J. his study was motivated by an increasing concern about national and local trends (PSRO, third-party payers, etc.) who have been considering elevation of length of stay to the status of a standard. This would be used to determine whether a particular diagnosis in a particular region is or is not in keeping with the average length of stay and possibly disallow reimbursement if the "standard" is exceeded. While these criteria are easily applicable to medicine and surgery they are highly problematic in clinical psychiatry, where in our efforts to refine our diagnoses we may have underestimated the complexities related to mu\tip\e variables pertaining to length of hospitaliza tion. This has made us realize Click 's prediction that there may be "a decade of reckoning for hospital psychiatry."'
In order to define and clarify these complexities we undertook this retrospective computer analysis of a number of variables. The data were collected at Hartford Hospital, a 1,000-bed general, medical -surgical hospital, from 765 charts representing admissions for 1976. The Department of Psychiatry consists of 40 inpatient beds, Emergency Crisis Intervention Services for children and adults, an Outpatient Clinic, a Partial Hospitalization Program and a Psychology Service. The department admits voluntary patients on an informal basis to its tenbed Ward Service or to its Private Service. The Ward Service is under supervision of a Director assisted by three full-time house staff, in affiliation with the University of Connecticut Department of Psychiatry. Admission procedures require a complete physical and psychiatric evaluation and documentation on a standardized form including a diagnostic formulation and treatment plan to be completed within the first 24 hours. Utilization Review is done on the 20th day following admission. Weekly chart reviews are conducted by attending psychiatrists who independently survey criteria, including diagnosis, treatment, length of stay, discharge plans, etc. The Unit is staffed by registered and licensed practical nurses, providing 3.7 hours per 24-hour patient period. In addition, there are three social workers, two clinical PhD psychologists and a full-time occupational therapist assisted by two occupational therapy assistants. Medical students, nursing students and social work students rotate through the Service. Average length of stay for the study year was 19 days.
Since our experience and a survey of the literature did not show significant relationship between length of stay and diagnosis in the current practice of hospital psychiatry, we designed a study which would test the statistical relationship between these two variables. We hoped to find any other obvious variables which might possibly influence the length of hospital iza ti on. As a spinoff we hoped the study would help refine our methods for selecting charts for further surveys.
The sample charts were used to study a variety of clinical and demographic variables. They included sex, age (established by birthdate), date of admission, date of discharge, marital status, doctor, the reason for admission, admission diagnosis, discharge diagnosis, employment status and use of medications (eg, major tranqu.ilizers, minor tranquilizers, antidepressants, and lithium). Data were also collected on electroshock treatments, ward status, source of payment, place of residence, admission to the emergency room, presence of concomitant medical diagnosis, existence of a language problem and discharge plans. These details included the place where the patient would be residing, such as home or institution, if they were leaving against advice or if they eloped. This information was punched on cards to create a computer file. A preliminary distribution for each variable was obtained and it showed that half of the patients had length of stay of 15 days orfewer. Hence, the patients were divided into two groups: a short-stay group with a length of stay of 1 5 days or fewer, and a long-stay group with a length of stay of 15 days or more. Chi squares were then used to determine whether the two groups differed significantly on the appropriate variables. Sex or marital status which yield group counts were nominal variables while others were continuous variables (ie, those for which values have achieved mathematical meaning). To determine whether these variables were related to length of stay, linear correlations were calculated between them and length of stay in days (rather than short or long). With Ii near regress ion, if two variables are correlated, one of the variables can be used to predict the other. In multiple regression, the use of additional variables improves the prediction. Our computer program allowed for choice of the predictor variables, had freedom to choose the predictor variables in the order in which they improved the prediction and used multiple regression to determine whether the most useful variables in combination could be used to predict length of stay. In the course of the examination of the data we recognized an association between longer stay and a .discrepancy between admission and discharge diagnoses. Similarly, we learned that a group of patients who were treated by certain doctors appeared to be staying longer than those treated by others. The final predictor variables selected in the regression were: 1) change in diagnosis, 2) doctor group, 3) presence of psychosis, 4) marital status variables, 5) sex, 6) employment status, 7) treatment with major tranquilizers, 8) minor tranquilizers, 9) presence of a medical diagnosis, 10) admission through the emergency room and, 1 1) discharge to home or day-care.
FINDINGS AND RESULTS
We found that length of stay was significantly associated with a number of variables. The short-stay group contained more me-n, married, divorced, separated people than would be expected, and fewer single and widowed people. There were more employed people than expected and fewer patients being treated with drug therapy or electroshock therapy; fewer patients with medical diagnosis; and fewer patients who were discharged to the home or to a day treatment facility. There was a suggestion of a significant association between length of stay and admission through the emergency room since more short-stay patients than expected were admitted in this way. A composite picture of the shortstay patient that emerged was that of an employed man who was or had been married, or who was admitted through the emergency room, and who had no physical illness and did not require or receive treatment with drugs or electroshock and who was not referred to day-care.
Special attention was given to the question of whether the short-stay group and the long-stay group differed significantly in admitting diagnosis. When the diagnoses were grouped in two large categories, psychosis and neurosis, it appeared that psychosis was significantly under-represented in the short-stay group and neurosis was significantly over-represented. The short-stay group had fewer patients than expected in each of the psychotic categories, more than expected in each of the neurotic categories while the reverse was true of the long-stay patients. Discrepancy was most marked in the neurotic and affective diagnoses (DSM-II). The 30Os were overrepresented in the short-stay group ( 1 85 as compared with an expected frequency of 173) and under-represented in the long-stay group (146 as compared with an expected frequency of 158); the 296 diagnoses were underrepresented in the short-stay group (21 as compared with an expected frequency of 30) and over-represented in the long-stay group (37 as compared with an expected frequency of 28). For these comparisons diagnoses of DSM 3I9 (non-diagnostic terms for administrative use) were not considered. (There were 1 4 patients in the shortstay group and 13 patients in the long-stay group with these diagnoses.)
The study raised questions found frequently in social sciences, in which predicting variables are to a certain extent correlated with each other and in which experimental control may be difficult or impossible. In this case, affective disorder was associated with long length of stay, and neurosis with short length of stay. Treatment withantidepressants, major tranquilizers or electroshock was correlated with long length of stay and treatment with minor tranquilizers was not related to length of stay. Treatment with antìdepressants, major tranquilizers or electroshock was most common in affective psychosis and treatment with minor tranquilizers was most common in neurosis. Long length of stay then correlated with treatment with major tranquilizers, antidepressants. electroshock or with the presence of affective disorders. Further, some of the other variables that were found to be associated with length of stay were correlated to some extent with each other. The strongest correlations were those between sex and employment status; between being unmarried and discharged to home; and between age and the use of antidepressants. Of the various possible techniques used in multiple regression, it was decided to let the computer choose first the variable that made the greatest contribution toward explaining the variance in length of stay (ie, in predicting length of stay), then choose the second most important variable, then the third, etc., until no further significant contribution was made to the explanation of length of stay. The first variable chosen by the computer was use of electroshock, and the remaining ones in order were: use of major tranquilizers, doctor group, use of minor tranquilizers, discharge to day-care, presence of a medical diagnosis, being unmarried, sex, presence of psychosis, use of antidepressants, change in diagnosis, age, employment status, admission through the emergency room, use of lithium, being divorced, being married, discharged to the home, and being separated. Thus, short length of stay was associated with not having a doctor associated with a short-stay, not having a medical diagnosis, and being separated from a spouse. Long-stay was associated with having a psychotic diagnosis, being female, and having a change in diagnosis. The regression technique showed in fact that treatment had a much stronger and more direct effect on length of stay than diagnosis, and that when the effect of treatment was controlled for, there were still a number of variables that had a stronger relationship with length of stay than diagnosis. If all of these variables were controlled for, there was still a relationship between diagnosis and length of stay, but it was a weak one.
Predictions of the length of stay of individual patients were made by the computer on the basis of these findings, and the residuals (differences between actual length of stay and predicted length of stay) were further examined. It appeared that the computer had more difficulty in making predictions for the long-stay patients, and made some extreme over-predictions and under-predictions for these patients. The nature of the errors suggested that they were not random and further research was needed to identify the factors as yet not included in the data that influences length of stay in these patients. Apart from the question of length of stay, it was considered interesting to determine whether ward patients and patients of private doctors receive differing treatments. It was found that there are significant differences between ward patients and private patients and the frequency which they are treated with minor tranquilizers and antidepressants, and the frequencies with which they were given electroshock. Of ward patients, 28.2% receive minor tranquilizers as compared with 41. 8% of the private patients; 31% of them receive antidepressants compared with 58% of the private patients; and 2.3% of them receive electroshock compared with 19% of the private patients. The ward patients receive almost a full share of major tranquilizers. Of the ward patients, 64.8% receive these drugs as compared with 68.7% of the private patients.
On the basis of this study, we note that diagnosis alone does not predict length of stay and that many other factors contribute to the prediction. When these variables were analyzed by multiple linear regression only 20% of the variance in length of stay could be accounted for. Of this, the first variable chosen by the computer was use of electroshock, use of major tranquilizers, doctor group, use of minor tranquilizers, discharge to day-care, presence of medical diagnosis and only in the bottom part, presence of psychosis, and change in diagnosis, etc. Though, in a field like psychiatry, where biological, psychological and social forces are interwoven in an intricate complex fabric, a 20% variance is acceptable, most certainly, there is room for additional studies. Previous studies attempting to link the diagnosis with length of stay indicate "influence" or "association" of primary psychiatric diagnosis with length of stay. One such study analyzed the data in five different psychiatric settings within the city of Tuscon, Arizona." Data on admission, discharge, diagnosis, age and sex were collected on psychiatric patients discharged from these hospitals. Diagnoses were standardized for comparison purposes and it was found that the hospital's length of stay did not differ significantly in the majority of comparisons, "except for two out of five hospitals. " This similarity was explained by the fact that all shared shortterm treatment goals. In a later article the same authors outlined problems with such studies, particularly the lack of standardization of terminology or data collecting and, hence, resulting in difficulties in communication and interpretation.3
One major problem is the variability in the definition of short-intermediate and long-term hospitalization which stresses the need for establishing a median rather than the mean for length of stay in certain hospitals. Data presented along those lines are clinically more useful because they are not affected by extremes. Heiman et al suggested the following topics for further study: I) the relationship among admission and discharge criteria, treatment goals, effectiveness and length of stay, 2) the relationship between the quality and quantity of psychiatric housestaff and length of stay, 3) the relationship of length of stay to referral source, distance of patients from hospital, and private and public patient status, 4) the effect of continuity of care and the availability of outpatient followup on length of stay, 5) the relationship of length of stay to the number of psychiatric beds per capita, and the utilization of these beds, and 6) geographic variables as they affect length of stay.3
Richman et al described the utilization review procedure for inpatient care involving the statement of therapeutic goals, prediction about the time required to reach these goals and evaluation of care in terms of whether these goals have been met.4 Monitoring extended stay resulted in the decrease in the number of patients with prolonged stays. The mean stay for ail discharges decreased one-fifth, from 26 days to 21 days, with no impact on the mean stay of short-stay patients which remained at 14 days. No one doubts that diagnosis contributes to the prediction of length of stay and that diagnosis and treatment have a relationship. Our analysis, however, indicates that they correlate poorly and that treatment factors, over diagnosis, were chosen by the computer.
Where do we go from here? Since this study, internal and external changes have taken place in our hospital, an eight-bed security sub-unit was carved out of the 40-bed inpatient psychiatric unit, which has shown an increase of five days of hospital stay. An alcoholism service has been added. DSM-III is now in use. A new study will include more variables such as chronicity and severity of illness. Beyond our particular setting we hope to compare our data with other facilities who have dissimilar patient populations and consequently, length of stay. As a result we look forward to utilizing data to foster optimal utilization and take into account bio psychosocial factors as well as fiscal and political realities which have an indirect but major impact on length of stay.
The authors want to acknowledge the help and assistance given by Helen Lansberg, Ph.D., Candidate, Director of the Medical Library at the Institute of Living, Hartford, Connecticut, for her invaluable assistance in the statistical analysis of the data.
1. Glick ID: Short-term intensive psychiatric hospital treatment: Which treatment and for whom'? Journal of the National Association of Privait Psychiatric Hospitals 9(I):S-11.
2. Heiman EM, S ha n field SB: Length of stay for patients in one city's hospitals wilh psychiatric units. Hasp Community Psychiatry 1980; 31:632-633,
3. Heiman EM. S ha n field SB: Reflections on length of stay. Psychiatric Annals 1981; 11:155-158,
4. Richman A. Pinsker H: Utilization review of psychiatric inpatient care. Am J Psychiatry 1973; 130:900-903.