Transcript Systematic Research Literature Reviewing

Social Epidemiologic Methods
in International Population
Health and Health Services
Research
A Research Agenda Using Cancer
Care as a Sentinel Indicator:
By Kevin M. Gorey
Kevin M. Gorey
Kevin is a social epidemiologist and
social welfare researcher interested
in advancing understandings about
how health care policies affect
health. He is particularly interested
in the impacts of various underand uninsured statuses in the US.
His web page is:
www.uwindsor.ca/gorey
Cancer Survival in Canadian
and United States
Metropolitan Areas: A Series
of Studies
Between-Country Effect
Modification by Socioeconomic
Status
(Health Insurance)
Research Team and Reports
Kevin Gorey, University of Windsor
Eric Holowaty & Gordon Fehringer, CCO
Erich Kliewer, Cancer Care Manitoba
Ethan Laukkanen, WRCC and Colleagues
Study series reports:
Am J Public Health 1997 & 2000
Can J Public Health 1998; Milbank Q 1999
J Public Health Med 2000
J Health Care Poor Underserved 2003
Ann Epidemiol 2003
Introduction
Mid-1980s to Mid-1990s:
Historical and Theoretical
Contexts
Historical Context
-
Canada: Universal single payer
-
US: Multi-tiered—uninsured and
underinsured, Medicaid, Medicare,
continuum of private coverages
-
Time of great systemic changes
- Managed care proliferation (US)
- Federal-provincial shift (Canada)
Politics Versus Science
-
Political debates tend to mythologize
anecdotal outcomes.
-
Rhetoric often not substantiated
(e.g., 2 Manitoba studies)
- Waits for 10 surgical procedures stable
or decreased 5 yrs post-downsizing
- Access to surgery actually increased
after hospital downsizing (maintaining
quality [mortality, readmissions])
Cancer Survival is a Sentinel
Health Care Outcome
-
Relatively common over the life course
Diverse constellation of diseases
Many with good prognoses and high
quality of survivable life
Diverse screens (including primary care)
and treatments exist and matter
Timely access, referral and follow-up
matter
Theoretical Context:
Systematic Literature Review
-
-
-
In the US, ethnicity and SES are strongly
associated with health insurance
statuses (odds ratios [OR] 2.0 to 15.0).
All are also strongly associated with
cancer screens, stages at diagnosis and
access to treatments (ORs 2.0 to 5.0).
Such Canadian associations tend to be
attenuated or nonexistent. For example:
- US SES-cancer survival OR = 1.56
- Canadian OR = 1.04 (NS) to 1.18
SES: A Key Effect Modifier?
Therefore, any Canada-US cancer outcome
study that does not incorporate SES is
unlikely to observe the truth.
-
SES is so intimately connected with
health in North America that it must be
incorporated into all such studies.
-
If an interaction exists, interpretations of
main effects alone can be misleading.
SES: An Effect Modifier? E.G.
-
One previous study of Canada-US
cancer survival (GAO, 1994)
- Found no between-country differences
- But, did not account for SES
-
We have observed a substantially
different picture within SES strata.
- Consistent Canadian advantages
within the lowest SES strata
A Country By SES Interaction
Hypothesis Guided Our Series
Relatively poor Canadian cancer patients
(better insured) would enjoy advantaged
survival over their similarly poor
counterparts in the United States.
-
We think this a better guide to policyinteresting and important research
questions in North America than those
provided by main effect country-based
hypotheses.
Methods
A Focused Series of Cancer
Survival Comparisons Among
Relatively Poor Residents of
Canadian and American
Metropolitan Areas
Comparative Series Overview
Toronto, Ontario vs Detroit, Michigan
An ecological exemplar
Toronto vs San Francisco, Seattle, Hartford
Adjustment for absolute income
Toronto vs Honolulu, HI
Health insurance hypothesis test
Winnipeg, Manitoba vs Des Moines, Iowa
Replicate among smaller cities
Comparisons of Subsamples < 65 yoa
Health insurance hypothesis test
Sampling—Persons/Cancer
Patients
-
-
Ontario and Manitoba Registries, SEER
First, primary invasive cancer cases
MC, not DC or autopsy only
With minimum 5 years follow-up
Began 15 most common cancers
- Since focused on most significant
Estimated case ascertainments, MC, and
follow-ups all > 95% (DCO/Autopsy < 1%)
- Even better among the most public
health-significant cancer types
Honolulu, Breast Cancer, 1986-1990
SES
MC%
DCO/Autopsy%
High
Low
100.0
100.0
100.0
100.0
100.0
100.0
100.0
98.5
97.9
98.9
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.7
0.0
0.0
Sampling—Places: Rationales
For Metropolitan Sampling
-
-
-
Maximize internal validity
- Higher: MC, follow-up, geocoding rates
- Lower: DCO or autopsy only
Maximize external validity
- Vast majority of NAs urban residents
- 1 of 3 Ontarians and 1 of 7 Canadians
reside in Toronto
Control for service availability
Sampling—Places: Ecological
Measures of SES
Neighborhoods
No NA registries coded personal SES.
- Census tracts joined cases at diagnosis
to income data (US Census, Stats Can)
- Neighborhood prevalence poor
- Theory, insurance, practical sig.
- Poverty (US), low income (Canada)
- Both household income-based and tied
to the consumer price index
- Though Canadian criterion more liberal
- Used to form relative SES quantiles
Comparison of SES Quintiles: 1990/91, US$
Winnipeg
Des Moines
SES
Mdn $
Mdn $
High
Low
$47,090
39,110
32,265
26,043
17,500
$44,050
36,370
30,165
26,890
19,570
Lowest US SES quintile: 20% poor, another
45% near poor; estimated (vs highest)
uninsured PR = 10.0, underinsured PR = 15.0
Results
Female Breast Cancer—5-Year
Survival—As Exemplar Throughout
SRRs With 95% CIs, 1984 to 1994
SES Toronto
Detroit
High
Low
1.00
…
1.00 (0.94,1.06)
0.98 (0.93,1.04)
1.00
…
0.94 (0.88,1.01)
0.80 (0.75,0.85)
No significant between-country differences
in the middle or high income areas
Low income areas: Between-country
SRR = 1.30 (1.23,1.38), Canadian patients
advantaged
SRRs With 95% CIs, 1986 to 1996
SES Toronto
Honolulu
High
Low
1.00
1.01
1.01
1.03
1.04
0.97
1.00
1.03
1.05
1.02
…
(0.93,1.10)
(0.95,1.08)
(0.96,1.11)
(0.97,1.12)
(0.90,1.04)
(0.81,1.24)
(0.95,1.11)
(0.98,1.13)
(0.95,1.10)
1.00
0.94
0.93
0.97
0.93
0.80
0.90
0.97
0.91
0.78
…
(0.82,1.07)
(0.81,1.06)
(0.86,1.09)
(0.81,1.07)
(0.69,0.93)
(0.79,1.02)
(0.87,1.09)
(0.80, 1.04)
(0.67,0.91)
Toronto-Honolulu BetweenCountry Survival Outcomes
The only significant decile difference was
for the lowest income area:
SRR = 1.20 (1.06, 1.36)
Canadian patients advantaged
Among those < 65 yoa:
SRR = 1.28 (1.07,1.53)
Discussion
The Screened/Developed
Health Insurance Hypothesis
Versus Alternative Explanations
Summary: Health Insurance
-
-
Consistent SES-cancer survival
associations in US, but not Canada
Consistent country-SES interactions
- Canada advantage lowest SES strata
- Particularly among those < 65 yoa
Consistency of pattern across diverse
contexts—people and places—points
toward a pervasive systemic effect
- 285 of 319 between-country
comparisons were in support of the
health insurance hypothesis
Alt1—Income Gap or Inequality
Larger in the United States?
-
For some of our studies, the
economic divide is actually larger in
the Canadian sample.
- E.g., Winnipeg vs Des Moines
Alt2—Ethnic or Cultural
Explanations?
-
Similar pattern of findings observed
among various ethnic mixes
-
North American studies of race/ethnicity
and cancer screening have implicated
knowledge (education), rather than race,
per se.
-
Consistent indictment of America:
Inequitable distribution of key social
resources—education and health care
Alt3—Lifestyle Factors (LS): Exercise, Diet,
BMI, Tobacco and Alcohol Consumption?
-
Associations with cancer survival tend
to be extremely small
- Larger associations with incidence
Survival findings consistent across
cancers with diverse component causes
-
-
-
Some LS factors very sig., others not
Income is associated with lifestyle in
both countries, but no income-survival
gradients were observed in Canada
Little to no Canada-US LS prevalence
differences (2%) have been observed
Alt4—Different Case Mixes by
Stage of Disease at Diagnosis?
-
Stage differences may account for some,
but probably not all of the betweencountry survival differences.
- In within-US stage-adjusted analyses,
treatment differences still account for
roughly 50% of survival variabilities.
Alt5—Cancer Registry Death Clearance?
National (US) vs Provincial (Canada)
-
Over the life of these studied cohorts,
only 1-3% of Toronto residents moved
out-of-province.
- Likely fewer chronically ill moved
-
Ontario Cancer Registry comparisons of
national and provincial death clearances
found inconsequential differences.
Alt6—Competing Causes of Death
(Observed vs Relative Survival)?
-
Life expectancy in Honolulu among
both women and men is close to 3
years greater than in Toronto
-
Therefore, our between-country
SRRs (Canadian advantage) may
actually underestimate the truth
Alt7—Lead Time Bias?
-
Our findings were fairly consistent
across different cancers probably with
various pre-clinical phase lengths.
-
A systematic review of 87 studies (with
adjustment for lead-time) observed stage
and treatment effects (Richards et al.,
1999, Lancet)
Alt8—Ecological Fallacy?
-
Even if it were merely an area effect, the
consistently observed residencesurvival association in the US, but not in
Canada would still be instructive.
-
The compositional measure (% poor and
near poor in neighborhoods) is well
known to be intimately associated with
under-and uninsured statuses in the US.
Future Research Needs
Health Insurance Hypothesis
Developed and Screened With An
Ecological—Income—Proxy:
More Definitive Testing Needed
Central Research Needs
-
Study more recent retrospective
and prospective cohorts
Perform stage-stratified analyses
Incorporate treatment variables
Extend generalizability to smaller
urban and rural-remote places
Develop construct validity of
ecological SES measures in Canada
Our Research Agenda Over
The Next 5 Years
Endeavoring to Filling Some of This
Field’s Central Knowledge Gaps
Social, Prognostic & Therapeutic
Factors Associated With Cancer
Survival in Canada and the US
Health Care Access and
Effectiveness in Diverse Urban
and Rural Contexts, 1985 to 2010
Research Team Co-Investigators
Kevin Gorey (PI) & Emma Bartfay
(Epidemiology)
Karen Fung (Biostatistics)
Isaac Luginaah (Geography)
Frances Wright (Surgical Oncology)
Caroline Hamm & Sindu Kanjeekal
(Medical Oncology)
Eric Holowaty & William Wright
(Cancer Surveillance & Registration)
To Address Identified Research
Needs, It Will:
-
-
Study more recent retrospective and
prospective cohorts
Perform stage-stratified analyses
Incorporate treatment variables
Extend generalizability to smaller urban
and rural-remote places
Develop construct & predictive validities
of ecological SES measures in Canada
Cohort Design
Incident cohorts: 1985-1990 & 1995-2000
Followed until:
2000
2010
Cox models over 1-, 3-, 5- to 10-years
In Canada and the US
During a policy-interesting period
- Federal-provincial shift in Canada
- For-profit managed care proliferation &
prevalent increases uninsured in US
Staged Analyses
No Canadian cancer registry routinely codes stage
of disease at diagnosis.
- Thus, no previous study in this field has been
able to account for case-mix.
Stage will be abstracted for this study’s samples.
Allowing for:
- More comparable between-country
comparisons
- Examination of the relative weightiness of
pre- (affect later diagnosis) and postdiagnostic (affect lack of access to best
treatments and follow-up) social forces
Incorporation of Treatments
No Canadian cancer registry routinely codes initial
treatments.
- Thus, no previous study in this field has been
able to account for them in survival analyses.
Detailed treatment variables will be abstracted for this
study’s samples.
- Surgery, radiation, chemotherapy and others
- Initial course and follow-up
- Type, dose, delays, timings/sequence
between various therapies
Extending Generalizability:
Contexualizing Knowledge
Systematic Replications in:
Large cities
Small cities
Ontario
California
Toronto San Fran/Oakland
Windsor
Salinas
Rural/remote areas of Ontario & California
1,060 breast and colon cancer cases for each
incident cohort in each type of place
Ecological Measurement Validity
Ontarian and Californian cancer cases will be
joined via their residential census tracts to
the following data:
- Income (poverty prevalence) and
- Physician supplies (count/10,000 pop)
- Primary care and specialists
This will provide opportunities to better
understand the meanings of such ecological
measures, particularly in Canada, where little
is yet known about them.
Hypotheses Related to Survival
1.
Significant country by SES interaction
(Canadian advantage low-income only)
1a. Advantage significantly increased over time
2. SES-survival significant in US (not in Canada)
2a. Age by SES interaction (Medicare advantage)
2b. US gradient significantly increased over time
3.
Physician supplies-survival associations
significant in both Canada & US (for both
primary care and specialists supplies)