Detecting signals of opioid analgesic abuse: application of a spatial mixed effect poisson regression model using data from a network of poison control centers.

PURPOSE

The recent rise in the non-medical use of opioid analgesics in the US has underscored the importance of comprehensive post-marketing surveillance of these products. To assist pharmacovigilance efforts, we developed a methodology for detecting geo-specific "signals" of potential outbreaks of prescription drug abuse by 3-digit ZIP (3DZ) code.

METHODS

The number of intentional exposure calls involving nine specific opioid analgesics were obtained from eight regional poison control centers between first quarter 2003 and fourth quarter 2004. The unit of analysis was a combination of drug-quarter/year-3DZ. We fitted an empirical Bayes mixed effects Poisson-Gamma regression model that adjusted for differences across 3DZs in opioid analgesic exposure. A relative report rate (RR) >or=3 at a probability of >0.95 was the signal threshold criterion.

RESULTS

A total of 15,769 valid drug-time-3DZ combinations were identified. Of these, 1.9% (n = 294) met the signal threshold criterion. The number of signals generated per drug-quarter/year-3DZ combination ranged from 0 to 13. The largest number of signals were those involving methadone (n = 71), hydrocodone (n = 57), and branded oxycodone extended-release (n = 45). Signals for methadone and branded oxycodone extended-release were predominantly clustered in Appalachia. Hydrocodone-related signals showed less geographic clustering with approximately 26% reported from California, and the remainder from other regions in the US.

CONCLUSIONS

Our results show marked regional differences in reported abuse of specific opioid analgesics. Additional research is needed to determine the sensitivity and specificity of signals obtained using this spatial mixed effect Poisson regression model.