The Role of Access to a Regular Primary Care Physician in Mediating Immigration-Based Disparities in Colorectal Screening: Application of Multiple Mediation Methods

In Canada, as in other developed nations (1, 2), immigrants are less likely than Canadian-born residents to have ever been screened for colorectal cancer (3, 4), which is currently the third most common cause of cancer death in the country (5). The gap in lifetime screening between those born in Canada and those born abroad is of approximately 10% (6). Since having never been screened is associated with later-stage at diagnosis and higher levels of colorectal cancer mortality (7), immigrants are likely to bear a disproportionate amount of the burden of colorectal cancer, in part because they are under-screened. Immigration-based screening disparities beg two questions: how do these disparities come to exist; and what interventions can be leveraged to reduce them?

Known social determinants of colorectal screening include lack of available free time (8), lack of high school graduation (4), lower income (3), rural residence (9), and not being exposed to an organized screening program (10). Beyond these determinants, one factor that is hypothesized to drive immigration-based inequalities in colorectal screening is the difficulty that many recent immigrants face in accessing primary health care services. In Canada, though immigrants granted permanent residency are entitled to universal health care coverage, linguistic, cultural and system-based barriers can make accessing health resources difficult (11, 12). For example, recent immigrants are less likely than non-immigrants to have a primary care physician (PCP; ref. 13). Because individuals without regular PCPs are less likely to be screened (3), the disparity in access to regular PCPs may explain, at least in part, this disparity in colorectal screening.

Although improving regular PCP access has been identified as a potential area of intervention to increase screening among recent immigrants, this potential mediating pathway between recent immigration and screening has yet to be formally empirically assessed. The aim of this study was therefore to assess if having a regular PCP mediates the disparity in lifetime colorectal screening between recent immigrants and non-immigrants in Canada, and if so, assess what proportion of the disparity would be eliminated if all had a regular PCP (referred to as the proportion eliminated or “PE”). Because recent immigrants' cancer screening habits and beliefs, and overall interactions with the health care system, can vary across racial and ethnic identities (14, 15), we wished to assess the PE across visible minority and white sub-populations. We did so using multiple techniques to compare the stability of findings.

Data from years 2003, 2005, 2007 to 2014 of the population-based, cross-sectional Canadian Community Health Study (CCHS) were used (16). The CCHS questionnaire was administered in English and French (with the possibility of completing the interview in an alternative language when necessary; ref. 17) via computer-assisted interviewing, either in-person or by telephone (40% and 60% of interviews, respectively; ref. 17). Response rates ranged from 80.7% in 2003 (18) to 65.6% in 2014 (17), with sampling weights adjusted for nonresponse. This study's target population was adults aged 50 to 75 years, who were either white and Canadian-born or had immigrated to Canada recently (≤10 years), and had no known risk factors or symptoms of colorectal cancer (19). Excluded from this study were respondents who reported screening due to “family history of colorectal cancer,” “follow-up of a problem,” and “follow-up of colorectal cancer treatment” (16), as well as longer-term immigrants (>10 years). The latter were excluded given their similar overall prevalence of never having been screened for colorectal cancer (45.6%; 95% CI, 44.4%–47.1%) as Canadian-born respondents (46.3%; 95% CI, 45.8%–46.8%).

Figure 1 describes the hypothesized relations between the following measures (additional theory and literature on which are described in detail in the Supplementary eMethods 1):

The CCHS questionnaire describes stool-tests and endoscopic examination (sigmoidoscopy, colonoscopy) and asks: “Have you ever had this test/either of these exams?” For immigrants, the question did not differentiate whether tests were conducted before or after arrival to Canada. Respondents were considered to have never been screened (Y = 1) if they reported to have never had any of these tests. We focused on lifetime screening as most Canadians have in fact never been screened in their lifetime (6) and since having never been screened is associated with later-stage cancer at diagnosis and higher risk of mortality (7).

The exposure of interest (A) was recent immigration experience (A = 1 for those who reported immigrating to Canada in the previous 10 years). A 10-year cut-off for recent immigration was used to reflect the observed period it takes, on average, for new residents to feel a sense of familiarity in the Canadian setting, and to report similar income earnings as average Canadian residents (20). To account for the intersecting experiences of recent immigration and racialization (21), recent immigrants and non-immigrants were stratified by visible minority status (yes or no). The CCHS' derived variable for visible minority status captures whether respondents' self-reported cultural and racial background is other than “white” (22). Screening among white recent immigrants and visible minority recent immigrants was compared to screening among white Canadian-born respondents to isolate associations for joint exposure (visible minority status, recent immigration) and single exposure (recent immigration).

The principal mediator of interest was access to a regular PCP. Respondents were considered to not have a regular PCP if they answered “No” to the question “Do you have a regular medical doctor?” And, as it is convention to consider other potential mediating factors in the analysis of direct and indirect effects (23), we identified two factors in the literature to be treated as additional potential mediators in the analyses: household income and area of residence. These factors were excluded from TE analyses and included as potential confounders of the mediator-outcome association in CDE analyses. Household income was categorized as quartile groupings (highest income [Quartile 4] as reference). Since missing income data were not imputed in the CCHS before 2005, income for CCHS 2003 was imputed based on individuals' age, sex, education, marital status, immigration status, and sampling weight, using hot deck imputation in Stata 14 (hotdeckvar; ref. 24). Area of residence was dichotomized (urban vs. rural). Urban classification in the CCHS is based on census population concentration (n ≥ 1,000 inhabitants) and density (n ≥ 400/km²). It includes urban core, urban fringe, secondary urban core, and suburban areas (22).

Covariates included were sex, age (50–59 years; 60–75 years), marital status (single; divorced, widowed, or separated; married or in a common-law relationship), educational attainment, and exposure to a provincially organized colorectal screening program. Because lack of high school graduation is an important predictor of non-recent screening participation across white and visible minority groups at other cancer sites (9) and has previously been associated with lower likelihood of colorectal screening (with nondifferential effect sizes for higher education groups, i.e., high school graduates, postsecondary attendees; ref. 25), those who had not completed high school were compared with those who had a high school diploma or more formal education (including college attendance). Residents of Manitoba from 2007, Ontario from 2008, Saskatchewan from 2009, Nova Scotia from 2009, and New Brunswick from 2014 were considered exposed to a mail-out based organized colorectal screening program (26), designed to promote screening. Exposure to these programs was included as a confounder (i.e., as a potential determinant of having a regular PCP, screening likelihood, and how immigration is experienced) and effect modifier (i.e., as a moderating factor of the associations between immigration experience and having a regular PCP, and between PCP access and screening).

Estimating the proportion of the disparity would be eliminated if all had a regular PCP (the “PE”), requires an estimation of (i) the total adjusted association (referred to as the total effect or “TE”) between recent immigration experience and lifetime colorectal screening, and (ii) the direct association between recent immigration and lifetime colorectal screening if all had a PCP (referred to as the controlled direct effect “CDE”). In a counterfactual framework (27), if we assume having measured all mediator-outcome confounders, the CDE can be defined as the remaining immigration-based disparity in lifetime screening prevalence had all individuals been assigned (possibly counterfactually) a regular PCP (28, 29). Measuring the CDE is particularly relevant when interested in assessing how a potential intervention on the mediator (here, assigning all a PCP) could influence a known inequality (28, 29). If the inequality in access to regular PCPs according to immigration status does explain, at least in part, immigration-based inequalities in screening, we would expect to see a proportion of the inequality eliminated.

Multiple approaches have been proposed to estimate the TE and CDE. Some use regression modeling (30, 31), whereas others combine regression modeling with inverse probability weighting techniques (32), or use a purely inverse probability weighting approach (23). The aim of using inverse probability weights is to create synthetic populations that are balanced in terms of the measured covariates, through which contrasts in average outcomes can be estimated (23). In this study, we applied three methods [summarized in detail in other texts (23, 31, 32) and in Table 1]—which, taken together, enable an assessment of the robustness of the findings. We describe the three methods here:

First, we used a regression-based product method (also referred to as the generalized product method; ref. 29) proposed by VanderWeele and Vansteelandt (31), which extends Baron and Kenny's product method (30) to allow for effect estimation in the presence of exposure-mediator interaction [Method 1]. This method requires the specification of two Poisson regression models for the screening outcome—one with, the other without, the mediator measure and its product term with the exposure (these are the CDE, and TE models, respectively). Second, we used an inverse probability-weighted marginal structural model approach [Method 2], described by VanderWeele (32), in which TE and CDE models (as in Method 1) are weighted using inverse probability weights for the exposure and mediator. These weights are constructed using propensity scores (predicted probabilities) for the exposure and mediator, given covariates and other mediator values, which are estimated using logistic models (details in Table 1 and the Supplement's eMethod 2). Finally, we used an inverse probability weighted approach for marginal effect estimation [Method 3], described by VanderWeele (23), in which screening prevalence is weighted (as in Method 2), and simple ratios of the average screening prevalence between the exposed and unexposed (for TE estimation), and between the exposed with a PCP and unexposed with a PCP (for CDE estimation) are computed. Inverse probability weights used in this method are also constructed using propensity scores for the exposure and mediator, estimated using logistic models (details in Table 1 and Supplement's eMethod 2). Estimates from IPW methods 2 and 3 can be interpreted as the average associations in the population, whereas method 1 associations are conditional on the strata of the variables in the models. With the TE and CDE, the proportion of the total effect explained by having a regular PCP (PE) was estimated on an excess relative risk scale [using prevalence ratio (PR) estimates] as follows: (PR^TE − PR^CDE)/(PR^TE − 1) (23). Confidence intervals (95%) for CDE, TE, and PE were estimated using the bootstrap method (500 replications; ref. 23). Analyses were conducted in R (version 3.4.1; ref. 33).

The analyses described above rely on two assumptions, the validity of which will be tested using sensitivity analyses. First, the validity of the CDE estimates (and consequently, of the PE estimates) relies on the assumption of controlled confounding for the mediator–outcome relationship (34). We apply formulas derived by VanderWeele (2015; ref. 35) to test the sensitivity of observed CDE estimates to unmeasured confounding of the mediator-outcome relationship. This approach estimates how large associations would have to be between an unmeasured factor and both the mediator and outcome for the true CDE estimates to be null (PR = 1) despite non-null estimates, or to be equivalent or smaller to the observed TEs (yielding null or positive PEs).

Second, estimating CDE requires both theoretical positivity of the mediator (i.e., that all respondents have a non-null probability of having a regular PCP) and—when using inverse probability weighting—practical positivity for the exposure and mediator [i.e., that propensity scores for these factors are neither 0 nor 1 (0% or 100% probability); ref. 36]. To assess this, we performed stratified, descriptive analyses of propensity scores for the exposure and mediator (23).

The study protocol was approved by the Ethical Review Board of the Centre de Recherche du Centre Hospitalier de l'Université de Montréal.

Of the total sample (102,366 of whom were white Canadian-born respondents, 659 were recent-immigrants of visible minorities, and 408 of whom were white recent-immigrants), 47% had never been screened in their lifetime and 9% did not have a regular PCP. The prevalence of never having been screened was 71% and 57% in visible minority and white recent immigrants, respectively, and 46% in white Canadian-born respondents (Table 2). Approximately 9% of white Canadian-born respondents did not have a regular PCP, compared with 18% among both visible minority and white recent immigrants, respectively (Table 2). Overall, the proportion of those who had never been screened was higher among those who were younger than 60 years, not partnered, had lower income (quartiles 1 and 2), had not obtained a high school diploma, did not have a PCP, resided in rural settings, and were not exposed to a provincial organized screening program (Table 2).

Overall, associations were observed between recent immigration, having a regular PCP, and screening. Adjusting for all factors, recent immigrants were less likely to have a PCP (Table 3) and more likely to have never been screened (Table 4). However, associations between having a regular PCP and screening were heterogeneous across strata of immigration status. Expressed as prevalence differences (PD), the adjusted difference in screening between those with and without a PCP was larger among white Canadian-born respondents [PD = 19% (95% CI, 17%–20%), i.e. prevalence of approximately 44% among those with a PCP, 68% among those without] than among white immigrants [PD = 4% (95% CI, 15%–24%), i.e. prevalence of approximately 54% among those with a regular PCP, 58% among those without] or among visible minority immigrants [PD = 8% (95% CI, 10%–27%), i.e., prevalence of approximately 71% among those with a PCP, 79% among those without; data not in table].

The TE, CDE, and PE estimates were largely consistent across all three mediation methods (Table 5). Large CDE estimates (between PR = 1.56 and 1.60 for visible minority recent immigrants, and between PR = 1.24 to 1.27 for white recent immigrants, depending on the method used) suggest that even if inequalities in access to a regular PCP were eliminated, a large disparity in lifetime screening would remain for recent immigrants across visible minority status. Most CDE estimates were larger than TE estimates, yielding null PE estimates for white recent immigrants, and negative PE estimates (i.e., exacerbated inequalities under mediator intervention) for visible minority recent immigrants (between −6% and −13%, depending on the method used; Table 2).

First, we found that the associations between the unmeasured factor and both the mediator and outcome would have to be at minimum PR = 2.5 for visible minority recent immigrants, and PR = 1.8 for white recent immigrants for the true CDE estimates to be null (PR = 1) despite non-null estimates; and would have to be at minimum PR = 1.3 and PR = 1.1 for visible minority and white recent immigrants, respectively, for true CDE estimates to be equivalent or smaller than observed TE estimates (yielding null or positive PE values; Supplementary Table S1 and S2). These are larger estimates than those observed for low education and not being exposed to an organized screening program (Tables 2 and 3). Nonetheless, the potential for unmeasured confounding remains. Second, analyses of propensity scores indicate good covariate balance between exposed and unexposed respondents after weighting, and of practical positivity for regular PCP access (Supplementary Table S3 and S4). However, the requirement of practical positivity for recent immigration may be violated (i.e., propensity scores—even when truncated at the 10th percentile—were close to 0). Lack of practical positivity may lead to potential instability of the weighting methods. Finally, results were largely consistent when accounting for effect modification by exposure to an organized screening program, with slightly attenuated direct effects (Supplementary Table S5).

The aim of this study was to assess whether having a regular PCP mediates the disparity in lifetime colorectal screening between recent immigrants and non-immigrants in Canada. In this sample, in which nearly half (47%) have never been screened, we observed large controlled direct effects between recent immigration and screening, as well as proportions eliminated that were either null or negative—indicating that improving access to regular PCPs may not reduce observed immigration-based screening inequalities. As the screening disparity between those with and without a regular PCP is larger among white Canadian-born respondents (PD = 19%) than among recent immigrants (PD = 8% and 4% among visible and white minority recent immigrants, respectively), having a PCP lead to larger increases in screening among Canadian-born individuals than among recent immigrants, thereby leaving the disparity untouched or exacerbated.

The observed associations between recent immigration and lifetime screening are consistent with those observed previously in Canada and North America (1). These associations may be explained by recent immigrants' more limited knowledge of and trust in the efficacy of the screening tests and the medical system, discomfort with the test itself, or lower perceived susceptibility to cancer (1, 37). Differences in effect sizes between white and visible minority recent immigrants may be explained by differences in ethno-cultural feelings of fatalism and helplessness with regards to colorectal cancer diagnosis and mortality (8) or in the acceptability of screening tests (15). On a more distal level, systemic discrimination (38), barriers to health care, and social stressors such as inadequate housing and precarious employment (39) are thought to explain, in part, why persons who immigrate to Canada —who, upon arrival are disproportionately healthy—see their health experience a decline in health over time, eventually converging with Canadian-born residents their age (11). These distal factors may also explain why such strong associations are observed between immigration experience and screening. However, it should be noted that Canada's immigrant and visible minority populations are highly heterogenous, and explanations for the observed findings may not hold across all subgroups.

Although having a regular PCP is an important enabling determinant of screening participation overall (1), our findings suggest increasing access to PCPs may lead to greater gains in screening for Canadian-born individuals than for recent immigrants. Several factors may explain this observation. First, recent immigrants may not systematically receive screening recommendations from their regular PCPs—as is observed for patients of visible minorities or lower socioeconomic status in Canada (40). Recent immigrants to Canada have also reported gaps in the cultural competency of health care providers, specifically with regards to cultural understandings of health and health care (41), and to language and communication (42). These limitations of the Canadian health care system, in conjunction with general logistic and psychological barriers to colorectal screening (such as unreachable or inadequate resources, lower social support, fear, embarrassment or anxiety of the test, its required preparation—especially for endoscopic procedures—or of its result (43)) may explain why simply having a regular PCP may not be sufficient to ensure screening uptake.

The implication of these findings is that improving individuals' access to regular PCPs may improve screening participation overall (namely through large gains among Canadian-born individuals) but fail to reduce screening disparities according to recent immigration. We recommend that future studies explore alternative areas of intervention to both reduce these inequalities and increase screening uptake overall.

These findings are bound by certain limitations. First, the broad categories of white versus visible minority immigrants may obscure sub-group heterogeneity in the associations measured, as has been observed in previous studies (44). These findings should therefore be interpreted as the average mediating role of PCPs in white and visible minority immigrants. Second, the cross-sectional data used required additional assumptions of the temporal ordering of associations between recent immigration, regular PCP access, and screening. Replication of these analyses using longitudinal data will be beneficial. Third, although we stated effect estimates in the causal language used in epidemiologic research, the validity of these assertions rely on the satisfaction of the causal assumptions underpinning each method (23). Sensitivity analyses suggest that some residual confounding is likely present. Among the unmeasured factors in this study (and indeed, in the CCHS) are concordance of individuals' and PCPs' gender identity or economic, linguistic, ethnic, or cultural background (45, 46). Future mediation studies may benefit from incorporating the latter measures, as well as exploring downstream mediators such as knowledge and cultural beliefs around cancer or cancer prevention, and the frequency or recency of PCP visits. Finally, because self-reported screening data tend to overestimate recent screening [i.e., previous two year fecal occult blood test (FOBT) sensitivity is 77.4% and specificity is 89.8%; ref. 47], and studies at other cancer sites have observed differential self-reported screening according to racial and ethnic subpopulations (47), it is possible that the screening gaps between immigrants and non-immigrants observed in this study may be underestimated.

In sum, almost half of adults aged 50 to 75 years in Canada have never been screened for colorectal cancer, and the prevalence of having never been screened is even higher for recent immigrants. This study suggests that increasing individuals' access to regular PCPs may increase screening overall, but not eliminate immigration-based disparities in screening. Other levers will be necessary to decrease these inequalities.

No potential conflicts of interest were disclosed.

Conception and design: A. Blair, L. Gauvin, M.E. Schnitzer, G.D. Datta

Development of methodology: A. Blair, M.E. Schnitzer, G.D. Datta

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Blair, M.E. Schnitzer, G.D. Datta

Writing, review, and/or revision of the manuscript: A. Blair, L. Gauvin, M.E. Schnitzer, G.D. Datta

Study supervision: G.D. Datta

Other (Cosupervision [with C.D. Datta] of first author [A. Blair] who is completing her doctoral work in public health at the Université de Montréal): L. Gauvin

This study was approved by the Quebec Inter-University Center for Social Statistics at the University of Montreal and the Comité Éthique à la Recherche of the Centre Hospitalier de l'Université de Montréal. The analyses presented in this paper were conducted at the Quebec Inter-university Centre for Social Statistics, which is part of the Canadian Research Data Centre Network (CRDCN). The services and activities provided by the QICSS are made possible by the financial or in-kind support of the Social Sciences and Humanities Research Council (SSHRC), the Canadian Institutes of Health Research (CIHR), the Canada Foundation for Innovation (CFI), Statistics Canada, the Fonds de recherche du Québec - Société et culture (FRQSC), the Fonds de recherche du Québec - Santé (FRQS), and Québec universities. The views expressed in this paper are those of the authors, and not necessarily those of the CRDCN or its partners.

A. Blair is supported by the Vanier Doctoral Scholarship Program and has travel support from the Quebec Inter-university Center for Social Statistics. G.D. Datta is the recipient of a career award from the Canadian Cancer Society (award # 703946). M.E. Schnitzer is the recipient of a New Investigator Salary Award from the Canadian Institutes for Health Research.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.