Evaluating Effects of Various Exposures on Mortality Risk of Opioid Use Disorders with Linked Administrative Databases

Administrative health records provide a rich source of information pertaining to various exposures, many of which are time-varying in nature. When internal time-varying covariates are included in a Cox regression model, likelihood-based inference procedures are no longer applicable to infer model parameters (Kalbfleisch and Prentice in The Statistical analysis of failure time data, Wiley, New York, 2002). Motivated by the ongoing opioid epidemic, we summarize an individual’s opioid agonist treatment (OAT) dispensation history and additional exposures with (i) a model-based summary, or (ii) its functional principal component scores. We show that the OAT dispensation proportion has a non-linear effect on the mortality hazard over time, and a significant interaction with time of birth. Particularly a clear protective effect against mortality for Millennials and Generation Z is revealed. Our approach is easy to implement by virtually any statistical software, and provides a risk assessment tool for utilizing available health records.