Accuracy of the Audio Computer Assisted Self Interview version of the Alcohol, Smoking and Substance Involvement Screening Test (ACASI ASSIST) for identifying unhealthy substance use and substance use disorders in primary care patients

Background An Audio Computer-assisted Self Interview (ACASI) version of the Alcohol, Smoking, and Substance Involvement Screening Test (ASSIST) could reduce barriers to substance use screening and assessment in primary care settings. This study evaluated the diagnostic accuracy of an ACASI ASSIST for identification of unhealthy substance use and substance use disorders (SUD). Methods 399 adult patients were consecutively recruited from an urban safety-net primary care clinic. ACASI ASSIST scores for tobacco, alcohol, marijuana, and cocaine were compared against reference standard measures to assess the instrument's diagnostic accuracy for identifying unhealthy use and SUD, first using empirically-derived optimal cutoffs, and second using the currently recommended ASSIST cutoffs. Results For identifying any unhealthy use, at the empirically-derived cutoffs the ACASI ASSIST had 93.6% sensitivity and 85.8% specificity (AUC = 0.90) for tobacco, 85.9% sensitivity and 60.3% specificity (AUC = 0.73), for alcohol in men, 100% sensitivity and 62.4% specificity (AUC = 0.81) for alcohol in women, 94.6% sensitivity and 81.6% specificity (AUC = 0.88) for marijuana, and 86.1% sensitivity, 84.0% specificity (AUC = 0.85) for cocaine. For SUD, sensitivity ranged from 79% (for alcohol in males), to 100% (for tobacco), and specificity was 83% or higher (AUCs ranged 0.83–0.91). For substances other than tobacco, empirically-derived cutoff scores were lower than the standard cutoffs, and resulted in higher sensitivity and lower specificity for identifying unhealthy substance use. Conclusions The ACASI ASSIST is a valid measure of unhealthy use and SUD for substances that are commonly used by primary care patients, and could facilitate effective and efficient screening for substance use in medical settings.