Optimization of Signal-Detection Methods in Novel Databases: a Focus on Direct-Acting Antivirals

Abstract

An adverse drug event (ADE) is an injury resulting from medical intervention related to a drug, and they include medication errors, adverse drug reactions, allergic reactions, and overdoses. In 2019, the Food and Drug Administration (FDA) identified the following outcomes as potential signals of serious risks associated with direct-acting antivirals (DAAs): angioedema, dysglycemia, and hepatic decompensation and hepatic failure. The risk of angioedema or dysglycemia while on DAA therapy was identified through post-market surveillance, and prior studies did not include these two outcomes and had mixed results for the risk of hepatic decompensation. The Food and Drug Administration (FDA) Adverse Event Reporting System (FAERS) is a database containing ADE reports that were submitted to the FDA. The reports need to be investigated, and when working with large databases, signal detection algorithms (SDA) are used to decide which reports are significant. Signal detection algorithms are statistical methods used to measure disproportionality, which quantifies unexpectedness. The purposes of SDAs are 1) flag potential signals that might be missed, 2) prioritize resources for signal detection and clinical review on most likely candidates, 3) detect more complex patterns in the data which are harder to detect via manual review, and 4) prioritization of potential signals. This dissertation will use the newly identified potential DAA signals to investigate and increase the efficiency of SDAs. It will assess the effects of subgroup analysis on SDAs and develop predictive models using FAERS data; and evaluate the association of DAA exposure and identified potential signals in a real-world population of commercially insured patients form the Merative MarketScan Research Databases. Subgroup analysis can address confounding and further classify signals to be investigated. Logistic regression-based signal detection algorithms are superior to disproportionality analysis due to their ability to account for potential confounders and masking factors. It is important to assess and compare the performance of other machine learning algorithms to logistic regression. Lastly, there is interest in pharmacovigilance and pharmacoepidemiology in using real-world data as a complementary data stream to spontaneous report databases. The current study used medical and pharmacy claims from the Merative MarketScan Commercial Databases to evaluate and validate the possible risks of angioedema, dysglycemia, and hepatic decompensation associated with DAAs identified through post-market surveillance.