For all primary analyses, time was treated as a continuous variable. To assess changes in the strength of associations between POUD and illegal behaviors over time while controlling for general population trends, each of the seven illegal behaviors was modeled using multivariable logistic regression. Unadjusted analyses modeled each illegal behavior as a function of POUD status, time, and an interaction term POUD status*time. Adjusted analyses controlled for covariates. The trend difference between POUD and No POUD groups for each illegal behavior was captured by the interaction term POUD status*time, which measured the adjusted interaction OR (AIOR). Thus, the AIOR represented the total proportional change over the 13-year time period in the odds ratio describing the association between POUD and the illegal behavior.
