A theoretical model for detecting drug interaction with awareness of timing of exposure

Drug-drug interaction-induced (DDI-induced) adverse drug event (ADE) is a significant public health burden. Risk of ADE can be related to timing of exposure (TOE) such as initiating two drugs concurrently or adding one drug to an existing drug. Thus, real-world data based DDI detection shall be expanded to investigate precise adverse DDI with a special awareness on TOE. We developed a Sensitive and Timing-awarE Model (STEM), which was able to optimize the probability of detection and control false positive rate for mining all two-drug combinations under case-crossover design, in particular for DDIs with TOE-dependent risk. We analyzed a large-scale US administrative claims data and conducted performance evaluation analyses. We identified signals of DDIs by using STEM, in particular for DDIs with TOE-dependent risk. We also observed that STEM identified significantly more signals than the conditional logistic regression model-based (CLRM-based) methods and the Benjamini-Hochberg procedure. In the performance evaluation, we found that STEM demonstrated proper false positive control and achieved a higher probability of detection compared to CLRM-based methods and the Benjamini-Hochberg procedure. STEM has a high probability to identify signals of DDIs in high-throughput DDI mining while controlling false positive rate, in particular for detecting signals of DDI with TOE-dependent risk.