Daniel Kim - New Horizons in Modeling and Simulation for Social Epidemiology and Public Health

A growing body of literature is attempting to reduce alternative explanations and other sources of bias in nonexperimental studies on the social determinants of health and more generally within public health. These novel approaches to strengthen causal inference include but are not limited to instrumental variable (IV) analysis, fixed effects analysis, propensity score analysis, inverse probability weighting, and natural experiments. By isolating random variation in the exposure, IV analysis can yield unbiased estimates of the causal association between an exposure and outcome, including through reducing attenuation bias due to measurement error and confounding bias due to both observed and unobserved factors (Kim 2016). Such approaches are increasingly being used to evaluate the causal roles of risk factors in public health, including obesity, neighborhood conditions, the social environment, and state policies (Davey Smith et al. 2009; Fish et al. 2010; Kim et al. 2011; Mojtabai and Crum 2013; Hawkins and Baum 2014; Kim 2016).

Similar to multivariable regression, propensity score analysis can control for imbalances between comparison groups and can thereby control for confounding. It has the advantage of being more efficient than traditional regression when there are relatively fewer events (Cepeda et al. 2003). However, like multivariable regression, propensity score analysis cannot control for unobserved or unmeasured confounders. Inverse probability weighting has also been used as an approach to estimate the counterfactual or potential outcome if all subjects were assigned to either exposure/treatment (Mansournia and Altman 2016). Finally, natural experiments or other quasi‐experimental designs such as regression discontinuity designs (Moscoe et al. 2015) can exploit random variation in exposures as in an experimental study and can thereby minimize confounding due to both observed and unobserved factors as a source of bias.

Results from individual studies can also be qualitatively reviewed in aggregate to identify existing gaps in methodological approaches, potential sources of bias, and similarities/differences in their results. Results across studies can be quantitatively summarized in meta‐analyses that yield overall point estimates of exposure–outcome associations, although, importantly, such estimates are only as good as the quality of the studies that are included in the meta‐analyses (Egger et al. 2001).

As Bambra et al. (2010) have noted, there are clear limitations to the existing evidence based on the social determinants of health. First, observational studies that dominate the literature can only hint at possible interventions and their associated health effects; causal inference is an inherent limitation. Second, there is still only sparse evidence on the impacts of interventions on the social determinants of health. Bambra et al. (2010) conducted an “umbrella review” of the existing systematic reviews of the evidence on specific interventions on the social determinants of health spanning housing/living environment, work environment, transportation, health and social care services, agriculture and food, and water and sanitation. They identified some suggestive evidence that certain categories of interventions may impact inequalities regarding the health of specific disadvantaged groups, particularly in the fields of housing and work environment. Yet in other areas, such as evidence on policies in education, the health system, food and agriculture, and more generally on the influences of macro‐level policies on health inequalities, the empirical literature on interventions was more limited (Bambra et al. 2010).

In a more recent umbrella review, Thomson et al. (2017) adopted a systematic review approach to summarize the state of knowledge on how public health policy interventions (e.g. taxation and educational campaigns) may impact health inequalities such as differential effects across socioeconomic groups or effects of interventions targeted at disadvantaged groups. After searching studies published up to May 2017 within 20 databases (e.g. Medline, EMBASE, CINAHL, PsycINFO, Social Science Citation Index, Sociological Abstracts, and the Cochrane Library), the authors identified 24 systematic reviews reporting 128 relevant primary studies. They then summarized the evidence on policies (fiscal, regulation, education, preventive treatment, and screening) across eight public health domains (tobacco; food and nutrition; the control of infectious diseases; screening; road traffic injuries; air, land, and water pollution; built environment; and workplace regulations). The systematic reviews were mixed in quality, and the results were mixed across public health domains. For the tobacco, food and nutrition, and control of infectious diseases domains, the authors found evidence to suggest that fiscal and regulation policies were more beneficial for reducing or preventing health inequalities than educational campaigns (Thomson et al. 2017).

In principle, intervening on the social determinants of health should have profound effects on population health outcomes and health equity. These outcomes include the numbers of lives saved and the occurrence of disease and other morbidity outcomes such as Disability‐Adjusted Life Years (DALYs) (Murray et al. 2015). If we consider the distal nature of these social determinants in Figure 1.2, the impacts of these determinants on population health may in fact be stronger than those of proximal biological and behavioral factors at the individual level (such as smoking and high cholesterol), because upstream social determinants likely shape many of these biological and behavioral factors.

Yet what does the empirical evidence show about the impacts of social determinants of health at a population level? Drawing on studies from the public health literature, the numbers of adult deaths attributable to six social determinants of health have been estimated (Galea et al., 2011): low education, poverty, low social support, area‐level poverty, income inequality, and racial segregation. The investigators calculated summary relative risk estimates of mortality, and used prevalence estimates for each of these social determinants to estimate the associated population attributable risks (PARs, the percentage of deaths attributed to each factor), and then project the total number of deaths attributable to each social determinant in the United States. Through this approach, the authors estimated that 245 000 deaths would have taken place among Americans in the year 2000 due to low education, 176 000 deaths to racial segregation, 162 000 deaths to low social support, 133 000 deaths to individual poverty, 119 000 deaths to income inequality, and 39 000 deaths to area‐level poverty. These estimates due to social determinants of health were comparable to the total numbers of deaths due to the leading pathophysiological causes such as heart attacks (192 898 deaths), strokes (167 661 deaths), and lung cancer (155 521 deaths) (Galea et al. 2011). To further put the size of these numbers into perspective, in the year 2000, it was estimated that smoking resulted in 269 655 deaths among men and 173 940 deaths among women in the United States (Centers for Disease Control and Prevention (CDC) 2008).

In another study, Krueger et al. (2015) estimated the mortality attributable to education under three hypothetical scenarios: (i) individuals having less than a high school degree, (ii) individuals having some college education but not completing a bachelor's degree, and (iii) individuals having any level of education but not completing a bachelor's degree. The authors used National Health Interview Survey data (1986–2004) linked to prospective mortality through 2006 and discrete‐time survival models to derive annual attributable mortality estimates. The estimated numbers of attributable deaths were striking: 45 243 deaths in the 2010 US population were attributed to individuals having less than a high school degree rather than a high school degree; 110 068 deaths were due to individuals having some college education; and 554 525 deaths were attributed to individuals having anything less than a bachelor's degree but not a bachelor's degree (Krueger et al. 2015). The total numbers of deaths due to having less than a high school degree was similar among women and men and among non‐Hispanic Blacks and Whites and was greater for cardiovascular disease than for cancer. Overall, these estimates point to the substantial impacts that policies that increase educational opportunities could have on reducing the burden of adult mortality (Krueger et al. 2015).