Details
Nearly 98% of medications lack safety evidence for pregnant patients and their offspring, leading to knowledge gaps and unmet need in pregnancy treatment and prevention. Post-marketing safety studies are crucial, but safety signal identification studies are in their infancy. TreeScan, a data-mining method, can help identify potential medication safety signals by evaluating thousands of outcomes simultaneously. However, the study aimed to evaluate sample size parameters for signal identification studies using the Poisson model with tree-based scan statistics. A maternal outcome tree was created using 1398 pregnancy-related maternal health complication codes from the International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) Chapter 15. The Poisson TreeScan analysis was used to compare observed and expected outcomes. Random datasets were generated using a Monte Carlo simulation process, and the log-likelihood ratio of each outcome was assigned a Monte Carlo-based p-value. The study used a prior empirical study to obtain background incidence proportions for maternal ICD-10-CM diagnoses. The power of TreeScan to detect a statistical alert depends on three parameters: sample size of the exposure group, incidence proportion for the outcome, and magnitude of the effect size. The study used Poisson TreeScan to analyze maternal outcomes in undersized control groups. The base case involved true background incidence proportions, and 1000 iterations were run. The study also used imprecise background incidence proportions to evaluate the effect of undersized control group populations. The results were compared and quantified.