PubMed1998
We apply pattern-recognizing artificial neural networks (ANNs) to the patients of two psychiatric hospitals, a private, for-profit hospital (PH) and a state hospital (SH), both serving the southern tier of Maine (approximately two-thirds of the state's population, i.e., 800,000 persons) over a 19-month period. In our data from the PH, N = 837 admissions, and at the SH, N = 834. Unique patient identifiers were assigned to patients so that their individual patterns of care could be incorporated into our ANNs. We used a previously reported methodology to measure quality of care (Q), and developed a measure of value of service (V) from the patients' perspective for both facilities. A random portion of the demographic and outcome data of patients from each hospital was sequestered as a test set, whereas the remainder was used to train ANNs with length-of-stay (LOS) as an outcome measure. Q, and V normalized for risk (RR), i.e., V/RR, were calculated for each test set, which included multiple admissions of individual patients to each hospital. The methodology for V accounts for the severity of illness with the calculation of a metric called U/G, for the differences in case-mix by exchanging "virtual patients" between ANNs, and for entropy in the health care system by using a metric called the risk ratio (RR). Results showed that V/RR was 2.4 times greater at the SH than at the PH. This advantage is likely due to prior knowledge of individual patient patterns of treatment by the SH's staff. Data sets from each hospital using only single admissions for each patient in the study period (thereby eliminating unique patient patterns of LOS), yielded a V/RR that was only 21% greater at the SH. We hypothesize that this difference is due to the SH's ability to use treatment and discharge based upon patient strengths and level of clinical improvement, unfettered by insurance deadlines. Approximately 5% of the admissions to our studied PH went on to our SH, namely, the most severely impaired and indigent patients. The SH not only had twice the value, but also had half the cost of the PH, despite the greater number of treatment non-responders and non-compliants at the SH. The reasons for this are skilled staff and specialized ancillary services, staff knowledge of patients' responses to previous therapy, and individualized LOSs. These are features that create a caring environment and better compliance with treatment. This study emphasizes the value of tertiary care psychiatric facilities in a comprehensive mental health care delivery system. There are few studies of the effects of excessive downsizing of SHs on the community, but reports from Massachusetts (in 1995) suggest a 79% increase in suicides when the closing of SH beds exceeded 98% of maximum historical census. Routine use of pattern-recognizing tools such as ANNs would serve to inform the public about the value of mental health services so that the most vulnerable in our society are not neglected.