Semiparametric Panel Count Model, With Applications to Signal Detection in Post-Market Drug Surveillance Systems

Panel count data occurs in a wide variety of applications, ranging from biomedical research to business, such as the number of accidents, product defects, and insurance claims. For such data under the FDA investigation, millions of reported adverse events (AEs) associated with thousands of drugs are monitored in the post-market drug safety surveillance systems worldwide. Evaluating the AEs of the associated drugs is an important public health concern and motivates our method. One statistical challenge in such systems is handling the excessive number of zero AE counts. Most existing methods utilize Poisson count models that cannot incorporate covariates nor account for the excessive zero counts adequately. This article proposes a novel semiparametric nonhomogeneous panel count model to detect AE signals by accounting for covariates, background AE occurrences, and excessive zero counts. The model is estimated using the Expectation-Maximization (EM) algorithm iteratively, where in each M-step, the maximization of the nonparametric component is reformulated as an optimization problem, as in the isotonic regression. The strong consistency and the asymptotic distributions of the estimators are formally derived. We conduct extensive simulation studies to evaluate the finite sample performance of the proposed method and to demonstrate the apparent advantage of the proposed method in signal detection with high power, high specificity, and sensitivity. We apply the method to a VigiBase dataset to detect the AE signals as an application of the proposed method.