Predicting symptoms and function for persons with severe and persistent mental illness.

摘要

This study explored the statistical approach to predict symptoms and function for people with severe and persistent mental illness. The three months predictions were constructed using data from the Crisis Alternatives Project (CAP). The CAP Project was a three-year longitudinal study that examined the cost-effectiveness of alternative treatments to standard 30-day hospitalization. Clients of a major community mental health agency were recruited into the study over the three years period. An independent team of trained assessors conducted face-to-face interviews with clients every three months. The interviews consisted of a structured assessment package that included a measure of symptom levels---the Brief Psychiatric Rating Scale (BPRS), and two measures of functional levels, the Colorado Client Assessment Ratings (CCAR) and the Global Assessment of Functioning (GAF). Data on demographic characteristics, psychiatric diagnosis, and treatment received were extracted from the agency's ongoing administrative database system. Hypotheses concerning the stability of symptoms and function were examined and tested. Cross sectional analyses were first conducted to identify relationships between symptoms and functions and other factors affecting those relationships. Next, multiple linear regression was used to built three months prediction models of symptoms and function with data from the baseline and 3-month assessments. Finally, the three months prediction model was evaluated with data from the 6-month and 9-month assessments. The three months prediction model of BPRS contained current BPRS scores, mental health service use, and stressful life events. Current CCAR scores, social relationships, and current GAF scores were significant variables in the three months prediction model of CCAR. The three months prediction model for the GAF contained current GAF scores, current CCAR scores, current BPRS scores, stressful life events, mental health service use, and social relationships. The three months prediction models for the BPRS, CCAR and GAF explained 34.2%, 43.0%, and 24.8% of the variances, respectively. The three models were fairly accurate in making predictions. The mean of the absolute differences between the model-generated predictions and the observed scores for the BPRS was .33 on a 7-point scale; for the CCAR was 2.98 on a 50-point scale; and for the GAF was 7.7 on a 90-point scale. The results showed that statistical predictions could be a viable alternative to clinical predictions.