Fig. 2. Overall accuracy predicting treatment outcome at 90 days.

We categorized patients into categories of abstinent, relapse, and dropout using only information available at baseline: the digital phenotype (DP) and the addiction severity index (ASI; the clinical measure), each alone, as well as combined (DP + ASI). ASI scores were from treatment intake, language was from two years before treatment until intake. AUC scores were evaluated out-of-sample using a 10-fold cross-validation of predictions from a random forest model. Difference between ASI and DP and ASI and DP + ASI was significant (single-tailed permutation test, p < 0.001). Difference between DP and DP + ASI was not significant. See Supplementary Table S1 for sensitivity and specificity.