Professor University of Texas Medical Branch Galveston, United States
Background: Utilization of administrative claims data as a complement to current pharmacovigilance practice has been limited, despite the potential to enhance timeliness of rare adverse event detection. One of the main limiting issues has been that most data mining methods cannot account for confounding. Sequence Symmetry Analysis uses a within subject design and is thus robust to time-invariant confounding.
Objectives: . In this study we aim to validate the use of Sequence Symmetry Analysis on large administrative claims data, by evaluating signals of adverse events after long-term opioid therapy initiation.
Methods: We used Texas-Medicare data to identify adults with at least 3 years continuous enrollment in Medicare part D, that initiated prescriptions for opioids lasting >= 90 consecutive days between 2018-2019. The opioid initiation date was set as the index date. All patients that already utilized opioids in 2017 or within the first 12 months of data coverage were excluded, in order to identify incident opioid users. Prescription sequence symmetry analysis (PSSA) was performed to explore association between opioid exposure and related adverse events treated by marker drugs. The observation period for sequences of incident marker drug prescriptions was limited to 12 months before and after the index date. Marker drugs were categorized based on the Anatomical Therapeutic Chemical (ATC) Classification System. The sequence symmetry ratio (SSR) was adjusted for temporal prescription patterns and 95% confidence intervals were constructed based on the binomial distribution.
Results: We identified a total of 6,625 incident opioid users, who were collectively prescribed incident marker drugs belonging to 145 distinct ATC classes. We found statistically significant (p-value < 0.05) signals of increased post-opioid prescriptions for propulsives, antiemetics, laxatives, cough suppressants, other analgesics and antipyretics, and drugs used in addictive disorders (e.g., naltrexone). Medications related to infections (e.g., tetracyclines and intestinal anti-infectives) were significant at the 0.1 level. No unexpected signals were found.
Conclusions: Medicare data can be a valid source of post-marketing drug surveillance. Although we found some anticipated signals for opioid adverse effects, further validation of the performance of PSSA on this dataset will be performed by clinical review of the findings and calculation of positive and negative predictive values. We will also explore the use of diagnosis codes in lieu of prescription medications for sequence symmetry analysis in this dataset.
