OR-2684 - Treescan For Active Surveillance Of The Safety Of Medications Used In Pregnancy

Krista F Huybrechts, Martin Kulldorff, Sonia Hernandez-Diaz, Brian T Bateman, Yanmin Zhu, Helen Mogun and Shirley V Wang

Background: When new drugs enter the market, there is little to no information about safety during pregnancy. Most post-market studies of drug safety in pregnancy focus on a single or selected outcomes. However, to make informed treatment decisions, pregnant women must balance the benefit of treatment against all possible adverse effects.

Objectives: To develop and evaluate tree-based scan statistic (TreeScan) methods as a safety surveillance approach that allows simultaneous evaluation of a comprehensive range of adverse pregnancy outcomes, while accounting for multiple testing and preserving the overall false positive rate.

Methods: We evaluated TreeScan with a cohort design and adjustment via propensity score matching and fine-stratification. We used two positive control test cases of drugs known to be associated with specific adverse pregnancy outcomes: 1) prescription opioids late in pregnancy and neonatal abstinence syndrome, and 2) valproate early in pregnancy and congenital malformations. These test cases were implemented in linked mother-infant pregnancy cohorts from the Medicaid Analytic eXtract (2000-2014; N=1,943,652) and the IBM MarketScan Research Database (2003-2015; N=1,251,451). We scanned across incident outcomes defined by hierarchical groupings of clinically related International Classification of Disease 9^th version codes. Statistical alerts were defined by a threshold of p<0.05. However, because adverse pregnancy outcomes tend to be rare, and multiplicity adjustment decreases power, outcomes with large magnitude of effect were also interpreted from a clinical and methodologic perspective.

Results: In both test cases, we identified known safety concerns without generating numerous false positives. For opioids, out of >8,000 outcomes evaluated, there was only 1 statistical alert, which was for the known adverse event of drug withdrawal in the newborn. For valproate, there were statistical alerts for several known associations, including spina bifida, several cardiac malformations, hypospadias, and polydactyly after scanning close to 700 hierarchically defined malformation outcomes. The alert threshold of p<0.05 was not reached for oral clefts, but a relative risk of 2 was observed. There were no alerts for other safety concerns.

Conclusions: This evaluation shows the promise of TreeScan based approaches for systematic drug safety monitoring in pregnancy. Such a targeted screening approach followed by deeper investigation to refine understanding of potential signals will ensure pregnant women and their physicians have access to the best available evidence to inform treatment decisions.