Zero-inflated Poisson model based likelihood ratio test for drug safety signal detection

Abstract

In recent decades, numerous methods have been developed for data mining of large drug safety databases, such as Food and Drug Administration's (FDA's) Adverse Event Reporting System, where data matrices are formed by drugs such as columns and adverse events as rows. Often, a large number of cells in these data matrices have zero cell counts and some of them are "true zeros" indicating that the drug-adverse event pairs cannot occur, and these zero counts are distinguished from the other zero counts that are modeled zero counts and simply indicate that the drug-adverse event pairs have not occurred yet or have not been reported yet. In this paper, a zero-inflated Poisson model based likelihood ratio test method is proposed to identify drug-adverse event pairs that have disproportionately high reporting rates, which are also called signals. The maximum likelihood estimates of the model parameters of zero-inflated Poisson model based likelihood ratio test are obtained using the expectation and maximization algorithm. The zero-inflated Poisson model based likelihood ratio test is also modified to handle the stratified analyses for binary and categorical covariates (e.g. gender and age) in the data. The proposed zero-inflated Poisson model based likelihood ratio test method is shown to asymptotically control the type I error and false discovery rate, and its finite sample performance for signal detection is evaluated through a simulation study. The simulation results show that the zero-inflated Poisson model based likelihood ratio test method performs similar to Poisson model based likelihood ratio test method when the estimated percentage of true zeros in the database is small. Both the zero-inflated Poisson model based likelihood ratio test and likelihood ratio test methods are applied to six selected drugs, from the 2006 to 2011 Adverse Event Reporting System database, with varying percentages of observed zero-count cells.

Signal detection on a patient cohort: A disproportionality analysis of the ANRS CO22 HEPATHER cohort to identify associations between direct acting antivirals and adverse events in patients with hepatitis C virus chronic infection

“…Although the gold standard for identifying and quantifying frequent ADRs is clinical trials, the sample sizes of trials are usually too small to detect ADRs when the frequencies of adverse events (AEs) following treatment are rare. To address this issue, spontaneous pharmacovigilance reporting systems have been set up to record incidental serious or unexpected AEs, and numerous analytical methods have been developed to link these events with current or past medication intake . Analyses of healthcare databases with thousands of patients exposed to a drug, eventually compared with thousands of unexposed patients, also allow the identification of associations between a drug and an AE, ie, “a signal.” Among various signal detection methods, disproportionality (DP) methods are commonly applied to spontaneous reporting systems or healthcare databases .…”

Signal detection in FDA AERS database using Dirichlet process

“…In addition, we notice that there is a large number of zero‐count cells in the post‐market safety databases. For example, the percentage of zero‐count cells by drug for the 2006–2011 AERS data ranges from 50% to 99.99% . The regular Poisson model is not suitable for modeling the data with a large number of zeros because some of the zeros are not from Poisson distribution but are true zeros (the cells that will never have counts bigger than 0).…”

“…The regular Poisson model is not suitable for modeling the data with a large number of zeros because some of the zeros are not from Poisson distribution but are true zeros (the cells that will never have counts bigger than 0). To address this problem, a zero‐inflated Poisson (ZIP) LRT method has been developed as a frequentist approach using the estimation–maximization (EM) algorithm. Here, we also propose ZIP‐sB and ZIP‐DP methods as Bayesian approaches for data with a large proportion of zero counts.…”

Propensity score‐adjusted three‐component mixture model for drug‐drug interaction data mining in FDA Adverse Event Reporting System

“…10 On the other hand, the likelihood ratio test (LRT) and its extension zero-inflated Poisson model-based LRT (ZIP-LRT) are able to test the associations between an ADE and many drugs. 11,12 The DPA method has also been used for investigating the DDI-induced ADEs by comparing the observed report frequency of a drug-drug-ADE combination to its baseline frequency assuming no interactions between two drugs. By considering the drug-drug pair as a "new drug", the observed report frequency can be obtained from the SRS database.…”