Colon cancer care and survival: income and insurance are more predictive in the USA, community primary care physician supply more so in Canada

We oversampled people who lived in poverty and were diagnosed with colon cancer in Ontario and California between 1996 and 2000 and followed them until enactment of the ACA in 2010. Cohorts were 4,574 people in California and 2,060 people in Ontario with the most treatable, non-metastasized colon cancers. Those diagnosed with stage IV disease according to the American Joint Committee on Cancer were excluded [34]. Ontario and California are among places with the most comprehensive and valid colon cancer registries in the world [35‐37]. National death indexes were used by both registries.

Since neither Ontario nor California registries collect income data, we joined them to neighborhood data via Statistics Canada (2001) and USA (2000) census tracts [38, 39]. A third of participants in California were randomly selected from high poverty neighborhoods where 30 % or more of the people were poor. Others were randomly selected from middle (5-29 % poor) and low poverty neighborhoods (<5 % poor). The Ontario sample similarly overrepresented the lowest income neighborhoods and then randomly selected from middle and high income neighborhoods. Lowest income neighborhoods in the two countries were quite similar. Median annual household incomes in US dollars were only slightly lower in California ($23,125) than Ontario ($24,175) [40]. Primary health insurers were defined in California as private (included Medicare with private supplemental coverage), public (Medicare alone or Medicaid) or none.

We then identified communities with relatively low to high health care endowments characterized by physician supplies. To do this we joined participants to county-level active physician data via the American Medical Association and Canadian Institute for Health Information databases (2000–2001) [41‐44]. PCPs were those who reported their specialty area as general or family practice. Consistent with validated definitions and practice patterns, general internists in the USA and emergency family medicine physicians in Canada were also included as PCPs. Physicians reporting the majority of their time in specialized practice or who were board certified in that specialty were so defined [45, 46]. Threshold effects, below which the participants were less likely to survive, were identified by exploring 0.5 physician increments [47]: < 7.5 PCPs per 10,000 or 2 gastroenterologists (GE) per 100,000 community inhabitants. Thresholds were not observed for the supplies of overall specialist physicians or for other specialists often involved in colon cancer care such as general surgeons or oncologists.

According to the Starfield-Gorey theory poverty is a strong predictor of early death in the USA, while adequate PCP supply better predicts long term survival in Canada. These hypotheses involve interaction effects: poverty-by-country and PCP supply-by-country. We used a logistic regression model to test them [47]. We estimated the strength of each predictor-10 year survival relationship (e.g., increased survival odds associated with PCP supply increases) adjusting for the effects of other predictors as well as potential confounds. We accounted for stage as well as initial surgery and chemotherapy which were available in California, but that we had to retrospectively collect from health records across Ontario at considerable cost. Therefore we oversampled California participants to serve as multiple “controls.” A strategy to efficiently increase analytic power, it meant that we could detect rate differences as small as 5 % (2-tailed α = 0.05; power_{1 - β} = 0.80) [48]. All variables, except three, had less than 1 % missing data: number of regional lymph nodes examined (2.4 %), wait time for treatment (5.4 %) and tumor grade (8.9 %). Odds ratios (OR) and confidence intervals (CI) were estimated from regression statistics.

We also described significant interactions by comparing within and between-country survival rates across poverty and physician supply strata. We directly adjusted rates using a standard population with the age and stage characteristics of this study’s combined Ontario-California sample. It was used to calculate age and stage-standardized rates for the Ontario and California samples. All rates were calculated per 100 participants and reported as percentages. We used rate ratios (RR) for comparisons with 95 % CIs derived from the Mantel-Haenszel χ ² test. Further methodological details have been published [13, 16, 49‐52].

Interactions are depicted in Table 2. As hypothesized, the inverse poverty-survival association was relatively strong in California (RR = 0.79, 95 % CI 0.73, 0.86), but not significant in Ontario. Consequently, among people who lived in poverty, those in Ontario were 15 percent more likely to survive for 10 years than were those in California (RR = 1.15, 95 % CI 1.02, 1.30). The Canadians were similarly advantaged when compared with their uninsured or publicly-insured counterparts in the USA (RR = 1.18, 95 % CI 1.09, 1.28). Again as hypothesized, the direct PCP-survival association was strong in Ontario (RR = 1.32, 95 % CI 1.19, 1.46), but modest in California (RR = 1.11, 95 % CI 1.00, 1.23). Consequently, in adequately supplied communities people in Ontario were 13 % more likely to survive than were their counterparts in California (RR = 1.13, 95 % CI 1.00, 1.28). High poverty neighborhoods in Ontario also had (mean = 8.6, SD = 2.4), on average, more than 2 more PCPs per 10,000 residents than did high poverty neighborhoods in California (mean = 6.4, SD = 1.8); F (1, 2186) = 595.77, p < .001.

The GE supply by country interaction is depicted in the bottom of Table 2. The direct GE supply-survival association was larger in Ontario (RR = 1.56, 95 % CI 1.28, 1.91) than in California (RR = 1.15, 95 % CI 1.03, 1.28). Among those who lived in adequately supplied communities with 2 or more GEs for every 100,000 inhabitants, Ontarians were much more likely to survive than Californians (RR = 1.40, 95 % CI 1.16, 1.70). Such communities in Ontario also had (mean = 9.0, SD = 1.9), on average, nearly 2 more PCPs per 10,000 inhabitants than similar communities in California (mean = 7.2, SD = 2.0); F (1, 4028) = 92.40, p < .001.

The primary care-based Canadian advantage on survival was largely mediated by access to a more effective initial course of treatment, typically provided during the first six months to a year after diagnosis. But primary care seemed to continue to be beneficial in both countries throughout this study’s remaining nine years of follow-up. This is consistent with studies in Canada and the USA that have observed associations between PCP visits and high quality follow-up care of cancer survivors [63‐66]. The cancer registries we studied, however, did not provide data on follow-up care so we were not able to study the effects of poverty, insurance and physician supplies on longer term, follow-up care. Such studies are needed to advance understandings of the entire colon cancer care process and to facilitate evidence-based decision making and policy planning in the post-ACA era in both countries.

Our research group has systematically replicated and affirmed the Starfield-Gorey theory for Canada-USA comparisons of colon and breast cancer care. We invite independent research teams to test the theory across other health care processes and outcomes. Such knowledge will help policy makers to empirically plan future PCP and specialist physician supplies, enabling the highest quality health care. We continue to study cancer care among those who live in poverty, now prospectively in diverse, including so-called “red” and “blue” states.

Our findings may not be generalizable to all Americans and Canadians. However, because one in three Canadians lives in Ontario and one in ten Americans lives in California, we think they have ample external validity [67, 68]. Admittedly, we purposefully oversampled those who lived in poverty so our findings are most representative of their experiences. Furthermore, California’s expanded Medicaid program is more liberal than most states’ so estimates of health care inequities among the poor there are probably underestimates of the nation’s [69‐71].

Some may question our choice of overall, rather than cancer-specific survival. Our rationales follow. First, cancer seemed the underlying cause of death among the vast majority of participants in Ontario and California. However, although vital status and length of survival are highly accurate in North American cancer registries, the underlying cause of death is not [37, 72‐74]. We avoided bias that would intrude through such, often substantial, death certificate error by focusing on overall survival. Second, the underlying cause of deaths not identified as cancer deaths can still be associated with lack of treatment or with their complications [75]. Furthermore, overall survival is the gold standard in most trials and the principle that guides most oncologic practice [76, 77]. Recall that participants were diagnosed with non-metastasized disease, imminently treatable in most cases. It seems likely that the guiding principle among most participants and their physicians was to maximally extend life. Therefore, we think overall survival is the most policy-telling, practical indicator of clinical and human significance. Finally, because our analysis had a hierarchical structure, with individuals nested within neighborhoods and counties, some readers may wonder why we did not use a multi-level regression model. Each key neighborhood and county-based variable was constituted by only two clusters. And germinal literature concurs that multi-level models should not be estimated with data consisting of fewer than 10 clusters. There has been some debate about the exact criterion of “too few,” but sensitivity analyses strongly suggested to us that two would definitely be too few [78, 79]. Other potential limitations have been discussed [13, 16, 52].

This study was reviewed and cleared by the University of Windsor’s research ethics board.