Understanding the role of hypertension in stroke outcomes using Bayesian analysis

Area of Science:

• Neurology
• Biostatistics
• Epidemiology

Background:

• Comorbid hypertension has been linked to survival benefits in stroke patients.
• The phenomenon of reverse epidemiology in stroke requires careful examination for potential biases.

Purpose of the Study:

• To investigate the impact of strong prior assumptions on hypertension's reverse association in stroke.
• To assess the sensitivity of reverse epidemiology findings to bias assumptions using Bayesian methods.

Main Methods:

• Analysis of stroke data (2014-2019, N=177,947) with random sampling.
• Bayesian multiple logistic mixed-effects regression modeling under three scenarios: informative priors, non-informative priors, and interaction effects (age*hypertension).
• Sensitivity analyses were conducted to evaluate the robustness of estimates to different prior choices.

Main Results:

• Elevated odds ratios (ORs) for hypertension were observed in small sample sizes (100-500) regardless of prior type.
• As sample size increased, ORs for hypertension decreased, plateauing below 1 for samples >5000.
• The age*hypertension interaction term showed an increasing effect with larger sample sizes, suggesting the reverse effect of hypertension diminishes with age.

Conclusions:

• Bayesian analysis can aid interpretation of reverse associations with limited data, but its influence wanes in large datasets.
• The observed patterns suggest reverse associations are more likely due to collider or selection bias than prior specification.
• Bayesian priors alone cannot fully mitigate design bias in large-scale epidemiological studies.