	  SG	  are	  valid	  instruments	  for	  preference	  elicitation	  for	  chronic	  health	  states,	  they	  were	  not	  developed	  for	  valuation	  of	  temporary	  health	  states.	  Therefore,	  preferences	  elicited	  for	  bleeding	  using	  TTO	  and	  SG	  in	  those	  22	  	  studies	  may	  not	  be	  valid.	  This	  warrants	  another	  approach	  to	  measuring	  not	  preferences	  limited	  by	  the	  cardinal	  valuation	  methods.	  	  1.5.3 Limited	  data	  comparing	  patient	  and	  physician	  preferences	  	  The	  argument	  for	  incorporating	  patient	  preference	  into	  clinical	  decision-­‐making	  is	  based	  on	  the	  notion	  that	  there	  is	  a	  discrepancy	  in	  preferences	  between	  the	  physicians	  and	  patients,	  such	  that	  physicians	  would	  be	  a	  poor	  proxy	  for	  making	  decisions	  for	  patients.	  While	  there	  is	  some	  data	  to	  substantiate	  this	  assumption,	  evidence	  regarding	  congruence	  of	  preferences	  between	  physicians	  and	  patients	  has	  been	  shown	  to	  be	  highly	  heterogeneous	  depending	  on	  the	  acuity	  of	  the	  condition	  and	  the	  elicitation	  methods	  used	  (113).	  Further,	  many	  of	  the	  studies	  aimed	  at	  comparing	  patient	  and	  physician	  preferences	  are	  qualitative	  in	  nature	  (114–118),	  and	  it	  was	  not	  until	  recently	  that	  more	  investigations	  in	  this	  area	  have	  been	  done	  using	  utility-­‐based	  preference	  instruments.	  Even	  then,	  the	  majority	  of	  these	  studies	  elicited	  preferences	  using	  cardinal	  methods,	  which	  can	  only	  study	  the	  preference	  of	  one	  health	  state/outcome	  of	  interest	  at	  a	  time,	  and	  have	  limited	  capacity	  to	  simulate	  real-­‐life	  decision-­‐making	  where	  trade-­‐off	  are	  being	  made	  between	  multiple	  features	  of	  an	  intervention	  simultaneously.	  This	  is	  an	  important	  knowledge	  gap	  to	  address.	  The	  aim	  of	  this	  study	  is	  to	  find	  out:	  1)	  whether	  there	  is	  a	  significant	  difference	  in	  patients	  and	  physician	  preferences;	  and	  2)	  does	  this	  difference	  in	  preference	  between	  the	  two	  parties,	  if	  any,	  may	  potentially	  affect	  their	  preference	  for	  selecting	  an	  oral	  antithrombotic	  for	  stroke	  prophylaxis.	  	  23	  	  1.6 Thesis	  objectives	  and	  organization	  To	  address	  the	  many	  knowledge	  gaps	  concerning	  preferences	  in	  patient-­‐centered	  decision-­‐making,	  the	  overall	  objective	  of	  this	  thesis’	  is	  to	  determine	  physicians’	  and	  patients’	  preferences	  for	  stroke	  prophylaxis	  in	  atrial	  fibrillation.	  The	  specific	  objectives	  for	  this	  thesis	  are	  outlined	  as	  the	  following:	  1. To	  quantify,	  using	  the	  time	  trade-­‐off	  (TTO)	  and	  the	  best	  worst	  scaling	  (BWS)	  choice	  experiment,	  AF	  patients’	  preferences	  for	  attributes	  related	  to	  stroke	  prophylaxis;	  2. To	  quantify,	  using	  the	  BWS	  choice	  experiment,	  physicians’	  preferences	  for	  antithrombotics	  in	  AF	  patients;	  	  3. To	  evaluate	  the	  factors,	  if	  present	  that	  are	  associated	  with	  patients’	  and	  physicians’	  preferences	  for	  stroke	  prophylaxis;	  4. To	  compare	  AF	  patients’	  and	  physicians’	  preferences	  for	  stroke	  prophylaxis	  and	  to	  evaluate	  how	  the	  difference	  in	  preference	  may	  affect	  therapeutic	  choices	  of	  currently	  available	  antithrombotics.	  	  In	  this	  chapter,	  I	  have	  described	  the	  current	  state	  of	  integrating	  patients’	  preferences	  in	  decision-­‐making.	  I	  have	  also	  outlined	  the	  theories	  and	  methods	  for	  preference	  elicitation.	  	  In	  chapter	  2,	  I	  presented	  the	  results	  of	  a	  literature	  review	  that	  examined	  studies	  comparing	  physicians’	  and	  patients’	  preferences	  in	  the	  context	  of	  clinical	  decision-­‐making.	  	  In	  chapter	  3,	  I	  reported	  the	  findings	  of	  the	  part	  of	  the	  study	  that	  measured	  the	  preferences	  of	  AF	  patients	  for	  health	  states	  and	  features	  related	  to	  stroke	  prophylaxis	  using	  the	  TTO	  and	  BWS	  24	  	  methods.	  In	  the	  same	  study,	  I	  also	  measured	  the	  utilities	  of	  temporary	  health	  states	  (i.e.	  major	  and	  minor	  bleed)	  associated	  with	  antithrombotics	  using	  a	  chained-­‐TTO	  method.	  	  	  In	  chapter	  4,	  I	  reported	  the	  findings	  of	  the	  part	  of	  the	  study	  that	  measured	  the	  preferences	  of	  physicians	  for	  health	  states	  and	  features	  related	  to	  stroke	  prophylaxis	  using	  the	  BWS	  method.	  	  	  In	  chapter	  5,	  I	  compared	  the	  preferences	  between	  the	  patients’	  and	  physicians’	  and	  explored	  how	  differences	  in	  preferences	  may	  lead	  to	  different	  decision	  in	  choosing	  an	  antithrombotics	  for	  stroke	  prevention.	  	  Finally,	  in	  chapter	  6,	  I	  provide	  a	  conclusion	  that	  summarizes	  the	  significance	  of	  these	  preference	  data	  using	  the	  valuation	  methods	  described.	  I	  also	  discuss	  how	  such	  data	  would	  impact	  integration	  of	  patient	  preferences	  in	  clinical	  decision-­‐making.	  	  	  	  	  	  	  	  	  25	  	  	  Table	  1.1	  Axioms	  of	  von	  Neumann-­‐Morgenstern	  utility	  theory	  (32)	  1. Preferences	  exist	  and	  are	  transitive.	  For	  any	  pair	  of	  risky	  prospects	  y	  and	  y’	  either	  y	  is	  preferred	  to	  y’,	  y’	  is	  preferred	  to	  y	  or	  the	  individual	  is	  indifferent	  between	  y	  	  and	  y’.	  In	  addition,	  for	  any	  three	  risky	  prospects,	  y,	  y’	  and	  y’’,	  if	  y	  is	  preferred	  to	  y’,	  and	  y’	  is	  preferred	  to	  y’’,	  then	  y	  is	  preferred	  to	  y’’;	  similarly,	  if	  y	  is	  indifferent	  to	  y’,	  and	  y’	  is	  indifferent	  to	  y’’,	  then	  y	  is	  indifferent	  to	  y’’.	  	  2. Independence.	  An	  individual	  should	  be	  indifferent	  between	  a	  two-­‐stage	  risky	  prospect	  and	  its	  probabilistically	  equivalent	  one-­‐stage	  counterpart	  derived	  using	  the	  ordinary	  laws	  of	  probability.	  For	  example,	  consider	  two	  risky	  prospects	  y	  and	  y’	  where	  y	  is	  made	  up	  of	  outcome	  x1	  with	  probability	  p1	  and	  outcome	  x2	  with	  probability	  (1-­‐p1),	  indicated	  symbolically	  as	  y={p1,	  x1,	  x2},	  and	  y’	  =	  {p2,	  x1,	  x2}.	  This	  axiom	  implies	  that	  an	  individual	  would	  be	  indifferent	  between	  the	  two-­‐stage	  risky	  prospect	  (p,	  y,	  y’),	  and	  its	  probabilistically	  equivalent	  one-­‐stage	  counterpart	  {pp1	  +	  (1-­‐p)p2,x1,x2}.	  	  3. Continuity	  of	  preferences.	  If	  there	  are	  three	  outcomes	  such	  that	  x1	  is	  preferred	  to	  x2,	  which	  is	  preferred	  to	  x3,	  there	  is	  some	  probability	  p	  at	  which	  the	  individual	  is	  indifferent	  between	  outcome	  x2	  with	  certainty	  or	  receiving	  the	  risky	  prospect	  made	  up	  of	  outcome	  x1	  with	  probability	  p	  and	  outcome	  x3	  with	  probability	  1-­‐p.	  	  	  	  	  	  	  	  	  	  	  	  26	  	  Table	  1.2	  CHADS2	  	  risk	  criteria	  	   Condition	   Points	  C	   Congestive	  heart	  failure	   1	  H	   Hypertension:	  blood	  pressure	  consistently	  above	  140/90	  mmHg	  (or	  on	  antihypertensives)	  1	  A	   Age	  ≥	  75	  years	   1	  D	   Diabetes	  mellitus	   1	  S2	   Prior	  stroke	  or	  TIA	  or	  thromboembolism	   1	  	  Table	  1.3	  CHA2DS2-­‐VASc	  risk	  criteria	  	   Condition	   Points	  C	   Congestive	  heart	  failure	  (or	  left	  ventricular	  systolic	  dysfunction)	  1	  H	   Hypertension:	  blood	  pressure	  consistently	  above	  140/90	  mmHg	  (or	  on	  antihypertensives)	  1	  A2	   Age	  ≥	  75	  years	   2	  D	   Diabetes	  mellitus	   1	  S2	   Prior	  stroke	  or	  TIA	  or	  thromboembolism	   1	  V	   Vascular	  disease	  (e.g.	  peripheral	  artery	  disease,	  myocardial	  infarction,	  aortic	  plaque)	  1	  A	   Age	  65-­‐74	  years	   1	  Sc	   Sex	  category	  (i.e.	  female)	   1	  	  	  27	  	  Chapter	  2: Comparison	  of	  patient	  preferences	  and	  physician	  judgments	  in	  clinical	  decision-­‐making:	  a	  literature	  review	  2.1 Introduction	  Patient	  involvement	  in	  the	  various	  levels	  of	  decision-­‐making	  has	  been	  increasingly	  recognized	  as	  an	  important	  element	  in	  the	  patient-­‐centered	  care	  model.	  Incorporation	  of	  patient	  values	  in	  making	  clinical	  decision	  is	  now	  being	  upheld	  as	  the	  ideal	  standard	  of	  shared	  decision-­‐making.	  In	  addition	  to	  inviting	  patient	  input	  at	  the	  point	  of	  care	  contact,	  some	  have	  also	  advocated	  for	  the	  incorporation	  of	  population	  data	  of	  patient	  values	  in	  the	  guideline	  development	  process	  (26,30),	  which	  is	  traditionally	  viewed	  as	  a	  paternalistic	  approach	  to	  guide	  clinician	  decision-­‐making	  using	  clinical	  trial-­‐based	  evidence	  (1,8).	  While	  the	  concept	  has	  been	  generally	  well	  received	  by	  healthcare	  policy	  makers	  and	  guideline	  authorities,	  it	  has	  not	  been	  done	  in	  practice.	  Aside	  from	  the	  absence	  of	  a	  systematic	  approach	  to	  incorporating	  patient	  values	  into	  guideline	  recommendations,	  there	  remain	  some	  fundamental	  questions	  surrounding	  the	  integration	  of	  preference	  data.	  While	  it	  is	  acknowledged	  that	  patient	  preference	  is	  important,	  other	  stakeholders,	  such	  as	  physicians	  should	  also	  be	  considered	  as	  their	  judgment	  plays	  a	  significant	  role	  in	  the	  decision-­‐making	  process.	  In	  other	  words,	  it	  is	  sensible	  to	  evaluate	  first	  how	  patient	  and	  physician	  preferences	  might	  differ,	  if	  at	  all	  before	  any	  universal	  approach	  to	  integrating	  patient	  preferences	  on	  a	  systematic	  level	  (i.e.	  in	  the	  clinical	  practice	  guidelines).	  This	  literature	  review	  will	  examine	  the	  evidence	  to	  date	  pertaining	  to	  this	  subject.	  	  	  28	  	  A	  review	  on	  this	  topic	  was	  recently	  published	  by	  Muhlbacher	  and	  Juhnke	  (113),	  however,	  several	  new	  studies	  have	  emerged	  since	  then.	  In	  addition	  to	  the	  inclusion	  of	  new	  studies,	  this	  review	  will	  also	  have	  the	  focus	  of	  evaluating	  only	  studies	  that	  elicited	  preferences	  using	  ordinal	  methods.	  The	  rationale	  is	  that	  with	  conventional	  cardinal	  scales,	  only	  preferences	  for	  one	  or	  a	  limited	  few	  health	  outcomes	  can	  be	  elicited	  at	  a	  time.	  It	  is	  apparent	  that	  patients’	  priorities	  and	  preferences	  are	  not	  limited	  to	  the	  clinical	  outcomes	  but	  also	  non-­‐health	  related	  features	  associated	  with	  an	  intervention	  or	  therapy.	  Ordinal	  elicitation	  methods	  have	  the	  advantage	  of	  evaluating	  trade-­‐offs	  the	  respondents	  are	  willing	  to	  make	  amongst	  multiple	  outcomes	  or	  features	  associated	  with	  the	  health	  states/services/products.	  Hence,	  this	  study	  has	  thus	  chosen	  to	  include	  only	  studies	  that	  utilized	  ordinal	  elicitation	  methods	  as	  they	  comprehensively	  examine	  the	  scope	  of	  preferences	  not	  limited	  to	  just	  one	  or	  two	  health	  outcomes.	  	  	  2.2 Methods	  A	  systematic	  search	  in	  Medline,	  PubMed,	  EMBASE,	  Google	  Scholar	  and	  the	  Cochrane	  database	  was	  carried	  out	  using	  the	  search	  terms	  ‘patient	  preferences’,	  ‘patient	  values’,	  ‘physician	  values’,	  ‘physician	  judgment’,	  ‘physician	  preferences’,	  ‘ordinal	  preferences’,	  ‘conjoint	  analysis’,	  ‘discrete	  choice	  experiment’,	  ‘choice	  experiment’,	  ‘best-­‐worst	  scaling’,	  starting	  from	  the	  inception	  of	  each	  databases	  up	  to	  August	  2013.	  Terms	  were	  initially	  combined	  to	  generate	  studies	  that	  included	  both	  patient	  and	  physician	  preferences;	  however,	  I	  found	  the	  result	  yielded	  to	  be	  restricted.	  Thus,	  the	  final	  search	  was	  done	  by	  collapsing	  the	  terms	  ‘physician’,	  ‘preferences’,	  ‘values’,	  and	  ‘judgment’.	  This	  also	  allowed	  me	  to	  capture	  studies	  that	  separately	  published	  29	  	  preference	  data	  for	  the	  physician	  and	  patients.	  I	  searched	  the	  World	  Wide	  Web	  for	  grey	  literature.	  Manual	  search	  of	  articles	  using	  references	  pulled	  were	  also	  done.	  	  	  I	  included	  studies	  published	  in	  English	  that	  compared	  physician	  and	  patient	  preferences	  using	  any	  one	  of	  the	  ordinal	  elicitation	  methods	  including	  conjoint	  analysis,	  discrete	  choice	  experiment,	  best-­‐worst	  scaling	  or	  other	  ranking	  exercises.	  A	  list	  of	  abstracts	  of	  all	  potential	  studies	  was	  first	  generated.	  All	  abstracts	  were	  reviewed	  and	  determined	  for	  inclusion	  in	  the	  final	  review.	  Studies	  that	  were	  selected	  for	  final	  inclusion	  in	  the	  review	  were	  then	  evaluated	  individually	  with	  regards	  to	  elicitation	  method	  and	  development,	  population	  recruitment,	  attribute	  and	  level	  selection,	  statistical	  analysis	  and	  results.	  I	  then	  extracted	  from	  each	  study	  the	  following:	  study	  location,	  study	  population,	  health	  condition/services/intervention	  of	  interest,	  attributes	  and	  levels,	  elicitation	  methods,	  statistical	  methods	  and	  results.	  	  	  2.3 Findings	  2.3.1 Included	  studies	  and	  overall	  findings	  I	  found	  a	  total	  of	  29	  studies	  that	  matched	  the	  inclusion	  criteria	  (119–147).	  Two	  studies	  published	  the	  preference	  results	  for	  physicians	  and	  patients	  separately,	  thus	  amounting	  to	  27	  comparisons	  between	  the	  two	  groups	  of	  interest.	  The	  studies	  are	  summarized	  in	  Table	  2.1.	  The	  studies	  were	  conducted	  in	  various	  countries,	  mostly	  in	  Western	  Europe,	  North	  America	  and	  Australia.	  Of	  the	  27	  comparisons,	  10	  and	  17	  studies	  elicited	  preferences	  using	  CAs	  and	  DCEs,	  respectively.	  No	  studies	  used	  the	  best-­‐worst	  scaling	  method.	  Most	  of	  the	  CAs	  employed	  30	  	  random	  effects	  probit	  models	  and	  most	  of	  the	  DCEs	  used	  conditional	  logit	  and	  mixed	  logit	  models.	  All	  the	  studies	  published	  after	  2010	  were	  DCEs.	  There	  were	  13	  comparisons	  that	  evaluated	  preferences	  related	  to	  acute	  medical	  conditions	  (121,124,126,127,129,132,138–141,144,145,147),	  three	  of	  which	  involved	  surgical	  interventions	  (126,139,140).	  Ten	  comparisons	  elicited	  preferences	  related	  to	  chronic	  conditions	  (119,125,127,128,130,131,133–135,137,142,146)	  and	  five	  comparisons	  evaluated	  preferences	  associated	  with	  diagnostics	  and	  screening	  tests	  (120,122,123,136,143).	  	  Most	  of	  the	  comparisons	  used	  the	  same	  survey	  design	  for	  patients	  and	  physicians,	  thus	  allowing	  direct	  comparison	  of	  preferences	  between	  the	  two	  groups.	  However,	  three	  studies	  (122,136,146)	  had	  different	  attributes	  and	  designs	  for	  the	  physician	  and	  patient	  surveys.	  In	  the	  study	  by	  Sassi	  and	  McDaid,	  two	  conjoint	  models	  were	  designed	  to	  assess	  the	  preference	  for	  cardiac	  risk	  assessment	  in	  asymptomatic	  patients	  in	  the	  United	  Kingdom	  and	  Italy	  (122).	  The	  authors	  concluded	  that	  in	  evaluating	  various	  aspects	  of	  a	  cardiac	  risk	  assessment,	  the	  patients	  placed	  significantly	  higher	  preference	  on	  accurate	  prediction	  of	  their	  cardiac	  risk	  than	  the	  physicians	  compared	  to	  their	  preference	  for	  clinical	  values.	  However,	  the	  magnitude	  of	  this	  observation	  may	  be	  challenged	  as	  direct	  comparison	  was	  not	  appropriate	  between	  the	  two	  groups	  as	  the	  surveys	  were	  of	  two	  different	  CA	  model	  designs.	  The	  general	  practitioners	  answered	  a	  conjoint	  analysis	  survey	  that	  included	  three	  attributes	  (perception	  of	  resource	  commitment,	  prognostic	  value,	  expected	  risk	  reduction	  after	  a	  preventive	  intervention)	  while	  the	  patient	  group	  participated	  in	  the	  survey	  that	  had	  four	  attributes	  (modality	  of	  the	  31	  	  assessment,	  preventive	  interventions,	  prognostic	  value,	  expected	  risk	  reduction	  after	  preventive	  intervention).	  	  Similar	  case	  applied	  to	  the	  study	  by	  Fiebig	  et	  al.,	  where	  female	  patients	  and	  general	  practitioners	  responded	  to	  discrete	  choice	  experiments	  evaluating	  their	  preferences	  for	  pap	  tests	  as	  screening	  tool	  for	  cervical	  cancer	  (136).	  Some	  attributes	  (time	  since	  last	  cervical	  screening	  test,	  recommended	  screening	  interval,	  cost,	  false	  negative	  and	  positive)	  were	  common	  for	  both	  groups	  while	  all	  other	  attributes	  were	  different	  (four	  additional	  ones	  for	  female	  patients	  and	  five	  for	  GPs).	  The	  investigators	  recognized	  that	  comparing	  the	  estimates	  of	  these	  common	  attributes	  would	  be	  inappropriate	  as	  difference	  in	  estimates	  may	  be	  due	  to	  differences	  in	  scale	  for	  each	  of	  the	  DCEs.	  They	  thus	  compared	  the	  attributes	  by	  ratios	  of	  coefficients	  and	  by	  calculating	  the	  marginal	  rates	  of	  substitution.	  Similarly,	  in	  the	  study	  by	  Chancellor	  et	  al.,	  two	  different	  DCEs	  were	  designed	  to	  evaluate	  the	  preferences	  of	  physicians	  and	  patients	  in	  the	  use	  of	  opioids	  for	  chronic	  pain	  (146).	  The	  authors	  acknowledged	  that	  since	  direct	  quantitative	  comparison	  between	  the	  two	  groups	  would	  not	  be	  appropriate,	  only	  qualitative	  observations	  can	  be	  made	  about	  the	  consistency	  in	  preference	  between	  the	  two	  groups.	  	  	  Overall,	  10	  comparisons	  showed	  disagreement	  between	  physician	  and	  patient	  preferences	  (122,125,126,128,135,137,140,141,143,144),	  seven	  comparison	  showed	  similar	  preferences	  (121,124,127,130–132,136,147),	  while	  the	  remaining	  10	  indicated	  similar	  ranking	  in	  preferences	  but	  differences	  in	  strength	  of	  preference	  between	  the	  two	  groups	  (119,120,123,129,133,134,138,139,142,145,146).	  Table	  2.2	  summarizes	  the	  congruence	  32	  	  between	  patient	  and	  physician	  preferences	  for	  each	  trial	  by	  acuity	  of	  the	  health	  conditions,	  where	  ‘D’	  denotes	  difference	  in	  preferences	  between	  patients	  and	  physicians	  as	  assessed	  by	  their	  ranking	  of	  attributes,	  ‘S’	  implies	  similar	  preferences	  between	  the	  two	  groups	  and	  ‘DS’	  means	  significant	  difference	  in	  strength	  of	  preferences	  between	  patients	  and	  physicians.	  	  	  2.3.2 Congruence	  of	  preferences	  by	  elicitation	  methods	  It	  is	  difficult	  to	  make	  any	  definitive	  conclusion	  about	  congruence	  of	  preferences	  by	  elicitation	  methods	  as	  the	  health	  conditions/intervention,	  attributes,	  study	  population,	  statistical	  analysis	  varied	  vastly	  between	  the	  included	  studies.	  All	  variables	  aside,	  it	  did	  not	  appear	  that	  congruence	  between	  patient	  and	  physician	  preference	  was	  associated	  with	  the	  methods	  of	  preference	  elicitation.	  Out	  of	  the	  ten	  comparisons	  using	  CAs,	  four	  demonstrated	  disagreement,	  three	  in	  agreement,	  and	  three	  comparisons	  that	  were	  similar	  in	  ranking	  but	  different	  in	  strength	  (119–128).	  For	  the	  17	  DCEs,	  six	  demonstrated	  disagreement,	  four	  in	  agreement	  and	  the	  remaining	  seven	  studies	  showed	  patient	  and	  physician	  preferences	  were	  similar	  in	  ranking,	  but	  different	  in	  strength	  (129–147).	  No	  apparent	  trend	  could	  be	  seen	  when	  comparing	  studies	  that	  evaluated	  preferences	  of	  same	  condition	  but	  with	  different	  elicitation	  methods.	  Both	  of	  the	  studies	  that	  examined	  pregnant	  women	  and	  physician	  preferences	  for	  Down’s	  syndrome	  screening	  test	  using	  CA	  showed	  similar	  ranking	  of	  attributes	  but	  different	  strengths	  in	  preference	  (120,123).	  On	  the	  other	  hand,	  one	  DCE	  by	  Hill	  et	  al.	  identified	  disagreement	  of	  preferences	  where	  pregnant	  women	  valued	  risk	  of	  miscarriage	  to	  be	  more	  important	  than	  accuracy	  of	  test	  whereas	  it	  was	  the	  reverse	  (143).	  Preference	  for	  chronic	  pain	  management	  33	  	  with	  opioids	  was	  evaluated	  by	  Gregorian	  et	  al.	  using	  CA,	  and	  Chancellor	  et	  al.	  using	  the	  DCE	  design	  (127,146).	  Results	  from	  the	  CA	  survey	  showed	  agreement	  in	  preference	  between	  physicians	  and	  patients	  where	  both	  nausea	  and	  vomiting	  were	  at	  least	  as	  important	  as	  pain	  relief	  to	  both	  physicians	  and	  patients	  (127).	  Qualitative	  comparison	  between	  the	  two	  groups	  from	  the	  DCE	  also	  suggested	  similar	  preferences	  but	  possibly	  with	  different	  strengths	  (146).	  This	  latter	  study	  by	  Chancellor	  et	  al.,	  however,	  had	  different	  survey	  design	  and	  attributes	  for	  the	  physician	  and	  patient	  groups,	  and	  the	  comparison	  between	  the	  two	  groups	  was	  at	  most	  speculative.	  	  	  2.3.3 Congruence	  of	  preferences	  by	  acuity	  of	  conditions	  	  The	  27	  comparisons	  were	  also	  grouped	  according	  to	  the	  acuity	  of	  health	  conditions	  associated	  with	  the	  preferences	  elicited	  and	  evaluated	  for	  any	  trend	  in	  preference	  agreement	  or	  disagreement	  between	  physicians	  and	  patients.	  Out	  of	  the	  13	  comparisons	  that	  evaluated	  preferences	  associated	  with	  acute	  conditions,	  four	  identified	  disagreement	  between	  physicians	  and	  patients,	  five	  in	  agreement	  and	  the	  remaining	  four	  in	  agreement	  but	  differences	  in	  strength	  of	  the	  preferences	  (121,124,126,127,129,132,138–141,144,145,147).	  Out	  of	  the	  nine	  comparisons	  that	  examined	  preferences	  in	  chronic	  conditions,	  four	  showed	  disagreement,	  two	  in	  agreement,	  and	  the	  remaining	  three	  in	  agreement	  but	  difference	  in	  strength	  of	  preferences	  (119,125,128,130,131,133–135,137,142,146).	  Similar	  to	  the	  assessment	  by	  elicitation	  methods,	  no	  apparent	  trend	  of	  agreement	  or	  disagreement	  can	  be	  observed	  based	  on	  the	  acuity	  of	  medical	  condition.	  Five	  comparisons	  elicited	  preferences	  related	  to	  risk	  assessment	  and	  34	  	  screening	  tests	  (120,122,123,136,143).	  Again,	  no	  trend	  of	  agreement	  or	  disagreement	  between	  physicians	  and	  patients	  can	  be	  made.	  When	  limiting	  the	  evaluation	  to	  comparisons	  that	  utilized	  the	  same	  methodology,	  no	  further	  information	  on	  preference	  congruence	  was	  gained	  either.	  One	  of	  the	  four	  CAs	  in	  acute	  conditions	  resulted	  in	  disagreement,	  while	  two	  of	  the	  four	  in	  chronic	  conditions	  resulted	  in	  disagreement.	  Similarly	  for	  DCEs,	  two	  of	  the	  six	  comparisons	  versus	  three	  of	  nine	  comparisons	  showed	  disagreement	  of	  preference	  in	  chronic	  and	  acute	  conditions,	  respectively.	  	  	  2.4 Discussion	  In	  the	  era	  of	  patient-­‐centered	  care,	  much	  of	  the	  discussion	  is	  focused	  on	  integrating	  patient	  preferences	  in	  making	  health	  policy	  decisions	  on	  all	  levels.	  However,	  it	  can	  be	  argued	  that	  the	  preferences	  of	  all	  decision	  stakeholders	  should	  be	  considered	  in	  the	  process.	  This	  review	  is	  an	  updated	  evaluation	  of	  the	  literature	  that	  has	  evaluated	  differences	  in	  preferences	  between	  physicians	  and	  patients,	  the	  two	  key	  decision	  stakeholders	  in	  making	  choices	  about	  the	  individual’s	  health	  interventions.	  Twenty-­‐seven	  comparisons	  that	  evaluated	  the	  congruence	  of	  patient	  and	  physician	  preferences	  using	  ordinal	  methods,	  either	  CA	  or	  DCE	  were	  identified.	  This	  review	  found	  that	  the	  published	  studies	  are	  highly	  heterogeneous	  in	  terms	  of	  population,	  disease	  states,	  attributes	  and	  the	  statistical	  analyses	  used,	  and	  it	  is	  therefore	  difficult	  to	  conduct	  a	  systematic	  review	  of	  the	  studies	  on	  a	  quantitative	  level.	  By	  means	  of	  qualitative	  observations,	  it	  did	  not	  appear	  that	  there	  was	  definitive	  congruence	  or	  incongruence	  between	  patient	  and	  physician	  preferences.	  However,	  it	  did	  seem	  that	  most	  studies	  showed	  some	  35	  	  difference	  in	  preferences	  between	  the	  two	  stakeholder	  groups,	  if	  not	  in	  ranking	  of	  attributes,	  then	  in	  the	  strengths	  of	  preference,	  compared	  to	  ten	  out	  of	  the	  27	  comparisons	  that	  suggested	  disagreement	  between	  patient	  and	  physician	  preferences.	  	  	  This	  review	  also	  explored	  the	  possibility	  of	  congruence	  in	  preference	  in	  accordance	  to	  the	  acuity	  of	  disease	  states	  associated	  with	  the	  preference	  experiment,	  but	  found	  no	  clear	  trend	  to	  describe	  the	  results.	  This	  observation	  suggests	  a	  departure	  from	  the	  construal	  level	  theory,	  which	  proposes	  that	  the	  way	  an	  individual	  makes	  decisions	  is	  dependent	  on	  the	  psychological	  distance	  from	  where	  they	  are	  to	  the	  time	  of	  the	  likely	  outcome	  or	  a	  health	  event	  (148).	  The	  further	  the	  event,	  say	  for	  cardiovascular	  events	  prevention	  in	  a	  newly	  diagnosed	  diabetic,	  the	  more	  abstract	  the	  decision	  becomes	  and	  the	  more	  objectivity	  is	  involved	  in	  making	  the	  decision.	  In	  contrast,	  the	  more	  immediate	  the	  associated	  health	  outcome,	  say	  for	  the	  management	  of	  postoperative	  pain	  in	  a	  patient	  waiting	  for	  surgery,	  the	  more	  concrete	  the	  decision	  and	  thus	  the	  more	  emotional	  the	  individual	  is	  in	  making	  the	  choice.	  Based	  on	  construal	  level	  theory,	  the	  hypothesis	  is	  that	  for	  chronic	  illnesses	  where	  the	  event	  is	  usually	  further	  away	  on	  the	  spectrum	  of	  the	  disease,	  that	  one	  could	  expect	  to	  see	  congruence	  in	  preference	  between	  patients	  and	  physicians	  as	  both	  of	  them	  would	  treat	  the	  decision	  in	  an	  abstract	  manner.	  On	  the	  contrary,	  when	  dealing	  with	  acute	  conditions	  or	  situations	  where	  the	  intervention	  is	  immediate,	  there	  could	  be	  a	  disagreement	  in	  preferences	  between	  the	  two	  stakeholders	  as	  physicians	  will	  still	  treat	  the	  decision	  abstractedly	  while	  the	  patients	  will	  take	  the	  decision-­‐making	  process	  on	  a	  more	  concrete	  level.	  The	  results	  of	  this	  review	  did	  not	  seem	  36	  	  to	  corroborate	  with	  the	  hypothesis	  of	  the	  construal	  level	  theory.	  Even	  when	  the	  results	  was	  further	  narrowed	  down	  to	  examining	  choice	  experiments	  of	  the	  same	  elicitation	  methods,	  it	  was	  still	  unclear	  whether	  the	  basis	  of	  the	  theory	  can	  be	  applied.	  	  	  Similarly,	  when	  evaluating	  congruence	  of	  preference	  by	  types	  of	  elicitation	  method,	  no	  apparent	  direction	  of	  agreement	  or	  disagreement	  between	  the	  patients	  and	  physicians	  was	  found,	  as	  both	  CAs	  and	  DCEs	  revealed	  about	  the	  same	  proportion	  of	  comparisons	  that	  resulted	  in	  agreement	  versus	  disagreement,	  bearing	  in	  mind	  that	  there	  were	  few	  studies	  available	  to	  make	  any	  quantitative	  comparisons.	  Nine	  new	  studies	  that	  were	  not	  included	  in	  the	  review	  by	  Muhlbacher	  et	  al.	  (113)	  were	  found.	  All	  of	  those	  studies	  were	  DCEs,	  suggesting	  the	  popularity	  the	  choice	  experiments	  have	  gained	  most	  recently.	  One	  can	  argue	  that	  DCE	  are	  inherently	  superior	  to	  CA	  as	  a	  preference	  elicitation	  method,	  as	  the	  former	  is	  based	  on	  the	  random	  utility	  theory	  (RUT),	  a	  well-­‐tested	  theory	  that	  describes	  the	  choice	  behavior.	  In	  fact,	  critiques	  of	  the	  CA	  design	  targeted	  the	  basis	  of	  CA,	  which	  was	  evolved	  from	  the	  theory	  of	  conjoint	  measurement	  that	  centers	  on	  the	  “behavior	  of	  sets	  of	  numbers	  in	  response	  to	  factorial	  manipulations	  of	  factor	  levels,”	  where	  “error	  components	  are	  largely	  ad	  hoc	  and	  lack	  clear	  interpretations.”	  (149).	  In	  contrast,	  the	  random	  utility	  theory	  that	  DCEs	  and	  BWS	  are	  based	  on	  is	  being	  seen	  as	  a	  comprehensive	  behavioral	  theory	  with	  the	  inclusion	  of	  a	  latent,	  unobservable	  utility	  with	  each	  choice	  made	  by	  the	  individual.	  This	  random,	  latent	  variable	  takes	  into	  account	  the	  unidentified	  factors	  that	  can	  affect	  choices,	  and	  therefore,	  make	  the	  utilities	  inherently	  stochastic.	  This	  reflects	  the	  real-­‐world	  decision	  process,	  where	  other	  variables,	  unobservable	  to	  37	  	  researchers	  and	  the	  individuals	  themselves,	  can	  impact	  an	  individual’s	  choices.	  	  Recent	  development	  in	  RUT-­‐based	  choice	  experiment	  has	  drawn	  the	  attention	  to	  BWS	  design,	  which	  can	  be	  seen	  as	  an	  extension	  of	  DCE	  with	  additional	  merits.	  	  Not	  only	  is	  BWS	  design	  deemed	  to	  be	  less	  cognitively	  burdensome,	  it	  also	  allows	  the	  direct	  comparison	  of	  preferences	  across	  all	  attribute	  levels	  (56,150,151).	  Interestingly,	  no	  study	  was	  found	  to	  evaluate	  congruence	  of	  preferences	  of	  physician	  and	  patients	  using	  this	  methodology.	  	  	  This	  review	  is	  one	  of	  the	  only	  two	  papers	  that	  have	  sought	  to	  describe	  the	  publication	  landscape	  for	  preference	  comparison	  of	  physicians	  and	  patients.	  However,	  this	  qualitative	  synthesis	  can	  only	  provide	  a	  preliminary	  overview	  of	  the	  current	  state	  of	  work	  done	  on	  this	  subject,	  as	  there	  were	  several	  limitations	  in	  the	  process	  of	  this	  review.	  Primarily,	  the	  heterogeneity	  of	  the	  published	  studies	  in	  terms	  of	  study	  population,	  social	  and	  cultural	  values,	  disease	  states/intervention/diagnostics,	  and	  attributes	  prohibited	  quantitative	  evaluation	  of	  how	  physician	  and	  patient	  preferences	  might	  be	  congruent.	  There	  is	  also	  no	  standardized	  approach	  to	  evaluating	  preference	  studies,	  thus	  making	  this	  review	  qualitative	  at	  best.	  However,	  each	  study	  was	  appraised	  thoroughly,	  evaluating	  any	  potential	  flaws	  that	  would	  make	  the	  results	  uninterpretable.	  	  	  	  Results	  from	  this	  review	  suggested	  that	  there	  is	  no	  clear	  pattern	  of	  preference	  congruence	  between	  patients	  and	  physicians.	  It	  did	  seem	  to	  indicate	  that	  on	  most	  occasions,	  there	  would	  be	  some	  disagreement	  between	  the	  two	  groups,	  if	  not	  in	  the	  ranking	  of	  attributes	  then	  the	  38	  	  strength	  of	  their	  preference.	  Agreement	  or	  disagreement	  in	  preference	  between	  the	  two	  decision	  stakeholders	  is	  likely	  population-­‐	  and	  disease-­‐specific.	  Thus,	  no	  assumption	  should	  be	  made	  about	  patient	  and	  physician	  preference	  without	  performing	  a	  preference	  study	  in	  the	  specific	  population	  of	  interest.	  39	  	  Table	  2.1:	  Summary	  of	  characteristics	  of	  included	  studies	  comparing	  physician	  and	  patient	  preferences	  Reference	  [Publication	  year]	   Country	   Medical	  condition/	  intervention/	  diagnostics	  analyzed	  Sample	  size	   Preference	  elicitation	  methods	  Attributes	  Meister	  et	  al.	  	  (119)	  [2002]	  	  Germany	   Hearing	  aid	   175	  hearing	  aid	  users,	  25	  clinicians	   CA	   Time,	  accuracy,	  risk	  Bishop	  et	  al.	  	  (120)	  [2004]	   United	  Kingdom	   Down’s	  syndrome	  screening	  test	   291	  pregnant	  women,	  98	  health	  professionals	  (obstetricians,	  house	  officers,	  midwives)	  	  CA	   Time	  at	  screening,	  detection	  rate,	  risk	  of	  miscarriage	  	  Lee	  et	  al.	  	  (121)	  [2005]	   Hong	  Kong,	  China	   Postoperative	  nausea	  and	  vomiting	  management	   200	  gynecological	  patients,	  52	  health	  care	  professionals	  CA	   Risk	  of	  PONV,	  level	  of	  pain,	  level	  of	  sedation,	  antiemetic	  management,	  efficacy	  in	  the	  first	  2hr	  after	  surgery,	  extra	  cost	  	  Sassi	  et	  al.	  	  (122)	  [2005]	   United	  Kingdom,	  Italy	   Cardiac	  risk	  assessment	  	  	   49/26	  males	  between	  40-­‐50	  years	  of	  age,	  29/15	  GPs	  (UK/Italy)	  CA	   GPs:	  Perception	  of	  resource	  commitment,	  prognostic	  value,	  expected	  risk	  reduction	  following	  preventive	  intervention	  	  Patients:	  modality	  of	  the	  assessment,	  preventive	  interventions,	  accuracy,	  expected	  risk	  reduction	  after	  preventive	  intervention	  	  Lewis	  et	  al.	  	  (123)	  [2006]	   Australia	   Down’s	  syndrome	  screening	  test	   322	  pregnant	  women,	  300	  health	  care	  professionals	  (obstetricians,	  midwives)	  	  CA	   Time	  at	  screening,	  detection	  rate,	  risk	  of	  miscarriage	  40	  	  Reference	  [Publication	  year]	   Country	   Medical	  condition/	  intervention/	  diagnostics	  analyzed	  Sample	  size	   Preference	  elicitation	  methods	  Attributes	  Vermeulen	  et	  al.	  (124)	  [2007]	   Netherlands	   Wound	  dressing	   74	  patients,	  50	  doctors,	  50	  nurses	   CA	   Wound	  healing	  time,	  pain	  costs	  	   	   	   	   	   	  Sampietro-­‐Colom	  et	  al.	  (125)	  [2008]	   Spain	   Waiting	  time	  for	  joint	  replacement	   96	  consultants,	  117	  allied	  health	  professionalsƚ,	  195	  patients,	  152	  relatives,	  300	  general	  public	  CA	   Disease	  severity,	  pain,	  recovery	  probability,	  difficulty	  in	  doing	  ADL,	  limitations	  on	  ability	  to	  work,	  has	  someone	  to	  look	  after	  the	  patient,	  be	  a	  caregiver	  	  Bedermen	  et	  al.	  (126)	  [2010]	   Canada	   Lumbar	  spinal	  surgery	   131	  surgeons,	  202	  GPs,	  164	  patients	   CA	   Duration	  of	  pain,	  severity,	  location	  of	  pain,	  onset	  of	  pain,	  neurological	  symptoms,	  walking	  tolerance	  	  Gregorian	  et	  al.	  	  (127)	  [2010]	   United	  States	   Opioids	  for	  acute	  and	  chronic	  pain	   618	  patients	  (302	  acute	  pain,	  316	  chronic	  pain);	  325	  physicians	  CA	   Pain	  relief,	  incidence	  rate	  of	  nausea,	  incidence	  rate	  of	  vomiting,	  incidence	  rate	  of	  constipation,	  incidence	  rate	  of	  drowsiness,	  incidence	  rate	  of	  pruritus	  	  Porzsolt	  et	  al.	  	  (128)	  [2010]	  	  Germany	   Diabetes	  management	   827	  diabetic	  patients,	  60	  physicians	   CA	   Main	  treatment	  effect,	  effect	  on	  body	  weight,	  mode	  of	  application,	  type	  of	  product	  	  Mantovani	  et	  al.	  (129)	  [2005]	   Italy	   Treatment	  for	  haemophilia	   178	  patients,	  69	  physicians,	  58	  pharmacists	  DCE	   Perceived	  viral	  safety,	  risk	  of	  inhibitor	  development,	  factor	  infusion	  frequency	  on	  prophylaxis,	  pharmaceutical	  dosage	  form,	  distribution	  modes,	  price	  	  Ashcroft	  et	  al.	  	  (130)	  [2006]	   United	  Kingdom	   Psoriasis	   227	  dermatologists	   DCE	   Time	  to	  moderate	  improvement,	  time	  to	  relapse,	  skin	  irritation,	  high	  blood	  pressure,	  20-­‐year	  risk	  of	  liver	  damage,	  20-­‐year	  risk	  of	  skin	  cancer	  41	  	  Reference	  [Publication	  year]	   Country	   Medical	  condition/	  intervention/	  diagnostics	  analyzed	  Sample	  size	   Preference	  elicitation	  methods	  Attributes	  Seston	  et	  al.	  	  (131)	  [2007]	   United	  Kingdom	  	  As	  above	   126	  patients	   As	  above	   As	  in	  above	  study	  by	  Ashcroft	  et	  al.	  	  Langer	  et	  al.	  	  (132)	  [2007]	   United	  States	   Chemotherapy-­‐related	  anemia	   467	  health	  care	  providers,	  438	  patients	   DCE	   Time	  to	  noticeable	  relief	  of	  fatigue,	  frequency	  of	  visits	  	  Mühlbacher	  et	  al.	  (133)	  [2008]	   Germany	   Multiple	  myeloma	   282	  patients	   DCE	   Life	  span,	  side	  effects,	  periods	  without	  treatment	  between	  individual	  lines	  of	  therapy,	  physical	  situation,	  emotional	  situation,	  social	  situation,	  medication,	  further	  therapies	  	  Mühlbacher	  et	  al.	  (134)	  [2011]	  	  Germany	   As	  above	   243	  physicians	   As	  above	  	   As	  above	  de	  Bekker-­‐Grob	  et	  al.	  (135)	  [2009]	  	   Netherlands	   Osteoporosis	   40	  GPs,	  120	  women	  with	  high	  fracture	  risk	   DCE	   Route	  of	  drug	  administration,	  effectiveness	  (10-­‐yr	  fracture	  risk	  reduction),	  Adverse	  effect	  (nausea	  within	  2	  hr	  of	  intake),	  total	  treatment	  duration,	  total	  cost	  to	  patient	  	  Fiebig	  et	  al.	  	  (136)	  [2009]	  	  	  	  	  	  	  	  	  Australia	   Pap	  test	   167	  women,	  215	  GPs	   DCE	   Women:	  screening	  interval,	  health	  care	  relationship	  with	  GP,	  GP’s	  sex,	  time	  since	  last	  cervical	  screening,	  GP’s	  recommendation,	  GP’s	  incentive	  payment,	  cost	  of	  test,	  chance	  of	  false	  negative,	  chance	  of	  false	  positive;	  GPs:	  reason	  for	  consultation,	  screening	  interval,	  health	  care	  relationship	  with	  patient,	  last	  cervical	  screening,	  patient’s	  age,	  42	  	  Reference	  [Publication	  year]	   Country	   Medical	  condition/	  intervention/	  diagnostics	  analyzed	  Sample	  size	   Preference	  elicitation	  methods	  Attributes	  Cont’d	  from	  previous	  page	   perception	  of	  patient’s	  socioeconomic	  status,	  reimbursement	  for	  test,	  cost	  of	  test,	  chance	  of	  false	  negative,	  chance	  of	  false	  positive	  	  	  	  Johnson	  et	  al.	  	  (137)	  [2010]	   United	  States	   Crohn’s	  disease	  management	   315	  gastroenterologists,	  580	  patients	  DCE	   Severity	  of	  daily	  symptoms,	  frequency	  of	  flare-­‐ups,	  prevention	  of	  serious	  disease	  complications,	  need	  for	  oral	  steroid,	  10-­‐year	  risks	  of	  death	  from	  PML,	  10-­‐year	  risks	  of	  death	  from	  infection,	  10-­‐year	  risks	  of	  death	  from	  lymphoma	  	  Shafey	  et	  al.	  	  (138)	  [2011]	   Canada	   Treatment	  for	  follicular	  lymphoma	   81	  patients,	  48	  physicians	   DCE	   Administration	  of	  treatment,	  survival	  free	  of	  relapse,	  side	  effects,	  health	  cost	  	  Thrumurthy	  et	  al.	  (139)	  [2011]	   United	  Kingdom	   Surgery	  for	  oesophagogastric	  cancer	  	   81	  patients,	  90	  physicians	   DCE	   Mortality,	  morbidity,	  QoL,	  cure	  rate,	  hospital	  type,	  surgeon’s	  reputation	  	  Faggioli	  et	  al.	  	  (140)	  [2011]	   Italy	   Surgery	  for	  abdominal	  aortic	  aneurysms	   160	  patients,	  102	  relatives,	  30	  surgeons	   DCE	   Anaesthesia,	  recovery	  time	  to	  basic	  everyday	  activities,	  risk	  of	  re-­‐intervention	  at	  5	  years,	  complexity	  of	  follow-­‐up,	  risk	  of	  major	  complications,	  additional	  cost	  	  	  Van	  Empel	  et	  al.	  (141)	  [2011]	   Netherlands	   Fertility	  care	   925	  patients,	  277	  physicians	   DCE	   Travel	  time	  to	  clinic,	  ongoing	  pregnancy	  rate	  per	  IVF-­‐cycle,	  physician’s	  attitude	  towards	  patients,	  information	  on	  treatment,	  continuity	  of	  physicians	  	  43	  	  Reference	  [Publication	  year]	   Country	   Medical	  condition/	  intervention/	  diagnostics	  analyzed	  Sample	  size	   Preference	  elicitation	  methods	  Attributes	  Pedersen	  et	  al.	  	  (142)	  [2012]	   Denmark	   Primary	  care	   969	  GPs,	  698	  patients	   DCE	   Telephone	  wait	  time,	  opening	  hours,	  appointment,	  distance,	  waiting	  room,	  consultation	  time,	  routine	  tasks	  	  Hill	  et	  al.	  	  (143)	  [2012]	   United	  Kingdom	   Down’s	  syndrome	  screening	  test	   335	  women	  (pregnant	  and	  non-­‐pregnant),	  181	  health	  professionals	  (midwives,	  obstetricians)	  	  DCE	   Accuracy,	  time	  of	  results,	  risk	  of	  miscarriage,	  information	  gained	  from	  the	  test	  Park	  et	  al.	  	  (144)	  [2012]	   South	  Korea	   Metastatic	  renal	  cell	  carcinoma	   120	  patients,	  52	  family	  members,	  39	  doctors,	  155	  nurses,	  78	  pharmacists	  DCE	   Progression-­‐free	  survival,	  bone	  marrow	  suppression,	  hand-­‐foot	  skin	  reaction,	  gastrointestinal	  perforation,	  bleeding,	  administration	  	  Regier	  et	  al.	  	  (145)	  [2012]	   Canada	   Antibiotic	  prophylaxis	  in	  pediatric	  oncology	   102	  parents,	  60	  healthcare	  providers	  (physicians,	  nurse	  practitioners,	  pharmacists,	  social	  workers)	  DCE	   Chance	  of	  infection,	  death	  and	  side	  effects,	  route	  of	  administration	  and	  cost	  of	  pharmacotherapy	  	  	  	  	  Chancellor	  et	  al.	  (146)	  [2012]	   France,	  Germany,	  Italy,	  Spain,	  Sweden,	  United	  Kingdom	  Opioids	  for	  chronic	  pain	   242	  patients,	  270	  physicians	   DCE	   Patients:	  efficacy,	  constipation	  and	  bowel	  problems,	  nausea	  and	  vomiting,	  alertness,	  energy;	  Physicians:	  range	  of	  dosage	  forms,	  proportion	  of	  patients	  achieving	  ≥50%	  pain	  reduction,	  constipation,	  nausea	  and	  vomiting,	  CNS	  side	  effects	  	  44	  	  Reference	  [Publication	  year]	   Country	   Medical	  condition/	  intervention/	  diagnostics	  analyzed	  Sample	  size	   Preference	  elicitation	  methods	  Attributes	  de	  Bekker-­‐Grob	  et	  al.	  (147)	  [2013]	   Netherlands	   Prostate	  cancer	   110	  patients,	  50	  urologists	   DCE	   Risk	  of	  urinary	  incontinence,	  risk	  of	  erection	  problems,	  risk	  of	  other	  permanent	  side	  effects,	  main	  aim	  is	  cure	  (i.e.	  expected	  outcome),	  frequency	  of	  PSA	  testing	  with	  risk	  of	  new	  biopsies	  CA	  conjoint	  analysis;	  DCE	  discrete	  choice	  experiment;	  GP	  general	  practitioner;	  PONV	  postoperative	  nausea	  and	  vomiting;	  CNS	  central	  nervous	  system;	  	  	  ƚ	  allied	  health	  professionals	  included	  GPs,	  nurses,	  social	  workers,	  and	  physiotherapists;	  consultants	  included	  orthopedic	  surgeons,	  rheumatologists,	  rehabilitators,	  and	  GPs.	  	  	  	  	  	  	  	  	  	  	  	  	  	  	  	  	  	  	  	  45	  	  Table	  2.2:	  Summary	  of	  results	  of	  included	  studies	  Reference	  [Publication	  year]	   Medical	  condition/	  intervention/	  diagnostics	  analyzed	  Preference	  elicitation	  methods	  Statistical	  analysis	   Results	   Congruence	  Acute	  conditions	   	   	   	   	   	  Lee	  et	  al.	  	  (121)	  [2005]	   Postoperative	  nausea	  and	  vomiting	  management	  CA	   Random	  effects	  probit	   Both	  physicians	  and	  patients	  showed	  similar	  preferences	  for	  management	  of	  PONV.	  	  S	  Vermeulen	  et	  al.	  (124)	  [2007]	   Wound	  dressing	   CA	   Fixed	  effects	  linear	  mixed	  model	  	  Similar	  preferences	  for	  wound	  dressing	  attributes	  amongst	  the	  three	  groups.	   S	  Bederman	  et	  al.	  (126)	  [2010]	   Lumbar	  spinal	  surgery	   CA	   Random	  effects	  probit	   Surgeons	  placed	  highest	  importance	  on	  the	  location	  of	  pain.	  FPs	  considered	  neurologic	  symptoms,	  walking	  tolerance,	  and	  severity	  to	  be	  of	  similar	  importance.	  Pain	  severity,	  walking	  distance,	  and	  duration	  of	  pain	  were	  the	  most	  important	  factors	  for	  patients.	  	  	  D	  Gregorian	  et	  al.	  (127)	  [2010]	   Opioids	  for	  acute	  and	  chronic	  pain	   CA	   Hierchical	  Bayesian	  estimation	  Similar	  preference	  between	  physicians	  and	  patients,	  regardless	  of	  acute	  or	  chronic	  pain;	  both	  nausea	  and	  vomiting	  were	  at	  least	  as	  important	  as	  pain	  relief	  to	  both	  physicians	  and	  patients	  in	  determining	  preferences	  for	  opioid	  medications.	  S	  Mantovani	  et	  al.	  (129)	  [2005]	   Treatment	  for	  haemophilia	   DCE	   Random	  effects	  probit	   Similar	  preferences	  for	  rankings	  of	  attributes	  (especially	  on	  outcome	  attributes).	  Differences	  in	  strength	  of	  preferences	  as	  well	  as	  some	  process	  of	  care-­‐related	  attributes.	  	  S;	  DS	  Langer	  et	  al.	  	  (132)	  [2007]	   Chemotherapy-­‐related	  anaemia	   DCE	   Multinomial	  logit	   Both	  health	  care	  providers	  and	  patients	  viewed	  effectiveness	  as	  more	  important	  than	  convenience.	  	  S	  46	  	  Reference	  [Publication	  year]	   Medical	  condition/	  intervention/	  diagnostics	  analyzed	  Preference	  elicitation	  methods	  Statistical	  analysis	   Results	   Congruence	  Shafey	  et	  al.	  	  (138)	  [2011]	   Treatment	  for	  follicular	  lymphoma	   DCE	   Random	  effects	  logit	  regression	   Both	  patients	  and	  physicians	  valued	  survival	  free	  of	  relapse	  to	  be	  an	  important	  positive	  influence	  on	  choice	  and	  toxicity	  of	  allo-­‐SCT	  a	  negative	  influence	  on	  choice	  relative	  to	  standard	  CT.	  Cost	  and	  toxicity	  of	  RIT	  relative	  to	  standard	  CT	  was	  not	  a	  significant	  factor	  affecting	  choice.	  Difference	  in	  strength	  of	  preference	  for	  patients	  and	  physicians.	  	  S;	  DS	  	  van	  Empel	  et	  al.	  (141)	  [2011]	   Fertility	  care	   DCE	   Multinomial	  logit	   pregnancy	  rates	  relatively	  more	  important	  to	  physicians.	  Patients	  assigned	  more	  value	  to	  patient-­‐centredness	  and	  were	  willing	  to	  trade-­‐off	  a	  higher	  pregnancy	  rate	  for	  patient-­‐centredness	  than	  physicians	  recommended	  them	  to	  do.	  	  D	  Thrumurthy	  et	  al.	  (139)	  [2011]	   Surgery	  for	  oesophagogastric	  cancer	  DCE	   Random	  effects	  probit	   no	  significant	  differences	  in	  ranking,	  but	  significant	  differences	  in	  strength;	  both	  groups	  placed	  QoL	  as	  the	  most	  important	  attribute	  and	  hospital	  type	  as	  the	  least	  important,	  however,	  physicians	  placed	  greater	  emphasis	  on	  mortality	  than	  morbidity.	  	  S;	  DS	  Faggioli	  et	  al.	  	  (140)	  [2011]	   Surgery	  abdominal	  aortic	  aneurysms	   DCE	   Conditional	  logit	   Similar	  preferences	  for	  most	  important	  attributes	  (major	  complications,	  re-­‐intervention	  risk)	  amongst	  all	  groups;	  patients	  and	  their	  relatives	  considered	  cost	  to	  be	  very	  important,	  whereas	  cost	  and	  type	  of	  anaesthesia	  were	  least	  important	  to	  surgeons.	  Also	  significant	  difference	  in	  preference	  for	  treated	  and	  untreated	  patient	  subgroups.	  	  	  D	  47	  	  Reference	  [Publication	  year]	   Medical	  condition/	  intervention/	  diagnostics	  analyzed	  Preference	  elicitation	  methods	  Statistical	  analysis	   Results	   Congruence	  Park	  et	  al.	  (144)	  [2012]	   Metastatic	  renal	  cell	  carcinoma	   DCE	   Conditional	  logit	  	   Substantial	  differences	  in	  preference	  between	  patient	  and	  health	  professional	  groups.	  Doctors	  prefer	  effective	  and	  orally	  administered	  drugs	  while	  patients	  show	  more	  reluctant	  attitudes	  about	  adverse	  events	  than	  do	  health	  care	  professionals.	  	  D	  Regier	  et	  al.	  	  (145)	  [2012]	   Antibiotic	  prophylaxis	  in	  pediatric	  oncology	   DCE	   Conditional	  logit	   Similar	  preference	  in	  ranking	  but	  differences	  in	  strength,	  where	  parents	  exhibited	  stronger	  positive	  preference	  for	  decreased	  chances	  of	  infection	  and	  death.	  	  S;	  DS	  de	  Bekker-­‐Grob	  et	  al.	  (147)	  [2013]	   Prostate	  cancer	   DCE	   Mixed	  logit	   Generally	  similar	  preference	  between	  patients	  and	  urologists	  with	  some	  different	  trade-­‐offs	  between	  various	  specific	  treatments	  aspects.	  	  S	  Chronic	  conditions	   	   	   	   	   	  Meister	  et	  al.	  	  (119)	  [2002]	   Hearing	  aid	   CA	   Rating,	  student	  t	  test	   Similar	  preference	  between	  hearing	  aid	  users	  and	  clinicians	  regarding	  importance	  of	  speech	  perception	  being	  the	  most	  valued	  attribute;	  strength	  of	  preference	  differ	  between	  groups	  	  S;	  DS	  Sampietro-­‐Colom	  et	  al.	  (125)	  [2008]	   Waiting	  time	  for	  joint	  replacement	   CA	   Rank-­‐ordered	  logit	   Patients	  and	  relatives	  rated	  pain	  higher	  whereas	  allied	  health	  professionals	  and	  consultants	  rated	  difficulty	  in	  doing	  activities	  to	  be	  higher	  	  	  D	  Gregorian	  et	  al.	  (127)	  [2010]	   Opioids	  for	  acute	  and	  chronic	  pain	   CA	   Hierchical	  Bayesian	  estimation	  Similar	  preference	  between	  physicians	  and	  patients;	  both	  nausea	  and	  vomiting	  were	  at	  least	  as	  important	  as	  pain	  relief	  to	  both	  physicians	  and	  patients	  in	  determining	  preferences	  for	  opioid	  medications.	  	  S	  48	  	  Reference	  [Publication	  year]	   Medical	  condition/	  intervention/	  diagnostics	  analyzed	  Preference	  elicitation	  methods	  Statistical	  analysis	   Results	   Congruence	  Porzsolt	  et	  al.	  	  (128)	  [2010]	   Diabetes	  management	   CA	   Main	  effects	  model	   Physicians	  valued	  the	  reduction	  of	  HbA1c	  and	  the	  reduction	  of	  body	  weight	  to	  be	  more	  important.	  Patients	  preferred	  original	  products,	  while	  physicians	  preferred	  generic	  products.	  D	  	   	   	   	   	   	  Ashcroft	  et	  al.	  	  (130)	  [2006]	  	  Seston	  et	  al.	  	  (131)	  [2007]	  	  Psoriasis	   DCE	   Random	  effects	  probit	   Both	  physicians	  and	  patients	  considered	  the	  risk	  of	  liver	  damage	  and	  risk	  of	  skin	  cancer	  to	  be	  the	  most	  important	  side-­‐effect.	  Similar	  preference	  for	  time	  to	  achieve	  moderate	  improvement	  in	  psoriasis	  over	  the	  time	  to	  relapse.	  	  S	  Mühlbacher	  et	  al.	  (133)	  [2008]	  	  Mühlbacher	  et	  al.	  (134)	  [2011]	  Multiple	  myeloma	   DCE	   Random	  effects	  probit	   Mostly	  agreement	  of	  attributes	  between	  groups;	  however,	  patients	  valued	  further	  therapy	  to	  be	  more	  important	  than	  gain	  in	  life	  span,	  whereas	  this	  was	  reversed	  for	  the	  physicians.	  Patients	  rated	  emotional	  quality	  of	  life	  over	  self-­‐treatment	  whereas	  this	  was	  reversed	  for	  physicians.	  Both	  groups	  did	  not	  rank	  side	  effects	  and	  physical	  quality	  of	  life	  to	  be	  particularly	  important.	  	  S;	  DS	  de	  Bekker-­‐Grob	  et	  al.	  (135)	  [2009]	   Osteoporosis	   DCE	   Conditional	  logit	   GPs	  had	  significantly	  less	  favourable	  attitude	  towards	  preventive	  osteoporosis	  drug	  treatment	  than	  patients;	  GPs	  placed	  higher	  relative	  values	  on	  effectiveness	  of	  drug	  treatment	  and	  shorter	  total	  treatment	  duration	  than	  patients.	  	  D	  Johnson	  et	  al.	  	  (137)	  [2010]	   Crohn’s	  disease	  management	   DCE	   Random-­‐parameters	  logit/	  mixed	  logit	  Gastroenterologists	  and	  patients	  disagreed	  about	  how	  much	  risk	  is	  tolerable	  for	  improvements	  in	  efficacy.	  In	  exchange	  for	  improvements	  from	  severe	  to	  moderate	  D	  49	  	  Reference	  [Publication	  year]	   Medical	  condition/	  intervention/	  diagnostics	  analyzed	  Preference	  elicitation	  methods	  Statistical	  analysis	   Results	   Congruence	  symptoms,	  gastroenterologists	  were	  significantly	  more	  tolerant	  than	  patients	  of	  treatment	  risk	  of	  PML,	  and	  serious	  infection,	  but	  not	  lymphoma.	  In	  contrast,	  patients	  were	  significantly	  more	  tolerant	  than	  gastroenterologists	  of	  treatment	  risks	  for	  serious	  infection,	  and	  lymphoma	  but	  not	  PML.	  	  Pedersen	  et	  al.	  (142)	  [2012]	   Primary	  care	   DCE	   Conditional	  logit/	  heteroscedastric	  conditional	  logit	  Patients	  considered	  organizational	  attributes	  (waiting	  time	  on	  the	  telephone,	  waiting	  time	  to	  the	  appointment,	  distance,	  consultation	  time)	  to	  be	  important,	  while	  GPs	  considered	  all	  attributes	  except	  for	  routine	  tasks	  to	  be	  important	  in	  patients’	  choice	  of	  a	  new	  GP.	  	  S;	  DS	  Chancellor	  et	  al.	  (146)	  [2012]	   Opioids	  for	  chronic	  pain	   DCE	   Multinomial	  logit/	  hierarchical	  Bayesian	  analysis	  Direct	  comparison	  not	  possible	  as	  different	  attributes	  and	  survey	  design;	  qualitative	  comparison	  suggest	  physicians	  may	  place	  more	  emphasis	  on	  pain	  response.	  	  S;	  DS	  Diagnostics/screening	  tests	   	   	   	   	  Bishop	  et	  al.	  	  (120)	  [2004]	   Down’s	  syndrome	  screening	  test	   CA	   Random	  effects	  probit	   Both	  pregnant	  women	  and	  health	  professionals	  showed	  similar	  preferences	  regarding	  importance	  of	  safe	  tests,	  conducted	  early	  and	  with	  high	  detection	  rates.	  Health	  care	  professionals	  valued	  earlier	  tests	  more	  highly.	  	  S;	  DS	  Lewis	  et	  al.	  	  (123)	  [2006]	   Down’s	  syndrome	  screening	  test	   CA	   Random	  effects	  probit	   Both	  pregnant	  women	  and	  health	  professionals	  showed	  similar	  preferences	  regarding	  importance	  of	  safe	  tests,	  conducted	  early	  and	  with	  high	  detection	  rates.	  Health	  care	  professionals	  valued	  earlier	  tests	  more	  highly.	  	  S;	  DS	  50	  	  Reference	  [Publication	  year]	   Medical	  condition/	  intervention/	  diagnostics	  analyzed	  Preference	  elicitation	  methods	  Statistical	  analysis	   Results	   Congruence	  Hill	  et	  al.	  	  (143)	  [2012]	   Down’s	  syndrome	  screening	  test	   DCE	   Conditional	  logit	   Women	  valued	  risk	  of	  miscarriage	  to	  be	  more	  important	  whereas	  for	  health	  professionals	  it	  was	  accuracy.	  	  D	  Sassi	  et	  al.	  	  (122)	  [2005]	   Cardiac	  risk	  assessment	   CA	   Nonlinear	  ordered	  probit	   general	  public	  rated	  prognostic	  value	  significantly	  higher	  than	  physicians	  	  D	  Fiebig	  et	  al.	  	  (136)	  [2009]	   Pap	  test	   DCE	   Random	  parameter	  or	  mixed	  logit	  	  Considerable	  commonality	  between	  GPs	  and	  patients	  for	  most	  test	  attributes.	   S	  CA	  conjoint	  analysis;	  DCE	  discrete	  choice	  experiment;	  GP	  general	  practitioner;	  PONV	  postoperative	  nausea	  and	  vomiting;	  CNS	  central	  nervous	  system;	  S	  similar	  in	  preference;	  D	  different	  in	  preference;	  DS	  difference	  in	  strength	  of	  preference51	  	  Chapter	  3: Patients’	  preferences	  for	  stroke	  prophylaxis	  in	  atrial	  fibrillation:	  time	  trade-­‐off	  (TTO)	  and	  best-­‐worst	  scaling	  (BWS)	  experiment	  3.1 Introduction	  Patient	  preferences	  have	  become	  increasingly	  important	  as	  healthcare	  moves	  to	  adopt	  a	  more	  patient-­‐centered	  care	  model.	  Good	  clinical	  decision-­‐making	  should	  integrate	  not	  only	  the	  best	  evidence,	  clinical	  expertise,	  and	  clinical	  uncertainty,	  but	  also	  patients’	  values	  and	  attitudes	  for	  health	  outcomes	  (35,37,152).	  It	  is	  recognized	  that	  patients'	  preferences	  are	  especially	  paramount	  in	  preference-­‐sensitive	  decisions,	  where	  the	  evidence	  for	  therapy	  is	  weak	  or	  conflicting,	  when	  several	  viable	  therapeutic	  options	  co-­‐exist,	  when	  there	  is	  a	  clear	  trade-­‐off	  between	  benefit	  and	  risk	  and	  when	  patients	  and	  clinicians	  may	  have	  disparate	  values	  for	  the	  health	  outcomes	  (6,26,30).	  	  Stroke	  prophylaxis	  in	  atrial	  fibrillation	  exemplifies	  a	  ‘preference-­‐sensitive’	  clinical	  decision-­‐making	  situation.	  With	  7	  possible	  therapeutic	  options,	  comparable	  efficacy	  amongst	  three	  of	  these	  regimens,	  clear	  trade-­‐offs	  between	  stroke	  prevention	  and	  bleeding	  risks,	  input	  of	  patient	  preferences	  in	  making	  decisions	  regarding	  stroke	  prophylaxis	  would	  be	  highly	  paramount.	  	  	  While	  there	  are	  studies	  that	  have	  addressed	  patient	  preferences	  in	  atrial	  fibrillation,	  very	  few	  provide	  quantitative	  preference	  data	  using	  utility-­‐based	  methods.	  The	  majority	  of	  the	  literature	  reporting	  patient	  preference	  used	  decision-­‐aids	  and	  probability	  trade-­‐off	  techniques	  as	  a	  means	  to	  elicit	  patient	  preferences	  in	  selecting	  antithrombotics	  (114,117,153–160).	  While	  the	  results	  of	  these	  analyses	  provide	  a	  general	  sense	  of	  patients’	  threshold	  for	  stroke	  and	  bleeding	  risk,	  52	  	  they	  do	  not	  determine	  how	  patients	  value	  the	  health	  outcomes	  related	  to	  stroke	  prophylaxis	  in	  AF.	  To	  date,	  two	  studies	  have	  measured	  patient	  preferences	  using	  utility-­‐based	  methods.	  However,	  several	  gaps	  remain	  to	  be	  explored	  in	  terms	  of	  utilities	  associated	  with	  stroke	  prophylaxis	  in	  AF.	  Gage	  et	  al.	  observed	  that	  patient	  preferences	  are	  highly	  variable;	  however,	  little	  is	  known	  about	  what	  factors	  influence	  this	  variability	  (110).	  Preference	  data	  on	  bleeding	  in	  stroke	  prophylaxis	  are	  also	  scarce	  for	  this	  population.	  Thomson	  et	  al.	  reported	  a	  utility	  associated	  with	  major	  bleed	  using	  the	  SG	  method	  (161),	  and	  a	  thorough	  literature	  search	  revealed	  no	  utility	  data	  for	  minor	  bleeds	  related	  to	  stroke	  prophylaxis	  in	  AF.	  A	  review	  of	  the	  cost-­‐utility	  analyses	  that	  examined	  new	  oral	  anticoagulants	  for	  stroke	  prevention	  in	  AF	  reveals	  that	  the	  majority	  of	  the	  cost-­‐effectiveness	  models	  assume	  a	  utility	  value	  for	  minor	  bleed	  to	  be	  the	  same	  as	  major	  bleed	  (162).	  This	  is	  concerning	  as	  it	  is	  quite	  plausible	  that	  the	  utility	  values	  for	  major	  and	  minor	  bleed	  would	  be	  different.	  Another	  knowledge	  gap	  of	  concern	  is	  that	  measuring	  utility	  of	  transient	  health	  states	  such	  as	  bleeding	  events,	  using	  the	  SG	  and	  TTO	  methods	  is	  a	  questionable	  approach,	  as	  these	  preference	  elicitation	  methods	  are	  developed	  and	  validated	  for	  quantifying	  utilities	  associated	  for	  chronic	  health	  states	  (e.g.	  post	  stroke)	  and	  not	  temporary	  health	  states.	  Most	  importantly,	  it	  would	  be	  of	  interest	  to	  clinicians	  and	  policy	  makers	  alike	  to	  find	  out	  what	  other	  determinants,	  other	  than	  the	  risk	  of	  stroke	  and	  bleed	  associated	  with	  stroke	  prophylaxis,	  could	  influence	  patients’	  decision	  on	  choosing	  an	  antithrombotic;	  for	  example,	  the	  frequency	  of	  blood	  test	  monitoring,	  drug	  interactions,	  or	  availability	  of	  reversibility	  agents	  in	  the	  event	  of	  a	  life-­‐threatening	  bleed.	  	  	  53	  	  This	  study	  is	  aimed	  at	  addressing	  the	  above	  knowledge	  gaps	  in	  patients’	  preferences	  for	  stroke	  prophylaxis.	  A	  TTO	  survey	  was	  administered	  to	  patients	  to	  elicit	  their	  preferences	  for	  non-­‐debilitating	  and	  debilitating	  stroke.	  A	  modified	  TTO	  using	  the	  chained	  method	  was	  developed	  to	  elicit	  their	  preferences	  for	  the	  temporary	  health	  states	  of	  major	  and	  minor	  bleed.	  Patient	  demographics	  will	  be	  used	  to	  explore	  possible	  factors	  that	  could	  explain	  differences	  in	  preferences.	  In	  addition,	  the	  study	  also	  measured	  the	  patients’	  preferences	  for	  other	  characteristics,	  or	  attributes	  associated	  with	  stroke	  prophylaxis	  that	  might	  influence	  their	  choices	  in	  selecting	  an	  antithrombotic.	  This	  was	  done	  using	  the	  best	  worst	  scaling	  (BWS)	  choice	  experiment	  to	  measure	  the	  relative	  preferences	  of	  different	  attribute	  levels	  of	  interest.	  	  	  3.2 Methods	  3.2.1 Study	  population	  Subjects	  were	  recruited	  through	  the	  atrial	  fibrillation	  clinic	  at	  Vancouver	  General	  Hospital	  and	  the	  inpatient	  cardiology	  units	  at	  St.	  Paul’s	  Hospital.	  Patients	  were	  invited	  to	  participate	  in	  the	  web-­‐based	  survey	  provided	  they	  had	  non-­‐valvular	  atrial	  fibrillation,	  were	  at	  least	  19	  years	  old,	  could	  comprehend	  English	  and	  had	  provided	  consent.	  The	  web-­‐based	  survey	  was	  administered	  on	  a	  laptop	  with	  the	  guidance	  of	  an	  interviewer.	  After	  giving	  consent,	  the	  subjects	  were	  directed	  to	  the	  interactive	  survey	  consisted	  of	  three	  sections.	  The	  first	  section	  asked	  questions	  related	  to	  the	  subjects’	  demographic	  information.	  The	  second	  section	  was	  a	  time	  trade-­‐off	  (TTO)	  questionnaire,	  followed	  lastly	  by	  a	  best	  worst	  scaling	  (BWS)	  choice	  experiment.	  The	  interviewer	  remained	  by	  the	  subjects’	  side	  throughout	  the	  process	  to	  answer	  questions	  and	  54	  	  provide	  assistance.	  	  Ethics	  approval	  was	  obtained	  from	  the	  Providence	  Health	  Care	  Research	  Ethics	  Board	  at	  the	  University	  of	  British	  Columbia.	  	  3.2.2 Experimental	  design Time	  trade-­‐off	  (TTO)	  questionnaire	  A	  questionnaire	  based	  on	  the	  TTO	  method	  (163)	  was	  developed	  to	  elicit	  utilities	  for	  four	  health	  states	  of	  interest	  related	  to	  stroke	  prophylaxis	  in	  atrial	  fibrillation:	  debilitating	  stroke,	  non-­‐debilitating	  stroke,	  major	  bleed	  and	  minor	  bleed.	  Subjects’	  preference	  for	  their	  current	  health	  state	  was	  also	  measured.	  Both	  health	  states	  related	  to	  stroke	  were	  described	  in	  detail	  including	  the	  fine	  and	  gross	  motor	  skills,	  spoken	  and	  written	  language,	  and	  cognitive	  and	  psychosocial	  function.	  For	  the	  major	  bleed	  health	  state,	  a	  scenario	  involving	  a	  gastrointestinal	  bleed	  requiring	  hospitalization	  was	  described.	  For	  the	  minor	  bleed	  health	  state,	  a	  description	  of	  frequent	  and	  prolonged	  nose-­‐bleed	  was	  provided	  to	  the	  respondents.	  	  	  Utilizing	  the	  TTO	  method,	  respondents	  were	  asked	  to	  trade-­‐off	  fewer	  life	  years	  in	  their	  current	  health	  (𝑥)	  versus	  longer	  time	  (𝑡)	  in	  the	  impaired	  health	  state.	  Using	  a	  slider	  embedded	  in	  the	  web-­‐based	  questionnaire,	  the	  respondents	  were	  asked	  to	  move	  the	  slider,	  varying	  time	  x	  in	  current	  health	  until	  they	  become	  indifferent	  between	  the	  time	  in	  their	  current	  health	  state	  and	  time	  in	  described	  health	  state.	  At	  this	  point,	  the	  health	  utility	  was	  then	  calculated	  as  ℎ? = 𝑥/𝑡.	  The	  utility	  obtained	  was	  anchored	  between	  0	  (death)	  and	  1	  (current	  health).	  For	  the	  stroke	  health	  states,	  the	  respondents	  were	  given	  the	  choice	  of	  living	  in	  the	  health	  state	  for	  10	  years	  55	  	  followed	  by	  painless	  death,	  or	  in	  current	  health	  for	  any	  amount	  of	  time	  less	  than	  10	  years.	  Figure	  3.1	  shows	  the	  TTO	  for	  debilitating	  stroke.	  	  	  To	  obtain	  the	  health	  utility	  for	  the	  temporary	  health	  states	  of	  major	  and	  minor	  bleed,	  a	  modified	  TTO	  known	  as	  the	  chained	  method	  was	  adopted	  (54).	  This	  approach	  was	  previously	  proposed	  by	  Torrance	  (49),	  and	  has	  been	  applied	  in	  measuring	  health	  utilities	  for	  temporary	  health	  states	  in	  other	  conditions	  (54,164).	  The	  chained	  method	  requires	  the	  set	  up	  of	  an	  additional	  hypothetical	  health	  state	  known	  as	  the	  anchor	  health	  state	  (ℎ?),	  which	  should	  have	  a	  utility	  lower	  than	  the	  temporary	  health	  state	  of	  interest	  (ℎ?)	  but	  not	  worse	  than	  death.	  The	  respondents	  were	  asked	  to	  choose	  between	  living	  a	  specified	  time	  in	  the	  temporary	  health	  state	  (ℎ?),	  or	  a	  shorter	  amount	  of	  time	  in	  the	  anchor	  health	  state	  (hA),	  following	  both	  of	  which	  they	  will	  live	  out	  the	  rest	  of	  their	  lives	  in	  their	  current	  health.	  Similar	  to	  TTO,	  the	  time	  in	  ℎ?	  is	  varied	  until	  a	  point	  of	  indifference,	  and	  the	  utility	  of	  the	  temporary	  health	  state	  Q	  is	  then	  calculated	  as	  ℎ? = 1− (1− ℎ?) ∗ 𝑥/𝑡.	  For	  this	  study,	  a	  hypothetical	  state	  of	  severe	  pneumonia	  was	  used	  as	  the	  anchor	  health	  state,	  the	  utility	  of	  which	  was	  first	  measured	  with	  the	  conventional	  TTO	  for	  chronic	  states:	  ℎ? = 𝑥/𝑡.	  Severe	  pneumonia	  was	  chosen	  as	  the	  anchor	  health	  state	  as	  it	  would	  require	  temporary	  mechanical	  ventilation	  and	  general	  anesthesia	  (see	  Appendix	  A	  for	  detailed	  description),	  and	  thus	  expected	  to	  have	  a	  utility	  estimate	  lower	  than	  the	  two	  temporary	  health	  states	  of	  interest	  (i.e.	  major	  and	  minor	  bleed).	  	  The	  respondents	  were	  then	  asked	  to	  trade-­‐off	  days	  (maximum	  of	  14	  days)	  in	  major	  or	  minor	  bleed	  to	  spend	  in	  fewer	  days	  with	  severe	  pneumonia,	  followed	  by	  their	  current	  health.	  In	  other	  words,	  56	  	  respondents	  are	  asked	  to	  trade	  off	  days	  in	  the	  temporary	  health	  states	  for	  shorter	  amount	  of	  time	  in	  a	  more	  severe	  anchor	  state,	  such	  that	  they	  can	  gain	  more	  days	  in	  their	  current	  health	  state. Best-­‐worst	  scaling	  (BWS)	  choice	  experiment	  Best-­‐worst	  scaling	  (BWS)	  is	  an	  adaptation	  of	  the	  discrete	  choice	  experiment	  (DCE)	  based	  on	  the	  random	  utility	  theory	  as	  described	  in	  Chapter	  1.	  In	  a	  BWS	  experiment,	  the	  respondent	  is	  shown	  a	  task	  with	  multiple	  attribute	  levels,	  and	  asked	  to	  choose	  the	  most	  preferred	  and	  least	  preferred	  attributes	  of	  the	  task.	  The	  pair	  of	  the	  most	  preferred	  and	  least	  preferred	  attribute	  levels	  is	  considered	  to	  be	  furthest	  apart	  on	  an	  underlying	  relative	  preference	  scale	  (56).	  After	  multiple	  scenario	  tasks	  where	  all	  possible	  pairs	  of	  attribute	  levels	  have	  been	  shown,	  the	  relative	  preference	  for	  that	  attribute	  level	  can	  be	  determined	  by	  the	  propensity	  it	  is	  chosen	  as	  the	  best	  (or	  least)	  relative	  to	  other	  attribute	  levels	  on	  a	  common	  underlying	  relative	  preference	  scale	  (56,165).	  This	  allows	  the	  direct	  comparison	  of	  relative	  preference	  across	  all	  attribute	  levels	  –	  a	  key	  advantage	  of	  BWS	  that	  sets	  this	  type	  of	  choice	  experiment	  apart	  from	  DCE.	  	  	  In	  this	  study,	  the	  case	  2	  (profile	  case)	  BWS	  was	  adopted	  to	  develop	  the	  experiment	  (60).	  Attribute	  and	  levels	  were	  selected	  through	  literature	  review	  of	  antithrombotics	  for	  stroke	  prophylaxis	  in	  AF	  as	  well	  as	  consultation	  with	  clinical	  experts.	  In	  consideration	  of	  optimizing	  attribute	  relevance	  and	  minimizing	  respondent	  burden,	  four	  attributes	  were	  selected	  for	  inclusion	  in	  BWS	  exercise:	  frequency	  of	  blood	  test	  monitoring,	  annual	  risk	  of	  stroke,	  annual	  risk	  57	  	  of	  major	  bleed,	  availability	  of	  reversal	  agents	  (Table	  3.1).	  Levels	  associated	  with	  each	  attribute	  were	  selected	  to	  reflect	  current	  practice.	  For	  example,	  the	  range	  of	  risk	  for	  annual	  risk	  of	  stroke	  were	  based	  on	  the	  population	  stroke	  risk	  in	  AF,	  and	  were	  chosen	  to	  cover	  a	  meaningful	  range	  of	  risk	  with	  or	  without	  oral	  antithrombotics.	  Prior	  to	  beginning	  the	  choice	  questionnaire,	  the	  respondents	  were	  provided	  with	  a	  detailed	  description	  of	  the	  attributes	  and	  shown	  a	  BWS	  task	  example.	  Each	  respondent	  were	  then	  asked	  to	  complete	  16	  BWS	  tasks	  (see	  Appendix	  A	  for	  an	  example	  of	  full	  BWS	  task)	  	  Sawtooth®	  software	  Webv6.0	  (Sawtooth	  Software	  Inc.	  Sequim,	  WA,	  USA)	  was	  used	  to	  design	  the	  BWS	  questionnaire	  using	  near	  optimal	  plans	  (166).	  While	  the	  design	  by	  SSI	  cannot	  ensure	  equal	  frequency	  of	  attribute	  pairs	  shown,	  near	  balance	  is	  obtained	  across	  multiple	  versions	  and	  respondents.	  For	  this	  study,	  one-­‐way	  and	  two-­‐way	  frequencies	  were	  shown	  to	  be	  optimally	  balanced	  across	  4	  versions	  of	  questionnaires,	  with	  a	  16	  total	  attribute	  levels	  and	  16	  BWS	  tasks	  per	  respondents.	  	  	  3.2.3 Data	  analysis	  Sample	  characteristics	  and	  time	  trade-­‐off	  utilities	  Descriptive	  statistics	  were	  computed	  to	  assess	  the	  characteristics	  of	  the	  survey	  respondents	  using	  SAS	  (v.9.2,	  SAS	  Institute	  Inc.,	  Cary,	  NC).	  For	  the	  TTO	  and	  chained	  TTO,	  means	  and	  medians	  of	  the	  utilities	  for	  each	  of	  the	  four	  health	  states	  of	  interest	  were	  calculated,	  in	  addition	  to	  the	  patients’	  preferences	  for	  their	  current	  health	  state.	  Wilcoxon-­‐signed	  rank	  test	  was	  used	  to	  58	  	  compare	  the	  utilities	  between	  the	  two	  stroke	  health	  states	  and	  the	  two	  bleeding	  health	  states,	  as	  well	  as	  to	  compare	  utilities	  by	  subgroups.	  	  	  Best-­‐worst	  scaling	  relative	  utilities	  Relative	  utilities	  based	  on	  the	  BWS	  choice	  experiment	  were	  first	  analyzed	  using	  conditional	  logistic	  regression	  where	  I	  modeled	  the	  effects	  of	  attribute	  levels	  on	  the	  ranked	  choice.	  The	  probability	  that	  case	  𝑖	  (respondent)	  chooses	  alternative	  𝑚  at	  replication	  𝑡  given	  attribute	  values	  Ζ™ ™? 	  and	  predictor	  values	  Ζ™ ™? 	  can	  be	  estimated	  using	  the	  following:	  𝑃 𝒴™ =   𝑚 Ζ™ ™? , Ζ™ ™? =    exp 𝜂? ? ™exp(?𝓂??? 𝜂?? ? ™ )	  (Eq.	  3.1)	  where	  𝜂? ? ™ 	  is	  the	  systematic	  component	  in	  the	  utility	  of	  choice	  𝑚  at	  replication	  𝑡	  for	  case	  𝑖.	  The	  parameter	  coefficient	  for	  each	  alternative	  𝑚	  describes	  the	  respondents’	  preference	  for	  that	  attribute	  level	  where	  the	  large	  estimates	  indicate	  positive	  impact	  of	  that	  attribute	  level	  on	  the	  ranked	  choice.	  	  In	  the	  case	  of	  this	  BWS	  experiment,	  it	  was	  planned	  a	  priori	  to	  set	  the	  most	  preferred	  level	  as	  the	  reference	  such	  that	  all	  parameter	  estimates	  would	  be	  negative	  and	  interpreted	  in	  relation	  to	  the	  most	  preferred	  level	  on	  a	  common	  scale.	  	  	  The	  conditional	  logit	  model	  is	  predisposed	  to	  biased	  preference	  estimation	  as	  it	  does	  not	  account	  for	  heterogeneity	  in	  respondent	  choices	  or	  variance	  scale.	  Thus,	  I	  also	  used	  a	  scale-­‐adjusted	  latent	  class	  analysis	  (LCA)	  to	  evaluate	  respondent	  preferences.	  	  The	  LCA	  is	  based	  on	  the	  framework	  that	  there	  are	  different	  discrete	  groups	  or	  ‘classes’	  of	  preferences,	  which	  can	  be	  59	  	  characterized	  by	  observed	  as	  well	  as	  unobserved	  (latent)	  variables.	  The	  class	  membership	  of	  each	  respondent	  can	  be	  assigned	  based	  on	  the	  structure	  of	  their	  preference.	  Thus,	  LCA	  allows	  a	  better	  profiling	  of	  the	  choice	  probability	  depending	  on	  the	  respondent’s	  class	  membership	  𝑥:	  	  	  𝑃 𝒴™ =   𝑚 𝑥, Ζ™ ™? , Ζ™ ™? , 𝑆™ =    exp(𝑆™    ∙   𝜂? ?,? ™ )exp(?𝓂? 𝑆™    ∙   𝜂?? ?,? ™ )	  (Eq.	  3.2)	  where	  the	  systematic	  component	  𝜂? ?,? ™ 	  has	  also	  now	  become	  class-­‐specific.	  The	  term	  𝑆™   is	  a	  scale	  factor	  assumed	  to	  be	  constant	  across	  alternatives	  within	  a	  replication.	  In	  the	  case	  of	  best-­‐worst	  choices,	  the	  best	  and	  worst	  choices	  take	  the	  scale	  factor	  of	  1	  and	  -­‐1,	  respectively.	  	  	  To	  determine	  the	  preference	  estimates	  of	  AF	  patient	  respondents,	  I	  effect	  coded	  the	  BWS	  responses	  and	  estimated	  models	  using	  the	  LCA	  for	  up	  to	  six	  classes.	  Model	  fit	  parameters	  including	  the	  log-­‐likelihood	  function	  (LL),	  the	  Bayesian	  Information	  Criteria	  (BIC),	  and	  the	  Akaike	  Information	  Criteria	  (AIC)	  were	  used	  to	  assess	  the	  optimal	  model	  fit.	  Using	  both	  univariate	  analyses	  and	  backwards	  selection	  methods,	  I	  included	  demographic	  covariates	  that	  potentially	  explain	  the	  difference	  across	  the	  latent	  classes.	  Similar	  to	  the	  conditional	  logit	  model,	  the	  most	  preferred	  attribute	  level	  was	  chosen	  as	  the	  reference	  level	  such	  that	  all	  parameter	  coefficients	  estimates	  could	  be	  interpreted	  relative	  to	  the	  reference.	  Latent	  Gold	  Choice	  (v.4.5,	  Statistical	  Innovations,	  Inc.,	  Belmont,	  MA)	  was	  used	  to	  carry	  out	  the	  LCA	  in	  this	  study.	  	  	  	  60	  	  Best-­‐worst	  scaling	  relative	  importance	  In	  interpreting	  the	  results	  for	  choice-­‐based	  response	  data,	  it	  is	  often	  helpful	  to	  characterize	  the	  relative	  importance	  of	  each	  attribute.	  Importance	  is	  essentially	  the	  maximum	  effect	  an	  attribute	  makes	  in	  the	  total	  utility	  of	  a	  product	  (167,168).	  This	  can	  be	  calculated	  as:	  𝑚𝑎𝑥𝑒𝑓𝑓™ = max 𝜂? ™ −𝑚𝑖𝑛 𝜂? ™ 	  (Eq.	  3.3)	  where	  𝑎	  denotes	  a	  level	  of	  attribute	  𝑝,	  and	  𝜂? ™ is	  the	  utility	  associated	  with	  the	  level	  𝑎	  for	  latent	  class	  𝑥.	  The	  difference	  between	  the	  maximum	  and	  minimum	  level	  for	  attribute	  𝑝	  for	  latent	  class	  𝑥	  thus	  defines  𝑚𝑎𝑥𝑒𝑓𝑓™ .	  The	  relative	  importance	  of	  the	  attribute	  𝑝  can	  then	  be	  obtained	  by	  the	  following	  equation	  to	  compare	  these	  maximum	  effects	  across	  attributes	  and	  latent	  classes:	  𝑟𝑒𝑙𝑒𝑓𝑓™ = 𝑚𝑎𝑥𝑒𝑓𝑓™𝑚𝑎𝑥𝑒𝑓𝑓™? 	  (Eq.	  3.4)	  Where	   𝑚𝑎𝑥𝑒𝑓𝑓™? 	  is	  the	  sum	  of	  the	  maximum	  effects	  of	  all	  attributes.	  	  3.3 Results	  3.3.1 Sample	  characteristics	  Eighty-­‐one	  potential	  subjects	  were	  invited	  to	  participate	  in	  the	  study	  of	  which	  58	  consented	  to	  and	  completed	  the	  study.	  Twenty-­‐three	  potential	  subjects	  did	  not	  consent	  to	  participation	  for	  the	  primary	  reasons	  of	  lack	  of	  time	  after	  their	  clinic	  visit.	  On	  average,	  it	  took	  the	  respondents	  45	  minutes	  to	  complete	  the	  survey.	  Respondent	  characteristics	  are	  summarized	  in	  Table	  3.2.	  61	  	  Fifty-­‐two	  out	  of	  the	  58	  participants	  were	  recruited	  through	  the	  AF	  clinic	  at	  Vancouver	  General	  Hospital.	  The	  mean	  age	  of	  participants	  was	  63.7	  ±	  11.8	  years	  (range	  28	  to	  90	  years)	  with	  a	  similar	  distribution	  in	  sex.	  Almost	  half	  of	  the	  participants	  were	  married	  (46.6%)	  and	  the	  majority	  did	  not	  have	  any	  dependents	  at	  the	  time	  of	  study	  (81.0%).	  The	  respondents	  were	  mostly	  at	  low	  risk	  for	  stroke	  with	  69%	  having	  a	  CHADS2	  score	  of	  0	  to	  1.	  Nine	  people	  had	  diabetes	  and	  ten	  reported	  having	  had	  at	  least	  one	  stroke	  or	  transient	  ischemic	  attack	  in	  the	  past.	  Majority	  of	  the	  respondents	  (69%)	  had	  paroxysmal	  AF.	  	  	  3.3.2 TTO	  Utilities	  –	  overall	  and	  in	  relevant	  subgroups	  Utilities	  for	  the	  relevant	  health	  states	  are	  as	  shown	  in	  Table	  3.3.	  The	  median	  utilities	  for	  debilitating	  and	  non-­‐debilitating	  stroke	  were	  0.23	  and	  0.80,	  respectively.	  These	  utilities	  for	  the	  stroke	  of	  two	  severities	  were	  distinct	  as	  shown	  by	  the	  non-­‐overlapping	  interquartile	  range	  as	  well	  as	  the	  result	  of	  Wilcoxon	  Signed	  Rank	  Test	  (p<0.001).	  Using	  the	  chained	  TTO	  method,	  the	  median	  utilities	  for	  major	  and	  minor	  bleed	  were	  calculated	  to	  be	  0.89	  and	  1.00,	  respectively.	  Utilities	  from	  individual	  respondents	  trended	  in	  the	  expected	  direction,	  which	  lends	  validity	  to	  participants’	  response.	  Of	  interest,	  the	  utility	  for	  severe	  pneumonia	  was	  0.29,	  lower	  than	  both	  of	  minor	  and	  major	  bleed	  and	  thus,	  a	  valid	  anchor	  health	  state	  for	  the	  chained	  TTO	  tasks.	  The	  utilities	  for	  the	  patients’	  current	  health	  were	  calculated	  to	  have	  a	  median	  of	  1.00,	  suggesting	  that	  most	  participants	  in	  the	  study	  perceived	  themselves	  to	  be	  in	  good	  health.	  	  	  62	  	  Utilities	  were	  calculated	  for	  relevant	  subgroups	  of	  interest	  (Table	  3.4).	  There	  was	  notable	  difference	  in	  median	  utilities	  between	  certain	  subgroups	  but	  those	  were	  not	  found	  to	  be	  statistically	  significant,	  except	  for	  the	  utilities	  for	  major	  bleed	  between	  the	  elderly	  and	  non-­‐elderly	  respondents.	  The	  median	  utilities	  were	  found	  to	  be	  0.75	  and	  0.92	  for	  respondents	  aged	  75	  years	  and	  above	  and	  those	  less	  than	  75	  years,	  respectively.	  	  3.3.3 BWS	  –	  relative	  utilities	  Model	  estimation	  Initial	  model	  estimation	  showed	  “annual	  risk	  of	  0%”	  as	  the	  most	  preferred	  attribute	  level;	  hence,	  all	  subsequent	  models	  were	  analyzed	  with	  that	  level	  as	  the	  reference.	  The	  estimated	  utilities	  for	  each	  attribute	  level	  relative	  to	  “annual	  risk	  of	  stroke	  of	  0%”	  are	  described	  in	  Table	  3.5.	  As	  expected,	  a	  0%	  stroke	  risk	  was	  the	  most	  valued	  attribute	  level,	  followed	  by	  “annual	  risk	  of	  major	  bleed	  of	  0%”	  with	  a	  mean	  relative	  utility	  estimate	  of	  -­‐2.29	  (0.38).	  Having	  a	  reversibility	  agent	  available	  and	  having	  laboratory	  tests	  done	  only	  once	  a	  year	  were	  the	  next	  most	  preferred	  attribute	  levels	  with	  relative	  utility	  estimates	  of	  -­‐3.46	  (0.40)	  and	  -­‐3.48	  (0.40),	  respectively.	  Those	  were	  followed	  by	  having	  laboratory	  testing	  done	  every	  6	  months	  and	  3	  months	  with	  mean	  relative	  utilities	  of	  -­‐4.68	  (0.41)	  and	  -­‐4.70	  (0.41),	  respectively.	  On	  the	  other	  hand,	  annual	  stroke	  risk	  of	  6%,	  8%	  and	  10%	  had	  the	  lowest	  mean	  relative	  utilities	  of	  -­‐8.44	  (0.44),	  -­‐9.75	  (0.46),	  and	  -­‐11.36	  (0.52),	  respectively.	  All	  utility	  estimates	  were	  significant	  at	  p<0.001,	  indicating	  that	  all	  attribute	  levels	  significantly	  impacted	  the	  respondent’s	  choice.	  Further,	  all	  parameter	  63	  	  estimates	  for	  the	  attribute	  levels	  are	  trending	  in	  the	  expected	  direction,	  lending	  validity	  to	  the	  choice	  data.	  	  	  Latent	  class	  analysis	  Comparison	  of	  six	  different	  latent	  class	  models	  using	  AIC,	  BIC	  and	  log-­‐likelihood	  estimates	  suggested	  that	  a	  three-­‐class	  model	  best	  represented	  the	  heterogeneity	  of	  respondents’	  preferences	  in	  this	  study	  sample.	  Inclusion	  of	  explanatory	  variables,	  specifically	  sex,	  age	  greater	  than	  75,	  hypertension,	  diabetes	  and	  current	  warfarin	  therapy	  further	  improved	  the	  model	  fit.	  Better	  model	  statistics	  of	  the	  three-­‐class	  models	  compared	  to	  the	  conditional	  logit	  model	  suggests	  heterogeneity	  in	  choice	  response.	  Utility	  estimates	  from	  the	  three-­‐class	  model	  are	  described	  in	  Table	  3.5.	  	  The	  chance	  of	  a	  respondent	  being	  assigned	  to	  class	  1	  is	  the	  highest	  with	  a	  probability	  of	  56.5%,	  followed	  by	  class	  2	  with	  a	  probability	  of	  31.2%.	  A	  0%	  risk	  of	  stroke	  remained	  the	  most	  valued	  attribute	  level	  across	  all	  three	  classes.	  However,	  those	  in	  class	  1	  had	  a	  much	  stronger	  negative	  preference	  for	  annual	  risk	  of	  stroke	  10%	  than	  respondents	  in	  class	  2	  and	  3	  with	  relative	  mean	  utilities	  of	  -­‐16.63	  (1.16),	  -­‐10.64	  (0.87)	  and	  -­‐10.61	  (1.45),	  respectively.	  This	  difference	  persisted	  for	  all	  other	  risk	  of	  stroke	  above	  2%	  and	  was	  most	  apparent	  in	  Figure	  3.1,	  suggesting	  participants	  in	  this	  class	  to	  be	  more	  stroke	  averse	  than	  those	  in	  class	  2	  and	  3.	  This	  was	  true	  for	  annual	  risk	  of	  major	  bleed	  as	  well,	  where	  relative	  utilities	  were	  much	  lower	  for	  those	  in	  class	  1	  compared	  to	  the	  other	  classes.	  Of	  note,	  those	  in	  class	  1	  also	  have	  the	  strongest	  preferences	  in	  general	  as	  shown	  by	  the	  largest	  range	  covered	  by	  the	  utility	  estimates,	  ranging	  from	  0	  to	  -­‐16.63.	  Those	  in	  class	  2	  had	  less	  distinguishable	  preferences	  between	  attribute	  levels	  64	  	  compared	  to	  class	  1	  and	  class	  3,	  with	  the	  smallest	  range	  of	  estimates.	  Although	  the	  probability	  of	  being	  assigned	  to	  class	  3	  was	  small	  (12.3%),	  those	  respondents	  had	  a	  strong	  negative	  preference	  for	  not	  having	  a	  reversibility	  agent	  available	  as	  indicated	  by	  a	  mean	  utility	  of	  -­‐11.18.	  Those	  in	  class	  3	  also	  preferred	  to	  have	  laboratory	  tests	  done	  more	  regularly,	  whereas	  those	  in	  class	  1	  and	  2	  preferred	  less	  frequent	  laboratory	  monitoring.	  	  	  The	  relative	  importance	  of	  attributes	  is	  presented	  as	  aggregates	  and	  illustrated	  in	  Figure	  3.2.	  As	  shown,	  annual	  risk	  of	  stroke	  was	  the	  most	  important	  attribute	  driving	  respondent	  choices	  across	  all	  class,	  where	  those	  in	  class	  1	  valued	  this	  attribute	  more	  compared	  to	  the	  other	  classes.	  Similarly,	  all	  classes	  found	  frequency	  of	  laboratory	  monitoring	  to	  be	  the	  least	  important	  attribute	  in	  making	  their	  choices	  for	  stroke	  prophylaxis.	  Those	  in	  class	  1	  and	  2	  did	  not	  find	  having	  reversibility	  agent	  to	  be	  an	  important	  attribute	  in	  making	  their	  choice.	  Interestingly,	  those	  in	  class	  3	  found	  reversibility	  agent	  to	  be	  an	  attribute	  more	  important	  than	  annual	  risk	  of	  major	  bleed	  and	  almost	  as	  important	  as	  annual	  risk	  of	  stroke.	  	  	  	  	  In	  fitting	  the	  latent	  class	  models,	  inclusion	  of	  some	  covariates	  were	  found	  to	  improve	  model	  fit	  parameters.	  Those	  covariates	  included	  current	  use	  of	  warfarin,	  hypertension,	  diabetes,	  sex	  and	  age.	  Those	  most	  stroke	  and	  bleed	  averse	  (i.e.	  class	  1)	  are	  more	  likely	  to	  be	  taking	  warfarin.	  Individuals	  belonging	  to	  class	  2	  are	  more	  likely	  to	  be	  hypertensive	  and	  taking	  warfarin	  at	  the	  time	  of	  the	  study.	  Lastly,	  those	  in	  class	  3	  are	  most	  likely	  to	  have	  hypertension	  but	  not	  on	  warfarin.	  Parameters	  for	  sex,	  age	  and	  diabetes	  were	  not	  found	  to	  be	  statistically	  significant;	  65	  	  however,	  these	  covariates	  are	  potentially	  important	  in	  explaining	  the	  variation	  amongst	  class	  memberships	  as	  model	  fit	  improved	  with	  the	  inclusion	  of	  these	  variables.	  	  	  3.4 Discussion	  The	  current	  study	  measured	  both	  the	  utilities	  associated	  with	  outcomes	  in	  stroke	  prophylaxis	  in	  AF	  using	  the	  TTO	  elicitation	  method	  as	  well	  as	  relative	  utilities	  of	  additional	  attributes	  related	  to	  stroke	  prophylaxis	  using	  the	  BWS	  choice	  experiment.	  From	  the	  TTO	  task,	  AF	  patients	  were	  found	  to	  have	  the	  highest	  disutility	  for	  debilitating	  stroke,	  followed	  by	  non-­‐debilitating	  stroke,	  major	  bleed	  and	  minor	  bleed.	  In	  the	  BWS	  choice	  experiment,	  I	  was	  able	  to	  further	  characterize	  AF	  patients’	  and	  physicians’	  preferences	  for	  attributes	  associated	  with	  stroke	  prophylaxis.	  	  To	  date,	  two	  other	  studies	  have	  used	  cardinal	  preferences	  to	  describe	  patient	  preferences	  for	  stroke.	  Gage	  et	  al.	  elicited	  patient	  preferences	  for	  three	  degrees	  of	  severity	  of	  stroke	  from	  70	  patients	  with	  AF	  (110).	  They	  reported	  mean	  utilities	  of	  0.76,	  0.39,	  0.11	  for	  mild,	  moderate	  and	  major	  stroke,	  respectively.	  In	  another	  study,	  Thomson	  et	  al.	  elicited	  preferences	  from	  57	  AF	  patients	  using	  the	  SG	  method	  and	  reported	  mean	  utilities	  of	  0.64	  and	  0.19	  for	  mild	  and	  severe	  stroke,	  respectively	  (161).	  The	  utilities	  for	  debilitating	  and	  non-­‐debilitating	  stroke	  in	  our	  study,	  which	  is	  deemed	  equivalent	  to	  a	  major	  stroke	  on	  the	  Rankin	  scale	  were	  found	  to	  be	  higher	  than	  both	  studies	  done	  previously.	  It	  was	  suspected	  that	  the	  difference	  is	  potentially	  influenced	  by	  the	  respondents’	  health	  state	  at	  the	  time	  of	  the	  survey;	  however,	  no	  association	  in	  utilities	  was	  found	  between	  their	  current	  health	  states	  and	  any	  of	  the	  health	  states	  of	  interest	  including	  66	  	  debilitating	  stroke.	  The	  difference	  between	  the	  studies	  is	  thus	  possibly	  attributed	  to	  the	  significantly	  younger	  patient	  population	  in	  this	  study	  sample.	  The	  mean	  age	  this	  sample	  population	  was	  63,	  compared	  to	  70	  and	  73	  in	  the	  study	  by	  Gage	  et	  al.	  and	  Thomson,	  respectively.	  However,	  similar	  to	  the	  previous	  two	  studies,	  many	  respondents	  (33%)	  valued	  debilitating	  stroke	  as	  equivalent	  to	  death	  with	  a	  utility	  of	  0.	  	  In	  terms	  of	  preference	  for	  major	  bleed,	  the	  utility	  result	  in	  this	  study	  using	  the	  chained	  TTO	  method	  corroborates	  the	  findings	  by	  Thomson	  et	  al.,	  which	  was	  the	  only	  other	  study	  that	  measured	  the	  utility	  for	  major	  bleed	  in	  this	  population	  using	  stated	  preference	  elicitation	  method	  on	  the	  cardinal	  scale.	  	  Comparable	  results	  to	  their	  utility	  for	  major	  bleed	  using	  standard	  gamble	  suggest	  that	  there	  may	  not	  be	  any	  difference	  in	  using	  the	  chained	  TTO	  for	  measuring	  utilities	  of	  temporary	  health	  states.	  However,	  there	  are	  few	  studies	  that	  had	  been	  done	  to	  validate	  this	  generalization	  and	  caution	  should	  be	  applied	  when	  measuring	  cardinal	  utilities	  for	  temporary	  health	  states.	  	  Regardless,	  this	  study	  is	  the	  first	  to	  report	  a	  utility	  for	  minor	  bleed	  in	  this	  population	  and	  found	  that	  patients	  did	  not	  value	  minor	  bleed	  as	  a	  significant	  detriment	  to	  health.	  	  	  In	  addition	  to	  measuring	  patients’	  utilities	  for	  health	  states	  associated	  with	  stroke	  prophylaxis	  in	  AF,	  this	  study	  also	  elicited	  the	  relative	  preference	  of	  other	  important	  attributes	  using	  a	  BWS	  choice	  experiment.	  To	  my	  knowledge,	  this	  is	  the	  first	  study	  that	  used	  a	  preference	  elicitation	  method	  on	  the	  ordinal	  scale	  to	  examine	  other	  important	  factors	  in	  selecting	  antithrombotics	  of	  67	  	  stroke	  prophylaxis.	  Overall,	  results	  of	  this	  study	  indicated	  that	  AF	  patients	  value	  annual	  risk	  of	  stroke	  to	  be	  the	  most	  important	  attribute	  in	  stroke	  prophylaxis	  compared	  to	  annual	  risk	  of	  major	  bleed,	  frequency	  of	  laboratory	  tests	  and	  availability	  of	  reversal	  agents.	  This	  was	  mostly	  consistent	  amongst	  subgroups	  discerned	  according	  to	  observed	  and	  latent	  factors.	  While	  “annual	  stroke	  risk	  of	  0%”	  is	  in	  theory	  an	  unrealistic	  attribute	  level,	  this	  attribute	  level	  was	  included	  it	  in	  the	  BWS	  task	  to	  test	  whether	  preferences	  for	  different	  stroke	  risk	  follow	  a	  linear	  relationship.	  The	  large	  decrement	  of	  preference	  weights	  after	  annual	  risk	  of	  stroke	  dropped	  below	  2%	  suggested	  that	  this	  was	  in	  fact	  a	  non-­‐linear	  relationship,	  and	  that	  the	  patients’	  threshold	  for	  an	  acceptable	  stroke	  risk	  is	  somewhere	  between	  0%	  and	  2%.	  Similar	  observation	  is	  made	  for	  annual	  risk	  of	  major	  bleed,	  except	  that	  the	  decrement	  in	  preference	  was	  not	  as	  significant	  with	  decreased	  risk	  of	  major	  bleed	  below	  2%.	  In	  considering	  other	  attributes	  for	  stroke	  prophylaxis,	  the	  study	  findings	  suggested	  that	  most	  patients	  placed	  the	  least	  importance	  on	  frequency	  of	  laboratory	  monitoring	  when	  compared	  to	  other	  attributes	  like	  risk	  of	  major	  bleed	  and	  availability	  of	  reversal	  agents.	  	  By	  using	  the	  latent	  class	  analysis,	  which	  can	  identify	  segments	  of	  individuals	  with	  similar	  preferences,	  this	  study	  was	  able	  to	  further	  explore	  the	  heterogeneity	  in	  respondent	  preferences	  as	  well	  as	  potential	  factors	  that	  identify	  the	  respondents	  in	  each	  group.	  Specifically,	  these	  finding	  suggested	  that	  half	  of	  the	  sample	  are	  very	  much	  stroke	  and	  bleed	  averse,	  as	  indicated	  by	  those	  in	  class	  1.	  	  Respondents	  in	  class	  2	  followed	  a	  similar	  ranking	  profile	  as	  class	  1,	  but	  with	  a	  smaller	  magnitude	  of	  preference	  strength	  compared	  to	  those	  in	  class	  1.	  Both	  class	  1	  68	  	  and	  2	  respondents	  were	  least	  concerned	  about	  whether	  a	  reversal	  agent	  is	  available	  relative	  to	  the	  other	  attributes.	  The	  exception	  is	  a	  small	  group	  of	  respondents	  in	  class	  3,	  who	  valued	  the	  availability	  of	  reversal	  agent	  more	  important	  than	  the	  risk	  of	  major	  bleed,	  and	  almost	  as	  important	  as	  the	  risk	  of	  stroke.	  Interestingly,	  while	  the	  chance	  of	  being	  in	  class	  3	  is	  small,	  respondents	  in	  this	  group,	  unlike	  respondents	  in	  the	  other	  classes,	  also	  preferred	  to	  have	  laboratory	  monitoring	  done	  more	  frequently.	  It	  is	  not	  known	  and	  difficult	  to	  speculate	  what	  characteristics	  in	  this	  group	  drove	  their	  preference.	  From	  the	  LCM,	  those	  in	  class	  3	  are	  less	  likely	  to	  be	  taking	  warfarin	  and	  more	  likely	  to	  have	  hypertension.	  It	  is	  possible	  that	  those	  who	  were	  not	  on	  warfarin	  had	  previously	  experienced	  difficulty	  maintaining	  INR	  or	  a	  bleeding	  events	  such	  that	  they	  were	  accustomed	  to	  the	  idea	  of	  having	  more	  frequent	  blood	  monitoring	  as	  the	  optimal	  care.	  However,	  prior	  history	  of	  bleeding	  was	  looked	  at	  as	  a	  potential	  covariate	  in	  LCM	  and	  did	  not	  seem	  to	  affect	  the	  choice	  model	  fit,	  so	  it	  would	  not	  appear	  that	  bleeding	  provide	  a	  secondary	  link	  to	  lack	  of	  warfarin	  use	  and	  preference	  of	  those	  in	  class	  3.	  For	  those	  in	  class	  1	  who	  are	  more	  stroke	  and	  bleed	  averse,	  respondents	  were	  more	  likely	  to	  have	  diabetes.	  This	  makes	  some	  clinical	  sense	  as	  those	  with	  diabetes	  in	  addition	  to	  AF	  had	  more	  awareness	  of	  their	  increased	  risk	  of	  having	  cardiovascular	  events,	  including	  stroke	  compared	  to	  those	  without	  DM.	  	  	  One	  potential	  limitation	  of	  this	  study	  is	  the	  conditional	  logit	  regression	  that	  was	  used	  to	  model	  the	  overall	  BWS	  choice	  data	  given	  this	  analysis	  may	  not	  account	  for	  the	  correlation	  of	  repeated	  choice	  tasks	  within	  an	  individual	  respondent	  as	  well	  as	  the	  heterogeneity	  amongst	  the	  respondents.	  One	  would	  argue	  that	  adopting	  a	  mixed	  logit	  model	  would	  be	  a	  more	  appropriate	  69	  	  analysis	  for	  this	  type	  of	  choice	  experiment.	  To	  test	  this	  assumption,	  the	  same	  data	  was	  analyzed	  using	  a	  mixed	  logit	  model	  in	  SAS	  and	  the	  parameter	  estimates	  and	  errors	  were	  found	  to	  be	  comparable	  on	  all	  parts.	  In	  addition,	  by	  using	  the	  latent	  class	  model	  I	  was	  able	  to	  account	  for	  any	  correlation	  within	  individuals	  as	  well	  as	  allowing	  unobserved	  variables	  to	  explain	  the	  heterogeneity	  in	  respondent	  choices,	  a	  feature	  that	  is	  not	  included	  in	  the	  mixed	  logit	  model.	  	  	  The	  study	  sample	  was	  recruited	  from	  two	  local	  institutions	  in	  BC;	  therefore,	  caution	  should	  be	  applied	  before	  generalizing	  the	  results	  to	  other	  populations.	  Also,	  there	  was	  inherent	  selection	  bias	  as	  I	  only	  included	  respondents	  who	  were	  willing	  to	  participate	  in	  the	  study,	  thus	  the	  preference	  expressed	  by	  patients	  in	  this	  sample	  may	  not	  be	  representative	  of	  all	  patients	  with	  AF.	  However,	  similar	  TTO	  results	  to	  previously	  published	  studies	  suggest	  validity	  of	  these	  study	  findings,	  and	  bias	  arising	  from	  sample	  selection	  should	  be	  negligible.	  	  	  	  This	  study	  is	  the	  first	  to	  utilize	  both	  cardinal	  and	  ordinal	  preference	  elicitation	  methods	  in	  the	  AF	  population	  with	  regards	  to	  stroke	  prophylaxis.	  Results	  from	  the	  TTO	  task	  corroborates	  with	  previous	  studies	  on	  major,	  minor	  stroke	  and	  major	  bleed.	  In	  addition,	  it	  was	  shown	  that	  the	  utility	  for	  minor	  bleed	  is	  not	  significantly	  different	  from	  optimal	  health,	  which	  is	  previously	  unknown	  and	  would	  be	  useful	  from	  the	  cost-­‐utility	  analysis	  perspective.	  Findings	  from	  this	  BWS	  choice	  experiment	  provided	  rich	  information	  on	  other	  attribute	  levels	  that	  are	  relevant	  to	  patient	  in	  stroke	  prophylaxis	  for	  AF.	  The	  results	  of	  this	  study	  further	  our	  understanding	  of	  patient	  preferences	  in	  this	  setting,	  and	  inform	  the	  health	  care	  system	  of	  patient	  values	  70	  	  important	  in	  maximizing	  the	  potential	  benefits	  for	  whom	  these	  class	  of	  therapeutic	  agents	  are	  indicated	  for.	  	  	  	  	  	  	  	  	  	  	  	  	  	  	  	  	  	  	  	  	  71	  	  Table	  3.1:	  Attributes	  and	  levels	  in	  BWS	  choice	  experiment	  Attributes	   Levels	  Frequency	  of	  laboratory	  test	   Once	  every	  month	  Once	  every	  3	  months	  Once	  every	  6	  months	  Once	  a	  year	  Annual	  risk	  of	  stroke	   0%	  2%	  4%	  6%	  8%	  10%	  Annual	  risk	  of	  major	  bleed	   0%	  2%	  4%	  6%	  Availability	  of	  reversibility	  agent	   Available	  Not	  available	  	  	  	  	  	  	  	  	  	  	  	  	  72	  	  Table	  3.2:	  Characteristics	  of	  patient	  participants	  (N=58)	  Characteristics	  	  	  	  	  	  	  	   	  Age	  (years)	   	  Mean	  (SD)	   63.7	  (11.8)	  Range	  (min-­‐max)	   28	  –	  90	  Sex	  (N,	  %)	   	  Male	   31	  (53.4)	  Female	   27	  (46.6)	  Annual	  Income	  (N,	  %)	   	  Less	  than	  $40,000	   15	  (25.9)	  $40,000-­‐$80,000	   16	  (27.6)	  $80,000+	   27	  (46.6)	  Highest	  Education	  Completed	  (N,	  %)	   	  High	  School	  diploma	   19	  (32.8)	  Post-­‐secondary	  or	  equivalent	   29	  (50.0)	  Graduate	  degree	   10	  (17.2)	  Marital	  Status	  (N,	  %)	   	  Single	   17	  (29.3)	  Married,	  common-­‐law	   27	  (46.6)	  Widowed	   7	  (12.1)	  Dependents	  (N,	  %)	   	  Yes	   9	  (15.5)	  No	   47	  (81.0)	  CHADS2	  Score	  (N,	  %)	   	  0	  or	  1	   40	  (69.0)	  2	   9	  (15.5)	  3-­‐6	   9	  (15.5)	  Other	  AF	  Risk	  Factors	  (N,	  %)	   	  Diabetes	   9	  (15.5)	  Heart	  failure	   4	  (6.9)	  Prior	  stroke	  or	  transient	  ischemic	  attack	   10	  (17.2)	  Hypertension	   23	  (39.7)	  Classification	  of	  atrial	  fibrillation	  (N,	  %)	   	  Paroxysmal	   40	  (69.0)	  Persistent	   11	  (19.0)	  Permanent	   3	  (5.17)	  Current	  oral	  antithrombotic	  (N,	  %)	   	  None	   3	  (5.2)	  ASA	   16	  (27.6)	  Warfarin	   9	  (15.5)	  Dabigatran	   16	  (27.6)	  Rivaroxaban	   12	  (20.7)	  Apixaban	   2	  (3.45)	  	  	  	  73	  	  Table	  3.3:	  TTO	  utilities	  for	  each	  health	  state	  (N=58)	  	   Median	  (IQR)	   Mean	  (SD)	  Current	  health	   1.00	  (0.80-­‐1.00)	   0.88	  (0.21)	  Debilitating	  stroke	   0.23	  (0.00-­‐0.50)	   0.31	  (0.31)	  Non-­‐debilitating	  stroke	   0.80	  (0.60-­‐1.00)	   0.79	  (0.24)	  Severe	  Pneumonia	   0.29	  (0.00-­‐0.64)	   0.35	  (0.33)	  Major	  bleed	   0.89	  (0.75-­‐1.00)	   0.84	  (0.16)	  Minor	  bleed	   1.00	  (1.00-­‐1.00)	   0.99	  (0.03)	  	  Table	  3.4:	  TTO	  utilities	  by	  subgroups	  	  Subgroups	  (N)	   Debilitating	  Stroke	   Major	  Bleed	  	   Median	  (IQR)	   Mean	  (SD)	   Median	  (IQR)	   Mean	  (SD)	  Age	  	   	   	   	   	  <75	  (51)	   0.30	  (0.00-­‐0.50)	   0.33	  (0.32)	   0.92	  (0.76-­‐1.00)*	   0.86	  (0.16)	  75+	  (7)	   0.05	  (0.00-­‐0.30)	   0.14	  (0.17)	   0.75	  (0.54-­‐0.86)*	   0.72	  (0.17)	  Sex	   	   	   	   	  Male	  (31)	   0.30	  (0.00-­‐0.50)	   0.37	  (0.32)	   0.92	  (0.74-­‐0.99)	   0.84	  (0.17)	  Female	  (27)	   0.10	  (0.00-­‐0.50)	   0.23	  (0.29)	   0.86	  (0.75-­‐1.00)	   0.85	  (0.15)	  Income	   	   	   	   	  <$40,000	  (15)	   0.30	  (0.00-­‐0.50)	   0.31	  (0.34)	   0.82	  (0.74-­‐1.00)	   0.83	  (0.13)	  $40,000-­‐$80,000	  (16)	   0.10	  (0.00-­‐0.30)	   0.23	  (0.30)	   0.87	  (0.66-­‐1.00)	   0.82	  (0.18)	  $80,000+	  (27)	   0.40	  (0.00-­‐0.50)	   0.35	  (0.30)	   0.93	  (0.75-­‐1.00)	   0.86	  (0.17)	  Marital	  Status	   	   	   	   	  Single,	  widowed	  (28)	   0.08	  (0.00-­‐0.50)	   0.27	  (0.34)	   0.85	  (0.71-­‐0.99)	   0.81	  (0.18)	  Married,	  common-­‐law	  (27)	   0.30	  (0.00-­‐0.50)	   0.33	  (0.28)	   0.94	  (0.79-­‐1.00)	   0.89	  (0.14)	  Dependent†	   	   	   	   	  Yes	  (9)	   0.20	  (0.00-­‐0.20)	   0.26	  (0.32)	   0.93	  (0.79-­‐0.97)	   0.76	  (0.13)	  No	  (47)	   0.30	  (0.00-­‐0.50)	   0.31	  (0.31)	   0.88	  (0.74-­‐1.00)	   0.84	  (0.17)	  CHADS2	  score	   	   	   	   	  0	  or	  1	  (40)	   0.20	  (0.00-­‐0.50)	   0.26	  (0.30)	   0.93	  (0.77-­‐1.00)	   0.86	  (0.18)	  2	  (9)	   0.50	  (0.40-­‐0.65)	   0.49	  (0.29)	   0.75	  (0.71-­‐0.86)	   0.80	  (0.09)	  3-­‐6	  (9)	   0.30	  (0.00-­‐0.50)	   0.31	  (0.32)	   0.85	  (0.79-­‐0.86)	   0.81	  (0.15)	  Prior	  stroke	  history	   	   	   	   	  Yes	  (10)	   0.40	  (0.00-­‐0.50)	   0.34	  (0.33)	   0.85	  (0.79-­‐0.93)	   0.84	  (0.12)	  No	  (48)	   0.20	  (0.00-­‐0.50)	   0.30	  (0.31)	   0.92	  (0.74-­‐1.00)	   0.85	  (0.17)	  CCS	  SAF	  score	   	   	   	   	  0	  or	  1	  (27)	   0.30	  (0.00-­‐0.50)	   0.32	  (0.31)	   0.93	  (0.71-­‐1.00)	   0.86	  (0.16)	  2+	  (31)	   0.20	  (0.00-­‐0.50)	   0.29	  (0.32)	   0.86	  (0.75-­‐0.98)	   0.83	  (0.17)	  *	  indicates	  statistically	  significant	  difference	  between	  groups	  	  †	  two	  missing	  observations	  	   	  74	  	  Table	  3.5:	  Relative	  preferences	  estimates	  from	  conditional	  logit	  and	  latent	  class	  models	  (BWS	  choice	  experiment)	  	   	   Overall	   Class	  1	   Class	  2	   Class	  3	   	  	   Class	  Size	   58	  (100%)	   33	  (56.5%)	   18	  (31.2%)	   7	  (12.3%)	   P	  value	  	   	   Coefficient	  (SE)	   Coefficient	  (SE)	   Coefficient	  (SE)	   Coefficient	  (SE)	   	  Frequency	  of	  Laboratory	  Test	   	   	   	   	   	  Every	  month	   -­‐5.53	  (0.41)	   -­‐6.15	  (0.64)	   -­‐6.46	  (0.72)	   -­‐4.28	  (1.20)	   <0.001	  Every	  3	  months	   -­‐4.70	  (0.41)	   -­‐4.90	  (0.62)	   -­‐5.56	  (0.72)	   -­‐4.76	  (1.20)	   <0.001	  Every	  6	  months	   -­‐4.68	  (0.41)	   -­‐4.37	  (0.59)	   -­‐5.29	  (0.71)	   -­‐5.55	  (1.21)	   <0.001	  Once	  a	  year	   -­‐3.48	  (0.40)	   -­‐3.15	  (0.56)	   -­‐3.38	  (0.68)	   -­‐6.09	  (-­‐4.91)	   <0.001	  Annual	  risk	  of	  stroke	   	   	   	   	   	  0%	   Ref	   Ref	   Ref	   Ref	   	  2%	   -­‐6.07	  (0.42)	   -­‐9.04	  (0.69)	   -­‐4.08	  (0.69)	   -­‐6.19	  (1.23)	   <0.001	  4%	   -­‐7.23	  (0.43)	   -­‐10.08	  (0.71)	   -­‐6.51	  (0.73)	   -­‐5.88	  (1.24)	   <0.001	  6%	   -­‐8.44	  (0.44)	   -­‐11.85	  (0.76)	   -­‐7.64	  (0.73)	   -­‐7.60	  (1.25)	   <0.001	  8%	   -­‐9.75	  (0.46)	   -­‐13.88	  (0.90)	   -­‐8.74	  (0.75)	   -­‐9.56	  (1.38)	   <0.001	  10%	   -­‐11.36	  (0.52)	   -­‐16.63	  (1.16)	   -­‐10.64	  (0.87)	   -­‐10.61	  (1.45)	   <0.001	  Annual	  risk	  of	  bleed	   	   	   	   	   	  0%	   -­‐2.29	  (0.38)	   -­‐2.59	  (0.55)	   -­‐2.04	  (0.64)	   -­‐2.05	  (1.10)	   <0.001	  2%	   -­‐5.51	  (0.41)	   -­‐7.08	  (0.66)	   -­‐5.22	  (0.71)	   -­‐5.23	  (1.21)	   <0.001	  4%	   -­‐6.52	  (0.42)	   -­‐8.44	  (0.71)	   -­‐6.27	  (0.72)	   -­‐6.53	  (1.25)	   <0.001	  6%	   -­‐7.73	  (0.43)	   -­‐10.23	  (0.70)	   -­‐7.46	  (0.71)	   -­‐7.86	  (1.30)	   <0.001	  Reversibility	  Agents	   	   	   	   	   	  Available	   -­‐3.46	  (0.40)	   -­‐4.10	  (0.60)	   -­‐3.33	  (0.69)	   -­‐1.96	  (1.21)	   <0.001	  Not	  Available	   -­‐7.39	  (0.42)	   -­‐8.98	  (0.69)	   -­‐7.17	  (0.72)	   -­‐11.18	  (1.48)	   <0.001	  	   	   	   	   	   	  Covariate	  parameter	  estimates	   	   	   	   	  Intercept	   	   3.26	  (1.22)	   1.93	  (1.21)	   -­‐5.16	  (1.93)	   	  Female	   	   -­‐0.09	  (0.50)	   -­‐0.88	  (0.53)	   0.97	  (0.75)	   0.26	  Age	  >75	   	   -­‐0.22	  (0.67)	   0.52	  (0.68)	   -­‐0.30	  (0.96)	   0.71	  Hypertension	   	   -­‐1.19	  (0.51)	   0.56	  (0.52)	   0.63	  (0.72)	   0.035	  Diabetes	   	   1.28	  (1.11)	   -­‐3.21	  (2.01)	   1.93	  (1.23)	   0.26	  Warfarin	   	   1.54	  (0.69)	   0.27	  (0.66)	   -­‐1.81	  (0.69)	   0.02	  	   	   	   	   	   	  R2	   0.21	   0.33	   0.20	   0.20	   	  R2	  (0)	   0.64	   0.77	   0.63	   0.63	   	  Number	  of	  respondents	   58	   33	   18	   7	   	  Number	  of	  observations	   1856	   1056	   576	   224	   	  	   	   	   	   	   	  Log-­‐likelihood	   -­‐926.62	   	   	   	   	  BIC	   1914.15	   	   	   	   	  Prediction	  error	   0.75	   	   	   	   	  	  	  75	  	  Figure	  3.1:	  Part-­‐worth	  utility	  estimates	  for	  attribute	  levels	  	  	  	  	  	  	  	  	  	  	  	  76	  	  Figure	  3.2:	  Relative	  importance	  of	  attributes	  	  	  	  	  	  	  	  77	  	  Chapter	  4: Physicians’	  preferences	  for	  stroke	  prophylaxis	  in	  atrial	  fibrillation	  4.1 Introduction	  Vitamin	  K	  antagonists	  such	  as	  warfarin	  have	  been	  the	  cornerstone	  for	  stroke	  prophylaxis	  in	  atrial	  fibrillation	  (AF)	  since	  the	  1950s.	  In	  2009,	  the	  promising	  results	  of	  the	  RE-­‐LY	  trial	  demonstrating	  comparable	  efficacy	  and	  better	  safety	  profile	  of	  dabigatran,	  a	  novel	  oral	  antithrombotic	  marked	  a	  milestone	  in	  the	  pharmacotherapeutic	  management	  of	  stroke	  prevention	  (89).	  Since	  then,	  two	  other	  oral	  antithrombotics,	  rivaroxaban	  and	  apixaban	  emerged	  as	  noteworthy	  competitors,	  shown	  by	  their	  non-­‐inferior	  and	  in	  the	  case	  of	  apixaban,	  superior	  efficacy	  and	  safety	  to	  warfarin	  (88,92).	  The	  most	  apparent	  advantage	  of	  these	  novel	  antithrombotics	  is	  the	  lack	  of	  regular	  blood	  monitoring	  requirement	  given	  their	  wider	  therapeutic	  window	  and	  predictable	  anticoagulant	  effect	  compared	  to	  warfarin	  (169).	  These	  new	  antithrombotics	  beckoned	  high	  expectation	  from	  the	  medical	  community	  and	  patients	  as	  they	  were	  thought	  to	  provide	  more	  options	  for	  stroke	  prophylaxis	  in	  AF	  and	  improve	  the	  management	  for	  AF	  patients,	  especially	  those	  who	  would	  be	  deemed	  unsuitable	  for	  warfarin	  therapy.	  	  	  	  Despite	  the	  advantages	  of	  these	  new	  agents,	  there	  remains	  an	  underutilization	  of	  oral	  antithrombotics	  in	  AF	  patients	  (170–173).	  Preliminary	  analysis	  of	  the	  GARFIELD	  registry,	  a	  global	  registry	  of	  more	  than	  55,000	  AF	  patients	  from	  up	  to	  50	  countries,	  showed	  that	  13%	  of	  patients	  at	  high	  risk	  for	  stroke	  were	  not	  receiving	  any	  antithrombotics	  (170).	  The	  PINNACLE	  registry	  with	  over	  120,000	  patients	  with	  AF	  in	  the	  United	  States	  also	  showed	  a	  low	  78	  	  anticoagulation	  rate	  of	  less	  than	  50%	  and	  a	  slow	  uptake	  of	  the	  new	  oral	  antithrombotics	  at	  12.6%	  as	  of	  late	  2011	  (173).	  Two	  studies	  that	  were	  part	  of	  the	  GARFIELD	  registry	  showed	  that	  one-­‐third	  of	  the	  high	  risk	  AF	  patients	  in	  Australia	  were	  not	  receiving	  oral	  antithrombotics.	  The	  investigators	  identified	  that	  the	  most	  common	  reason	  (30%)	  for	  non-­‐prescription	  was	  due	  to	  physician	  preference	  (171,172).	  It	  was	  not	  clear,	  however,	  what	  the	  components	  of	  those	  preferences	  were	  and	  how	  they	  affect	  decision-­‐making	  for	  stroke	  prophylaxis.	  	  	  It	  would	  be	  of	  benefit	  to	  identify	  the	  determinants	  of	  physician	  preferences	  as	  they	  impact	  patient	  care	  at	  the	  clinical	  level,	  as	  well	  as	  drug	  development	  and	  guideline	  formulation	  on	  a	  systematic	  scale.	  Specifically,	  it	  would	  be	  paramount	  to	  quantify	  explicitly	  the	  features	  of	  stroke	  prophylaxis	  that	  determine	  physician	  preferences,	  which	  may	  impact	  their	  choice	  of	  oral	  antithrombotics.	  It	  would	  also	  be	  valuable	  to	  examine	  the	  heterogeneity,	  if	  present,	  amongst	  the	  physicians,	  as	  it	  is	  quite	  plausible	  that	  physicians	  from	  different	  specialties	  might	  have	  different	  preferences	  for	  stroke	  prophylaxis.	  For	  instance,	  emergency	  medicine	  physicians	  most	  often	  encounter	  and	  manage	  major	  bleeds	  as	  a	  consequence	  of	  oral	  antithrombotics	  and	  may	  therefore	  be	  more	  bleeding-­‐averse	  than	  cardiologists	  and	  neurologists,	  who	  might	  be	  more	  stroke	  averse.	  	  	  To	  date,	  there	  is	  limited	  preference	  data	  that	  describe	  physician	  preferences	  quantitatively	  and	  explicitly	  in	  this	  area.	  	  A	  thorough	  literature	  search	  identified	  no	  studies	  that	  specifically	  addressed	  physician	  preferences	  in	  stroke	  prophylaxis.	  There	  were,	  however,	  some	  data	  on	  79	  	  physician	  preferences	  from	  studies	  that	  aimed	  at	  comparing	  physician	  and	  patient	  preferences.	  However,	  all	  of	  these	  studies	  were	  either	  observational,	  or	  focused	  primarily	  on	  stroke	  and	  bleeding	  utilizing	  conjoint	  analysis	  (114,174,175).	  Using	  a	  web-­‐based	  best-­‐worst	  scaling	  (BWS)	  choice	  experiment	  that	  allows	  explicit	  estimation	  of	  preferences	  on	  a	  relative	  scale,	  the	  objective	  of	  this	  study	  was	  to	  quantify	  the	  impact	  of	  important	  features	  of	  stroke	  prophylaxis	  on	  physician	  preference.	  	  	  4.2 Methods	  4.2.1 Study	  population	  Physicians	  were	  invited	  to	  participate	  and	  take	  the	  online	  questionnaire	  through	  emails	  via	  a	  local	  health	  authority’s	  listserv	  and	  research	  institution	  broadcast.	  Physicians	  licensed	  to	  practice	  in	  the	  province	  of	  British	  Columbia,	  excluding	  those	  undergoing	  residency	  programs	  or	  fellowships,	  were	  eligible.	  Emails	  embedded	  with	  a	  password-­‐protected	  URL	  were	  sent	  to	  each	  potential	  participant.	  After	  giving	  consent	  online,	  the	  participants	  were	  directed	  to	  begin	  the	  web-­‐based	  survey.	  The	  first	  section	  asked	  questions	  related	  to	  the	  subjects’	  demographic	  information.	  The	  second	  section	  was	  a	  best	  worst	  scaling	  (BWS)	  choice	  experiment.	  	  	  4.2.2 Experimental	  design	  Details	  of	  the	  best-­‐worst	  scaling	  (BWS)	  choice	  experiment	  were	  described	  in	  Chapters	  1	  and	  3.	  Similar	  to	  the	  BWS	  experiment	  administered	  to	  the	  patient	  respondents	  (see	  Chapter	  3),	  the	  physician	  respondent	  was	  also	  shown	  a	  task	  with	  multiple	  attribute	  levels,	  and	  asked	  to	  choose	  80	  	  the	  most	  preferred	  and	  least	  preferred	  attributes	  of	  the	  task.	  The	  pair	  of	  the	  most	  preferred	  and	  least	  preferred	  attribute	  levels	  is	  considered	  to	  be	  furthest	  apart	  (56).	  After	  multiple	  scenario	  tasks	  where	  all	  possible	  pairs	  of	  attribute	  levels	  have	  been	  shown,	  the	  impact	  of	  attribute	  levels	  can	  be	  determined	  by	  the	  propensity	  it	  is	  chosen	  as	  the	  best	  (or	  worst)	  relative	  to	  other	  attribute	  levels	  on	  a	  common	  underlying	  relative	  preference	  scale	  (56,165).	  This	  allows	  the	  direct	  comparison	  of	  relative	  preference	  across	  all	  attribute	  levels	  –	  a	  key	  advantage	  of	  BWS	  that	  sets	  this	  type	  of	  choice	  experiment	  apart	  from	  DCE.	  	  	  Similar	  to	  the	  patient	  version	  of	  the	  BWS,	  the	  same	  attributes	  and	  levels	  were	  used	  in	  the	  BWS	  task	  for	  physicians:	  frequency	  of	  blood	  test	  monitoring,	  annual	  risk	  of	  stroke,	  annual	  risk	  of	  major	  bleed,	  availability	  of	  reversibility	  agents.	  Table	  4.1	  lists	  the	  attributes	  and	  their	  associated	  levels.	  Prior	  to	  beginning	  the	  choice	  questionnaire,	  the	  respondents	  were	  provided	  with	  a	  detailed	  description	  of	  the	  attributes	  and	  shown	  a	  BWS	  task	  example.	  Each	  respondent	  was	  then	  asked	  to	  complete	  20	  BWS	  tasks.	  The	  BWS	  task	  asked	  the	  physician	  respondent	  to	  choose	  a	  best	  and	  a	  worst	  feature	  of	  the	  available	  attributes	  for	  a	  hypothetical	  patient	  newly	  diagnosed	  with	  AF.	  See	  Appendix	  B	  for	  a	  sample	  physician	  questionnaire.	  	  	  Sawtooth®	  software	  Webv6.0	  (Sawtooth	  Software	  Inc.	  Sequim,	  WA,	  USA)	  was	  used	  to	  design	  the	  BWS	  questionnaire	  using	  near	  optimal	  plans	  (166).	  While	  the	  design	  by	  SSI	  cannot	  ensure	  equal	  frequency	  of	  attribute	  pairs	  shown,	  near	  balance	  is	  obtained	  across	  multiple	  versions	  and	  respondents.	  For	  this	  study,	  one-­‐way	  and	  two-­‐way	  frequencies	  were	  shown	  to	  be	  optimally	  81	  	  balanced	  across	  2	  versions	  of	  questionnaires,	  with	  a	  16	  total	  attribute	  levels	  and	  20	  BWS	  tasks	  per	  respondents.	  	  	  4.2.3 Data	  analysis	  Best-­‐worst	  scaling	  relative	  utilities	  Relative	  utilities	  based	  on	  the	  BWS	  choice	  experiment	  were	  first	  analyzed	  by	  the	  conditional	  logistic	  regression	  where	  I	  modeled	  the	  effects	  of	  attribute	  levels	  on	  the	  ranked	  choice.	  The	  probability	  that	  case	  𝑖	  (respondent)	  chooses	  alternative	  𝑚  at	  replication	  𝑡  given	  attribute	  values	  Ζ™ ™? 	  and	  predictor	  values	  Ζ™ ™? 	  can	  be	  estimated	  using	  the	  following:	  𝑃 𝒴™ =   𝑚 Ζ™ ™? , Ζ™ ™? =    exp 𝜂? ? ™exp(?𝓂??? 𝜂?? ? ™ )	  (Eq.	  3.1)	  where	  𝜂? ? ™ 	  is	  the	  systematic	  component	  in	  the	  utility	  of	  choice	  𝑚  at	  replication	  𝑡	  for	  case	  𝑖.	  The	  parameter	  coefficient	  for	  each	  alternative	  𝑚	  describes	  the	  respondents’	  preference	  for	  that	  attribute	  level	  where	  the	  large	  estimates	  indicate	  positive	  impact	  of	  that	  attribute	  level	  on	  the	  ranked	  choice.	  	  In	  the	  case	  of	  this	  BWS	  experiment,	  the	  attribute	  level	  “annual	  stroke	  risk	  of	  0%”	  was	  chosen	  as	  the	  reference	  level,	  such	  that	  all	  parameter	  estimates	  directly	  describe	  their	  respective	  preference	  in	  relation	  to	  that	  reference	  level	  on	  a	  common	  scale.	  	  	  The	  conditional	  logit	  model	  is	  predisposed	  to	  biased	  preference	  estimation	  as	  it	  does	  not	  account	  for	  heterogeneity	  in	  respondent	  choices	  or	  variance	  scale.	  Thus,	  I	  also	  used	  a	  scale-­‐adjusted	  latent	  class	  analysis	  (LCA)	  to	  evaluate	  respondent	  preferences.	  	  The	  LCA	  is	  based	  on	  82	  	  the	  framework	  that	  there	  are	  different	  discrete	  groups	  or	  ‘classes’	  of	  preferences,	  which	  can	  be	  characterized	  by	  observed	  as	  well	  as	  unobserved	  (latent)	  variables.	  The	  class	  membership	  of	  each	  respondent	  can	  be	  assigned	  based	  on	  the	  structure	  of	  their	  preference.	  Thus,	  LCA	  allows	  a	  better	  profiling	  of	  the	  choice	  probability	  depending	  on	  the	  respondent’s	  class	  membership	  𝑥:	  	  	  𝑃 𝒴™ =   𝑚 𝑥, Ζ™ ™? , Ζ™ ™? , 𝑆™ =    exp(𝑆™    ∙   𝜂? ?,? ™ )exp(?𝓂? 𝑆™    ∙   𝜂?? ?,? ™ )	  (Eq.	  3.2)	  where	  the	  systematic	  component	  𝜂? ?,? ™ 	  has	  also	  now	  become	  class-­‐specific.	  The	  term	  𝑆™   is	  a	  scale	  factor	  assumed	  to	  be	  constant	  across	  alternatives	  within	  a	  replication.	  In	  the	  case	  of	  best-­‐worst	  choices,	  the	  best	  and	  worst	  choices	  take	  the	  scale	  factor	  of	  1	  and	  -­‐1,	  respectively.	  	  	  To	  determine	  the	  preference	  estimates	  of	  the	  physician	  respondents,	  I	  effect	  coded	  the	  BWS	  responses	  and	  estimated	  models	  using	  the	  LCA	  up	  to	  six	  classes.	  Model	  fit	  parameters	  including	  the	  log-­‐likelihood	  function	  (LL),	  the	  Bayesian	  Information	  Criteria	  (BIC),	  and	  the	  Akaike	  Information	  Criteria	  (AIC)	  were	  used	  to	  assess	  the	  optimal	  class	  model.	  Using	  both	  univariate	  analysis	  and	  backwards	  selection	  method,	  demographic	  covariates	  were	  included	  in	  the	  models	  to	  explore	  possible	  explanation	  for	  the	  difference	  across	  the	  latent	  classes.	  Similar	  to	  the	  conditional	  logit	  model,	  the	  most	  preferred	  attribute	  level	  was	  chosen	  as	  the	  reference	  level	  such	  that	  all	  parameter	  coefficients	  estimates	  could	  be	  interpreted	  relative	  to	  the	  reference.	  	  Latent	  Gold	  Choice	  (v.4.5,	  Statistical	  Innovations,	  Inc.,	  Belmont,	  MA)	  was	  used	  to	  carry	  out	  the	  LCA	  in	  this	  study.	  	  83	  	  Best-­‐worst	  scaling	  relative	  importance	  In	  interpreting	  the	  results	  for	  choice-­‐based	  response	  data,	  it	  is	  often	  helpful	  to	  characterize	  the	  relative	  importance	  of	  each	  attribute.	  Importance	  is	  essentially	  the	  maximum	  effect	  an	  attribute	  make	  in	  the	  total	  utility	  of	  a	  product	  (167,168).	  This	  can	  be	  calculated	  as:	  𝑚𝑎𝑥𝑒𝑓𝑓™ = max 𝜂? ™ −𝑚𝑖𝑛 𝜂? ™ 	  (Eq.	  3.3)	  where	  𝑎	  denotes	  a	  level	  of	  attribute	  𝑝,	  and	  𝜂? ™ is	  the	  utility	  associated	  with	  the	  level	  𝑎	  for	  latent	  class	  𝑥.	  The	  difference	  between	  the	  maximum	  and	  minimum	  level	  for	  attribute	  𝑝	  for	  latent	  class	  𝑥	  thus	  defines  𝑚𝑎𝑥𝑒𝑓𝑓™ .	  The	  relative	  importance	  of	  the	  attribute	  𝑝  can	  then	  be	  obtained	  by	  the	  following	  equation	  to	  compare	  these	  maximum	  effects	  across	  attributes	  and	  latent	  classes:	  𝑟𝑒𝑙𝑒𝑓𝑓™ = 𝑚𝑎𝑥𝑒𝑓𝑓™𝑚𝑎𝑥𝑒𝑓𝑓™? 	  (Eq.	  3.4)	  Where	   𝑚𝑎𝑥𝑒𝑓𝑓™? 	  is	  the	  sum	  of	  the	  maximum	  effects	  of	  all	  attributes.	  	  4.3 Results	  4.3.1 Sample	  characteristics	  450	  email	  invitations	  were	  sent	  out	  to	  physician	  from	  across	  the	  province.	  12.6%	  physicians	  provided	  consent	  and	  started	  the	  web-­‐based	  survey.	  7.4%	  physicians	  completed	  the	  survey.	  On	  average,	  it	  took	  the	  respondents	  13	  minutes	  to	  complete	  the	  survey.	  Respondent	  characteristics	  are	  summarized	  in	  Table	  4.3.	  The	  mean	  age	  of	  participants	  was	  44.7	  ±	  10.2	  years	  84	  	  (range	  29	  to	  63	  years)	  with	  similar	  distribution	  by	  sex.	  The	  majority	  of	  the	  physicians	  were	  located	  in	  the	  Greater	  Vancouver	  area	  (69.7%)	  with	  a	  hospital-­‐based	  practice.	  	  Family	  practice,	  internal	  medicine	  and	  emergency	  medicine	  physicians	  dominated	  the	  participant	  pool.	  Almost	  all	  physicians	  had	  treated	  patients	  who	  presented	  with	  bleeding	  while	  on	  oral	  antithrombotics.	  	  4.3.2 BWS	  –	  relative	  utilities	  Model	  estimation	  The	  estimated	  utilities	  for	  each	  attribute	  level	  relative	  to	  “annual	  risk	  of	  stroke	  of	  0%”	  are	  presented	  in	  Table	  4.4.	  As	  expected,	  0%	  stroke	  risk	  was	  the	  most	  valued	  attribute	  level,	  followed	  by	  0%	  risk	  of	  major	  bleed	  with	  a	  mean	  relative	  utility	  estimate	  of	  -­‐1.14.	  Having	  laboratory	  test	  done	  only	  once	  a	  year,	  reversal	  agent	  available	  and	  an	  annual	  stroke	  risk	  of	  2%	  were	  the	  next	  most	  preferred	  attribute	  levels	  with	  mean	  relative	  utilities	  of	  -­‐2.23,	  -­‐2.24,	  -­‐2.40,	  respectively.	  On	  the	  other	  hand,	  annual	  stroke	  risk	  of	  8%	  and	  10%	  had	  the	  lowest	  mean	  relative	  utilities	  of	  -­‐7.10	  and	  -­‐8.25,	  respectively.	  All	  utility	  estimates	  were	  significant	  at	  p<0.001,	  indicating	  that	  all	  attribute	  levels	  significantly	  impacted	  the	  respondent’s	  choice.	  Further,	  all	  parameter	  estimates	  for	  the	  attribute	  levels	  are	  trending	  in	  the	  expected	  direction,	  lending	  face	  validity	  to	  the	  choice	  data.	  Figure	  4.1	  is	  a	  graphical	  representation	  of	  the	  part-­‐worth	  utility	  estimates	  for	  the	  four	  attributes	  and	  associated	  levels.	  For	  the	  attribute	  “annual	  risk	  of	  stroke”,	  steeper	  slope	  was	  noted	  between	  0%	  to	  6%	  stroke	  risk	  compared	  to	  the	  slope	  between	  6%	  to	  8%,	  suggesting	  a	  higher	  preference	  change	  with	  risk	  increase	  in	  the	  lower	  range.	  	  Similar	  observation	  was	  noted	  for	  the	  attribute	  “annual	  risk	  of	  major	  bleed”	  where	  the	  sharper	  slope	  between	  0%	  and	  4%	  flattened	  out	  slightly	  in	  the	  higher	  range.	  For	  the	  attribute	  on	  laboratory	  85	  	  monitoring,	  the	  slope	  did	  not	  change	  markedly	  over	  the	  different	  frequency	  levels,	  suggesting	  that	  physicians’	  preference	  on	  this	  attribute	  was	  not	  particular	  strong.	  This	  was	  reflected	  in	  the	  relative	  importance	  showing	  that	  the	  maximum	  effect	  of	  frequency	  of	  laboratory	  monitoring	  was	  the	  least	  important	  compared	  to	  the	  remaining	  attributes	  (Figure	  4.2).	  	  	  Comparison	  of	  six	  different	  latent	  class	  models	  using	  AIC,	  BIC	  and	  log-­‐likelihood	  estimates	  suggested	  that	  a	  two-­‐class	  model	  best	  fit	  the	  physicians’	  choice	  data	  and	  pointed	  two	  distinct	  latent	  classes.	  However,	  class	  profile	  of	  the	  two-­‐class	  model	  revealed	  that	  only	  4%	  of	  the	  study	  sample	  was	  assigned	  to	  class	  2,	  making	  the	  validity	  of	  the	  model	  questionable.	  Hence,	  it	  was	  decided	  not	  to	  use	  the	  LCM	  to	  describe	  the	  respondent	  sample.	  	  Consequently,	  it	  was	  not	  possible	  to	  explore	  and	  explain	  potential	  class	  heterogeneity	  by	  known	  covariates.	  In	  addition,	  the	  study	  sample	  size	  was	  too	  small	  to	  allow	  the	  exploration	  of	  covariates	  that	  might	  influence	  physician	  respondent	  choices	  (e.g.	  mixed	  logit	  model).	  	  Figure	  4.2	  shows	  the	  relative	  importance	  of	  attributes.	  Annual	  risk	  of	  stroke	  was	  the	  most	  important	  attribute,	  followed	  by	  annual	  risk	  of	  major	  bleed,	  and	  availability	  of	  reversal	  agent.	  Frequency	  of	  laboratory	  had	  the	  least	  impact	  on	  respondent	  choices	  compared	  to	  the	  other	  attributes.	  	  	  86	  	  4.4 Discussion	  By	  using	  a	  BWS	  choice	  experiment,	  this	  study	  elicited	  the	  relative	  preferences	  of	  physicians	  for	  the	  relevant	  attributes	  in	  stroke	  prophylaxis.	  Overall,	  findings	  from	  the	  study	  indicate	  that	  annual	  risk	  of	  stroke	  had	  the	  highest	  impact	  on	  respondent	  choices	  for	  stroke	  prophylaxis,	  followed	  by	  annual	  risk	  of	  major	  bleed.	  Not	  only	  does	  this	  study	  finding	  affirm	  the	  common	  perception	  that	  clinicians	  are	  in	  general	  more	  concerned	  about	  clinical	  outcomes	  in	  making	  medical	  decisions	  than	  non-­‐outcome	  based	  attributes,	  the	  utility	  estimates	  from	  the	  elicitation	  allow	  the	  explicit	  quantification	  of	  their	  relative	  preference	  over	  a	  range	  of	  stroke	  and	  bleeding	  risks.	  	  The	  steep	  slope	  of	  risk	  of	  stroke	  and	  risk	  of	  major	  bleed	  over	  the	  risk	  range	  showed	  that	  physicians’	  preferences	  for	  stroke	  prophylaxis	  were	  sensitive	  to	  the	  incremental	  change	  of	  risk,	  particularly	  between	  0%	  and	  6%	  for	  annual	  risk	  of	  stroke	  and	  between	  0%	  and	  4%	  for	  annual	  risk	  of	  major	  bleed,	  both	  of	  which	  suggested	  that	  physician	  respondents	  were	  stroke	  and	  bleed	  averse.	  It	  also	  appears	  that	  the	  physician	  preferences	  for	  stroke	  and	  bleeding	  risk	  follow	  an	  almost	  linear	  relationship.	  Consequently,	  this	  would	  allow	  the	  approximation	  of	  physicians’	  relative	  preferences	  for	  stroke	  and	  bleeding	  risk	  given	  a	  risk	  probability.	  This	  is	  to	  be	  discussed	  in	  detail	  in	  the	  following	  chapter	  when	  comparing	  physician	  and	  patient	  preferences	  and	  their	  clinical	  implications.	  	  	  87	  	  The	  results	  also	  suggest	  that	  physicians	  did	  not	  place	  as	  high	  an	  importance	  on	  whether	  or	  not	  a	  reversal	  agent	  was	  available	  in	  stroke	  prophylaxis.	  	  There	  also	  did	  not	  appear	  to	  be	  a	  strong	  preference	  for	  the	  frequency	  of	  laboratory	  test	  with	  a	  narrow	  range	  of	  preference	  estimates	  between	  having	  to	  do	  blood	  test	  every	  month	  to	  every	  year.	  It	  is	  consistent	  with	  the	  observation	  that	  the	  study	  sample	  was	  largely	  dominated	  by	  physicians	  with	  a	  hospital-­‐based	  practice,	  who	  were	  not	  normally	  responsible	  for	  patients’	  long-­‐term	  anticoagulation	  monitoring	  out	  in	  the	  community.	  Had	  the	  sample	  recruited	  more	  GPs	  to	  participate	  in	  the	  elicitation	  task,	  the	  preference	  profile	  may	  have	  differed	  with	  stronger	  preference	  (or	  wider	  utility	  range)	  for	  this	  particular	  attribute.	  	  	  While	  there	  is	  no	  similar	  preference	  study	  available	  for	  comparison	  with	  the	  result	  from	  this	  study,	  findings	  from	  this	  elicitation	  task	  do	  offer	  corroboration	  with	  observations	  in	  current	  practice	  settings.	  Strong	  relative	  preferences	  for	  stroke	  risk,	  followed	  by	  major	  bleeding	  risk	  as	  elicited	  in	  this	  study	  are	  reflective	  of	  how	  decisions	  regarding	  new	  therapeutic	  agents	  are	  evaluated,	  where	  efficacy/effectiveness	  and	  safety	  are	  the	  priority	  criteria.	  This	  preference	  is	  prevalent	  throughout	  decision	  making	  processes	  where	  physicians	  are	  important	  stakeholders,	  including	  clinical	  practice	  guideline	  development	  and	  at	  the	  patient	  care	  level.	  Most	  often,	  it	  is	  the	  efficacy	  and	  safety	  profiles	  that	  dominate	  the	  discussion	  and	  contribute	  to	  the	  recommendations	  being	  made.	  Other	  non-­‐outcome	  based	  attributes,	  such	  as	  reversal	  agents	  or	  laboratory	  monitoring	  are	  brought	  up	  as	  accessory	  information	  in	  making	  the	  decision.	  The	  relatively	  higher	  preferences	  for	  clinical	  efficacy/effectiveness	  and	  safety	  are	  also	  apparent	  at	  88	  	  the	  drug	  approval	  and	  drug	  reimbursement	  level	  where	  physicians	  offer	  key	  input	  in	  health	  policy	  decisions.	  	  	  One	  major	  shortcoming	  of	  this	  study	  was	  the	  small	  sample	  size	  which	  limited	  the	  condition	  for	  a	  more	  vigorous	  statistical	  analysis	  such	  as	  a	  mixed-­‐logit	  model,	  and	  thus	  curtailed	  the	  initial	  intent	  of	  exploring	  subgroup	  preference	  differences.	  The	  conditional	  logit	  regression	  is	  an	  appropriate	  model	  for	  this	  study	  design	  and	  sample	  size,	  however,	  it	  may	  not	  account	  for	  the	  correlation	  of	  repeated	  choice	  tasks	  within	  an	  individual	  respondent	  as	  well	  as	  the	  heterogeneity	  amongst	  the	  respondents.	  One	  would	  argue	  that	  adopting	  a	  mixed	  logit	  model	  would	  be	  a	  more	  appropriate	  analysis	  for	  this	  type	  of	  choice	  experiment.	  However,	  previous	  experience	  with	  analyzing	  BWS	  data	  demonstrated	  that	  the	  parameter	  estimates	  and	  errors	  using	  a	  conditional	  logit	  models	  are	  comparable	  to	  that	  of	  a	  mixed	  logit	  model.	  To	  substantiate	  any	  subgroup	  differences	  in	  preferences	  amongst	  the	  physicians,	  the	  study	  would	  need	  to	  recruit	  a	  larger	  sample	  size.	  	  	  This	  study	  sample	  consisted	  of	  physicians	  in	  BC;	  therefore,	  caution	  should	  be	  applied	  before	  generalizing	  results	  to	  population	  in	  other	  jurisdiction.	  There	  is	  also	  the	  concern	  of	  inherent	  selection	  bias	  as	  the	  study	  only	  included	  respondents	  who	  were	  willing	  to	  participate	  in	  the	  survey,	  and	  that	  the	  preference	  expressed	  in	  this	  sample	  may	  not	  be	  representative	  of	  all	  physicians.	  However,	  this	  limitation	  would	  be	  inherent	  in	  any	  type	  of	  preference	  studies	  and	  is	  89	  	  not	  expected	  to	  pose	  significant	  threat	  to	  the	  validity	  of	  study	  findings	  nor	  undermine	  utility	  data	  on	  physician	  preferences	  for	  stroke	  prophylaxis.	  	  	  To	  my	  knowledge,	  this	  is	  the	  first	  study	  that	  utilized	  a	  preference	  elicitation	  method	  on	  the	  ordinal	  scale	  to	  examine	  other	  important	  factors	  for	  stroke	  prophylaxis	  in	  the	  AF	  population.	  Findings	  from	  this	  BWS	  choice	  experiment	  provided	  information	  on	  physician	  preferences	  for	  attributes	  (both	  outcome	  and	  non-­‐outcome	  based)	  that	  are	  relevant	  in	  stroke	  prophylaxis.	  The	  results	  of	  this	  study	  affirm	  the	  general	  perception	  that	  physicians	  would	  generally	  exhibit	  stronger	  preferences	  for	  attributes	  related	  to	  clinical	  outcomes	  (i.e.	  efficacy	  and	  safety).	  Not	  only	  does	  this	  study	  illustrate	  physician	  preferences	  for	  the	  relevant	  attribute	  explicitly,	  it	  also	  shows	  the	  relative	  magnitude	  of	  preferences	  across	  different	  levels	  of	  attributes	  of	  interest.	  This	  is	  valuable	  insights	  that	  can	  be	  utilized	  further	  to	  illuminate	  how	  physician	  preferences	  may	  influence	  medical	  decision	  making	  in	  the	  setting	  of	  stroke	  prophylaxis	  in	  AF,	  both	  at	  a	  clinical	  and	  global	  level.	  	  	  	  	  	  	  	  	  90	  	  Table	  4.1:	  Attributes	  and	  levels	  in	  BWS	  choice	  experiment	  Attributes	   Levels	  Frequency	  of	  laboratory	  test	   Once	  every	  month	  Once	  every	  3	  months	  Once	  every	  6	  months	  Once	  a	  year	  Annual	  risk	  of	  stroke	   0%	  2%	  4%	  6%	  8%	  10%	  Annual	  risk	  of	  major	  bleed	   0%	  2%	  4%	  6%	  Availability	  of	  reversibility	  agent	   Available	  Not	  available	  	  	  	  	  	  	  	  	  	  	  	  	  91	  	  Table	  4.2:	  Characteristics	  of	  physician	  participants	  (N=33)	  Characteristics	  Age	  (years)	   	  Mean	  (SD)	   44.7	  (10.2)	  Range	  (min-­‐max)	   29-­‐63	  Sex	  (N,	  %)	   	  Male	   17	  (51.5)	  Female	   16	  (48.5)	  Practice	  duration	  (N,	  %)	   	  <	  5	  years	   10	  (30.3)	  6	  to	  10	  years	   3	  (9.1)	  11	  to	  20	  years	   12	  (36.4)	  >20	  years	   8	  (24.2)	  City	  of	  Practice	  (N,	  %)	   	  Greater	  Vancouver	   23	  (69.7)	  Others	   10	  (30.3)	  Primary	  practice	  Setting	  (N,	  %)	   	  Tertiary	  care	   12	  (36.4)	  Community	  hospital	   11	  (33.3)	  Ambulatory,	  outpatient	  clinics	   4	  (12.1)	  Community	   6	  (18.2)	  Practice	  Specialty	  (N,	  %)	   	  Family	  practice	   10	  (30.3)	  Internal	  medicine	   7	  (21.2)	  Emergency	  medicine	   6	  (18.2)	  Neurology	   2	  (6.1)	  Cardiology	   3	  (9.1)	  Geriatrics	   3	  (9.1)	  History	  of	  treating	  patients	  with	  bleed	  while	  on	  OAC	  (N,	  %)	   	  Yes	   30	  (90.9)	  No	   3	  (9.1)	  OAC	  oral	  anticoagulant	  	  	  	  	  	  	  	  92	  	  Table	  4.3:	  Relative	  preference	  estimates	  from	  conditional	  logit	  model	  	  	   	   Coefficient	  (SE)	   P	  value	  Frequency	  of	  Laboratory	  Test	   	   	  Every	  month	   -­‐4.04	  (0.31)	   <0.001	  Every	  3	  months	   -­‐3.29	  (0.30)	   <0.001	  Every	  6	  months	   -­‐3.03	  (0.30)	   <0.001	  Once	  a	  year	   -­‐2.23	  (0.29)	   <0.001	  Annual	  risk	  of	  stroke	   	   	  0%	   Ref	   	  2%	   -­‐2.40	  (0.29)	   <0.001	  4%	   -­‐4.58	  (0.32)	   <0.001	  6%	   -­‐6.20	  (0.34)	   <0.001	  8%	   -­‐7.10	  (0.35)	   <0.001	  10%	   -­‐8.25	  (0.39)	   <0.001	  Annual	  risk	  of	  bleed	   	   	  0%	   -­‐1.14	  (0.28)	   <0.001	  2%	   -­‐3.26	  (0.30)	   <0.001	  4%	   -­‐5.31	  (0.33)	   <0.001	  6%	   -­‐6.36	  (0.34)	   <0.001	  Reversibility	  Agents	   	   	  Available	   -­‐2.24	  (0.29)	   <0.001	  Not	  Available	   -­‐5.57	  (0.32)	   <0.001	  	   	   	  R2	   0.1471	   	  R2	  (0)	   0.5382	   	  Number	  of	  respondents	   33	   	  Number	  of	  observations	   1320	   	  	   	   	  Log-­‐likelihood	   -­‐818.1	   	  BIC	   1688.56	   	  Prediction	  Error	   0.75	   	  	  	  	  	  	  	  	  93	  	  Figure	  4.1:	  Part-­‐worth	  utility	  estimates	  for	  attribute	  levels	  	  	  	  	  	  	  	  	  	  	  94	  	  Figure	  4.2:	  Relative	  importance	  of	  attributes	  	  	  95	  	  Chapter	  5: Comparison	  of	  patients’	  and	  physicians’	  preferences	  for	  stroke	  prophylaxis	  in	  atrial	  fibrillation	  5.1 Introduction	  In	  the	  era	  of	  patient-­‐centered	  care,	  a	  significant	  focus	  has	  been	  placed	  on	  engaging	  patients	  in	  the	  decision-­‐making	  process	  such	  that	  their	  values	  are	  being	  taken	  into	  consideration	  in	  making	  decisions	  about	  their	  care.	  In	  addition	  to	  involving	  patient	  input	  at	  the	  point-­‐of-­‐care	  setting,	  advocacy	  for	  incorporating	  patient	  preferences	  on	  a	  systematic	  level	  is	  also	  emerging.	  Many	  authors	  have	  emphasized	  the	  need	  to	  move	  beyond	  evidence-­‐based	  medicine	  and	  incorporate	  patient	  values	  in	  the	  guideline	  development	  process	  (6,9,26,30,38).	  Krahn	  has	  on	  more	  than	  one	  occasion,	  noted	  the	  importance	  of	  including	  population-­‐based	  preference	  data	  in	  formulating	  guideline	  recommendations	  and	  pointed	  out	  the	  gap	  that	  has	  yet	  to	  be	  met	  on	  this	  objective	  by	  guideline	  committees	  around	  the	  world	  (26,30).	  Such	  a	  proposition	  is	  based	  on	  the	  attractive	  reasoning	  that	  every	  decision	  is	  value-­‐laden	  and	  thus	  value	  is	  inherent	  in	  every	  recommendation,	  either	  explicitly	  or	  implicitly	  (18,30).	  	  	  Countering	  these	  ideals,	  there	  is	  some	  skepticism	  regarding	  why	  and	  whether	  patient	  values	  should	  be	  considered	  in	  guideline	  recommendations.	  Umscheid	  argued	  that	  population-­‐based	  data	  on	  patient	  values	  are	  not	  necessary	  in	  guideline	  recommendations,	  as	  individual	  patient	  preferences	  will	  contribute	  to	  their	  clinical	  care	  regardless	  of	  population	  preference	  data	  (176).	  It	  is	  apparent	  that	  such	  statements	  do	  not	  recognize	  the	  complexity	  of	  ‘value’	  and	  overlook	  the	  impracticality	  of	  eliciting	  individual	  patient	  preferences	  for	  each	  scenario.	  The	  fact	  that	  a	  96	  	  person’s	  preference	  also	  has	  an	  implicit	  component,	  it	  would	  be	  difficult	  to	  ask	  the	  patients	  to	  reveal	  their	  complete	  preferences	  when	  they	  themselves	  might	  not	  know	  what	  their	  preferences	  are.	  	  	  While	  the	  argument	  Umscheid	  and	  other	  opponents	  made	  is	  not	  a	  strong	  one,	  the	  question	  of	  whether	  population	  preference	  data	  should	  become	  part	  of	  the	  guideline	  development	  is	  an	  issue	  worthy	  of	  further	  in-­‐depth	  contemplation.	  Even	  though	  it	  sounds	  intuitive	  to	  incorporate	  patient	  preferences	  in	  guideline	  recommendations,	  one	  must	  also	  bear	  in	  mind	  that	  each	  recommendation	  is	  also	  laden	  with	  the	  values	  of	  the	  guideline	  developers,	  or	  clinicians.	  The	  proposition	  for	  integrating	  population	  preference	  data	  into	  the	  guideline	  process	  largely	  hinges	  on	  the	  assumption	  that	  patient	  preferences	  are	  different	  from	  physicians	  and	  that	  their	  values	  have	  been	  absent	  in	  formulating	  the	  recommendations.	  As	  shown	  by	  existing	  literature	  and	  the	  systematic	  review	  in	  Chapter	  2,	  physician-­‐patient	  value	  concordance	  is	  highly	  variable	  depending	  on	  the	  patient	  population	  (i.e.	  disease	  severity,	  acuity,	  cultural	  and	  social	  background)	  and	  the	  preference	  elicitation	  methods	  used	  (111).	  To	  date,	  no	  studies	  have	  examined	  how	  physician	  and	  patients	  might	  be	  different	  in	  atrial	  fibrillation	  with	  regards	  to	  stroke	  prophylaxis	  using	  either	  cardinal	  or	  ordinal	  methods.	  Thus,	  in	  order	  to	  validate	  the	  argument	  for	  integrating	  population	  preference	  data	  in	  practice	  guidelines,	  this	  part	  of	  the	  study	  investigated	  to	  what	  extent,	  if	  any,	  do	  physician	  and	  patient	  preferences	  differ	  for	  stroke	  prophylaxis.	  And	  if	  such	  a	  difference	  exists,	  whether	  it	  will	  bring	  forth	  any	  clinical	  implications.	  In	  other	  words,	  it	  would	  be	  meaningful	  to	  explore	  whether	  a	  significant	  difference	  in	  preference	  97	  	  between	  patients	  and	  physicians	  translates	  into	  difference	  in	  making	  clinical	  decisions	  with	  regards	  to	  stroke	  prophylaxis.	  This	  part	  of	  the	  study	  aimed	  to	  achieve	  the	  above	  objectives	  by	  comparing	  the	  data	  obtained	  from	  the	  BWS	  choice	  experiments	  in	  as	  described	  in	  chapters	  3	  and	  4.	  	  	  5.2 Methods	  5.2.1 Study	  population	  and	  experimental	  design	  Study	  population	  consisted	  of	  58	  AF	  patients	  and	  33	  physicians,	  and	  the	  development	  and	  administration	  of	  the	  BWS	  task	  are	  as	  described	  in	  chapters	  3	  and	  4.	  	  	  5.2.2 Data	  analysis	  Comparison	  of	  relative	  preferences	  by	  best-­‐worst	  scores	  Because	  the	  BWS	  choice	  design	  for	  the	  physicians	  and	  the	  patients	  were	  different	  (the	  former	  had	  2	  versions	  of	  profile	  sets,	  each	  with	  20	  choice	  tasks	  and	  the	  latter	  with	  4	  versions,	  each	  with	  16	  choice	  tasks),	  it	  was	  not	  possible	  to	  combine	  the	  choice	  data	  from	  both	  experiments	  and	  compare	  the	  two	  groups	  directly	  with	  the	  same	  regression	  model.	  One	  approach	  to	  discern	  the	  differences	  between	  the	  two	  groups	  is	  to	  compare	  the	  preference	  weights	  of	  the	  physicians	  and	  the	  patients	  obtained	  from	  the	  conditional	  logit	  models	  (see	  Chapter	  3	  and	  4	  results).	  These	  preference	  weights,	  however,	  do	  not	  take	  into	  account	  the	  heterogeneity	  of	  preferences	  across	  the	  respondents	  and	  comparing	  the	  preference	  weights	  on	  an	  aggregate	  level	  might	  lead	  to	  important	  loss	  of	  information	  (177).	  To	  mitigate	  the	  variability	  of	  preferences	  across	  98	  	  population,	  this	  study	  adopted	  a	  method	  that	  was	  previously	  developed	  by	  Finn	  and	  Louviere	  that	  involved	  computing	  a	  best-­‐worst	  score	  (B-­‐W	  score)	  for	  each	  attribute	  level	  (61).	  The	  B-­‐W	  scores	  were	  calculated	  by	  counting	  the	  number	  of	  times	  each	  attribute	  level	  was	  picked	  as	  being	  ‘best’	  minus	  the	  number	  of	  times	  each	  attribute	  level	  was	  picked	  as	  being	  ‘worst’	  at	  the	  individual	  level.	  	  This	  would	  allow	  one	  to	  obtain	  preference	  weights	  that	  have	  been	  shown	  to	  be	  similar	  to	  those	  from	  the	  logit	  models,	  while	  retaining	  the	  distribution	  of	  preferences	  for	  each	  item.	  Best-­‐worst	  score	  for	  each	  attribute	  for	  the	  patients	  and	  the	  physicians	  was	  calculated	  separately,	  then	  compared	  using	  Wilcoxon	  signed-­‐rank	  test.	  	  	  Exploratory	  analysis	  of	  oral	  antithrombotic	  choices	  according	  to	  stated	  preference	  To	  determine	  whether	  differences	  in	  preferences	  would	  lead	  to	  differences	  in	  patients’	  and	  physicians’	  choices	  of	  oral	  antithrombotics,	  an	  exploratory	  analysis	  was	  done	  by	  computing	  a	  preference	  score	  consisting	  of	  the	  relative	  utility	  estimates	  obtained	  in	  the	  BWS	  tasks,	  based	  on	  the	  baseline	  stroke	  risk	  and	  the	  oral	  antithrombotic	  of	  interest.	  The	  preference	  score	  can	  be	  simply	  illustrated	  by	  the	  following	  equation:	  𝑈?,? =   𝑈™???? ?,? + 𝑈™ ⌣? ? +   𝑈™? (?) + 𝑈™?????? (?)	  (Eq.	  5.1)	  where	  𝑈™???? ?,? 	  denotes	  the	  relative	  utility	  estimate	  for	  stroke	  based	  on	  a	  baseline	  stroke	  𝑥	  and	  oral	  antithrombotic	  𝑘,	  𝑈™ ⌣? ? 	  denotes	  the	  relative	  utility	  estimate	  for	  major	  bleed	  give	  oral	  antithrombotic	  𝑘,	  𝑈™? (?)  refers	  to	  the	  relative	  utility	  estimate	  for	  the	  recommended	  laboratory	  monitoring	  frequency	  for	  oral	  antithrombotic	  𝑘,	  and	  𝑈™?????? (?)	  represents	  the	  99	  	  utility	  for	  availability	  of	  reversal	  agent	  for	  oral	  antithrombotic	  𝑘.	  Essentially,	  the	  preference	  score	  𝑈?,?	  for	  a	  baseline	  CHADS2	  or	  CHA2DS2-­‐Vasc	  score	  𝑥	  on	  oral	  antithrombotic	  𝑘	  is	  the	  sum	  of	  the	  relative	  utility	  estimates	  extrapolated	  from	  the	  stated	  preference	  task.	  Utility	  estimates	  for	  stroke	  and	  major	  bleed	  were	  derived	  from	  the	  slope	  of	  a	  best-­‐fitted	  line	  of	  the	  relative	  utilities	  obtained	  from	  the	  BWS	  choice	  experiment	  for	  patients	  and	  physicians.	  The	  preference	  score	  for	  patients	  and	  physicians	  was	  computed	  for	  each	  CHADS2	  and	  CHA2DS2-­‐Vasc	  score	  and	  the	  five	  different	  oral	  antithrombotic	  regimens	  including	  warfarin,	  dabigatran	  150	  mg	  twice	  daily,	  dabigatran	  110	  mg	  twice	  daily,	  rivaroxaban	  20	  mg	  daily,	  and	  apixaban	  5	  mg	  daily.	  	  The	  preference	  scores	  were	  adjusted	  with	  warfarin	  as	  the	  reference	  score,	  such	  that	  a	  final	  positive	  preference	  score	  (i.e.	  𝑈?,?	  >	  0)	  would	  imply	  that	  oral	  antithrombotic	  𝑘	  was	  preferred	  over	  warfarin	  at	  baseline	  stroke	  risk	  𝑥.	  There	  were	  several	  assumptions	  being	  made	  in	  this	  exploratory	  analysis,	  the	  most	  crucial	  one	  being	  that	  the	  choices	  of	  oral	  antithrombotics	  for	  stroke	  prevention	  depend	  exclusively	  on	  the	  four	  attributes	  evaluated	  in	  this	  study.	  The	  other	  assumption	  to	  note	  was	  that	  the	  risk	  of	  major	  bleed	  for	  the	  antithrombotic	  of	  interest	  would	  remain	  constant	  regardless	  of	  the	  baseline	  stroke	  risk.	  	  5.3 Results	  Best-­‐worst	  scores	  comparison	  Best-­‐worst	  scores	  for	  each	  attribute	  levels	  for	  the	  physician	  and	  patient	  groups	  are	  presented	  in	  Table	  5.1.	  As	  seen	  with	  the	  relative	  utility	  estimates	  from	  the	  conditional	  logit	  and	  latent	  class	  models	  in	  chapters	  3	  and	  4,	  the	  BW	  scores	  trended	  in	  the	  expected	  direction	  within	  each	  100	  	  attribute.	  As	  mentioned	  previously,	  the	  BW	  score	  was	  computed	  by	  taking	  the	  number	  of	  times	  the	  attribute	  level	  was	  chosen	  as	  the	  best	  minus	  the	  number	  of	  times	  it	  was	  chosen	  as	  the	  worst.	  In	  other	  words,	  a	  positive	  BW	  score	  would	  suggest	  that	  it	  was	  picked	  as	  best	  more	  often	  than	  worst,	  and	  that	  the	  larger	  the	  score,	  the	  higher	  the	  preference	  for	  that	  attribute	  level	  relative	  to	  others.	  	  	  While	  the	  general	  direction	  of	  attribute	  ranking	  was	  similar	  between	  patients	  and	  physicians,	  results	  from	  the	  Wilcoxon	  signed-­‐rank	  test	  showed	  that	  patients	  and	  physician	  had	  significantly	  different	  preferences	  for	  several	  attribute	  levels	  including	  “laboratory	  test	  every	  month”,	  “annual	  stroke	  risk	  0%”,	  “annual	  stroke	  risk	  2%”,	  “annual	  stroke	  risk	  8%”,	  “annual	  stroke	  risk	  10%”,	  “annual	  bleeding	  risk	  2%”,	  “annual	  bleeding	  risk	  4%”,	  “annual	  bleeding	  risk	  6%”,	  and	  “no	  reversibility	  agent	  available”.	  	  This	  suggested	  that	  even	  though	  there	  was	  a	  general	  agreement	  in	  preference	  between	  patients	  and	  physicians,	  the	  strength	  in	  preferences	  differed	  significantly.	  Specifically,	  the	  results	  point	  out	  that	  while	  both	  groups	  had	  a	  strong	  preference	  for	  0%	  stroke	  risk,	  the	  physicians	  valued	  this	  attribute	  level	  to	  be	  much	  more	  desirable	  than	  the	  patients.	  The	  same	  observation	  applies	  to	  stroke	  risk	  of	  10%,	  where	  physicians	  expressed	  much	  stronger	  negative	  preferences	  for	  the	  attribute	  level	  compared	  to	  the	  patients,	  as	  well	  as	  the	  range	  of	  annual	  bleeding	  rates,	  suggesting	  that	  physicians	  in	  general,	  are	  more	  stroke	  and	  bleed	  averse	  than	  the	  patients.	  Interestingly,	  it	  also	  suggests	  that	  physicians	  did	  not	  favor	  frequent	  laboratory	  test	  on	  a	  monthly	  interval,	  whereas	  it	  was	  the	  reverse	  trend	  for	  the	  patients.	  	  	  101	  	  Oral	  antithrombotic	  choices	  according	  to	  stated	  preferences	  Preference	  scores	  for	  patients	  and	  physicians	  for	  each	  oral	  antithrombotic	  were	  computed	  for	  all	  CHADS2	  and	  CHA2DS2-­‐Vasc	  risk	  levels.	  	  Figure	  5.1	  illustrates	  the	  adjusted	  preference	  scores	  relative	  to	  warfarin	  for	  the	  patients.	  As	  shown,	  based	  on	  the	  stated	  preferences	  obtained	  from	  the	  BWS	  task,	  patients	  preferred	  warfarin	  to	  all	  the	  other	  oral	  antithrombotics	  regardless	  of	  the	  baseline	  stroke	  risk.	  Next	  to	  warfarin,	  it	  appeared	  the	  next	  oral	  antithrombotic	  of	  choice	  would	  be	  dabigatran	  150	  mg	  twice	  daily	  or	  apixaban	  5	  mg	  once	  daily	  depending	  on	  the	  baseline	  stroke	  risk.	  Rivaroxaban	  20	  mg	  remained	  the	  least	  favored	  antithrombotic	  regardless	  of	  the	  baseline	  stroke	  risk.	  Similar	  observations	  were	  found	  for	  the	  physician	  group	  as	  illustrated	  in	  Figure	  5.2.	  Again,	  warfarin	  remained	  the	  antithrombotic	  of	  choice	  for	  most	  baseline	  stroke	  risks,	  with	  the	  exception	  of	  apixaban	  being	  more	  favored	  for	  CHADS2	  score	  levels	  1,	  5	  and	  6.	  However,	  the	  actual	  differences	  in	  preference	  score	  between	  apixaban	  and	  warfarin	  were	  small	  and	  likely	  insignificant.	  This	  preference	  for	  apixaban	  was	  also	  not	  observed	  when	  using	  the	  CHA2DS2-­‐Vasc	  risk	  scores.	  	  	  5.4 Discussion	  This	  study	  used	  a	  BWS	  experiment	  which	  allowed	  the	  evaluation	  of	  relative	  preferences	  for	  multiple	  attributes	  related	  to	  stroke	  prophylaxis	  in	  addition	  to	  stroke	  and	  major	  bleed,	  the	  two	  primary	  outcomes	  associated	  with	  taking	  oral	  antithrombotics.	  By	  utilizing	  the	  relative	  preference	  data	  from	  the	  BWS	  experiments,	  this	  study	  was	  also	  able	  to	  quantitatively	  evaluate	  and	  compare	  the	  preferences	  of	  patients	  and	  physicians	  for	  stroke	  prophylaxis,	  and	  to	  examine	  102	  	  whether	  this	  difference	  in	  preference,	  if	  any,	  would	  have	  meaningful	  implications	  in	  the	  clinical	  setting.	  	  	  By	  calculating	  the	  BW	  scores	  for	  each	  attribute	  levels,	  this	  study	  demonstrated	  that	  while	  patients	  and	  physicians	  showed	  similar	  preferences	  for	  stroke	  prophylaxis,	  the	  strength	  of	  their	  preferences	  differed	  significantly.	  Specifically,	  it	  suggested	  that	  physicians	  were	  more	  stroke	  and	  bleed	  averse	  than	  patients.	  In	  addition,	  physicians	  were	  found	  not	  to	  favor	  laboratory	  monitoring	  at	  one-­‐month	  interval.	  On	  the	  contrary,	  physicians	  and	  patients	  had	  similar	  preferences	  for	  frequency	  of	  laboratory	  monitoring	  from	  once	  every	  3	  months	  to	  once	  every	  year.	  There	  was	  no	  difference	  in	  preference	  for	  having	  a	  reversal	  agent	  available;	  however,	  the	  results	  also	  suggested	  physicians	  would	  react	  strongly	  if	  a	  reversal	  agent	  were	  not	  available.	  	  	  Interestingly,	  the	  difference	  in	  preference	  strength	  did	  not	  appear	  to	  translate	  into	  differences	  in	  choosing	  oral	  antithrombotics	  based	  on	  the	  exploratory	  analysis	  using	  preference	  scores.	  It	  was	  shown	  in	  the	  study	  that	  warfarin	  remained	  the	  oral	  antithrombotic	  of	  choice	  for	  both	  patients	  and	  physicians,	  regardless	  of	  the	  baseline	  stroke	  risk.	  This	  suggests	  that	  perhaps	  there	  is	  overall	  agreement	  on	  stroke	  prophylaxis	  between	  patients	  and	  physicians	  and	  that	  integration	  of	  patient	  preferences	  on	  a	  population	  level	  is	  not	  as	  important	  as	  it	  was	  proposed.	  The	  lack	  of	  impact	  on	  oral	  antithrombotic	  choices	  despite	  the	  significant	  difference	  in	  preference	  strength	  also	  poses	  an	  interesting	  question	  of	  whether	  difference	  noted	  in	  preference	  studies	  have	  significant	  real-­‐world	  implications.	  Caution	  has	  to	  be	  applied	  here;	  103	  	  however,	  that	  conclusion	  made	  above	  cannot	  be	  readily	  made	  as	  the	  analysis	  done	  in	  this	  section	  of	  the	  study	  was	  purely	  exploratory	  and	  dependent	  on	  several	  assumptions	  that	  were	  likely	  violated	  in	  real	  practice.	  To	  further	  validate	  whether	  the	  difference	  in	  preference	  strength	  noted	  in	  this	  study	  would	  lead	  to	  differences	  in	  oral	  antithrombotic	  choices	  would	  require	  a	  discrete	  choice	  experiment	  or	  a	  best-­‐worst	  discrete	  choice	  experiment	  (see	  chapter	  1),	  where	  respondents	  are	  asked	  to	  make	  discrete	  choices	  between	  alternatives.	  Another	  possibility	  to	  examine	  the	  difference	  in	  making	  oral	  antithrombotic	  choices	  is	  through	  the	  revealed	  preference	  method	  where	  choice	  data	  are	  gathered	  through	  observing	  the	  actual	  choices	  physicians	  and	  patients	  make	  in	  clinical	  practice.	  	  	  Another	  limitation	  of	  this	  study	  is	  that	  it	  did	  not	  include	  other	  attributes	  that	  would	  also	  be	  deemed	  relevant	  to	  respondent	  choices	  in	  the	  selection	  of	  oral	  antithrombotics.	  For	  example,	  attributes	  such	  as	  drug-­‐drug	  interaction,	  single	  versus	  twice	  daily	  dosing	  and	  drug-­‐food	  interaction	  would	  be	  potentially	  relevant	  to	  include	  as	  part	  of	  the	  choice	  task.	  These	  were	  left	  out	  of	  the	  elicitation	  task	  out	  of	  the	  concern	  for	  increased	  respondent	  burden	  for	  both	  the	  physicians	  and	  patients	  and	  associated	  loss	  of	  choice	  data	  validity.	  I	  am	  confident,	  however,	  that	  the	  four	  attributes	  chosen	  were	  the	  most	  relevant	  attributes	  related	  to	  stroke	  prophylaxis,	  as	  suggested	  by	  a	  thorough	  literature	  review	  and	  expert	  opinions.	  Another	  shortcoming	  of	  this	  comparison	  study	  lies	  in	  the	  statistical	  analysis	  that	  was	  used.	  Even	  though	  the	  attributes	  and	  levels	  were	  identical	  for	  the	  choice	  tasks	  for	  the	  two	  groups,	  the	  BWS	  for	  each	  group	  was	  of	  different	  design	  such	  that	  choice	  data	  from	  the	  two	  groups	  could	  not	  be	  combined	  directly	  to	  104	  	  derive	  preference	  weights	  using	  the	  same	  model.	  As	  a	  result,	  the	  B-­‐W	  score	  method	  was	  adopted	  to	  compare	  the	  individual	  level-­‐derived	  preference	  weights	  between	  the	  physicians	  and	  patients.	  While	  this	  method	  allowed	  the	  estimates	  to	  retain	  individual	  error	  distribution,	  and	  is	  thus	  better	  than	  directly	  comparing	  the	  conditional	  logit	  preference	  estimates,	  the	  B-­‐W	  score	  comparison	  methods	  may	  not	  account	  for	  the	  heterogeneity	  between	  the	  two	  groups.	  However,	  similar	  preferences	  from	  the	  B-­‐W	  scores	  compared	  to	  the	  conditional	  logit	  model	  (see	  chapter	  3	  and	  4)	  were	  observed,	  suggesting	  the	  B-­‐W	  scores	  were	  an	  appropriate	  and	  an	  equivalently	  valid	  method	  to	  evaluate	  the	  preferences	  between	  the	  two	  groups,	  as	  previously	  demonstrated	  by	  Finn	  and	  Louviere	  (61).	  	  The	  results	  from	  this	  part	  of	  the	  study	  further	  emphasize	  the	  need	  to	  conduct	  preference	  studies	  using	  rigorous	  designs	  to	  allow	  explicit	  evaluation	  of	  decision	  stakeholders’	  preferences.	  The	  significant	  difference	  in	  strength	  found	  in	  this	  study	  suggests	  that	  population	  data	  of	  patient	  preferences	  should	  be	  considered	  for	  integration	  on	  a	  systematic	  level	  when	  making	  health	  policy	  decisions	  or	  clinical	  practice	  guideline	  recommendations.	  However,	  this	  consideration	  is	  highly	  dependent	  on	  whether	  the	  difference	  in	  preference	  found	  would	  impact	  oral	  antithrombotic	  choices	  in	  the	  clinical	  setting.	  To	  properly	  incorporate	  this	  type	  of	  preference	  data	  into	  the	  guidelines,	  the	  next	  step	  would	  be	  to	  conduct	  a	  formal	  net	  risk-­‐benefit	  analysis	  and	  evaluate	  whether	  this	  difference	  in	  value	  would	  translate	  into	  therapeutic	  decision-­‐making	  difference.	  This	  study	  has	  laid	  down	  the	  foundation	  for	  a	  more	  rigorous	  and	  105	  	  transparent	  approach	  to	  including	  population	  preference	  data	  for	  making	  health	  policy	  and	  decision	  in	  this	  era	  of	  patient-­‐centered	  care.	  	  	  	  	  	  	  	  	  	  	  	  	  	  	  	  	  	  	  	  106	  	  Table	  5.1:	  Best-­‐worst	  score	  for	  patients	  and	  physicians	  	   Patients	   Physicians	   	  Attribute	  Levels	   Mean	  (SD)	   Mean	  (SD)	   P-­‐value	  Frequency	  of	  laboratory	  monitoring	   	   	   	  Every	  1	  month	   0.26	  (0.98)	   -­‐0.27	  (1.13)	   0.0298	  Every	  3	  months	   1.09	  (1.20)	   0.79	  (1.05)	   0.2279	  Every	  6	  months	   0.97	  (0.94)	   0.91	  (0.95)	   0.8422	  Every	  year	   1.88	  (1.33)	   2.06	  (1.30)	   0.7650	  Annual	  risk	  of	  stroke	   	   	   	  0%	   3.86	  (0.44)	   4.36	  (1.48)	   <0.0001	  2%	   -­‐0.36	  (1.74)	   1.94	  (1.50)	   <0.0001	  4%	   -­‐0.98	  (1.08)	   -­‐0.52	  (1.09)	   0.0815	  6%	   -­‐2.14	  (0.78)	   -­‐2.33	  (1.67)	   0.1517	  8%	   -­‐2.93	  (1.02)	   -­‐3.33	  (1.16)	   0.013	  10%	   -­‐3.72	  (0.85)	   -­‐4.21	  (1.90)	   <0.0001	  Annual	  risk	  of	  major	  bleed	   	   	   	  0%	   2.88	  (0.68)	   3.33	  (1.16)	   0.0536	  2%	   0.16	  (1.06)	   0.79	  (1.36)	   0.018	  4%	   -­‐0.47	  (0.82)	   -­‐1.21	  (1.02)	   0.0005	  6%	   -­‐1.43	  (0.96)	   -­‐2.58	  (0.97)	   <0.0001	  Availability	  of	  reversal	  agent	   	   	   	  Available	  	   2.17	  (1.22)	   2.24	  (1.25)	   0.8985	  Not	  available	   -­‐1.22	  (1.24)	   -­‐1.97	  (1.47)	   0.0171	  107	  	  Figure	  5.1:	  Patient	  choices	  of	  oral	  antithrombotics	  relative	  to	  warfarin	  based	  on	  stated	  preferences	  according	  to	  baseline	  stroke	  risk	  	  	  	  	  -­‐3	  -­‐2	  -­‐1	  0	  1	  0	   1	   2	   3	   4	   5	   6	  Preference	  Score	  CHADS2	  Score	  DABI150	  DABI110	  RIVA20	  APIX5	  -­‐3	  -­‐2	  -­‐1	  0	  1	  0	   1	   2	   3	   4	   5	   6	   7	   8	   9	  Preference	  Score	  CHA2DS2-­‐Vasc	  Score	  DABI150	  DABI110	  RIVA20	  APIX5	  108	  	  Figure	  5.2:	  Physician	  choices	  of	  oral	  antithrombotics	  relative	  to	  warfarin	  based	  on	  stated	  preferences	  according	  to	  baseline	  stroke	  risk	  	  	  	  	  	  -­‐3	  -­‐2	  -­‐1	  0	  1	  0	   1	   2	   3	   4	   5	   6	  Preference	  Score	  CHADS2	  Score	  -­‐5	  -­‐4	  -­‐3	  -­‐2	  -­‐1	  0	  1	  0	   1	   2	   3	   4	   5	   6	   7	   8	   9	  Preference	  Score	  CHA2DS2-­‐Vasc	  Score	  DABI150	  DABI110	  RIVA20	  APIX5	  109	  	  Chapter	  6: Discussion	  6.1 Summary	  of	  major	  findings	  and	  study	  implications	  Health	  care	  in	  the	  last	  few	  decades	  have	  seen	  major	  transformation	  from	  evidence-­‐based	  medicine	  to	  patient-­‐centered	  medicine,	  where	  patient	  preference	  is	  now	  deemed	  a	  requisite	  in	  health	  decision-­‐making.	  However,	  much	  vagueness	  remains	  when	  it	  comes	  to	  how	  preference	  should	  be	  defined,	  how	  it	  should	  be	  measured	  and	  how	  and	  when	  it	  should	  be	  applied.	  	  Proponents	  have	  advocated	  for	  the	  incorporation	  of	  preferences	  into	  clinical	  practice	  guidelines	  on	  a	  population	  level.	  However,	  it	  is	  not	  known	  if	  patient	  and	  physician	  preferences	  differ	  to	  the	  extent	  such	  that	  population	  patient	  preference	  data	  should	  be	  considered	  systematically	  at	  the	  guideline	  development	  stage.	  This	  study	  aimed	  at	  answering	  some	  of	  these	  questions.	  Specifically,	  the	  main	  objective	  was	  to	  investigate	  the	  preferences	  of	  patients	  and	  physicians,	  the	  two	  major	  stakeholders	  in	  the	  decision-­‐making	  process,	  using	  utility-­‐based	  preference	  elicitation	  methods.	  	  	  Best-­‐worst	  scaling	  (BWS)	  is	  an	  ordinal	  preference	  elicitation	  method	  that	  allows	  direct	  and	  simultaneous	  preference	  estimation	  of	  multiple	  levels	  of	  multiple	  attributes.	  Comparing	  the	  preferences	  of	  patients	  and	  physicians	  was	  of	  especial	  interest	  in	  stroke	  prophylaxis	  in	  patients	  with	  atrial	  fibrillation,	  as	  this	  has	  always	  been	  recognized	  as	  a	  preference-­‐sensitive	  decision	  where	  the	  risks	  of	  stroke	  and	  bleeding	  events	  must	  be	  weighed	  cautiously.	  To	  date,	  the	  very	  few	  studies	  that	  have	  attempted	  to	  investigate	  how	  preferences	  might	  differ	  between	  patients	  and	  physicians	  for	  stroke	  prophylaxis	  in	  this	  population	  utilized	  mostly	  qualitative	  methods	  or	  110	  	  at	  best,	  cardinal	  methods	  that	  were	  limited	  to	  assessing	  preferences	  for	  major	  clinical	  outcomes	  only	  (i.e.	  stroke,	  major	  bleed).	  It	  has	  become	  apparent	  that	  other	  features	  of	  the	  stroke	  prophylaxis	  therapies	  might	  influence	  how	  patients	  and	  physicians	  make	  decisions	  about	  their	  antithrombotic	  choices.	  Thus,	  this	  research	  was	  carried	  out	  to	  examine	  and	  compare	  the	  preferences	  of	  physicians	  and	  patients	  for	  stroke	  prophylaxis	  using	  a	  BWS	  choice	  experiment.	  	  In	  the	  second	  chapter	  of	  this	  thesis,	  a	  qualitative	  review	  was	  conducted	  to	  evaluate	  all	  the	  published	  literature	  that	  compared	  physician	  and	  patient	  preferences	  using	  the	  ordinal	  elicitation	  methods.	  From	  the	  review,	  no	  conclusions	  can	  be	  drawn	  about	  the	  congruence	  of	  patient	  and	  physician	  preferences.	  There	  was	  a	  comparable	  proportion	  of	  studies	  that	  showed	  preference	  congruence,	  preference	  disagreement	  and	  similar	  preference	  but	  difference	  in	  strength.	  Pattern	  of	  patient	  and	  physician	  preference	  congruence	  remained	  inconclusive	  when	  the	  review	  was	  limited	  to	  the	  type	  of	  elicitation	  methods	  used	  or	  the	  acuity	  of	  the	  medical	  conditions.	  This	  could	  be	  largely	  due	  to	  the	  fact	  that	  the	  studies	  that	  examined	  preference	  congruence	  between	  the	  different	  decision	  stakeholders	  were	  highly	  heterogeneous	  in	  terms	  of	  disease	  states,	  patient	  population,	  elicitation	  methods,	  attributes	  and	  analytical	  methods.	  Findings	  of	  this	  study	  were	  consistent	  with	  a	  previously	  published	  review,	  suggesting	  that	  preference	  congruence	  between	  physicians	  and	  patients	  is	  most	  likely	  circumstance-­‐specific	  (i.e.	  dependent	  on	  the	  disease	  states,	  patient	  population	  and	  study	  setting)	  and	  in	  most	  cases	  is	  different	  in	  terms	  of	  strengths,	  if	  not	  in	  terms	  of	  ranking.	  	  An	  important	  finding	  from	  this	  review	  111	  	  revealed	  that	  no	  studies	  have	  tried	  to	  compare	  patient	  and	  physician	  preferences	  using	  BWS,	  which	  substantiates	  one	  of	  the	  knowledge	  gap	  this	  research	  was	  hoping	  to	  address.	  	  The	  third	  chapter	  of	  this	  thesis	  was	  a	  major	  section	  of	  this	  research,	  where	  patient	  preferences	  for	  stroke	  prophylaxis	  were	  elicited	  using	  both	  the	  time	  trade-­‐off	  (TTO)	  and	  BWS	  methods.	  Using	  the	  TTO	  method,	  utilities	  for	  debilitating	  and	  non-­‐debilitating	  stroke	  were	  measured	  from	  58	  AF	  patients.	  This	  elicitation	  task	  also	  adopted	  a	  chained-­‐TTO	  to	  measure	  the	  utilities	  for	  major	  and	  minor	  bleed.	  This	  was	  done	  in	  recognition	  that	  conventional	  TTO	  or	  other	  cardinal	  methods	  were	  derived	  for	  the	  eliciting	  preferences	  of	  chronic	  conditions	  and	  its	  utilization	  in	  temporary	  health	  states	  had	  been	  challenged	  in	  the	  past	  .	  (54).	  This	  was	  the	  first	  study	  that	  measured	  utilities	  for	  minor	  bleed	  in	  the	  AF	  population.	  The	  TTO	  utilities	  for	  debilitating	  and	  non-­‐debilitating	  stroke	  were	  found	  to	  be	  significantly	  higher	  than	  previous	  studies,	  and	  the	  most	  probable	  explanation	  for	  this	  was	  the	  younger	  patient	  population	  enrolled	  in	  this	  study.	  On	  the	  contrary,	  the	  utility	  for	  major	  bleed	  in	  this	  study	  was	  not	  significantly	  different	  from	  the	  only	  other	  published	  study	  by	  Thomsen	  et	  al.	  (161).	  	  	  The	  relative	  preferences	  from	  the	  BWS	  experiment	  indicated	  that	  annual	  stroke	  risk	  had	  the	  highest	  impact	  on	  patient	  respondent	  choices	  compared	  to	  the	  other	  attributes	  (annual	  risk	  of	  major	  bleed,	  frequency	  of	  laboratory	  monitoring,	  availability	  of	  reversibility	  agents).	  Using	  latent	  class	  analysis,	  it	  was	  found	  that	  the	  respondents	  largely	  fall	  into	  three	  distinct	  classes	  112	  	  according	  to	  the	  observed	  and	  unobserved	  variables,	  where	  risk	  of	  stroke	  dominated	  the	  preferences	  regardless	  of	  class	  memberships.	  	  	  	  Next,	  the	  relative	  preference	  weights	  for	  stroke	  prophylaxis	  were	  elicited	  from	  33	  physicians	  using	  a	  BWS	  experiment	  with	  the	  same	  attributes	  as	  the	  ones	  in	  the	  patient	  group.	  Again,	  the	  annual	  risk	  of	  stroke	  was	  the	  most	  important	  attribute	  driving	  respondent	  choices.	  Unlike	  the	  patient	  population,	  choice	  response	  in	  the	  physician	  group	  was	  found	  to	  be	  mostly	  homogeneous	  where	  no	  latent	  classes	  identified	  to	  explain	  the	  preference	  responses.	  No	  subgroup	  difference	  according	  to	  known	  characteristics	  from	  the	  physician	  sample	  was	  found,	  either.	  However,	  this	  could	  have	  been	  due	  to	  the	  small	  study	  sample.	  	  	  In	  chapter	  5,	  relative	  preferences	  of	  patients	  and	  physicians	  were	  compared	  by	  using	  the	  BW	  scores.	  The	  resulting	  scores	  demonstrated	  that	  patient	  and	  physician	  preferences	  for	  stroke	  prophylaxis	  were	  consistent	  with	  the	  relative	  utilities	  from	  the	  conditional	  logit	  and	  latent	  class	  models.	  Comparing	  the	  BW	  scores	  of	  patients	  and	  physicians,	  it	  was	  shown	  that	  preferences	  of	  the	  two	  groups	  were	  overall	  similar	  in	  terms	  of	  ranking,	  but	  significantly	  different	  in	  strengths	  for	  several	  attribute	  levels.	  More	  importantly,	  it	  also	  showed	  that	  physicians,	  in	  general,	  were	  more	  stroke	  and	  bleed	  averse	  than	  the	  patients	  as	  exemplified	  by	  their	  wider	  range	  of	  preference	  scores.	  The	  second	  part	  of	  this	  chapter	  attempted	  to	  explore	  whether	  the	  difference	  in	  preference	  strength	  would	  result	  in	  meaningful	  difference	  in	  making	  oral	  antithrombotic	  choices.	  Utilizing	  the	  relative	  utility	  estimates	  obtained	  from	  the	  BWS	  elicitation	  tasks	  as	  113	  	  described	  in	  chapters	  3	  and	  4,	  there	  did	  not	  appear	  to	  be	  any	  difference	  between	  patients	  and	  physicians	  in	  choosing	  oral	  antithrombotics,	  where	  both	  seemed	  to	  favour	  warfarin	  as	  the	  antithrombotic	  of	  choice	  regardless	  of	  the	  baseline	  stroke	  risk.	  However,	  this	  exploratory	  analysis	  was	  embedded	  with	  multiple	  assumptions	  and	  proper	  preference	  study	  designs	  allowing	  discrete	  choices	  need	  to	  be	  employed	  to	  test	  this	  hypothesis.	  	  	  The	  study	  findings	  presented	  in	  this	  thesis	  are	  significant	  on	  several	  fronts.	  Both	  the	  chained-­‐TTO	  and	  BW	  scores	  provided	  methodological	  insights	  in	  the	  field	  of	  preference	  studies.	  While	  this	  study	  was	  not	  designed	  to	  validate	  the	  chained	  TTO	  method	  for	  measuring	  utilities	  for	  temporary	  health	  states,	  it	  proved	  the	  feasibility	  of	  carrying	  out	  such	  a	  task	  and	  provided	  the	  utility	  for	  minor	  bleed	  in	  AF	  patients,	  which	  was	  previously	  absent	  in	  the	  literature.	  The	  BW	  scores	  was	  developed	  in	  the	  early	  days	  of	  BWS	  design	  but	  have	  not	  been	  utilized	  extensively	  by	  health	  researchers	  since	  conditional	  logit	  and	  multinomial	  logit	  models	  became	  the	  standard	  analysis	  of	  max-­‐diff	  and	  choice-­‐based	  data.	  However,	  as	  seen	  in	  this	  study	  findings,	  BW	  scores	  were	  able	  to	  provide	  reliable	  preference	  data	  without	  having	  to	  involve	  complex	  statistical	  modeling,	  and	  should	  be	  given	  more	  merit	  as	  an	  alternative	  analysis	  of	  BW	  choice	  data.	  	  	  On	  a	  decision-­‐making	  level,	  the	  relative	  utilities	  obtained	  impart	  important	  information	  about	  the	  preferences	  of	  patients	  and	  physicians	  for	  stroke	  prophylaxis	  in	  AF.	  The	  results	  confirmed	  that	  while	  risk	  of	  stroke	  and	  major	  bleed	  were	  the	  attributes	  with	  the	  dominant	  impact	  on	  patient	  and	  physician	  respondent	  preferences,	  other	  non-­‐clinical	  outcome	  attributes	  such	  as	  114	  	  availability	  of	  reversal	  agents	  and	  frequency	  of	  laboratory	  monitoring	  also	  significantly	  affected	  respondent	  choices	  in	  stroke	  prophylaxis.	  An	  interesting,	  though	  not	  surprising	  observation	  was	  that	  the	  patients’	  choice	  data	  was	  much	  more	  heterogeneous	  than	  the	  physician	  choice	  data.	  This	  was	  consistent	  with	  the	  known	  fact	  that	  patient	  preference	  can	  be	  highly	  variable	  given	  the	  diverse	  demographics,	  cultural	  beliefs,	  knowledge	  of	  disease	  states	  etc.,	  whereas	  physicians	  are	  health	  professionals	  trained	  to	  provide	  care	  and	  meet	  certain	  health	  outcome	  goals	  on	  an	  objective,	  uniform	  construct	  of	  evidence-­‐based	  medicine.	  	  	  Comparison	  of	  physician	  and	  patient	  preferences	  for	  stroke	  prophylaxis	  is	  the	  highlight	  of	  this	  study,	  when	  shows	  that	  the	  two	  different	  decision	  stakeholders	  are	  similar	  in	  preferences,	  yet	  significantly	  different	  in	  terms	  of	  their	  preference	  strengths.	  Specifically,	  it	  showed	  that	  while	  stroke	  and	  bleeding	  risk	  were	  both	  important	  attributes	  in	  determining	  patient	  and	  physician	  preferences,	  the	  physicians	  were	  found	  to	  be	  much	  more	  stroke	  and	  bleed	  averse	  compared	  to	  the	  patients.	  This	  observation	  suggests	  that	  physicians,	  as	  health	  care	  providers,	  tend	  to	  make	  decisions	  that	  would	  maximize	  the	  benefit	  and	  minimize	  the	  harm	  in	  terms	  of	  clinical	  outcomes.	  In	  contrast,	  the	  patient	  respondents	  weigh	  in,	  to	  a	  greater	  extent	  on	  other	  non-­‐clinical	  outcome	  attributes	  in	  their	  preferences.	  What	  is	  perhaps	  the	  most	  intriguing	  finding	  in	  this	  study	  is	  the	  exploratory	  analysis	  in	  chapter	  5	  that	  suggests	  that	  patients	  and	  physicians	  would	  not	  choose	  oral	  antithrombotic	  differently	  despite	  the	  significant	  differences	  in	  preference	  strength	  as	  noted	  in	  the	  BW	  scores.	  As	  mentioned	  before,	  this	  finding	  needs	  to	  be	  further	  validated	  by	  another	  study	  involving	  respondents	  making	  discrete	  choices	  between	  antithrombotic	  115	  	  alternatives.	  However,	  if	  this	  observation	  is	  confirmed,	  then	  the	  recommendation	  to	  make	  integrating	  patient	  preferences	  on	  a	  systematic	  level	  would	  seem	  unfounded.	  	  	  	  The	  findings	  and	  interpretation	  above	  may	  seem	  self-­‐evident;	  however,	  this	  was	  the	  first	  study	  undertaken	  to	  explicitly	  demonstrate	  the	  preference	  profile	  of	  patients	  and	  physicians	  in	  this	  clinical	  context.	  By	  using	  an	  ordinal	  scale	  elicitation	  method,	  findings	  of	  this	  study	  shed	  light	  on	  important	  knowledge	  gaps	  regarding	  preferences	  of	  decision	  stakeholders	  in	  this	  population	  and	  contributed	  valuable	  groundwork	  for	  future	  incorporation	  of	  preferences	  on	  a	  systematic	  level.	  	  	  6.2 Limitations	  	  Some	  limitations	  of	  this	  study	  must	  be	  considered	  in	  the	  applications	  of	  these	  study	  findings.	  In	  chapter	  2,	  the	  review	  of	  published	  literature	  evaluating	  congruence	  of	  patient	  and	  physician	  preferences	  was	  not	  a	  meta-­‐analysis.	  Thus,	  conclusions	  were	  made	  based	  on	  the	  qualitative	  assessment	  of	  the	  included	  studies.	  No	  existing	  guideline	  on	  appraising	  preference	  literature	  is	  available	  and	  such	  qualitative	  review	  could	  be	  susceptible	  to	  reviewer	  bias.	  However,	  I	  tried	  to	  be	  as	  systematic	  in	  the	  assessment	  as	  possible,	  focusing	  on	  evaluating	  the	  validity	  of	  the	  elicitation	  methods	  and	  design	  of	  the	  individual	  studies.	  It	  was	  apparent	  during	  the	  review,	  however,	  that	  the	  studies	  were	  highly	  heterogeneous	  and	  a	  meta-­‐analysis	  to	  meet	  the	  scope	  of	  the	  research	  question	  would	  be	  quite	  difficult	  until	  a	  substantial	  preference	  data	  with	  overlapping	  disease	  states	  have	  emerged.	  	  116	  	  	  A	  major	  shortcoming	  of	  this	  study	  was	  the	  lack	  of	  a	  focus	  group	  to	  identify	  the	  study	  attributes	  for	  the	  BWS	  elicitation	  task.	  This	  could	  have	  potentially	  led	  to	  missing	  important	  attributes	  in	  the	  choice	  experiments	  such	  as	  drug	  interaction,	  frequency	  of	  short-­‐term	  laboratory	  monitoring,	  dietary	  modification,	  and	  perioperative	  management.	  Given	  the	  limited	  funding,	  I	  was	  not	  able	  to	  conduct	  a	  focus	  group	  to	  appropriately	  identify	  the	  attributes	  of	  interest.	  However,	  I	  am	  quite	  confident	  that	  the	  attributes	  selected,	  based	  on	  clinical	  expertise	  and	  literature	  review,	  are	  relevant	  attributes	  most	  likely	  to	  influence	  respondent	  choices	  for	  antithrombotics.	  It	  was	  also	  decided	  that	  the	  attributes	  for	  the	  physician	  and	  patient	  group	  should	  be	  kept	  identical	  to	  allow	  preference	  comparison	  of	  the	  two	  groups.	  Had	  a	  focus	  group	  been	  carried	  out,	  there	  is	  the	  possibility	  that	  two	  different	  sets	  of	  attributes	  would	  be	  identified	  for	  the	  different	  groups.	  This	  would	  have	  made	  direct	  preference	  comparison	  between	  physicians	  and	  patients	  difficult.	  The	  number	  of	  attributes	  was	  kept	  to	  four	  primary	  because	  of	  the	  concern	  with	  respondent	  burden.	  Given	  that	  the	  patient	  respondents	  also	  had	  to	  complete	  the	  TTO	  tasks,	  the	  number	  of	  BWS	  was	  minimized	  without	  compromising	  design	  efficiency.	  For	  the	  same	  reason,	  it	  was	  felt	  that	  the	  four	  attributes	  would	  be	  acceptable	  for	  the	  physician	  respondents	  in	  terms	  of	  time	  spent	  on	  answering	  a	  survey	  without	  any	  incentive.	  	  	  A	  major	  limitation	  of	  the	  case	  2	  BWS	  design	  adopted	  in	  this	  study	  was	  that	  it	  could	  not	  allow	  the	  evaluation	  of	  welfare	  estimates	  such	  as	  marginal	  rate	  of	  substitution,	  or	  maximum	  acceptable	  risk	  or	  minimum	  acceptable	  benefit	  that	  are	  often	  inferred	  in	  discrete	  choice	  117	  	  experiments.	  The	  reason,	  as	  discussed	  in	  chapter	  1,	  is	  that	  in	  BWS	  the	  respondents	  are	  prompted	  to	  make	  discrete	  choices	  within	  alternatives	  and	  not	  between.	  Case	  2	  BWS	  was	  chosen	  as	  the	  design	  for	  this	  study’s	  elicitation	  tasks	  as	  it	  was	  decided	  at	  the	  study	  design	  stage,	  that	  relative	  preferences	  of	  all	  attribute	  levels	  on	  a	  common	  scale	  were	  more	  important	  given	  the	  existing	  knowledge	  gap.	  Consideration	  of	  respondent	  burden	  with	  a	  DCE	  or	  BWSDCE	  was	  also	  a	  factor	  in	  deciding	  against	  the	  more	  complex	  choice	  experiments.	  While	  the	  BWS	  task	  in	  this	  study	  was	  not	  able	  to	  provide	  welfare	  estimates,	  I	  attempted	  to	  explore	  how	  relative	  utilities	  obtained	  might	  translate	  into	  real	  world	  decisions	  when	  choosing	  an	  oral	  antithrombotic.	  	  	  Because	  of	  the	  number	  of	  attributes	  and	  levels	  selected	  for	  the	  elicitation	  task,	  a	  perfectly	  balanced	  orthogonal	  design	  was	  not	  possible	  and	  consequently,	  design	  efficiency	  might	  be	  an	  issue.	  However,	  as	  described	  in	  chapter	  3	  and	  4,	  Sawtooth®	  Software	  was	  used	  to	  generate	  a	  BWS	  design	  that	  was	  optimal.	  This	  was	  confirmed	  by	  checking	  the	  one-­‐	  and	  two-­‐way	  frequency	  to	  ensure	  that	  adequate	  efficiency	  was	  in	  place	  before	  administering	  the	  BWS	  tasks.	  While	  the	  sample	  size	  for	  both	  patients	  and	  physicians	  was	  deemed	  adequate	  as	  demonstrated	  by	  the	  statistical	  significance	  of	  all	  attribute	  levels,	  the	  sample	  may	  not	  have	  been	  large	  enough	  to	  explore	  how	  preference	  amongst	  patients	  or	  physicians	  might	  be	  differentiated	  according	  to	  individual	  characteristics.	  I	  hope	  to	  see	  a	  future	  study	  with	  larger	  sample	  size	  and	  possibly	  with	  more	  attributes	  to	  explore	  further	  questions	  related	  to	  preferences	  in	  stroke	  prophylaxis.	  	  118	  	  6.3 Directions	  of	  future	  research	  Findings	  of	  this	  study	  are	  only	  the	  first	  step	  in	  exploring	  how	  and	  whether	  patient	  preference	  data	  can	  be	  incorporated	  on	  a	  systematic	  level.	  This	  study	  has	  demonstrated	  that	  while	  relative	  preferences	  for	  stroke	  prophylaxis	  are	  largely	  the	  same	  between	  physicians	  and	  AF	  patients,	  there	  are	  notable	  differences	  in	  strengths	  between	  the	  two	  groups.	  This	  difference	  may,	  or	  may	  not	  lead	  to	  different	  choices	  of	  antithrombotics.	  Related	  to	  the	  limitations	  pointed	  out	  in	  the	  previous	  section,	  I	  had	  hoped	  to	  identify	  factors	  related	  to	  the	  respondents	  that	  would	  better	  characterize	  their	  preference.	  This	  was	  apparent	  in	  the	  BWS	  analysis	  for	  the	  patient	  group,	  where	  three	  distinct	  latent	  classes	  were	  found	  to	  explain	  the	  differences	  in	  respondent	  choices.	  However,	  there	  was	  no	  way	  to	  identify	  those	  respondents	  in	  the	  respective	  classes,	  as	  some	  of	  those	  factors	  are	  latent	  and	  not	  observed	  in	  the	  demographic	  data.	  This	  remains	  to	  be	  elucidated	  in	  future	  studies	  that	  can	  enroll	  a	  large	  enough	  sample	  to	  meet	  this	  objective.	  	  	  From	  the	  methodological	  perspective,	  BWS	  is	  still	  a	  relatively	  new	  preference	  elicitation	  method.	  It	  has	  been	  shown	  to	  confer	  comparable	  preference	  estimates	  to	  other	  ordinal	  elicitation	  methods	  such	  as	  the	  discrete	  choice	  experiments	  and	  has	  been	  proposed	  to	  have	  the	  theoretical	  advantage	  of	  direct	  comparison	  across	  attribute	  levels.	  However,	  like	  other	  ordinal	  methods,	  major	  questions	  remain	  on	  how	  preference	  weights	  from	  those	  elicitation	  methods	  might	  be	  related	  to	  the	  von	  Neuman-­‐Morgenstern	  utilities,	  and	  how	  they	  can	  be	  used	  in	  cost-­‐utility	  analysis.	  This	  also	  leads	  to	  the	  other	  question	  of	  how	  guideline	  development	  committees	  119	  	  and	  other	  regulatory	  authorities	  can	  utilize	  such	  relative	  preference	  estimates	  in	  making	  health	  policies.	  	  	  6.4 Knowledge	  translation	  I	  hope	  to	  disseminate	  the	  results	  of	  this	  research	  to	  academic,	  health	  policy	  and	  clinician	  communities	  through	  publication	  and	  conference	  presentations.	  Practice	  guideline	  development	  committees	  might	  be	  interested	  in	  the	  results	  from	  this	  study.	  In	  particular,	  findings	  of	  patients	  being	  less	  stroke	  and	  bleed	  averse	  than	  physicians	  should	  be	  considered	  in	  coming	  up	  with	  therapeutic	  recommendations.	  Policy	  makers	  will	  also	  want	  to	  consider	  patient	  and	  physician	  preferences	  in	  making	  drug	  reimbursement	  decisions	  regarding	  stroke	  prophylaxis,	  and	  give	  weight	  to	  the	  fact	  that	  the	  relative	  utilities	  obtained	  in	  this	  study	  is	  representative	  of	  a	  patient	  population	  of	  interest,	  rather	  than	  the	  limited	  input	  from	  patient	  or	  public	  engagement	  as	  seen	  in	  many	  regulatory	  decision-­‐making	  process.	  	  	  	  6.5 Conclusion	  In	  evaluating	  and	  comparing	  preferences	  for	  stroke	  prophylaxis	  in	  physicians	  and	  AF	  patients,	  this	  study	  was	  able	  to	  characterize	  some	  of	  the	  factors	  important	  in	  making	  stroke	  prevention	  decisions	  for	  the	  two	  main	  stakeholders.	  Results	  from	  the	  preference	  elicitation	  task	  suggested	  that	  there	  was	  general	  congruence	  in	  preference	  between	  patients	  and	  physicians;	  however,	  there	  was	  a	  significant	  difference	  in	  preference	  strength.	  Findings	  of	  this	  study	  will	  be	  of	  importance	  and	  interest	  to	  academia,	  regulatory	  authorities,	  clinicians	  and	  patients	  alike,	  and	  120	  	  this	  is	  another	  step	  towards	  improving	  health	  outcomes	  in	  the	  era	  of	  patient-­‐centered	  health	  care.	  	  	  121	  	  Bibliography	  1.	  	   Bensing	  J.	  Bridging	  the	  gap:	  The	  separate	  worlds	  of	  evidence-­‐based	  medicine	  and	  patient-­‐centered	  medicine.	  Patient	  Educ	  Couns.	  2000;39(1):17–25.	  	  2.	  	   Rogers	  WA.	  Evidence-­‐based	  medicine	  in	  practice:	  limiting	  or	  facilitating	  patient	  choice?	  Health	  Expect.	  2002;5(2):95–103.	  	  3.	  	   Halpern	  J.	  Can	  the	  Development	  of	  Practice	  Guidelines	  Safeguard	  Patient	  Values.	  J	  Law	  Med	  Ethics.	  1995;23:75.	  	  4.	  	   Sackett,	  David	  L.	  The	  Doctor’s	  (ethical	  and	  Economic)	  Dilemma:	  OHE	  annual	  lecture	  series.	  London:	  Office	  of	  Health	  Economics;	  1996.	  	  5.	  	   Sackett	  DL,	  Rosenberg	  WM,	  Gray	  JA,	  Haynes	  RB,	  Richardson	  WS.	  Evidence	  based	  medicine:	  what	  it	  is	  and	  what	  it	  isn’t.	  BMJ.	  1996;312(7023):71–2.	  	  6.	  	   Weijden	  T	  van	  der,	  Légaré	  F,	  Boivin	  A,	  Burgers	  JS,	  Veenendaal	  H	  van,	  Stiggelbout	  AM,	  et	  al.	  How	  to	  integrate	  individual	  patient	  values	  and	  preferences	  in	  clinical	  practice	  guidelines?	  A	  research	  protocol.	  Implement	  Sci.	  2010;5(1):10.	  	  7.	  	   Guyatt	  GH,	  Oxman	  AD,	  Vist	  GE,	  Kunz	  R,	  Falck-­‐Ytter	  Y,	  Alonso-­‐Coello	  P,	  et	  al.	  GRADE:	  an	  emerging	  consensus	  on	  rating	  quality	  of	  evidence	  and	  strength	  of	  recommendations.	  BMJ.	  2008;336(7650):924–6.	  	  8.	  	   Sweeney	  KG,	  MacAuley	  D,	  Gray	  DP.	  Personal	  significance:	  the	  third	  dimension.	  Lancet.	  1998;351(9096):134–6.	  	  9.	  	   Bovenkamp	  HM,	  Trappenburg	  MJ.	  Reconsidering	  Patient	  Participation	  in	  Guideline	  Development.	  Health	  Care	  Anal.	  2008;17(3):198–216.	  	  10.	  	   Bardes	  CL.	  Defining	  “Patient-­‐Centered	  Medicine.”N	  Engl	  J	  Med.	  2012;366(9):782–3.	  	  11.	  	   Laine	  C,	  Davidoff	  F.	  Patient-­‐centered	  medicine:	  A	  professional	  evolution.	  JAMA.	  1996	  	  10;275(2):152–6.	  	  12.	  	   National	  Research	  Council.	  Crossing	  the	  Quality	  Chasm:	  A	  New	  Health	  System	  for	  the	  21st	  Century.	  Washington,	  DC:	  The	  National	  Academies	  Press;	  2001.	  360	  p.	  	  13.	  	   Lewin	  S,	  Skea	  Z,	  Entwistle	  VA,	  Zwarenstein	  M,	  Dick	  J.	  Interventions	  for	  providers	  to	  promote	  a	  patient-­‐centred	  approach	  in	  clinical	  consultations.	  Cochrane	  Database	  of	  Systematic	  Reviews	  [Internet].	  John	  Wiley	  &	  Sons,	  Ltd;	  1996	  [cited	  2013	  Aug	  1].	  Available	  from:	  122	  	  14.	  	   De	  Haes	  H.	  Dilemmas	  in	  patient	  centeredness	  and	  shared	  decision	  making:	  A	  case	  for	  vulnerability.	  Patient	  Educ	  Couns.	  2006;62(3):291–8.	  	  15.	  	   Cohen	  J.	  Are	  clinical	  practice	  guidelines	  impartial?	  Int	  J	  Technol	  Assess	  Health	  Care.	  2004;20(4):415–20.	  	  16.	  	   Giacomini	  MK,	  Cook	  DJ,	  Streiner	  DL,	  Anand	  SS.	  Using	  practice	  guidelines	  to	  allocate	  medical	  technologies.	  An	  ethics	  framework.	  Int	  J	  Technol	  Assess	  Health	  Care.	  2000;16(4):987–1002.	  	  17.	  	   Hofmann	  B.	  Toward	  a	  procedure	  for	  integrating	  moral	  issues	  in	  health	  technology	  assessment.	  Int	  J	  Technol	  Assess	  Health	  Care.	  2005;21(3):312–8.	  	  18.	  	   Schünemann	  HJ,	  Fretheim	  A,	  Oxman	  AD.	  Improving	  the	  use	  of	  research	  evidence	  in	  guideline	  development:	  10.	  Integrating	  values	  and	  consumer	  involvement.	  Health	  Res	  Policy	  Syst.	  2006;4(1):22.	  	  19.	  	   Morris	  AH,	  Ioannidis	  JPA.	  LImitations	  of	  medical	  research	  and	  evidence	  at	  the	  patient-­‐clinician	  encounter	  scale.	  CHEST.	  2013;143(4):1127–35.	  	  20.	  	   Grol	  R.	  Improving	  the	  quality	  of	  medical	  care:	  Building	  bridges	  among	  professional	  pride,	  payer	  profit,	  and	  patient	  satisfaction.	  JAMA.	  2001;286(20):2578–85.	  	  21.	  	   Barry	  MJ.	  Involving	  patients	  in	  medical	  decisions:	  how	  can	  physicians	  do	  better?	  JAMA.	  1999;282(24):2356–7.	  	  22.	  	   Kassirer	  JP.	  Incorporating	  patients’	  preferences	  into	  medical	  decisions.	  N	  Engl	  J	  Med.	  1994;330(26):1895–6.	  	  23.	  	   Eddy	  DM.	  Clinical	  decision	  making:	  from	  theory	  to	  practice.	  Rationing	  by	  patient	  choice.	  JAMA.	  1991;265(1):105–8.	  	  24.	  	   Gilmore	  A.	  Clinical	  practice	  guidelines:	  weapons	  for	  patients,	  or	  shields	  for	  MDs?	  CMAJ.	  1993;148(3):429–31.	  	  25.	  	   Nease	  RF,	  Owens	  DK.	  A	  Method	  for	  Estimating	  the	  Cost-­‐	  Effectiveness	  of	  Incorporating	  Patient	  Preferences	  into	  Practice	  Guidelines.	  Med	  Decis	  Making.	  1994;14(4):382–92.	  	  26.	  	   Krahn	  M,	  Naglie	  G.	  The	  next	  step	  in	  guideline	  development.	  JAMA.	  2008;300(4):436–8.	  	  27.	  	   Chong	  CA,	  Chen	  I,	  Naglie	  G,	  Krahn	  MD.	  How	  Well	  Do	  Guidelines	  Incorporate	  Evidence	  on	  Patient	  Preferences?	  J	  Gen	  Intern	  Med.	  2009;24(8):977–82.	  	  28.	  	   Shaneyfelt	  TM,	  Centor	  RM.	  Reassessment	  of	  clinical	  practice	  guidelines:	  go	  gently	  into	  that	  good	  night.	  JAMA.	  2009	  Feb	  25;301(8):868–9.	  	  123	  	  29.	  	   Vranceanu	  A-­‐M,	  Cooper	  C,	  Ring	  D.	  Integrating	  Patient	  Values	  into	  Evidence-­‐Based	  Practice:	  Effective	  Communication	  for	  Shared	  Decision-­‐Making.	  Hand	  Clin.	  2009;25(1):83–96.	  	  30.	  	   Krahn	  M.	  “New”	  Evidence	  for	  Clinical	  Practice	  Guidelines.	  Patient	  Patient-­‐Centered	  Outcomes	  Res.	  2010;3(2):71–7.	  	  31.	  	   Brennan	  PF,	  Strombom	  I.	  Improving	  Health	  Care	  by	  Understanding	  Patient	  Preferences	  The	  Role	  of	  Computer	  Technology.	  J	  Am	  Med	  Inform	  Assoc.	  1998;5(3):257–62.	  	  32.	  	   Drummond,	  Michael	  F.,	  Sculpher,	  Mark	  J,	  Torrance,	  George	  W,	  O’Brien,	  Bernie	  J,	  Stoddart,	  Greg	  L.	  Methods	  for	  the	  Economic	  Evaluation	  of	  Health	  Care	  Programmes.	  3rd	  ed.	  Oxford:	  Oxford	  University	  Press;	  2005.	  	  33.	  	   Von	  Neumann	  J,	  Morgenstern	  O.	  Theory	  of	  Games	  and	  Economic	  Behavior.	  New	  York:	  John	  Wiley	  &	  Sons,	  Ltd;	  1964.	  	  34.	  	   Ali	  S,	  Ronaldson	  S.	  Ordinal	  preference	  elicitation	  methods	  in	  health	  economics	  and	  health	  services	  research:	  using	  discrete	  choice	  experiments	  and	  ranking	  methods.	  Br	  Med	  Bull.	  2012;103(1):21–44.	  	  35.	  	   Detsky	  AS,	  Naglie	  G,	  Krahn	  MD,	  Naimark	  D,	  Redelmeier	  DA.	  Primer	  on	  Medical	  Decision	  Analysis:	  Part	  1—Getting	  Started.	  Med	  Decis	  Making.	  1997;17(2):123–5.	  	  36.	  	   Rb	  H,	  Pj	  D,	  Gh.	  G.	  Clinical	  Expertise	  in	  the	  Era	  of	  Evidence-­‐Based	  Medicine	  and	  Patient	  Choice	  (editorial).	  Vox	  Sang.	  2002;83:383–6.	  	  37.	  	   Towle	  A,	  Godolphin	  W,	  Greenhalgh	  T,	  Gambrill	  J.	  Framework	  for	  teaching	  and	  learning	  informed	  shared	  decision	  making	  Commentary:	  Competencies	  for	  informed	  shared	  decision	  making	  Commentary:	  Proposals	  based	  on	  too	  many	  assumptions.	  BMJ.	  1999;319(7212):766–71.	  	  38.	  	   Schünemann	  HJ,	  Hill	  SR,	  Kakad	  M,	  Vist	  GE,	  Bellamy	  R,	  Stockman	  L,	  et	  al.	  Transparent	  Development	  of	  the	  WHO	  Rapid	  Advice	  Guidelines.	  PLoS	  Med.	  2007;4(5):e119.	  	  39.	  	   Qaseem	  A,	  Snow	  V,	  Owens	  DK,	  Shekelle	  P.	  The	  Development	  of	  Clinical	  Practice	  Guidelines	  and	  Guidance	  Statements	  of	  the	  American	  College	  of	  Physicians:	  Summary	  of	  Methods.	  Ann	  Intern	  Med.	  2010;153(3):194–9.	  	  40.	  	   Woolf	  S,	  Schünemann	  HJ,	  Eccles	  MP,	  Grimshaw	  JM,	  Shekelle	  P.	  Developing	  clinical	  practice	  guidelines:	  types	  of	  evidence	  and	  outcomes;	  values	  and	  economics,	  synthesis,	  grading,	  and	  presentation	  and	  deriving	  recommendations.	  Implement	  Sci.	  2012;7(1):61.	  	  124	  	  41.	  	   Elwyn	  G.,	  Edwards	  A.,	  Kinnersley	  P.,	  Grol	  R.	  Shared	  decision	  making	  and	  the	  concept	  of	  equipoise:	  the	  competences	  of	  involving	  patients	  in	  healthcare	  choices.	  Br	  J	  Gen	  Pract.	  2000;50(460):892–9.	  	  42.	  	   Ryan	  M,	  Scott	  DA,	  Reeves	  C,	  Bate	  A,	  van	  Teijlingen	  ER,	  Russell	  EM,	  et	  al.	  Eliciting	  public	  preferences	  for	  healthcare:	  a	  systematic	  review	  of	  techniques.	  Health	  Technol	  Assess	  Winch	  Engl.	  2001;5(5):1–186.	  	  43.	  	   Llewellyn-­‐Thomas	  H,	  Sutherland	  HJ,	  Tibshirani	  R,	  Ciampi	  A,	  Till	  JE,	  Boyd	  NF.	  The	  measurement	  of	  patients’	  values	  in	  medicine.	  Med	  Decis	  Mak.	  1982;2(4):449–62.	  	  44.	  	   Ryan	  M,	  Farrar	  S.	  Using	  conjoint	  analysis	  to	  elicit	  preferences	  for	  health	  care.	  BMJ.	  2000;320(7248):1530–3.	  	  45.	  	   Ryan	  M.	  Using	  conjoint	  analysis	  to	  take	  account	  of	  patient	  preferences	  and	  go	  beyond	  health	  outcomes:	  an	  application	  to	  in	  vitro	  fertilisation.	  Soc	  Sci	  Med.	  1999;48(4):535–46.	  	  46.	  	   Kjaer,	  T.	  A	  review	  of	  the	  discrete	  choice	  experiment	  -­‐	  with	  emphasis	  on	  its	  application	  in	  health	  care.	  University	  of	  Southern	  Denmark;	  2005.	  	  47.	  	   Louviere	  JJ.	  Conjoint	  Analysis	  Modelling	  of	  Stated	  Preferences:	  A	  Review	  of	  Theory,	  Methods,	  Recent	  Developments	  and	  External	  Validity.	  J	  Transp	  Econ	  Policy.	  1988;22(1):93–119.	  	  48.	  	   Schoemaker	  PJH.	  The	  Expected	  Utility	  Model:	  Its	  Variants,	  Purposes,	  Evidence	  and	  Limitations.	  J	  Econ	  Lit.	  1982;20(2):529–63.	  	  49.	  	   Torrance	  GW.	  Measurement	  of	  health	  state	  utilities	  for	  economic	  appraisal.	  J	  Health	  Econ.	  1986;5(1):1–30.	  	  50.	  	   Brazier	  J,	  Rowen	  D,	  Yang	  Y,	  Tsuchiya	  A.	  Using	  rank	  and	  discrete	  choice	  data	  to	  estimate	  health	  state	  utility	  values	  on	  the	  QALY	  scale	  [Internet].	  University	  Library	  of	  Munich,	  Germany;	  2009.	  Report	  No.:	  29891.	  Available	  from:	  51.	  	   Bleichrodt	  H.	  A	  new	  explanation	  for	  the	  difference	  between	  time	  trade-­‐off	  utilities	  and	  standard	  gamble	  utilities.	  Health	  Econ.	  2002;11(5):447–56.	  	  52.	  	   Bala	  MV,	  Wood	  LL,	  Zarkin	  GA,	  Norton	  EC,	  Gafni	  A,	  O’Brien	  BJ.	  Are	  health	  states	  “timeless”?	  The	  case	  of	  the	  standard	  gamble	  method.	  J	  Clin	  Epidemiol.	  1999;52(11):1047–53.	  	  53.	  	   Cohen	  BJ.	  Is	  Expected	  Utility	  Theory	  Normative	  for	  Medical	  Decision	  Making?	  Med	  Decis	  Making.	  1996;16(1):1–6.	  	  125	  	  54.	  	   Wright	  DR,	  Wittenberg	  E,	  Swan	  JS,	  Miksad	  RA,	  Prosser	  LA.	  Methods	  for	  measuring	  temporary	  health	  States	  for	  cost-­‐utility	  analyses.	  PharmacoEconomics.	  2009;27(9):713–23.	  	  55.	  	   Ratcliffe	  J,	  Brazier	  J,	  Tsuchiya	  A,	  Symonds	  T,	  Brown	  M.	  Using	  DCE	  and	  ranking	  data	  to	  estimate	  cardinal	  values	  for	  health	  states	  for	  deriving	  a	  preference-­‐based	  single	  index	  from	  the	  sexual	  quality	  of	  life	  questionnaire.	  Health	  Econ.	  2009;18(11):1261–76.	  	  56.	  	   Flynn	  TN,	  Louviere	  JJ,	  Peters	  TJ,	  Coast	  J.	  Best–worst	  scaling:	  What	  it	  can	  do	  for	  health	  care	  research	  and	  how	  to	  do	  it.	  J	  Health	  Econ.	  2007;26(1):171–89.	  	  57.	  	   Salomon	  JA.	  Reconsidering	  the	  use	  of	  rankings	  in	  the	  valuation	  of	  health	  states:	  a	  model	  for	  estimating	  cardinal	  values	  from	  ordinal	  data.	  Popul	  Health	  Metr.	  2003;1(1):12.	  	  58.	  	   Ryan	  M,	  Gerard	  K,	  Amaya-­‐Amaya	  M.	  Using	  Discrete	  Choice	  Experiments	  to	  Value	  Health	  and	  Health	  Care.	  Springer;	  2007.	  	  59.	  	   Lancsar	  E,	  Louviere	  J,	  Donaldson	  C,	  Currie	  G,	  Burgess	  L.	  Best	  worst	  discrete	  choice	  experiments	  in	  health:	  Methods	  and	  an	  application.	  Soc	  Sci	  Med.	  2013;76:74–82.	  	  60.	  	   Flynn	  TN.	  Valuing	  citizen	  and	  patient	  preferences	  in	  health:	  recent	  developments	  in	  three	  types	  of	  best–worst	  scaling.	  Expert	  Rev	  Pharmacoecon	  Outcomes	  Res.	  2010;10(3):259–67.	  	  61.	  	   Finn	  A,	  Louviere	  JJ.	  Determining	  the	  Appropriate	  Response	  to	  Evidence	  of	  Public	  Concern:	  The	  Case	  of	  Food	  Safety.	  J	  Public	  Policy	  Mark.	  1992;11(2):12–25.	  	  62.	  	   Jordan	  J	  Louviere	  TNF.	  Using	  best-­‐worst	  scaling	  choice	  experiments	  to	  measure	  public	  perceptions	  and	  preferences	  for	  healthcare	  reform	  in	  australia.	  The	  patient.	  2010;3(4):275–83.	  	  63.	  	   Lancsar	  E,	  Louviere	  J,	  Flynn	  T.	  Several	  methods	  to	  investigate	  relative	  attribute	  impact	  in	  stated	  preference	  experiments.	  Soc	  Sci	  Med.	  2007;64(8):1738–53.	  	  64.	  	   De	  Bekker-­‐Grob	  EW,	  Ryan	  M,	  Gerard	  K.	  Discrete	  choice	  experiments	  in	  health	  economics:	  a	  review	  of	  the	  literature.	  Health	  Econ.	  2012;21(2):145-­‐72.	  	  65.	  	   Goekoop-­‐Ruiterman	  YPM,	  de	  Vries-­‐Bouwstra	  JK,	  Allaart	  CF,	  Kerstens	  PJSM,	  Grillet	  BAM,	  de	  Jager	  MH,	  et	  al.	  Patient	  preferences	  for	  treatment:	  report	  from	  a	  randomised	  comparison	  of	  treatment	  strategies	  in	  early	  rheumatoid	  arthritis	  (BeSt	  trial).	  Ann	  Rheum	  Dis.	  2007;66(9):1227–32.	  	  66.	  	   O’Brien	  BJ,	  Elswood	  J,	  Calin	  A.	  Willingness	  to	  accept	  risk	  in	  the	  treatment	  of	  rheumatic	  disease.	  J	  Epidemiol	  Community	  Health.	  1990;44(3):249–52.	  	  126	  	  67.	  	   Thompson	  MS.	  Willingness	  to	  pay	  and	  accept	  risks	  to	  cure	  chronic	  disease.	  Am	  J	  Public	  Health.	  1986;76(4):392–6.	  	  68.	  	   Ho	  M,	  Lavery	  B,	  Pullar	  T.	  The	  risk	  of	  treatment.	  A	  study	  of	  rheumatoid	  arthritis	  patients’	  attitudes.	  Rheumatology.	  1998;37(4):459–60.	  	  69.	  	   Lane	  DA,	  Hutchinson	  TA.	  The	  notion	  of	  “acceptable	  risk”:	  The	  role	  of	  utility	  in	  drug	  management.	  J	  Chronic	  Dis.	  1987;40(6):621–5.	  	  70.	  	   Louviere	  JJ,	  Lancsar	  E.	  Choice	  experiments	  in	  health:	  the	  good,	  the	  bad,	  the	  ugly	  and	  toward	  a	  brighter	  future.	  Health	  Econ	  Policy	  Law.	  2009;4(04):527–46.	  	  71.	  	   Kerr,	  C,	  Roy,	  D.	  Canadian	  Cardiovascular	  Society	  Consensu	  Conference:	  Atrial	  Fibrillation.	  Can	  J	  Cardiol.	  2005;21	  Suppl	  B:9B–73B.	  	  72.	  	   Go	  AS,	  Hylek	  EM,	  Phillips	  KA,	  et	  al.	  Prevalence	  of	  diagnosed	  atrial	  fibrillation	  in	  adults:	  National	  implications	  for	  rhythm	  management	  and	  stroke	  prevention:	  the	  anticoagulation	  and	  risk	  factors	  in	  atrial	  fibrillation	  (atria)	  study.	  JAMA.	  2001;285(18):2370–5.	  	  73.	  	   DeWilde	  S,	  Carey	  IM,	  Emmas	  C,	  Richards	  N,	  Cook	  DG.	  Trends	  in	  the	  prevalence	  of	  diagnosed	  atrial	  fibrillation,	  its	  treatment	  with	  anticoagulation	  and	  predictors	  of	  such	  treatment	  in	  UK	  primary	  care.	  Heart.	  2006;92(8):1064–70.	  	  74.	  	   Lip	  GYH,	  Brechin	  CM,	  Lane	  DA.	  The	  global	  burden	  of	  atrial	  fibrillation	  and	  stroke:	  A	  systematic	  review	  of	  the	  epidemiology	  of	  atrial	  fibrillation	  in	  regions	  outside	  north	  america	  and	  europe.	  CHEST.	  2012;142(6):1489–98.	  	  75.	  	   Kirchhof	  P,	  Auricchio	  A,	  Bax	  J,	  Crijns	  H,	  Camm	  J,	  Diener	  H-­‐C,	  et	  al.	  Outcome	  parameters	  for	  trials	  in	  atrial	  fibrillation:	  executive	  summary	  Recommendations	  from	  a	  consensus	  conference	  organized	  by	  the	  German	  Atrial	  Fibrillation	  Competence	  NETwork	  (AFNET)	  and	  the	  European	  Heart	  Rhythm	  Association	  (EHRA).	  Eur	  Heart	  J.	  2007;28(22):2803–17.	  	  76.	  	   Friberg	  J,	  Scharling	  H,	  Gadsbøll	  N,	  Truelsen	  T,	  Jensen	  GB.	  Comparison	  of	  the	  impact	  of	  atrial	  fibrillation	  on	  the	  risk	  of	  stroke	  and	  cardiovascular	  death	  in	  women	  versus	  men	  (The	  Copenhagen	  City	  Heart	  Study).	  Am	  J	  Cardiol.	  2004;94(7):889–94.	  	  77.	  	   Miyasaka	  Y,	  Barnes	  ME,	  Bailey	  KR,	  Cha	  SS,	  Gersh	  BJ,	  Seward	  JB,	  et	  al.	  Mortality	  Trends	  in	  Patients	  Diagnosed	  With	  First	  Atrial	  FibrillationA	  21-­‐Year	  Community-­‐Based	  Study.	  J	  Am	  Coll	  Cardiol.	  2007;49(9):986–92.	  	  78.	  	   Conen	  D,	  Chae	  CU,	  Glynn	  RJ,	  et	  al.	  RIsk	  of	  death	  and	  cardiovascular	  events	  in	  initially	  healthy	  women	  with	  new-­‐onset	  atrial	  fibrillation.	  JAMA.	  2011;305(20):2080–7.	  	  127	  	  79.	  	   Gattellari	  M,	  Goumas	  C,	  Aitken	  R,	  Worthington	  JM.	  Outcomes	  for	  Patients	  with	  Ischaemic	  Stroke	  and	  Atrial	  Fibrillation:	  The	  PRISM	  Study	  (A	  Program	  of	  Research	  Informing	  Stroke	  Management).	  Cerebrovasc	  Dis.	  2011;32(4):370–82.	  	  80.	  	   Wolf	  PA,	  Abbott	  RD,	  Kannel	  WB.	  Atrial	  fibrillation	  as	  an	  independent	  risk	  factor	  for	  stroke:	  the	  Framingham	  Study.	  Stroke.	  1991;22(8):983–8.	  	  81.	  	   Alberts	  MJ,	  Eikelboom	  JW,	  Hankey	  GJ.	  Antithrombotic	  therapy	  for	  stroke	  prevention	  in	  non-­‐valvular	  atrial	  fibrillation.	  Lancet	  Neurol.	  2012;11(12):1066–81.	  	  82.	  	   Potpara	  TS,	  Polovina	  MM,	  Licina	  MM,	  Marinkovic	  JM,	  Prostran	  MS,	  Lip	  GYH.	  Reliable	  Identification	  of	  “Truly	  Low”	  Thromboembolic	  Risk	  in	  Patients	  Initially	  Diagnosed	  With	  “Lone”	  Atrial	  Fibrillation	  The	  Belgrade	  Atrial	  Fibrillation	  Study.	  Circ	  Arrhythm	  Electrophysiol.	  2012;5(2):319–26.	  	  83.	  	   Potpara	  TS,	  Stankovic	  GR,	  Beleslin	  BD,	  Polovina	  MM,	  Marinkovic	  JM,	  Ostojic	  MC,	  et	  al.	  A	  12-­‐year	  follow-­‐up	  study	  of	  patients	  with	  newly	  diagnosed	  lone	  atrial	  fibrillation:	  Implications	  of	  arrhythmia	  progression	  on	  prognosis:	  the	  belgrade	  atrial	  fibrillation	  study.	  CHEST.	  2012;141(2):339–47.	  	  84.	  	   Sacco	  RL,	  Adams	  R,	  Albers	  G,	  Alberts	  MJ,	  Benavente	  O,	  Furie	  K,	  et	  al.	  Guidelines	  for	  Prevention	  of	  Stroke	  in	  Patients	  With	  Ischemic	  Stroke	  or	  Transient	  Ischemic	  Attack	  A	  Statement	  for	  Healthcare	  Professionals	  From	  the	  American	  Heart	  Association/American	  Stroke	  Association	  Council	  on	  Stroke:	  Co-­‐Sponsored	  by	  the	  Council	  on	  Cardiovascular	  Radiology	  and	  Intervention:	  The	  American	  Academy	  of	  Neurology	  affirms	  the	  value	  of	  this	  guideline.	  Circulation.	  2006;113(10):e409–e449.	  	  85.	  	   Hart	  RG,	  Benavente	  O,	  McBride	  R,	  Pearce	  LA.	  Antithrombotic	  therapy	  to	  prevent	  stroke	  in	  patients	  with	  atrial	  fibrillation:	  a	  meta-­‐analysis.	  Ann	  Intern	  Med.	  1999;131(7):492–501.	  	  86.	  	   Copland	  M,	  Walker	  ID,	  Tait	  R.	  ORal	  anticoagulation	  and	  hemorrhagic	  complications	  in	  an	  elderly	  population	  with	  atrial	  fibrillation.	  Arch	  Intern	  Med.	  2001	  Sep;161(17):2125–8.	  	  87.	  	   Go	  AS,	  Hylek	  EM,	  Chang	  Y,	  et	  al.	  Anticoagulation	  therapy	  for	  stroke	  prevention	  in	  atrial	  fibrillation:	  How	  well	  do	  randomized	  trials	  translate	  into	  clinical	  practice?	  JAMA.	  2003;290(20):2685–92.	  	  88.	  	   Connolly	  SJ,	  Eikelboom	  J,	  Joyner	  C,	  Diener	  H-­‐C,	  Hart	  R,	  Golitsyn	  S,	  et	  al.	  Apixaban	  in	  Patients	  with	  Atrial	  Fibrillation.	  N	  Engl	  J	  Med.	  2011;364(9):806–17.	  	  89.	  	   Connolly	  SJ,	  Ezekowitz	  MD,	  Yusuf	  S,	  Eikelboom	  J,	  Oldgren	  J,	  Parekh	  A,	  et	  al.	  Dabigatran	  versus	  Warfarin	  in	  Patients	  with	  Atrial	  Fibrillation.	  N	  Engl	  J	  Med.	  2009;361(12):1139–51.	  	  128	  	  90.	  	   Effect	  of	  Clopidogrel	  Added	  to	  Aspirin	  in	  Patients	  with	  Atrial	  Fibrillation.	  N	  Engl	  J	  Med.	  2009;360(20):2066–78.	  	  91.	  	   Clopidogrel	  plus	  aspirin	  versus	  oral	  anticoagulation	  for	  atrial	  fibrillation	  in	  the	  Atrial	  fibrillation	  Clopidogrel	  Trial	  with	  Irbesartan	  for	  prevention	  of	  Vascular	  Events	  (ACTIVE	  W):	  a	  randomised	  controlled	  trial.	  Lancet.	  2006;367(9526):1903–12.	  	  92.	  	   ROCKET	  AF	  Study	  Investigators.	  Rivaroxaban—Once	  daily,	  oral,	  direct	  factor	  Xa	  inhibition	  Compared	  with	  vitamin	  K	  antagonism	  for	  prevention	  of	  stroke	  and	  Embolism	  Trial	  in	  Atrial	  Fibrillation:	  Rationale	  and	  Design	  of	  the	  ROCKET	  AF	  study.	  Am	  Heart	  J.	  2010;159(3):340–347.	  	  93.	  	   Cairns	  JA,	  Connolly	  S,	  McMurtry	  S,	  Stephenson	  M,	  Talajic	  M.	  Canadian	  Cardiovascular	  Society	  Atrial	  Fibrillation	  Guidelines	  2010:	  Prevention	  of	  Stroke	  and	  Systemic	  Thromboembolism	  in	  Atrial	  Fibrillation	  and	  Flutter.	  Can	  J	  Cardiol.	  2011;27(1):74–90.	  	  94.	  	   Hughes	  S.	  Experts	  compare	  new	  oral	  anticoagulatns	  preparing	  to	  battle	  for	  the	  warfarin	  market.	  [Internet].	  2010	  Dec	  14	  [cited	  2011	  Jan	  31];	  Available	  from:	  95.	  	   Fraenkel	  L,	  Fried	  TR.	  Individualized	  medical	  decision	  making:	  Necessary,	  achievable,	  but	  not	  yet	  attainable.	  Arch	  Intern	  Med.	  2010;170(6):566–9.	  	  96.	  	   McAlister	  F,	  Laupacis	  A,	  Man-­‐Song-­‐Hing	  M.	  Choices:	  an	  atrial	  fibrillation	  treatment	  decision-­‐making	  aid	  [Internet].	  Ottawa:	  Canadian	  Stroke	  Network;	  1996.	  Available	  from:	  97.	  	   McAlister	  FA,	  Man-­‐Son-­‐Hing	  M,	  Straus	  SE,	  Ghali	  WA,	  Gibson	  P,	  Anderson	  D,	  et	  al.	  A	  randomized	  trial	  to	  assess	  the	  impact	  of	  an	  antithrombotic	  decision	  aid	  in	  patients	  with	  nonvalvular	  atrial	  fibrillation:	  the	  DAAFI	  trial	  protocol	  [ISRCTN14429643].	  BMC	  Cardiovasc	  Disord.	  2004;4(1):5.	  	  98.	  	   Hagerty	  RG,	  Butow	  PN,	  Ellis	  PA,	  Lobb	  EA,	  Pendlebury	  S,	  Leighl	  N,	  et	  al.	  Cancer	  Patient	  Preferences	  for	  Communication	  of	  Prognosis	  in	  the	  Metastatic	  Setting.	  J	  Clin	  Oncol.	  2004;22(9):1721–30.	  	  99.	  	   Kaplowitz	  SA,	  Campo	  S,	  Chiu	  WT.	  Cancer	  Patients’	  Desires	  for	  Communication	  of	  Prognosis	  Information.	  Health	  Commun.	  2002;14(2):221–41.	  	  100.	  	   Swenson	  SL,	  Buell	  S,	  Zettler	  P,	  White	  M,	  Ruston	  DC,	  Lo	  B.	  Patient-­‐centered	  Communication.	  J	  Gen	  Intern	  Med.	  2004;19(11):1069–79.	  	  101.	  	   Arora	  NK,	  McHorney	  CA.	  Patient	  Preferences	  for	  Medical	  Decision	  Making:	  Who	  Really	  Wants	  to	  Participate?	  Med	  Care.	  2000;38(3):335–41.	  	  129	  	  102.	  	   Jones	  DJ,	  Barkun	  AN,	  Lu	  Y,	  Enns	  R,	  Sinclair	  P,	  Martel	  M,	  et	  al.	  Conflicts	  of	  Interest	  Ethics:	  Silencing	  Expertise	  in	  the	  Development	  of	  International	  Clinical	  Practice	  Guidelines.	  Ann	  Intern	  Med.	  2012;156(11):809–16.	  	  103.	  	   Norris	  SL,	  Burda	  BU,	  Holmer	  HK,	  Ogden	  LA,	  Fu	  R,	  Bero	  L,	  et	  al.	  Author’s	  specialty	  and	  conflicts	  of	  interest	  contribute	  to	  conflicting	  guidelines	  for	  screening	  mammography.	  J	  Clin	  Epidemiol.	  2012;65(7):725–33.	  	  104.	  	   Neuman	  J,	  Korenstein	  D,	  Ross	  JS,	  Keyhani	  S.	  Prevalence	  of	  financial	  conflicts	  of	  interest	  among	  panel	  members	  producing	  clinical	  practice	  guidelines	  in	  Canada	  and	  United	  States:	  cross	  sectional	  study.	  BMJ.	  2011;343:d5621.	  	  105.	  	   Tricoci	  P,	  Allen	  JM,	  Kramer	  JM,	  Califf	  RM,	  Smith	  SC.	  SCientific	  evidence	  underlying	  the	  acc/aha	  clinical	  practice	  guidelines.	  JAMA.	  2009;301(8):831–41.	  	  106.	  	   O’Connor	  AM,	  Stacey	  D,	  Entwistle	  V,	  Llewellyn-­‐Thomas	  H,	  Rovner	  D,	  Holmes-­‐Rovner	  M,	  et	  al.	  Decision	  aids	  for	  people	  facing	  health	  treatment	  or	  screening	  decisions.	  Cochrane	  Database	  Syst	  Rev.	  2003;(2):CD001431.	  	  107.	  	   Graham	  ID,	  Beardall	  S,	  Carter	  AO,	  Tetroe	  J,	  Davies	  B.	  The	  state	  of	  the	  science	  and	  art	  of	  practice	  guidelines	  development,	  dissemination	  and	  evaluation	  in	  Canada.	  J	  Eval	  Clin	  Pract.	  2003;9(2):195–202.	  	  108.	  	   Protheroe	  J,	  Fahey	  T,	  Montgomery	  AA,	  Peters	  TJ,	  Smeeth	  L.	  The	  impact	  of	  patients’	  preferences	  on	  the	  treatment	  of	  atrial	  fibrillation:	  observational	  study	  of	  patient	  based	  decision	  analysisCommentary:	  patients,	  preferences,	  and	  evidence.	  BMJ.	  2000;320(7246):1380–4.	  	  109.	  	   Man-­‐Son-­‐Hing	  M,	  Laupacis	  A.	  BAlancing	  the	  risks	  of	  stroke	  and	  upper	  gastrointestinal	  tract	  bleeding	  in	  older	  patients	  with	  atrial	  fibrillation.	  Arch	  Intern	  Med.	  2002;162(5):541–50.	  	  110.	  	   Gage	  BF,	  Cardinalli	  AB,	  Owens	  DK.	  THe	  effect	  of	  stroke	  and	  stroke	  prophylaxis	  with	  aspirin	  or	  warfarin	  on	  quality	  of	  life.	  Arch	  Intern	  Med.	  1996;156(16):1829–36.	  	  111.	  	   MacLean	  S,	  Mulla	  S,	  Akl	  EA,	  Jankowski	  M,	  Vandvik	  PO,	  Ebrahim	  S,	  et	  al.	  Patient	  values	  and	  preferences	  in	  decision	  making	  for	  antithrombotic	  therapy:	  A	  systematic	  review:	  antithrombotic	  therapy	  and	  prevention	  of	  thrombosis,	  9th	  ed:	  american	  college	  of	  chest	  physicians	  evidence-­‐based	  clinical	  practice	  guidelines.	  CHEST	  J.	  2012;141(2_suppl):e1S–e23S.	  	  112.	  	   Lynd	  LD,	  O’Brien	  BJ.	  Advances	  in	  risk-­‐benefit	  evaluation	  using	  probabilistic	  simulation	  methods:	  an	  application	  to	  the	  prophylaxis	  of	  deep	  vein	  thrombosis.	  J	  Clin	  Epidemiol.	  2004;57(8):795–803.	  	  130	  	  113.	  	   Mühlbacher	  AC,	  Juhnke	  C.	  Patient	  Preferences	  Versus	  Physicians’	  Judgement:	  Does	  it	  Make	  a	  Difference	  in	  Healthcare	  Decision	  Making?	  Appl	  Health	  Econ	  Health	  Policy.	  2013;11(3):163–80.	  	  114.	  	   Devereaux	  PJ,	  Anderson	  DR,	  Gardner	  MJ,	  Putnam	  W,	  Flowerdew	  GJ,	  Brownell	  BF,	  et	  al.	  Differences	  between	  perspectives	  of	  physicians	  and	  patients	  on	  anticoagulation	  in	  patients	  with	  atrial	  fibrillation:	  observational	  study.	  BMJ.	  2001;323(7323):1218.	  	  115.	  	   Slomka	  J,	  Hoffman-­‐Hogg	  L,	  Mion	  LC,	  Bair	  N,	  Bobek	  MB,	  Arroliga	  AC.	  Influence	  of	  clinicians’	  values	  and	  perceptions	  on	  use	  of	  clinical	  practice	  guidelines	  for	  sedation	  and	  neuromuscular	  blockade	  in	  patients	  receiving	  mechanical	  ventilation.	  Am	  J	  Crit	  Care.	  2000;9(6):412–8.	  	  116.	  	   Uhlmann	  RF,	  Pearlman	  RA,	  Cain	  KC.	  Understanding	  of	  elderly	  patients’	  resuscitation	  preferences	  by	  physicians	  and	  nurses.	  West	  J	  Med.	  1989;150(6):705–7.	  	  117.	  	   Alonso-­‐Coello	  P,	  Montori	  VM,	  Solà	  I,	  Schünemann	  HJ,	  Devereaux	  PJ,	  Charles	  C,	  et	  al.	  Values	  and	  preferences	  in	  oral	  anticoagulation	  in	  patients	  with	  atrial	  fibrillation,	  physicians’	  and	  patients’	  perspectives:	  protocol	  for	  a	  two-­‐phase	  study.	  BMC	  Health	  Serv	  Res.	  2008;8(1):221.	  	  118.	  	   Gage	  BF,	  Cardinalli	  AB,	  Owens	  DK.	  Cost-­‐Effectiveness	  of	  Preference-­‐Based	  Antithrombotic	  Therapy	  for	  Patients	  With	  Nonvalvular	  Atrial	  Fibrillation.	  Stroke.	  1998	  Jun;29(6):1083–91.	  	  119.	  	   Meister	  H,	  Lausberg	  I,	  Kiessling	  J,	  Walger	  M,	  von	  Wedel	  H.	  Determining	  the	  Importance	  of	  Fundamental	  Hearing	  Aid	  Attributes.	  [Miscellaneous	  Article].	  Otol	  Neurotol.	  2002;23(4):457–62.	  	  120.	  	   Bishop	  AJ,	  Marteau	  TM,	  Armstrong	  D,	  Chitty	  LS,	  Longworth	  L,	  Buxton	  MJ,	  et	  al.	  Women	  and	  health	  care	  professionals’	  preferences	  for	  Down’s	  Syndrome	  screening	  tests:	  a	  conjoint	  analysis	  study.	  Int	  J	  Obstet	  Gynaecol.	  2004;111(8):775–9.	  	  121.	  	   Lee	  A,	  Gin	  T,	  Lau	  ASC,	  Ng	  FF.	  A	  Comparison	  of	  Patients’	  and	  Health	  Care	  Professionals’	  Preferences	  for	  Symptoms	  During	  Immediate	  Postoperative	  Recovery	  and	  the	  Management	  of	  Postoperative	  Nausea	  and	  Vomiting.	  Anesth	  Analg.	  2005;100(1):87–93.	  	  122.	  	   Sassi	  F,	  McDaid	  D,	  Ricciardi	  W.	  Conjoint	  analysis	  of	  preferences	  for	  cardiac	  risk	  assessment	  in	  primary	  care.	  Int	  J	  Technol	  Assess	  Health	  Care.	  2005;21(02):211–8.	  	  123.	  	   Lewis	  SM,	  Cullinane	  FM,	  Carlin	  JB,	  Halliday	  JL.	  Women’s	  and	  health	  professionals’	  preferences	  for	  prenatal	  testing	  for	  Down	  syndrome	  in	  Australia.	  Aust	  N	  Z	  J	  Obstet	  Gynaecol.	  2006;46(3):205–11.	  	  131	  	  124.	  	   Vermeulen	  H,	  Ubbink	  DT,	  De	  Zwart	  F,	  Goossens	  A,	  De	  Vos	  R.	  Preferences	  of	  patients,	  doctors,	  and	  nurses	  regarding	  wound	  dressing	  characteristics:	  A	  conjoint	  analysis.	  Wound	  Repair	  Regen.	  2007;15(3):302–7.	  	  125.	  	   Sampietro-­‐Colom	  L,	  Espallargues	  M,	  Rodríguez	  E,	  Comas	  M,	  Alonso	  J,	  Castells	  X,	  et	  al.	  Wide	  Social	  Participation	  in	  Prioritizing	  Patients	  on	  Waiting	  Lists	  for	  Joint	  Replacement:	  A	  Conjoint	  Analysis.	  Med	  Decis	  Making.	  2008;28(4):554–66.	  	  126.	  	   Bederman	  SS,	  Mahomed	  NN,	  Kreder	  HJ,	  McIsaac	  WJ,	  Coyte	  PC,	  Wright	  JG.	  In	  the	  Eye	  of	  the	  Beholder:	  Preferences	  of	  Patients,	  Family	  Physicians,	  and	  Surgeons	  for	  Lumbar	  Spinal	  Surgery.	  Spine.	  2010;35(1):108–15.	  	  127.	  	   Gregorian	  Jr.	  RS,	  Gasik	  A,	  Kwong	  WJ,	  Voeller	  S,	  Kavanagh	  S.	  Importance	  of	  Side	  Effects	  in	  Opioid	  Treatment:	  A	  Trade-­‐Off	  Analysis	  With	  Patients	  and	  Physicians.	  J	  Pain.	  2010;11(11):1095–108.	  	  128.	  	   Porzsolt	  F,	  Clouth	  J,	  Deutschmann	  M,	  Hippler	  H-­‐J.	  Preferences	  of	  diabetes	  patients	  and	  physicians:	  A	  feasibility	  study	  to	  identify	  the	  key	  indicators	  for	  appraisal	  of	  health	  care	  values.	  Health	  Qual	  Life	  Outcomes.	  2010;8(1):125.	  	  129.	  	   Mantovani	  LG,	  Monzini	  MS,	  Mannucci	  PM,	  Scalone	  L,	  Villa	  M,	  Gringeri	  A,	  et	  al.	  Differences	  between	  patients’,	  physicians’	  and	  pharmacists’	  preferences	  for	  treatment	  products	  in	  haemophilia:	  a	  discrete	  choice	  experiment.	  Haemophilia.	  2005;11(6):589–97.	  	  130.	  	   Ashcroft	  DM,	  Seston	  E,	  Griffiths	  CEM.	  Trade-­‐offs	  between	  the	  benefits	  and	  risks	  of	  drug	  treatment	  for	  psoriasis:	  a	  discrete	  choice	  experiment	  with	  U.K.	  dermatologists.	  Br	  J	  Dermatol.	  2006;155(6):1236–41.	  	  131.	  	   Seston	  EM,	  Ashcroft	  DM,	  Griffiths	  CM.	  Balancing	  the	  benefits	  and	  risks	  of	  drug	  treatment:	  A	  stated-­‐preference,	  discrete	  choice	  experiment	  with	  patients	  with	  psoriasis.	  Arch	  Dermatol.	  2007;143(9):1175–9.	  	  132.	  	   Langer	  CJ,	  Fastenau	  JM,	  Forlenza	  JB,	  Tak	  Piech	  C,	  Bolge	  SC,	  Gano	  Allen	  J,	  et	  al.	  Effectiveness	  versus	  convenience:	  patient	  preferences	  for	  an	  erythropoietic	  agent	  to	  treat	  cancer-­‐related	  anemia.	  Curr	  Med	  Res	  Opin.	  2007;23(1):85–92.	  	  133.	  	   Muhlbacher	  AC,	  Lincke	  H-­‐J,	  Nubling	  M.	  Evaluating	  patients’	  preferences	  for	  multiple	  myeloma	  therapy,	  a	  Discrete-­‐Choice-­‐Experiment.	  Psychosoc	  Med.	  2008;5:1-­‐11.	  134.	  	   Mühlbacher	  AC,	  Nübling	  M.	  Analysis	  of	  physicians’	  perspectives	  versus	  patients’	  preferences:	  direct	  assessment	  and	  discrete	  choice	  experiments	  in	  the	  therapy	  of	  multiple	  myeloma.	  Eur	  J	  Health	  Econ.	  2011;12(3):193–203.	  	  132	  	  135.	  	   Bekker-­‐Grob	  EW	  de,	  Essink-­‐Bot	  M-­‐L,	  Meerding	  WJ,	  Koes	  BW,	  Steyerberg	  EW.	  Preferences	  of	  GPs	  and	  Patients	  for	  Preventive	  Osteoporosis	  Drug	  Treatment.	  Pharmacoeconomics.	  2009;27(3):211–9.	  	  136.	  	   Fiebig	  DG,	  Haas	  M,	  Hossain	  I,	  Street	  DJ,	  Viney	  R.	  Decisions	  about	  Pap	  tests:	  What	  influences	  women	  and	  providers?	  Soc	  Sci	  Med.	  2009;68(10):1766–74.	  	  137.	  	   Johnson	  FR,	  Hauber	  B,	  Özdemir	  S,	  Siegel	  CA,	  Hass	  S,	  Sands	  BE.	  Are	  gastroenterologists	  less	  tolerant	  of	  treatment	  risks	  than	  patients?	  Benefit-­‐risk	  preferences	  in	  Crohn’s	  disease	  management.	  J	  Manag	  Care	  Pharm.	  2010;16(8):616–28.	  	  138.	  	   Shafey	  M,	  Lupichuk	  SM,	  Do	  T,	  Owen	  C,	  Stewart	  DA.	  Preferences	  of	  patients	  and	  physicians	  concerning	  treatment	  options	  for	  relapsed	  follicular	  lymphoma:	  a	  discrete	  choice	  experiment.	  Bone	  Marrow	  Transplant.	  2011;46(7):962–9.	  	  139.	  	   Thrumurthy	  SG,	  Morris	  JJA,	  Mughal	  MM,	  Ward	  JB.	  Discrete-­‐choice	  preference	  comparison	  between	  patients	  and	  doctors	  for	  the	  surgical	  management	  of	  oesophagogastric	  cancer.	  Br	  J	  Surg.	  2011;98(8):1124–31.	  	  140.	  	   Faggioli	  G,	  Scalone	  L,	  Mantovani	  LG,	  Borghetti	  F,	  Stella	  A.	  Preferences	  of	  Patients,	  Their	  Family	  Caregivers	  and	  Vascular	  Surgeons	  in	  the	  Choice	  of	  Abdominal	  Aortic	  Aneurysms	  Treatment	  Options:	  The	  PREFER	  Study.	  Eur	  J	  Vasc	  Endovasc	  Surg.	  2011;42(1):26–34.	  	  141.	  	   Empel	  IWH	  van,	  Dancet	  EAF,	  Koolman	  XHE,	  Nelen	  WLDM,	  Stolk	  EA,	  Sermeus	  W,	  et	  al.	  Physicians	  underestimate	  the	  importance	  of	  patient-­‐centredness	  to	  patients:	  a	  discrete	  choice	  experiment	  in	  fertility	  care.	  Hum	  Reprod.	  2011;26(3):584–93.	  	  142.	  	   Pedersen	  LB,	  Kjær	  T,	  Kragstrup	  J,	  Gyrd-­‐Hansen	  D.	  Do	  general	  practitioners	  know	  patients’	  preferences?	  An	  empirical	  study	  on	  the	  agency	  relationship	  at	  an	  aggregate	  level	  using	  a	  discrete	  choice	  experiment.	  Value	  Health	  J	  Int	  Soc	  Pharmacoeconomics	  Outcomes	  Res.	  2012;15(3):514–23.	  	  143.	  	   Hill	  M,	  Fisher	  J,	  Chitty	  LS,	  Morris	  S.	  Women/’s	  and	  health	  professionals/’	  preferences	  for	  prenatal	  tests	  for	  Down	  syndrome:	  a	  discrete	  choice	  experiment	  to	  contrast	  noninvasive	  prenatal	  diagnosis	  with	  current	  invasive	  tests.	  Genet	  Med.	  2012;14(11):905–13.	  	  144.	  	   Park	  M-­‐H,	  Jo	  C,	  Bae	  EY,	  Lee	  E-­‐K.	  A	  Comparison	  of	  Preferences	  of	  Targeted	  Therapy	  for	  Metastatic	  Renal	  Cell	  Carcinoma	  between	  the	  Patient	  Group	  and	  Health	  Care	  Professional	  Group	  in	  South	  Korea.	  Value	  Health.	  2012;15(6):933–9.	  	  145.	  	   Regier	  DA,	  Diorio	  C,	  Ethier	  M-­‐C,	  Alli	  A,	  Alexander	  S,	  Boydell	  KM,	  et	  al.	  Discrete	  Choice	  Experiment	  to	  Evaluate	  Factors	  That	  Influence	  Preferences	  for	  Antibiotic	  Prophylaxis	  in	  Pediatric	  Oncology.	  PLoS	  ONE.	  2012;7(10):e47470.	  	  133	  	  146.	  	   Chancellor	  J,	  Martin	  M,	  Liedgens	  H,	  Baker	  MG,	  Müller-­‐Schwefe	  GHH.	  Stated	  Preferences	  of	  Physicians	  and	  Chronic	  Pain	  Sufferers	  in	  the	  Use	  of	  Classic	  Strong	  Opioids.	  Value	  Health.	  2012;15(1):106–17.	  	  147.	  	   De	  Bekker-­‐Grob	  EW,	  Bliemer	  MCJ,	  Donkers	  B,	  Essink-­‐Bot	  M-­‐L,	  Korfage	  IJ,	  Roobol	  MJ,	  et	  al.	  Patients’	  and	  urologists’	  preferences	  for	  prostate	  cancer	  treatment:	  a	  discrete	  choice	  experiment.	  Br	  J	  Cancer.	  2013;109(3):633–40.	  	  148.	  	   Trope	  Y,	  Liberman	  N.	  Construal-­‐Level	  Theory	  of	  Psychological	  Distance.	  Psychol	  Rev.	  2010;117(2):440–63.	  	  149.	  	   Louviere	  JJ,	  Flynn	  TN,	  Carson	  RT.	  Discrete	  Choice	  Experiments	  Are	  Not	  Conjoint	  Analysis.	  J	  Choice	  Model.	  2010;3(3):57–72.	  	  150.	  	   Flynn	  TN,	  Louviere	  JJ,	  Peters	  TJ,	  Coast	  J.	  Estimating	  preferences	  for	  a	  dermatology	  consultation	  using	  Best-­‐Worst	  Scaling:	  Comparison	  of	  various	  methods	  of	  analysis.	  BMC	  Med	  Res	  Methodol.	  2008;8(1):76.	  	  151.	  	   Flynn	  DTN.	  Using	  Conjoint	  Analysis	  and	  Choice	  Experiments	  to	  Estimate	  QALY	  Values.	  Pharmacoeconomics.	  2010;28(9):711–22.	  	  152.	  	   Haynes	  RB,	  Devereaux	  PJ,	  Guyatt	  GH.	  Clinical	  expertise	  in	  the	  era	  of	  evidence-­‐based	  medicine	  and	  patient	  choice.	  Evid	  Based	  Med.	  2002;7(2):36–8.	  	  153.	  	   Fuller	  R,	  Dudley	  N,	  Blacktop	  J.	  Avoidance	  hierarchies	  and	  preferences	  for	  anticoagulation—semi-­‐qualitative	  analysis	  of	  older	  patients’	  views	  about	  stroke	  prevention	  and	  the	  use	  of	  warfarin.	  Age	  Ageing.	  2004;33(6):608–11.	  	  154.	  	   Howitt	  A,	  Armstrong	  D.	  Implementing	  evidence	  based	  medicine	  in	  general	  practice:	  audit	  and	  qualitative	  study	  of	  antithrombotic	  treatment	  for	  atrial	  fibrillation.	  BMJ.	  1999;318(7194):1324–7.	  	  155.	  	   Holbrook	  A,	  Labiris	  R,	  Goldsmith	  CH,	  Ota	  K,	  Harb	  S,	  Sebaldt	  RJ.	  Influence	  of	  decision	  aids	  on	  patient	  preferences	  for	  anticoagulant	  therapy:	  a	  randomized	  trial.	  Can	  Med	  Assoc	  J.	  2007;176(11):1583–7.	  	  156.	  	   Holbrook	  A,	  Schulman	  S,	  Witt	  DM,	  Vandvik	  PO,	  Fish	  J,	  Kovacs	  MJ,	  et	  al.	  Evidence-­‐based	  management	  of	  anticoagulant	  therapy:	  Antithrombotic	  therapy	  and	  prevention	  of	  thrombosis,	  9th	  ed:	  american	  college	  of	  chest	  physicians	  evidence-­‐based	  clinical	  practice	  guidelines.	  CHEST.	  2012;141(2_suppl):e152S–e184S.	  	  157.	  	   Man-­‐Son-­‐Hing	  M,	  Laupacis	  A,	  O’Connor	  AM,	  et	  al.	  A	  patient	  decision	  aid	  regarding	  antithrombotic	  therapy	  for	  stroke	  prevention	  in	  atrial	  fibrillation:	  A	  randomized	  controlled	  trial.	  JAMA.	  1999;282(8):737–43.	  	  134	  	  158.	  	   Man-­‐Son-­‐Hing	  M,	  O’Connor	  AM,	  Drake	  E,	  Biggs	  J,	  Hum	  V,	  Laupacis	  A.	  The	  effect	  of	  qualitative	  vs.	  quantitative	  presentation	  of	  probability	  estimates	  on	  patient	  decision-­‐making:	  a	  randomized	  trial.	  Health	  Expect.	  2002;5(3):246–55.	  	  159.	  	   Sudlow	  M,	  Thomson	  R,	  Kenny	  RA,	  Rodgers	  H.	  A	  community	  survey	  of	  patients	  with	  atrial	  fibrillation:	  associated	  disabilities	  and	  treatment	  preferences.	  Br	  J	  Gen	  Pract.	  1998;48(436):1775–8.	  	  160.	  	   Thomson	  RG,	  Eccles	  MP,	  Steen	  IN,	  Greenaway	  J,	  Stobbart	  L,	  Murtagh	  MJ,	  et	  al.	  A	  patient	  decision	  aid	  to	  support	  shared	  decision-­‐making	  on	  anti-­‐thrombotic	  treatment	  of	  patients	  with	  atrial	  fibrillation:	  randomised	  controlled	  trial.	  Qual	  Saf	  Health	  Care.	  2007;16(3):216–23.	  	  161.	  	   Thomson	  R,	  Parkin	  D,	  Eccles	  M,	  Sudlow	  M,	  Robinson	  A.	  Decision	  analysis	  and	  guidelines	  for	  anticoagulant	  therapy	  to	  prevent	  stroke	  in	  patients	  with	  atrial	  fibrillation.	  The	  Lancet.	  2000;355(9208):956–62.	  	  162.	  	   Limone	  BL,	  Baker	  WL,	  Kluger	  J,	  Coleman	  CI.	  Novel	  Anticoagulants	  for	  Stroke	  Prevention	  in	  Atrial	  Fibrillation:	  A	  Systematic	  Review	  of	  Cost-­‐Effectiveness	  Models.	  PLoS	  ONE.	  2013;8(4):e62183.	  	  163.	  	   Torrance	  GW.	  Social	  preferences	  for	  health	  states:	  An	  empirical	  evaluation	  of	  three	  measurement	  techniques.	  Socioecon	  Plann	  Sci.	  1976;10(3):129–36.	  	  164.	  	   Jansen	  SJT,	  Stiggelbout	  AM,	  Wakker	  PP,	  Vlieland	  TPMV,	  Leer	  J-­‐WH,	  Nooy	  MA,	  et	  al.	  Patients’	  Utilities	  for	  Cancer	  Treatments	  A	  Study	  of	  the	  Chained	  Procedure	  for	  the	  Standard	  Gamble	  and	  Time	  Tradeoff.	  Med	  Decis	  Making.	  1998;18(4):391–9.	  	  165.	  	   Marley	  AAJ,	  Louviere	  JJ.	  Some	  probabilistic	  models	  of	  best,	  worst,	  and	  best–worst	  choices.	  J	  Math	  Psychol.	  2005;49(6):464–80.	  	  166.	  	   The	  MaxDiff	  System	  Technical	  Paper.	  Orem,	  Utah:	  Sawtooth	  Software	  Inc.;	  2013.	  Report	  No.:	  Version	  8.	  	  167.	  	   Orme	  B.	  Interpreting	  the	  Results	  of	  Conjoint	  Analysis.	  Getting	  Started	  with	  Conjoint	  Analysis:	  Strategies	  for	  Product	  Design	  and	  Pricing	  Research.	  2nd	  ed.	  Madison,	  Wis:	  Research	  Publishers	  LLC;	  2010.	  	  168.	  	   Vermunt	  JK,	  Magidson	  J.	  Technical	  Guide	  for	  Latent	  GOLD	  Choice	  4.0:	  Basic	  and	  Advanced.	  Belmont	  Massachusetts:	  Statistical	  Innovations	  Inc.;	  2005.	  	  169.	  	   Spinler	  SA,	  Shafir	  V.	  New	  Oral	  Anticoagulants	  for	  Atrial	  Fibrillation.	  Circulation.	  2012;126(1):133–7.	  	  135	  	  170.	  	   Lip	  G,	  Bassand	  J-­‐P,	  Fitzmaurice	  D,	  Goldhaber	  S,	  Goto	  S,	  Verheugt	  F,	  et	  al.	  Inappropriate	  utilization	  of	  anticoagulation	  in	  patients	  with	  atrial	  fibrillation:	  the	  global	  anticoagulant	  registry	  in	  the	  field	  (GARFIELD)	  registry.	  J	  Am	  Coll	  Cardiol.	  2012;59(13s1):E670–E670.	  	  171.	  	   Gibbs	  H,	  Kakkar	  A.	  One	  year	  Australian	  outcome	  results	  of	  cohort	  1	  of	  the	  Global	  Anticoagulant	  Registry	  in	  the	  Field	  (GARFIELD)	  study.	  Heart	  Lung	  Circ.	  2013;22(Supplement	  1):S221.	  	  172.	  	   Gibbs	  H,	  Kakkar	  A.	  Atrial	  fibrillation	  in	  Australia	  -­‐	  results	  from	  the	  GARFIELD	  study.	  Heart	  Lung	  Circ.	  2012;21(Supplement1):S243.	  	  173.	  	   American	  College	  of	  Cardiology.	  Registry	  data	  shows	  early	  patterns	  for	  new	  atrial	  fibrillation	  treatments	  [Internet].	  2012	  Aug.	  Available	  from:­‐Media/Media-­‐Center/News-­‐Releases/2012/08/PINN-­‐AF.aspx	  174.	  	   Okumura	  K,	  Inoue	  H,	  Yasaka	  M,	  Gonzalez	  JM,	  Hauber	  AB,	  Iwamoto	  K,	  et	  al.	  PCV101	  Japanese	  Patients	  and	  Physicians	  Preferences	  for	  Anticoagulants	  Use	  in	  Atrial	  Fibrillation	  -­‐	  Results	  From	  a	  Conjoint-­‐Analysis	  Study.	  Value	  Health.	  2012;15(7):A380.	  	  175.	  	   Levitan	  B,	  Yuan	  Z,	  González	  JM,	  Hauber	  AB,	  Lees	  M,	  Piccini	  JP,	  et	  al.	  RM1	  -­‐	  Patient	  And	  Physician	  Preferences	  In	  The	  United	  States	  For	  Benefits	  And	  Risks	  Of	  Anticoagulant	  Use	  In	  Atrial	  Fibrillation	  –	  Results	  From	  A	  Conjoint-­‐Analysis	  Study.	  Value	  Health.	  2013;16(3):A11.	  	  176.	  	   Umscheid	  CA.	  Should	  Guidelines	  Incorporate	  Evidence	  on	  Patient	  Preferences?	  J	  Gen	  Intern	  Med.	  2009;24(8):988–90.	  	  177.	  	   Marti	  J.	  A	  best–worst	  scaling	  survey	  of	  adolescents’	  level	  of	  concern	  for	  health	  and	  non-­‐health	  consequences	  of	  smoking.	  Soc	  Sci	  Med.	  2012;75(1):87–97.	  	  	  136	  	  Appendix	  A:	  Patient	  Questionnaire	  	    Next    0%     100%  137	  	  	  Principal  Investigator:Dr.  Larry  Lynd,  Associate  Professor,  Faculty  of  Pharmaceutical  Sciences,  University  of  British  Columbia  (604)  827-­3397Co-­Investigators:Dr.  Karin  Humphries,  Associate  Professor,  Faculty  of  Medicine,  University  of  British  ColumbiaDr.  Robert  Boone,  Assistant  Professor,  Faculty  of  Medicine,  Department  of  Cardiology,  University  of  British  ColumbiaDr.  Ross  Tsuyuki,  Professor,  Faculty  of  Medicine,  Department  of  Cardiology,  University  of  AlbertaDr.  Carlo  Marra,  Professor,  Faculty  of  Pharmaceutical  Sciences,  University  of  British  ColumbiaDr.  I  fan  Kuo,  Postdoctoral  Fellow,  Faculty  of  Pharmaceutical  Sciences,  University  of  British  ColumbiaSponsors:Canadian  Institutes  of  Health  Research  (CIHR)Heart  and  Stroke  Foundation  of  Canada  (HSFC)  Next    0%     100%  138	  	  	  INVITATIONYou  are  being  invited  to  take  part  in  this  research  study  because  you  have  a  medical  condition  known  as  atrial  fibrillation  (AF),or  a  type  of  irregular  heart  rhythm.  People  with  this  irregular  heart  rhythm  are  more  likely  to  experience  stroke,  a  debilitatingevent  that  could  cause  death  or  greatly  decrease  a  person’s  quality  of  life.  To  reduce  the  risk  of  stroke,  people  with  AF  are  usuallyprescribed  blood  thinners,  or  oral  anticoagulants,  that  must  be  taken  for  the  rest  of  their  life.Background  Information:Currently,  there  are  a  few  available  blood  thinners  that  are  effective  for  reducing  the  risk  of  stroke.  However,  each  blood  thinnerhas  slightly  different  characteristics,  and  the  most  appropriate  drug  therapy  for  a  person  with  AF  is  not  always  clear.  For  instance,some  blood  thinners  might  be  more  likely  to  cause  bleeding,  and  some  blood  thinners  might  require  the  individual  to  go  to  alaboratory  regularly  to  have  blood  tests  done.  The  purpose  of  our  study  is  to  find  out  which  of  those  characteristics  are  importantto  people  with  AF.  For  example,  we  are  interested  in  finding  out  what  risk  of  a  bleed  you  might  be  willing  to  accept  in  order  toprevent  a  stroke.  In  other  words,  we  would  like  to  evaluate  patients’  preferences  for  a  blood  thinner  therapy.To  achieve  our  study  goal,  we  will  be  asking  you  to  complete  a  survey  that  should  take  approximately  20  minutes.  Details  of  thequestionnaires  will  be  further  explained  in  the  following  section.  Please  click  'next'  to  proceed.  Next    0%     100%  139	  	  	  Study  Procedures:Once  you  have  agreed  to  participate  in  this  study,  you  will  be  directed  to  take  the  study  survey  online.  We  will  begin  the  survey  byasking  you  a  few  questions  about  your  background  information,  including  age,  gender,  socioeconomic  status,  and  prior  experiencewith  blood  thinners.  All  the  information  you  provide  us  with  will  be  used  strictly  for  research  purpose  and  will  not  be  shared  (seefollowing  page  on  confidentiality).The  survey  will  then  proceed  to  present  you  with  a  few  hypothetical  scenarios,  and  ask  you  to  trade  off  life  years  remaining  forperfect  health.  This  will  be  explained  in  further  detail  later  on.In  the  last  part  of  the  survey,  we  will  ask  you  to  make  choices  through  a  series  of  16  questions.  Each  question  will  present  youwith  a  set  of  characteristics  associated  with  the  blood  thinners.  Specifically,  the  four  characteristics  we  are  interested  in  are:annual  chance  of  stroke,  annual  chance  of  bleeding,  the  need  of  long-­term  blood  monitoring,  and  the  availability  of  a  reversalagent  in  case  of  a  life-­threatening  bleed.  Each  of  these  characteristics  will  also  vary  in  degrees  of  descriptors.  In  each  question,we  would  like  you  to  choose  what  you  think  the  worst  characteristic  is,  and  what  the  best  characteristic  is.Please  remember  that  these  treatment  options  are  hypothetical  and  may  not  exist  in  real  life.  Therefore,  the  option  you  end  upchoosing  should  be  based  on  the  treatment  characteristics  presented  in  the  question  and  not  on  the  treatment  you  might  becurrently  taking  at  home.Potential  Risks:There  is  no  health  risk  involved  with  your  participation  in  the  study  as  it  is  entirely  survey-­based.Potential  Benefits:There  are  no  direct  or  immediate  benefits  to  you  as  a  participant.  However,  your  participation  in  the  study  may  contribute  tobetter  treatment  choices  and  therefore  improved  health  outcomes  of  other  patients  in  the  future.Confidentiality:Your  confidentiality  will  be  respected.  No  information  that  discloses  your  identity  will  be  released  or  published  without  yourconsent.  All  files  will  not  be  identifiable  by  name  or  other  personal  information.  No  records  that  identify  you  by  name  will  leavethe  principal  investigator’s  office.  Materials  related  to  this  research  for  publication  or  reports  will  not  contain  any  participantinformation.  Your  rights  to  privacy  are  also  protected  by  the  Freedom  of  Information  and  Protection  of  Privacy  Act  of  BritishColumbia.  This  act  lays  down  rules  for  the  collection,  protection,  and  retention  of  your  personal  information  by  public  bodies,  suchas  the  University  of  British  Columbia  and  its  affiliated  teaching  hospitals.140	  	  	  	  Remuneration/Compensation:As  a  participant,  you  will  receive  a  $5  gift  card  as  a  small  token  of  our  appreciation.Contact  for  information  about  the  study:If  you  have  any  questions  or  desire  further  information  with  respect  to  this  study,  you  may  contact  the  study  coordinator,  Dr.  Ifan  Kuo  at  604-­827-­3674,  or  the  Principal  Investigator,  Dr.  Larry  Lynd  at  604-­827-­3397.  Contact  for  concerns  about  the  rights  of  research  subjects:If  you  have  any  concerns  about  your  treatment  or  rights  as  a  research  subject,  you  may  contact  the  Research  SubjectInformation  Line  in  the  UBC  Office  of  Research  Services  at  604-­822-­8598  or  if  long  distance  email  to  Consent  to  Participate:Your  participation  is  entirely  voluntary,  so  it  is  up  to  you  to  decide  whether  or  not  to  take  part  in  this  study.  By  clicking  "next"  anddoing  the  survey,  you  are  giving  consent.You  have  10  days  to  decide  whether  or  not  you  consent  to  participating  in  this  study.  After  10  days,  your  login  id  will  be  invalid,and  at  which  point,  if  you  wish  to  participate  in  this  study,  you  will  need  to  contact  the  study  coordinator,  Dr.  I  fan  Kuo,  at  604-­827-­3674  or  email:  to  get  a  new  id  and  password.  If  you  do  not  wish  to  continue  with  the  survey,  you  canwithdraw  at  any  time  and  there  will  be  no  penalty  or  loss  of  benefit  to  which  you  are  entitled,  and  your  future  medical  care  willnot  be  affected  in  any  way.Click  "next"  if  you  wish  to  give  consent  and  proceed  with  the  survey.  Next    0%     100%  141	  	  	  	  Thank  you  for  your  consent.  Please  click  "next"  again  to  begin  the  survey.    Next    0%     100%  142	  	  	  Section  I:  DemographicsIn  this  section,  we  would  like  to  get  some  information  about  your  background.1.  What  is  your  Gender?  MaleFemale2.  Please  enter  your  date  of  birthday    month    year3.  What  is  the  distance  from  your  residence  to  the  closest  hospital  with  an  emergency  department?  less  than  50  km50  km  or  more4.  What  is  your  annual  household  income?  <$20,000$20,000-­$40,000$40,000-­$60,000$60,000-­$80,000>$80,0005.  What  is  the  highest  level  of  education  you  have  completed?  No  formal  educationHigh  school  diplomaPost-­secondary  degree  (e.g.  bachelor,  college  diploma,  trades  certificate)Graduate  (e.g.  MSc/PhD)Professional  Degree  (e.g.  MD,  DVM)6.  What  is  your  current  marital  status?143	  	  	    SingleMarried,  or  common-­lawWidowedSeparated7.  Do  you  currently  have  any  dependent  children  under  the  age  of  19?  YesNo  Next    0%     100%  144	  	  	  	  8.  How  long  ago  were  you  diagnosed  with  atrial  fibrillation?  less  than  a  year1  to  5  years6  to  10  yearsmore  than  10  yearsI  do  not  know9.  Are  you  currently  on,  or  have  you  taken  blood  thinners  in  the  past?  (Examples  of  blood  thinners  include:  warfarin  orCoumadin®,  aspirin  or  ASA,  clopidogrel  or  Plavix®,  dabigatran  or  Pradax®,  rivaroxaban  or  Xarelto®)  YesNo  Next    0%     100%  145	  	  	  	  	  14.  Have  you  ever  had  a  stroke?  YesNo  Next    0%     100%  146	  	  	  	  Section  II:  Time  Trade-­off  (TTO)  QuestionnaireTime   trade-­off   is   a   method   we   use   to   determine   how   people   value   different   health   outcomes.   In   general,   we   ask   peoplehypothetically,  how  much  of  their  life  would  they  be  willing  to  give  up  to  avoid  experiencing  something  bad,  for  example,  having  astroke.  In  this  study,  we  would  like  to  ask  you  hypothetically,  how  many  years  of  your  remaining  lifespan  would  you  be  willing  totrade  off  to  prevent  specific  medical  events.  This  allows  researchers,  clinicians  and  policy  makers  to  understand  how  people  valuequality   of   life   associated  with   a  medical   condition.   In   other  words,   the   impact   a   certain   illness   has   on   individuals   and   to  whatextent   they   would   like   to   avoid   it.   The   following   is   a   sample   question   to   guide   you   through   the   format   of   this   part   of   thequestionnaire.  Next    0%     100%  147	  	  	  The following scenario is intended to help you become familiar with the format of questions in this section and is NOT an actualquestion of the questionnaire. The scenario uses rheumatoid arthritis (RA) as an example. Do not worry if you do not know whatthis condition is about.SAMPLE QuestionRheumatoid arthritis (RA) is a disease that causes pain and swelling in the joints. People with RA often start with a mild form ofthe disease with sore joints, which can progress to severe disease marked by chronic pain, immobility, and decreased quality oflife. RA is a chronic illness and people with RA often go through periods of flare-ups. There is no treatment that can cure RA butthere are drugs people can take to reduce the number or flare-ups.Now, imagine that you have RA and the chronic joint pain has become more bothersome such that you can no longer climb thestairs, or bend down to pick up grocery bags without pain. You used to play a lot of golf with your friends but now you can nolonger join them because of the pain. Imagine that you will live for another 10 years with the disease after which you willexperience a painless death.How many years of those remaining 10 years would you be willing to give up to live in perfect health for a shorter amount of time,following which you will experience a painless death? For instance, let's say you are willing to trade off 3 years. This would implythat you are willing to live for 7 years in perfect health rather than 10 years with the pain and activity limitations associated withhaving RA. Likewise, if you are willing to trade off 5 years, this would imply that you would accept living 5 years in perfect healthinstead of 10 years with RA.Using the slider below, please move the marker along the ruler to indicate how many years of living with RA you would be willingto trade off to live in a shorter life in perfect health.I would be willing to give up0  years of the remaining 10 years with RAto live for 10 years in perfect health148	  	  	  	  	  The   following   few  questions  will   be  very   similiar   in   format   to   the   sample  question  you  have   just   answered.  Click  next   to  beginthose  questions.  Next    0%     100%  149	  	  	  	  	  StrokeUsing  this  time-­trade  off  question,  we  would  like  to  determine  how  much  of  your  remaining  life  you  would  hypothetically  be  willingto  give  up  to  avoid  having  a  stroke.  A  stroke  is  a  serious  medical  event  that  requires  urgent  medical  care.  A  stroke  can  occurwhen  blood  stops  flowing  to  a  specific  part  of  the  brain  as  a  result  of  either  a  blood  clot  (which  blocks  the  blood  flow),  or  aruptured  blood  vessel.  As  a  result,  brain  cells  in  the  area  may  die,  leading  to  permanent  damage.  The  brain  damage  could  causethe  person  to  lose,  or  have  impaired,  eyesight,  speech,  body  coordination  and  balance.  The  person  will  often  experience  severeheadaches  and  memory  loss.  In  serious  cases,  a  stroke  will  cause  coma.  Stroke  can  range  in  severity.  Some  people  will  justexperience  what  is  known  as  a  'mini-­stroke'  or  a  'transient  ischemic  attack  (TIA)'  and  recover  fully  from  any  symptoms  within  24hours,  while  some  people  will  become  permanently  disabled  where  they  can  no  longer  take  care  of  themselves.  In  the  mostserious  cases,  people  will  die  immediately  following  the  stroke.  People  who  survive  with  complications  of  strokes  generally  have  amuch  lower  quality  of  life.  Next    0%     100%  150	  	  	  	  151	  	  	  	  152	  	  	  153	  	  	  	  154	  	  	  	  155	  	  	  	  156	  	  	  157	  	  	  	  158	  	  	  	  159	  	  	  	  160	  	  	  	  	  	  	  	  161	  	  	  162	  	  	  163	  	  	  164	  	  	  	  	  165	  	  	  	  	  166	  	  	  	  167	  	  	  168	  	  	  169	  	  	  170	  	  	  171	  	  	  172	  	  	  173	  	  	  174	  	  	  175	  	  	  176	  	  	  177	  	  	  	  178	  	  	  	  	  179	  	  Appendix	  B:	  Physician	  Questionnaire	  	  	  180	  	  	  	  	  	  181	  	  	  	  	  	  	  182	  	  	  	  	  183	  	  	  	  	  184	  	  	  	  	  185	  	  	  186	  	  	  	  	  	  187	  	  188	  	  	  	  	  	  	  	  	  189	  	  	  	  	  190	  	  	  	  	  191	  	  	  	  	  	  	  	  192	  	  	  	  	  193	  	  	  	  	  194	  	  	  	  	  195	  	  	  	  	  196	  	  	  	  	  197	  	  	  	  	  198	  	  	  	  	  199	  	  	  	  200	  	  	  	  	  201	  	  	  	  	  202	  	  	  	  	  203	  	  	  	  204	  	  	  	  	  205	  	  	  	  206	  	  	  	  207	  	  	  	  208	  	  	  	  209	  	  	  	  	  210	  	  	  	  	  	  


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items