Abstract: Objective To establish the model of Bayesian instrumental variable analysis in the active surveillance data of adverse drug reactions for controlling unmeasurable confounding and acquiring the accurate causal relation between the drug and adverse reaction. Methods Hamilton Markov Chain Monte Carlo method was used to perform data simulation and parameter estimation. Further, the established model was compared with traditional models on bias and accuracy to assess the performance of different methods. Results Bayesian instrumental variable analysis performed well and was the optimal method under the small sample, weak instrumental variable, strong unmeasurable confounding and rare treatments and outcomes. Conclusion Bayesian instrumental variable analysis could improve bias and accuracy compared with traditional instrumental variable methods in active surveillance data of adverse drug reactions.

Key words: Bayesian instrumental variable, adverse drug reaction, active surveillance