Associate Professor The Ohio State University The Ohio State university Columbus, United States
Background: Polypharmacy is very high and is associated with higher risks of adverse drug events (ADEs) among older adults. Knowledge on the ADE-risk level of exposure to different drug combinations is critical for safe polypharmacy practice yet approaches for this type of knowledge discovery are limited.
Objectives: To apply an innovative data mining approach to discover both high-risk high-order drug combinations (3- and 4-drug combinations).
Methods: A cohort of older adults (≥65 years) who visited an ED were identified from Medicare fee-for-service (FFS) and MarketScan Medicare Supplemental data. A nested case-control study design was used to implement the analysis. First, we defined cases as ADEs potentially induced by anticoagulants, antidiabetic agents, or opioids that were recorded during the first observed ED visit per patient – thus, three sets of cases were created. For each ADE case, we selected up to 50 controls from the pool of patients without an ADE during the ED visit. Cases and controls were matched on age (±5 years) and gender. We defined a 30-day risk window prior to the ED visit date for measuring the concurrent exposure to ≥3 unique drugs; concurrent exposure was operationalized based on overlap of prescription fill dates and the day’s supply within the risk window, thus patients were required to be continuously enrolled during the risk window. Second, we used the mixture drug-count response model (MDRM) to mine drug combinations associated with an increased risk of ADE. Under the MDRM, we defined the baseline risk as the ADE risk of using a single drug. We assumed the ADE risk of a high-order drug combination was either similar to the baseline risk or higher. We included 5 parameters characterizing constant risk, drug-count response risk, and probabilities to follow drug-count risk for two-drug combinations, three-drug combinations, and four-drug combinations. After clearly specifying the null distribution, we were able to identify adverse high-order drug combinations at a low false positive rate (FDR) under the empirical Bayesian framework.
Results: For brevity, we present the results from the Medicare FFS sample only. Of the 5.1 million ED visits, 148098, 124194 and 146245 were due to anticoagulant, antidiabetic and opioid-induced ADEs, respectively. Warfarin was most frequently involved in 48.4% and 73.9% of the observed 3- and 4-drug combinations, respectively, for the anticoagulant-induced ADE set. The most frequent drug classes involved with both 3- and 4-drug combinations for both the antidiabetic drug-induced and opioid-induced ADEs were: anticholinesterases, other anti-dementia drugs, digitalis glycosides, selective serotonin reuptake inhibitors, and thyroid hormones.
Conclusions: We have successfully demonstrated the application of a data-mining technique for the discovery of high-order drug combinations associated with specific drug-induced ADEs. This approach can be scaled to investigate other ADEs, especially serious ADEs that result in ED visits.
