Understanding the role of hypertension in stroke outcomes using Bayesian analysis.

A comorbid hypertension was previously associated with survival advantages in patients with stroke. We aimed to explore how strong priors for the hypertension covariate affect the reverse association, as a way to test the sensitivity of reverse epidemiology findings to bias assumptions. The authors used stroke data from 2014 to 2019 (N = 177,947) and subsequently performed random sampling from a population of various sizes. The data were analyzed using Bayesian multiple logistic mixed-effects regression, which was further modelled in three scenarios: with informative (strong) priors, non-informative priors, and accounting for the interaction mechanism (age*hypertension). In addition, we perform a series of sensitivity analyses to check the robustness of the estimates to different prior choices. Both informative and non-informative priors demonstrated elevated posterior odds ratios (ORs) for hypertension in low sample fractions (n = 100-500). As the sample size increased, the ORs declined (below 1) for each subsequently larger samples. The ORs plateaued as the sample exceeded 5000 and became similar for both the modeling scenarios. Conversely, the interaction term revealed inverse patterns, increasing in effect as the sample size grew large. Thus, the reverse effect of hypertension diminishes with age. Although further modifications of prior precision revealed somewhat higher ORs for hypertension covariate, the estimates mostly overlapped. Bayesian analysis may improve the interpretation of reverse associations when data are limited; however, in large datasets, their influence diminishes. This pattern suggests that reverse associations reflect collider or selection bias rather than prior choice and that Bayesian priors alone cannot address design bias.