< 0.001) in models with only PRS as features (Supplementary Fig. S1). No difference was found in the females’ groups between the two types of ancestry definition. The AA male group combined feature model of PRS, EEG-FC, marital status, and employment status achieved the highest accuracy of 86.04% (specificity = 85.83%, sensitivity = 86.257%, AUC = 0.97). The AA female group combined feature model of PRS, EEG-FC, and depression medication also achieved high accuracy of 85.43% (specificity = 80.66%, sensitivity = 86.19%, AUC = 0.9). The models of EA male and female groups achieved AUC >0.74 for the model with combined features of PRS, EEG-FC, and medication (accuracy 64.96%, EA males) and PRS, EEG-FC, and marital status (accuracy 63.60%, EA females) (Table 1). Adding discriminatory features to the models increased accuracy, specifically EEG-FC was the most discriminative feature category for all groups (p < 0.001).Table 1Selected models predicting AUD remission stratified by ancestry and sex.Model [# features]Specificity (%)STDSensitivity (%)STDAccuracy (%)STDAUCSTDEA male PRS, EEG, Other & sleep meds874.960.554.821.864.960.90.740.0EA female PRS, EEG, Marital status1064.382.062.7912.363.601.30.770.0AA male PRS, EEG, Marital, Employment status885.836.886.256.486.045.40.970.0AA female PRS, EEG, Depression meds780.664.990.4763.185.433.20.980.0Values are means ± standard deviation (STD). Alcohol Use Disorder (AUD), European Ancestry (EA), African ancestry (AA). Meds—Medication.
