Causal inference challenges in social epidemiology: Bias, specificity, and imagination

Considerations for Subgroup Analyses in Cluster-Randomized Trials Based on Aggregated Individual-Level Predictors

In research assessing the effect of an intervention or exposure, a key secondary objective often involves assessing differential effects of this intervention or exposure in subgroups of interest; this is often referred to as assessing effect modification or heterogeneity of treatment effects (HTE). Observed HTE can have important implications for policy, including intervention strategies (e.g., will some patients benefit more from intervention than others?) and prioritizing resources (e.g., to reduce observed health disparities). Analysis of HTE is well understood in studies where the independent unit is an individual. In contrast, in studies where the independent unit is a cluster (e.g., a hospital or school) and a cluster-level outcome is used in the analysis, it is less well understood how to proceed if the HTE analysis of interest involves an individual-level characteristic (e.g., self-reported race) that must be aggregated at the cluster level. Through simulations, we show that only individual-level models have power to detect HTE by individual-level variables; if outcomes must be defined at the cluster level, then there is often low power to detect HTE by the corresponding aggregated variables. We illustrate the challenges inherent to this type of analysis in a study assessing the effect of an intervention on increasing COVID-19 booster vaccination rates at long-term care centers.

Sexual harassment, sexual violence and subsequent depression and anxiety symptoms among Swedish university students: a cohort study

PURPOSE

To determine the gender-specific impact of recent exposure to different forms of sexual harassment and sexual violence (SHV) on depression and anxiety symptoms three, six, and nine months later.

METHODS

We recruited 2229 women and 1274 men studying at Swedish universities and followed them with web-surveys every three months over one year. We estimated mean differences (MDs) of depression and anxiety symptoms between exposed and unexposed at each follow-up, adjusting for prior SHV, prior depression and anxiety symptoms and potential confounders.

RESULTS

For women, sexual harassment (wide subjective definition) was associated with higher symptom levels of depression (MD 1.0 [95% CI: 0.3; 1.7]) and anxiety (MD 0.8 [95% CI: 0.3; 1.4]) three months later. Unwanted sexual attention was associated with higher symptom levels of anxiety three (MD 0.5 [95% CI: 0.1; 0.8]) and six months later (MD 0.4 [95% CI: 0.0; 0.7]). Exposure to sex against ones will was associated with higher depression symptoms three (MD 1.7 [95% CI: 0.1;3.4]), and six months later (MD 3.1 [95% CI: 1.0; 5.2]). Trends indicated that associations with subsequent mental health differed between forms of SHV among women, and that most associations were more pronounced in temporal proximity to the exposures. For men, we refrain from interpreting the results since they showed high variability and were not robust to sensitivity analyses using multiple imputation to account for missing outcome data.

CONCLUSIONS

Among women, several forms of SHV were associated with higher subsequent depression and anxiety symptoms.

A quantitative approach to the intersectional study of mental health inequalities during the COVID-19 pandemic in UK young adults

PURPOSE

Mental health inequalities across social identities/positions during the COVID-19 pandemic have been mostly reported independently from each other or in a limited way (e.g., at the intersection between age and sex or gender). We aim to provide an inclusive socio-demographic mapping of different mental health measures in the population using quantitative methods that are consistent with an intersectional perspective.

METHODS

Data included 8,588 participants from two British cohorts (born in 1990 and 2000-2002, respectively), collected in February/March 2021 (during the third UK nationwide lockdown). Measures of anxiety and depressive symptomatology, loneliness, and life satisfaction were analysed using Multilevel Analysis of Individual Heterogeneity and Discriminatory Accuracy (MAIHDA) models.

RESULTS

We found evidence of large mental health inequalities across intersectional strata. Large proportions of those inequalities were accounted for by the additive effects of the variables used to define the intersections, with some of the largest gaps associated with sexual orientation (with sexual minority groups showing substantially worse outcomes). Additional inequalities were found by cohort/generation, birth sex, racial/ethnic groups, and socioeconomic position. Intersectional effects were observed mostly in intersections defined by combinations of privileged and marginalised social identities/positions (e.g., lower-than-expected life satisfaction in South Asian men in their thirties from a sexual minority and a disadvantaged childhood social class).

CONCLUSION

We found substantial inequalities largely cutting across intersectional strata defined by multiple co-constituting social identities/positions. The large gaps found by sexual orientation extend the existing evidence that sexual minority groups were disproportionately affected by the pandemic. Study implications and limitations are discussed.

Avoiding overadjustment bias in social epidemiology through appropriate covariate selection: a primer

Obtaining accurate estimates of the causal effects of socioeconomic position (SEP) on health is important for public health interventions. To do this, researchers must identify and adjust for all potential confounding variables while avoiding inappropriate adjustment for mediator variables on a causal pathway between the exposure and outcome. Unfortunately, 'overadjustment bias' remains a common and under-recognised problem in social epidemiology. This paper offers an introduction to selecting appropriate variables for adjustment when examining effects of SEP on health, with a focus on overadjustment bias. We discuss the challenges of estimating different causal effects including overadjustment bias, provide guidance on overcoming them, and consider specific issues including the timing of variables across the life-course, mutual adjustment for socioeconomic indicators, and conducting systematic reviews. We recommend three key steps to select the most appropriate variables for adjustment. First, researchers should be clear about their research question and causal effect of interest. Second, using expert knowledge and theory, researchers should draw causal diagrams representing their assumptions about the interrelationships between their variables of interest. Third, based on their causal diagram(s) and causal effect(s) of interest, researchers should select most appropriate set of variables, which maximises adjustment for confounding while minimising adjustment for mediators.

Marginal Structural Models for Life-Course Theories and Social Epidemiology: Definitions, Sources of Bias, and Simulated Illustrations

Social epidemiology aims to identify social structural risk factors, thus informing targets and timing of interventions. Ascertaining which interventions will be most effective and when they should be implemented is challenging because social conditions vary across the life course and are subject to time-varying confounding. Marginal structural models (MSMs) may be useful but can present unique challenges when studying social epidemiologic exposures over the life course. We describe selected MSMs corresponding to common theoretical life-course models and identify key issues for consideration related to time-varying confounding and late study enrollment. Using simulated data mimicking a cohort study evaluating the effects of depression in early, mid-, and late life on late-life stroke risk, we examined whether and when specific study characteristics and analytical strategies may induce bias. In the context of time-varying confounding, inverse-probability-weighted estimation of correctly specified MSMs accurately estimated the target causal effects, while conventional regression models showed significant bias. When no measure of early-life depression was available, neither MSMs nor conventional models were unbiased, due to confounding by early-life depression. To inform interventions, researchers need to identify timing of effects and consider whether missing data regarding exposures earlier in life may lead to biased estimates.

Longitudinal association between oral status and cognitive decline by fixed-effects analysis

Background

Although the feasibility of randomized trials for investigating the long-term association between oral health and cognitive decline is low, deriving causal inferences from observational data is challenging. We aimed to investigate the association between poor oral status and subjective cognitive complaints (SCC) using fixed-effects model to eliminate the confounding effect of unobserved time-invariant factors.

Methods

We used data from Japan Gerontological Evaluation Study (JAGES) which was conducted in 2010, 2013, and 2016. β regression coefficients and 95% confidence intervals [CIs] were calculated using fixed-effects models to determine the effect of deteriorating oral status on developing SCC. Onset of SCC was evaluated using the Kihon Checklist-Cognitive function score. Four oral status variables were used: awareness of swallowing difficulty, decline in masticatory function, dry mouth, and number of teeth.

Results

We included 13,594 participants (55.8% women) without SCC at baseline. The mean age was 72.4 (standard deviation [SD], 5.1) years for men and 72.4 (SD, 4.9) years for women. Within the 6-year follow-up, 26.6% of men and 24.9% of women developed SCC. The probability of developing SCC was significantly higher when participants acquired swallowing difficulty (β = 0.088; 95% CI, 0.065–0.111 for men and β = 0.077; 95% CI, 0.057–0.097 for women), decline in masticatory function (β = 0.039; 95% CI, 0.021–0.057 for men and β = 0.030; 95% CI, 0.013–0.046 for women), dry mouth (β = 0.026; 95% CI, 0.005–0.048 for men and β = 0.064; 95% CI, 0.045–0.083 for women), and tooth loss (β = 0.043; 95% CI, 0.001–0.085 for men and β = 0.058; 95% CI, 0.015–0.102 for women).

Conclusion

The findings suggest that good oral health needs to be maintained to prevent the development of SCC, which increases the risk for future dementia.

Win-Win: Reconciling Social Epidemiology and Causal Inference

Social epidemiology is concerned with the health effects of forces that are "above the skin." Although causal inference should be a key goal for social epidemiology, social epidemiology and quantitative causal inference have been seemingly at odds over the years. This does not have to be the case and, in fact, both fields stand to gain through a closer engagement of social epidemiology with formal causal inference approaches. We discuss the misconceptions that have led to an uneasy relationship between these 2 fields, propose a way forward that illustrates how the 2 areas can come together to inform causal questions, and discuss the implications of this approach. We argue that quantitative causal inference in social epidemiology is an opportunity to do better science that matters, a win-win for both fields.

Acute Changes in Community Violence and Increases in Hospital Visits and Deaths From Stress-responsive Diseases

BACKGROUND

Community violence may affect a broad range of health outcomes through physiologic stress responses and changes in health behaviors among residents. However, existing research on the health impacts of community violence suffers from problems with bias.

METHODS

We examined the relations of acute changes in community violence with hospital visits and deaths due to stress-responsive diseases (mental, respiratory, and cardiac conditions) in statewide data from California 2005-2013. The community violence exposure was measured as both binary spikes and continuous acute changes. We applied a combined fixed-effects and time-series design that separates the effects of violence from those of community- and individual-level confounders more effectively than past research. Temporal patterning was removed from community violence rates and disease rates in each place using a Kalman smoother, resulting in residual rates. We used linear regression with place fixed-effects to examine within-place associations of acute changes in community violence with residual rates of each outcome, controlling for local time-varying covariates.

RESULTS

We found acute increases in hospital visits and deaths due to anxiety disorders (0.31 per 100,000; 95% confidence interval [CI] = 0.02, 0.59), substance use (0.47 per 100,000; 95% CI = 0.14, 0.80), asthma (0.56 per 100,000; 95% CI = 0.16, 0.95), and fatal acute myocardial infarction (0.09 per 100,000; 95% CI = 0.00, 0.18) co-occurring with violence spikes. The pattern of findings was similar for the exposure of continuous acute violence changes.

CONCLUSIONS

Although the associations were small, the identified increases in stress-responsive conditions suggest the possibility of health impacts of acute changes in community violence.

Mobility limitation as determinant of primary care use and ambulatory care sensitive conditions

PURPOSE

First, to investigate associations between mobility limitations and use of general practitioners and hospitalizations of acute care sensitive conditions, respectively. Second, to investigate whether these associations vary by socio-demographic factors.

METHODS

The study included 3574 females and males aged 75 or 80 years from the Danish Intervention Study on Preventive Home Visits. Fixed-effects logistic and poisson regression models were applied to study the relationship between mobility limitations (measured two-four times) and general practitioner consultations and hospitalizations with acute care sensitive conditions each subsequent year, respectively.

RESULTS

Each additional mobility limitation was associated with 15% higher odds of general practitioner home consultation (Odds ratio 1.15, 95% CI 1.07;1.23) and 4% increased incidence rate of general practitioner consultations among those with ≥ 1 consultation (Incidence rate ratio 1.04, 95% CI 1.03;1.04). There were no associations between mobility limitations and whether older adults had at least one general practitioner consultation nor acute care sensitive condition hospitalization. Test of interactions (p < 0.03) showed that more mobility limitations were associated with greater incidence rate of general practitioner consultations among males compared to females, married compared to unmarried, and older adults with high compared to low financial assets.

CONCLUSIONS

Older adults with more mobility limitations had more often a general practitioner home consultation. Mobility limitations were not associated with whether older adults had at least one general practitioner consultation, but increased mobility limitations were associated with higher contact rate among those who had ≥ 1 consultation, especially among males and older adults who were married or had high financial assets.

Socioeconomic status indicators and common mental disorders: Evidence from a study of prenatal depression in Pakistan

There is growing interest in the relationship between socioeconomic status (SES), poverty, and mental health in low and middle-income countries (LMIC). However, it is not clear whether a gradient approach focused on a wider SES distribution or a binary poverty approach is more salient for mental health in LMIC. Yet this distinction has implications for interventions aimed at improving population health. We contribute to the literature by examining how multiple indicators of socioeconomic status, including gradient SES and binary poverty indicators, contribute to prenatal depression symptoms in a LMIC context. Prenatal depression is an important public health concern with negative sequela for the mother and her children. We use data on assets, education, food insecurity, debt, and depression symptoms from a sample of 1154 pregnant women residing in rural Pakistan. Women who screened positive for depression participated in a cluster randomized controlled trial of a perinatal depression intervention; all women were interviewed October 2015-February 2016, prior to the start of the intervention. Cluster-specific sampling weights were used to approximate a random sample of pregnant women in the area. Findings indicate that fewer assets, experiencing food insecurity, and having household debt are independently associated with worse depression symptoms. The association with assets is linear with no evidence of a threshold effect, supporting the idea of a gradient in the association between levels of SES and depression symptoms. A gradient was also initially observed with woman’s educational attainment, but this association was attenuated once other SES variables were included in the model. Together, the asset, food insecurity, and debt indicators explain 14% of the variance in depression symptoms, more than has been reported in high income country studies. These findings support the use of multiple SES indicators to better elucidate the complex relationship between socioeconomic status and mental health in LMIC.

•

We compare the association between poverty, gradient measures of SES, and depression.

•

Above the poverty level, SES predict depression in a low income country.

•

Relying on simple, dichotomous, poverty measures in studies of mental health is not recommended.

Absolute rather than relative income is a better socioeconomic predictor of chronic obstructive pulmonary disease in Swedish adults

BACKGROUND

While psychosocial theory claims that socioeconomic status (SES), acting through social comparisons, has an important influence on susceptibility to disease, materialistic theory says that socioeconomic position (SEP) and related access to material resources matter more. However, the relative role of SEP versus SES in chronic obstructive pulmonary disease (COPD) risk has still not been examined.

METHOD

We investigated the association between SES/SEP and COPD risk among 667 094 older adults, aged 55 to 60, residing in Sweden between 2006 and 2011. Absolute income in five groups by population quintiles depicted SEP and relative income expressed as quintile groups within each absolute income group represented SES. We performed sex-stratified logistic regression models to estimate odds ratios and the area under the receiver operator curve (AUC) to compare the discriminatory accuracy of SES and SEP in relation to COPD.

RESULTS

Even though both absolute (SEP) and relative income (SES) were associated with COPD risk, only absolute income (SEP) presented a clear gradient, so the poorest had a three-fold higher COPD risk than the richest individuals. While the AUC for a model including only age was 0.54 and 0.55 when including relative income (SES), it increased to 0.65 when accounting for absolute income (SEP). SEP rather than SES demonstrated a consistent association with COPD.

CONCLUSIONS

Our study supports the materialistic theory. Access to material resources seems more relevant to COPD risk than the consequences of low relative income.

[Causal inference in epidemiology]

This essay makes a brief account of the historical development of epidemiology as a fundamental element for understanding the development of thought and causality models. Subsequently, the theoretical foundations that support the identification of causal relationships and the available models and methods of analysis are exposed, providing some examples of their application. Finally, the strengths and limitations of this epidemiological analysis during the identification of causal relationships are presented.

Intersectionality and risk for ischemic heart disease in Sweden: Categorical and anti-categorical approaches

Intersectionality theory can contribute to epidemiology and public health by furthering understanding of power dynamics driving production of health disparities, and increasing knowledge about heterogeneities within, and overlap between, social categories. Drawing on McCall, we relate the first of these potential contributions to categorical intersectionality and the second to anti-categorical intersectionality. Both approaches are used in study of risk of ischemic heart disease (IHD), based on register data on 3.6 million adults residing in Sweden by 2010, followed for three years. Categorical intersectionality is here coupled with between-group differences in average risk calculation, as we use intersectional categorizations while estimating odds ratios through logistic regressions. The anti-categorical approach is operationalized through measurement of discriminatory accuracy (DA), i.e., capacity to accurately categorize individuals with or without a certain outcome, through computation of the area under the curve (AUC). Our results show substantial differences in average risk between intersectional groupings. The DA of social categorizations is found to be low, however, due to outcome variability within and overlap between categories. We argue that measures of DA should be used for proper interpretation of differences in average risk between social (or any other) categories. Tension between average between-group risk and the DA of categorizations, which can be related to categorical and anti-categorical intersectional analyses, should be made explicit and discussed to a larger degree in epidemiology and public health.