(calculated within each ancestry group). The cluster option was used to account for the presence of related individuals in the samples by generating robust variance estimates. OD and CD TD analyses were performed separately and without considering an individual’s status on exposure to or dependence on the other drug. Analyses in the Yale-Penn and COGA samples were stratified by population and genotyping chip (in Yale-Penn) and all results were combined via inverse variance weighted meta-analysis using METAL (https://genome.sph.umich.edu/wiki/METAL) [41]. Analyses were stratified by cohort and ancestry (where applicable) and combined with inverse variance weighted meta-analysis. Variants with P-values < 1.0 × 10−5 were tested in the replication samples. The proportional hazards assumption was checked by verifying that the Schoenfeld residuals for each term in the model were independent of time. Both age at interview and age at fist cocaine/opioid use were significantly associated with TD in Yale-Penn but both violated the proportional hazards assumption and were not included in the final analysis in Yale-Penn. The top results were very similar with and without each age term in the model, but including an age by time interaction term attenuated them substantially.
