Causal Inference in Epidemiology: Estimating the average treatment effect of high education on hypertension prevalence using datafrom the Västerbotten Intervention Programme

Sammanfattning

An inverse relationship between educational attainment and hypertension is well documented, but the causal effect remains unclear due to, for example, unmeasured confounding and the difficulty of defining education as a well defined exposure. The goal of this study is to estimate the average treatment effect (ATE) of high versus low educational attainment on hypertension prevalence among 50 year-olds in Västerbotten County, within the potential outcomes framework. Data was drawn from the Västerbotten Intervention Programme and restricted to 50 year-olds examined between 1995 and 2000 (n = 6,150). Hypertension was defined as SBP ≥ 140 mmHg or DBP ≥ 90 mmHg. Propensity scores were estimated by logistic regression as a function of the participants sex, country of birth, family history of cardiovascular disease and family history of diabetes. Stabilized inverse probability weighting was used to estimate the prevalence difference, prevalence ratio and odds ratio. Confidence intervals were obtained by non-parametric bootstrap (B = 10,000). Two sensitivity analyses were conducted. After IPW adjustment, hypertension prevalence was 28.0% in the high education group and 34.6% in the low education group. The estimated prevalence difference was -6.6 percentage points (95% CI: -9.1; -4.3), the prevalence ratio was 0.81 (95% CI: 0.75; 0.87) and the odds ratio was 0.73 (95% CI: 0.65; 0.82). Covariate balance was achieved with all absolute SMDs below 0.005. Results were robust across both sensitivity analyses. We thus conclude that high educational attainment was associated with lower hypertension prevalence after adjustment for measured baseline covariates. We are, however, careful about the causal interpretation of the estimate due to factors such as unmeasured confounding and the difficulty of clearly defining education as an intervention.