Bridging Disciplinary Boundaries (January 11 - 14, 2007)

Purpose: Epidemiological studies have well established that individuals in the criminal justice system have serious mental illnesses (SMI) at rates that exceed those in the general population. Unfortunately, many if not most individuals with SMI do not get psychiatric treatment while under criminal justice supervision. Although many factors likely underlie the gap between need and services, a contributing issue has been reliance on non-standardized screening tools based largely on self-reported psychiatric treatment history as the main indicator for further assessment and treatment. The accuracy of using psychiatric treatment indicators either alone or in conjunction with a standardized screening tool, however, has not yet been determined. This study investigated the accuracy of a recently developed screen – the K6 – for detecting SMI in criminal justice populations and sought to assess the affects of combining the K6 with indicators for recent psychiatric treatment on overall screening accuracy.

Methods: We analyzed data from the 2001 and 2002 National Household Survey on Drug Use and Health (NSDUH) for participants 18 or older, who reported being arrested or on probation or parole in the past year(N = 4,884). The gold standard for SMI was results from the Composite International Diagnostic Interview Short Form (CIDI-SF), which was embedded in the NSDUH surveys in 2001 and 2002. We aggregated the CIDI-SF results for 8 past-year psychiatric disorders into a single category indicating any past-year DSM-IV disorder. Using logistic regression models, we then regressed the CIDI-SF diagnostic results on four sets of predictors: 1) scores on the K6 alone, which was also embedded in the NSDUH questionnaires; 2) K6 scores plus indicators for any past-year inpatient or outpatient psychiatric treatment; 3) the K6 plus any psychiatric medication in the past year; and 4) the K6 plus any psychiatric treatment or medication in the past year. We then compared the receiver operating characteristic curves area under curves (ROC-AUC) for the predicted scores for each model. Finally, we generated sensitivity, specificity, LR+, and LR- statistics for each model using the K6 scores dichotomized at the recommended threshold (>= 13) with and without the psychiatric treatment indicators. Analyses were run for all participants as well as separately for men (N = 3,534) and women (N = 1,350).

Results: The K6 accurately screened for any SMI when compared with the CIDI-SF. ROC-AUCs for the K6 were .88 for both men and women. Sensitivity was about 66% and specificity was about 89% for all participants with negligible differences between men and women. Past-year indicators for psychiatric treatment did not significantly improve upon the screening accuracy of the K6 for either men or women or for the aggregate sample.

Implications: The K6 appears to be a highly accurate screen for use with both men and women under criminal justice supervision. Adding questions about psychiatric treatment history does not improve screening accuracy beyond that achieved using the K6 alone. However, it may still be desirable to include treatment history questions in screening protocols for other reasons such as to insure continuity of care.