Abstract
The Addiction Severity Index (ASI) is a reliable and valid measure of problem severity among addicted patients. Concerns have been raised about the reliability of the Interviewer Severity Rating (ISR), a summary score for each of 7 domains. As part of an effort to build a computer-administered ASI, regression equations were developed to predict the ISR. Repeated resampling of a large dataset, consisting of 1,124 ASIs conducted by trained interviewers, permitted derivation of stable regression equations predicting the ISR for each ASI domain from patients' answers to preselected interview items. The resulting 7 Predicted Severity Ratings (PSRs) were tested on 8, standardized vignettes, with "gold standard," expert-generated ISRs. Reliabilities compared well with those of intensively trained interviewers. The PSRs could provide an alternative to potentially unreliable interviewer ratings, enhancing the ASI's role in treatment planning and treatment matching and make possible a computer-administered version of the ASI.
