Using self-reported data on the social determinants of health in primary care to identify cancer screening disparities: opportunities and challenges

The social determinants of health (SDOH), such as income, race/ethnicity, education, and working and living conditions, are responsible for substantial morbidity and mortality worldwide, with income being perhaps the most important determinant [1‐8]. The World Health Organization defines the SDOH as “the conditions in which people are born, grow, work, live and age, and the wider set of forces and systems shaping the conditions of daily life” [9]. The SDOH influence one’s health-related behaviours, health literacy levels, sense of self-control and self-efficacy, and access to societal resources including access to healthcare [8, 10]. The Institute of Medicine is moving to capture SDOH in electronic medical records, due to increased recognition that data on the SDOH can provide crucial information useful for treatment choices, health care system design and targeted innovations [11].

Quality primary care has been associated with an attenuation of the health effects of income inequalities [12, 13], making primary care a highly appropriate healthcare setting in which to measure and intervene on the SDOH, especially with the expansion of multidisciplinary patient-centred teams [14, 15]. However, individual-level data and practice-level data on the SDOH are rarely available within the primary care context and the best way to systematically collect and use these data in the primary care setting is not yet clear. Improving colorectal, cervical and breast cancer screening uptake for patients is an example of an area in primary care where data on the SDOH could be valuable, as disparities in cancer screening at the population level are well established, particularly related to income and foreign-born status [16‐27].

In our large urban primary care practice in Ontario, Canada, we now routinely collect data on the SDOH from our patients using a self-completed survey in the waiting room. These data allow us to identify potential gaps in care related to SDOH and target our quality improvement efforts [28]. In this study, we aimed to understand: i) the relationship between the SDOH, particularly income, and colorectal, cervical and breast cancer screening, and ii) challenges and opportunities of using SDOH data from a self-administered patient survey.

We analysed data for 5766 patients who answered at least one question on the sociodemographic survey (Fig. 1). Table 1 describes the sociodemographic characteristics of the study participants, including missing data, as well as the proportion of patients up-to-date on the three screening types by sociodemographic subgroup. These categories were not mutually exclusive; 1207 women were eligible for all three screening types. There was a large amount of missing data for the LICO-BT variable, approximately 25% across the three cohorts.

Patients who completed the sociodemographic survey were significantly more likely to be up-to-date on colorectal, cervical and breast cancer screening than those who did not complete the sociodemographic survey (Fig. 2). Study participants were most likely to be up-to-date on cervical cancer screening (78.7%) and least likely for CRC screening (72.9%), which also held true for the general practice population. No significant differences were observed for any of the screening modalities based on age or neighbourhood income quintile (data not shown) between participants who completed the sociodemographic survey and those who did not complete the sociodemographic survey. Among those eligible for colorectal cancer screening, female patients were more likely to complete the sociodemographic survey than male patients.

In bivariate analyses (Table 2), age, immigrant status, income and housing showed statistically significant associations with cancer screening. Patients who were up-to-date with colorectal cancer screening were significantly older than those were overdue, whereas those who were up-to-date with cervical cancer screening were significantly younger than those were overdue. Foreign-born patients were more likely to be up-to-date on colorectal screening than their Canadian-born peers but showed no significant differences in breast or cervical cancer screening. Although there were no significant differences for any of the screening modalities based on neighbourhood income quintile, the derived LICO-BT variable was associated with all three screening types. People living below the LICO-BT were less likely to be up-to-date on screening for all screening types and this difference was most marked for breast cancer screening. Housing status was also associated with all three cancer screening types. Screening was less common among patients who rented a home and those who reported their housing as “other” than among those who owned a home. There were differences in breast and cervical cancer screening between heterosexual and non-heterosexual women but these were not statistically significant.

We did not conduct multivariable analyses because of the large amount of missing data for income (approximately 25%), which was our main variable of interest. It was not possible to use neighbourhood income as a proxy in multivariable analysis because there was not a high level of agreement between neighbourhood income and the LICO. For example, 50.9% of participants in the lowest income quintile and 65.5% of participants in the second lowest income quintile lived above the LICO.

To better understand our results, we looked at the relationship between housing status and LICO-BT status. Of those who were homeowners, the vast majority lived above the LICO-BT (90.1% for the CRC screening cohort, 90.9% for cervical cohort, 89.4% for breast cohort). Similarly, of those who lived in “other” housing i.e. were neither homeowners nor renters, most were below the LICO-BT (81.1% for CRC cohort, 76.2% for cervical cohort, 76.6% for breast cohort).

We used self-reported SDOH data within the primary care setting to describe the association between the SDOH and clinical outcomes: colorectal, cervical and breast cancer screening. We found that participants living below a low-income threshold indicative of poverty and people who did not own their homes were significantly less likely to be screened for cancers than their more advantaged peers. However, we also found that patients who responded to the sociodemographic survey were not typical of all patients in the practice, with higher cancer screening rates than patients who did not respond to the survey. Although we were able to use survey data to formulate a low-income cut-off based on the Statistics Canada definition, there was a large amount of missing income information, approximately 25% across the three cancer screening cohorts, precluding us from conducting multivariable analyses. It is possible that income was on the causal pathway of the relationship we observed between housing status and screening. Other variables (e.g. age) may have also confounded the relationships we observed between income and screening and housing and screening or acted as intermediate factors.

Our study highlights opportunities in the use of self-reported SDOH data in primary care. We found a significant association between the low-income cut-off variable and cancer screening despite the high amount of missing income data, and despite neighbourhood income quintile not being associated with cancer screening. Most studies on income inequities in health use neighbourhood income as a proxy for individual income, as the latter is often not available [19, 23, 25, 27, 40‐44]. Our results demonstrate that neighbourhood income may under-represent income-related inequalities and were not in high agreement with the individual-level LICO. This may be especially true in urban centres like Toronto where the demographics of neighbourhoods are changing rapidly and faster than a quinquennial census can accurately track [45].

We also found intriguing results based on housing status. We found no other studies that examined the relationship between housing status and cancer screening. However, housing status has been associated with other aspects of cancer care in the international literature. A recent U.S. study showed that both renters and people in unstable housing had delays in the time to diagnostic resolution after a cancer screening abnormality when compared to homeowners, potentially due to barriers such as coordination and scheduling, transportation, childcare and low literacy [46]. Hagedoorn et al. found that home ownership (versus tenancy) provided a beneficial effect on lung cancer mortality in Belgium [47]. Housing status may simply be a marker of income, but it may also be a marker of wealth or social stability. Future research should further explore the causal pathway that lies between housing and cancer screening, taking into account sociodemographic variables that have been previously found to be associated with cancer screening, such as age, income, immigrant status, and co-morbidities, and that could serve as confounders for observed associations [19, 23, 26, 27, 48, 49].

Our results also highlight several challenges in collecting individual-level income status using self-reported data. One-quarter of participating patients did not provide a household income; missing data for other social determinants were otherwise less than 10%. Our findings of missing income are in line with previous literature. Kirst et al. found that 67% of Ontario residents were uncomfortable with disclosing household income and there was a general lack of awareness of the importance of collecting sociodemographic data [50]. Similarly, a national survey conducted in 2009 found that 65% of Canadian respondents were uncomfortable with the collection of household income in healthcare settings [51]. Respondents were most comfortable with sociodemographic data being directly collected by their family physician and certain groups were more comfortable with the collection of income, such as visible minorities and males [51].

It is possible that asking about socioeconomic status in different ways might have been more acceptable to our survey participants and might have led to less missing data. The question “Do you have difficulty making ends meet at the end of the month?” has been found to be a good predictor in a Canadian primary care population of being below the low-income cut-off, with a sensitivity of 98% and specificity of 40%, with the caveat that non-English speakers may not understand this colloquial term [52]. Future research could test such questions for their acceptability and validity in our setting [15]. Interestingly, in the pilot testing of our sociodemographic survey, only 10.1% of patients did not answer the income question [28], stressing the importance of ongoing post-implementation surveillance of such initiatives.

We also experienced several other challenges with collecting and using SDOH data. First, although there were a large number of categories for ethnicity in the survey, many of these categories had quite small cell sizes. We also had data on indigenous persons that we did not feel comfortable using due to a lack of adherence to Ownership, Control, Access and Possession principles [38]. We ultimately collapsed these categories into White vs. non-White for bivariate analyses. However, different categorizations might have led to more meaningful ethnicity-based screening results. How to best categorize ethnicity while maximizing both sample size and granularity needs to be further explored. Second, our survey was only available in English or French, which may explain the low proportion of patients in our survey who reported their preferred language as one other than English. This may have dissuaded some foreign-born patients from participating in the survey, and may partially explain why we saw no disparities based on immigrant status. Third, study participants were not representative of all patients in the practice, as noted by their higher screening rates. Participants were all willing to answer a waiting room survey, so it is plausible that they may have been more likely to come in for visits than other patients and that they may also have been more likely to be intrinsically motivated to participate in prevention and screening manoeuvres. These latter two points may partially explain why we found no disparities in screening for foreign-born patients despite well-documented disparities at the provincial level [23, 26, 44]. Fourth, although a sizeable proportion of our patients have completed the survey, detecting differences in care among some patient groups (e.g. non-heterosexual patients, certain ethnic groups) would likely require a larger sample. Fifth, the survey did not include a question on education, another known social determinant of health. Prior research demonstrates that higher education levels are positively associated with colorectal, cervical cancer, and breast cancer screening adherence [53‐55]. Sixth, we were unable to track family history of cancer, which has been shown to be a motivator for cancer screening [56‐59]. Seventh, younger participants in our study had less opportunity for cancer screening, which may at least partially explain our findings on age and colorectal cancer screening. Finally, it is plausible that the missing data encountered in all questions was non-random and potentially biased our findings.

Our experiences highlight key considerations for primary care providers and researchers interested in collecting and intervening on SDOH in primary care. While we were able to show that neighbourhood income might under-estimate income-related disparities in screening, individual-level income was also the most challenging variable to collect. Future work in this area should focus on targeted interventions that reduce income-related disparities in cancer screening [60] and simultaneously continue to explore how best to collect socioeconomic data, particularly measures of poverty. Further research is also needed to understand whether housing is an independent risk factor for under-screening.