Economic Grand Rounds: Estimation of Costs of Public Psychiatric Treatment

Many attempts have been made to understand factors influencing the cost of care in the public psychiatric system. Perhaps the best known is the approach using diagnosis-related groups (DRGs), initially designed to classify hospital episodes for medical and surgical care. However, this system has proved difficult to apply to mental health care (1). A critical limitation is that DRGs are weak predictors of costs (2).

Virtually all studies concur that hospital care is more expensive than community care. Some studies found hospital costs to be two to three times higher than costs for community care (3), while others reported hospital costs eight to nine times higher (4,5). Given this high level of disparity and the need to develop models for estimating costs that are valid across service settings, it is important to clarify this issue. This study compared inpatient and outpatient costs and examined their relationship to patients' diagnosis and functioning.

Methods

The study sample consisted of 200 adult clients who were registered between December 1991 and August 1993 with the public psychiatric service and who had an address in a local government area of suburban Melbourne. Their mean±SD age was 40± 11.6 years, and 108 clients were male. A total of 127 had never been married, and 183 were unemployed when they entered the study.

Diagnostic information was obtained for 199 clients. Treating physicians indicated whether the clients had any of the following problems: schizophrenia or another psychosis, any neurosis, a personality disorder, alcohol or substance abuse, an intellectual disability, and any significant medical problem. Multiple diagnoses were permitted. A total of 158 clients were diagnosed as having schizophrenia, 90 had a significant medical problem, 49 had alcohol or substance abuse, 47 had a personality disorder, 37 had a neurosis, and 21 had an intellectual disability.

Clients were evaluated by mental health professionals at three-month intervals over a 21-month period. Functional level was assessed using the Life Skills Profile (LSP) (6). Data on the psychometric properties of this 39-item scale have been published (7,8). A later version of the LSP (8) presented an alternative set of five subscales: antisocial, bizarre, compliance, withdrawal, and self-care.

Costs (in Australian dollars) were estimated separately for hospital and community care. Hospital costs were estimated by multiplying the number of days spent in different hospital wards—acute locked, acute open, rehabilitation locked, rehabilitation open, and hospital hostel—by the cost of a bed day for that location: $462.45, $332.52, $229, $306.27, and $216.56, respectively.

For community costs, the total yearly cost to operate the clinic, including overhead costs, was first obtained. Second, the total number of clinic visits and the durations of these visits were obtained from the central computerized database, in which contact durations are coded by ranges. The duration of each contact was assumed to be the midpoint of its range. Finally, the total yearly cost of running the clinic was divided by the total yearly contact time, yielding a cost of $2.21 per contact minute. A client's community costs for each three-month interval were calculated from the client's total contact time. Adjustments were made when the intervals between assessments were not exactly 90 days. A community episode was defined as a three-month period in which a client had no hospital costs. A hospital episode was defined as a three-month period during which a client accrued no community costs. A mixed episode consisted of both community and hospital costs.

Results

Table 1 presents average costs for the three types of care episodes. Costs rose steeply with increasing hospital involvement. All distributions were very positively skewed. To overcome this skew, which precludes many statistical procedures, the logarithms of the costs were used.

Figure 1 confirms the sharp separation of the costs of the three types of episodes while demonstrating that the typically skewed distributions of raw costs can be rendered more amenable to statistical analysis by logarithmic transformation. These results suggest that the prediction of costs rests largely on the prediction of hospitalization.

At first assessment, 140 clients were in the community and 60 were in the hospital. Thus a simple prediction that all clients would be in the community yielded a base correct allocation rate of 70 percent. It is relative to this base rate that any contribution of the predictive variables must be judged.

Logistic regression was used to relate diagnostic and functional information to locus of care. The first analysis was a two-stage regression in which the diagnostic variables were included at stage 1 and the scores on the five LSP subscales and the total score were included at stage 2. Variables were introduced in a forward stepwise fashion at each stage.

Three variables entered the equation—a diagnosis of schizophrenia among the diagnostic variables and antisocial and bizarre behavior among the functional variables. Having a diagnosis of schizophrenia and more severe antisocial and bizarre behavior were significantly associated with a hospital locus of care. The model correctly allocated 74.9 percent of cases to their actual locus of care.

A question arose about whether the diagnostic or functional variables alone could do as well as the two sets of variables combined. Therefore, two additional one-stage logistic regressions were performed, one using only the diagnostic variables and the other only the functional variables. The first of these produced the base 70 percent prediction, with a diagnosis of schizophrenia as the sole predictor variable. In contrast, the second regression produced a prediction rate of 74.4 percent, with antisocial and bizarre behavior as the predictors. These results suggest that antisocial and bizarre behavior, independent of diagnosis, are moderately effective predictors of a hospital locus of care.

To show whether these behavioral factors were associated with costs without the intervening variable of hospitalization, a linear regression was performed, with the logarithm of costs as the dependent variable, and the scores on the antisocial and bizarre behavior subscales as the independent variables. The linear regression with both independent variables entered explained 14.9 percent of the cost variance. The total regression was highly significant (F=11.74, df=2,134, p<.001). Antisocial behavior alone was highly significant (t= 3.12, df=135, p=.002), and bizarre behavior approached significance (t=1.79, df=135, p=.075). Another regression analysis using only antisocial behavior explained 12.9 percent of the variance (F=19.93, df=1,135, p<.001).

Thus it appears that antisocial behavior was associated with hospitalization and accounted for nearly 13 percent of the variance in total costs. Bizarre behavior, which was correlated with antisocial behavior and which was also associated with hospitalization, accounted for an additional 2 percent of the variance.

These analyses relied on the large differences in estimated hospital and community costs and implicitly on our assumptions about relative costs. Because most clients in the Melbourne public psychiatric system are treated in the community, it was of interest to determine whether the same variables were predictive of costs for clients who were treated only in the community and who were not hospitalized in the three-month period.

To this end, a similar linear regression analysis, with log costs as the dependent variable and diagnostic and functional data as the independent variables, was conducted. This analysis accounted for 13.9 percent of log costs (F=6.08, df=3, 113, p<.001), with three variables entering the equation—a diagnosis of schizophrenia, a diagnosis of personality disorder, and social withdrawal. The strongest single predictor was personality disorder. The mean raw costs of community care of those with and without a personality disorder were $912 and $447, respectively. The presence of a personality disorder diagnosis alone explained 6.2 percent of the variance in log costs, withdrawal alone accounted for 5.8 percent, and schizophrenia alone for 2.6 percent.

These results suggest that cost determinants in a purely outpatient group were quite different from those in a group with inpatient contacts. The best single determinant of cost in the group that received only outpatient care was the presence of a personality disorder. A diagnosis of schizophrenia and an above-average level of social withdrawal also explained some cost variance, but withdrawal explained more than did a diagnosis of schizophrenia. This finding suggests that it is not simply the diagnosis of schizophrenia that is associated with higher costs but that form of the illness in which withdrawal is evident—schizophrenia with prominent negative symptoms.

Discussion and conclusions

Our first finding for this sample of 200 clients in the Melbourne public psychiatric system was that costs for hospital care were far greater than for mixed hospital and community care, which in turn were far greater than costs for community care only. Of interest are the relative magnitudes of the costs: a three-month episode of hospital care cost nearly 40 times as much as a similar episode of outpatient care. An episode of mixed inpatient and outpatient care cost about 20 times more than one of outpatient care alone.

A corollary of the wide differences between inpatient and outpatient costs is that from the perspective of the total psychiatric system, accurate estimation of costs depends largely on the prediction of hospitalization. Antisocial and bizarre behavior and a diagnosis of schizophrenia were associated with hospitalized status, and additional analyses suggested that antisocial and bizarre behavior were the primary determinants. The fact that measures of functioning were able to account for nearly 15 percent of the variance in log-transformed costs is particularly impressive given that use of DRGs to account for cost variance in mental health care has rarely exceeded 20 percent, and then usually by the exclusion of outliers, or cases for which costs far exceed the usual limits.

In the study reported here, when the focus of attention was restricted to episodes of community care without hospital involvement, the correlates of cost changed markedly. The most time-consuming and therefore most costly clients treated in the community were those with a personality disorder, closely followed by those with a high level of social withdrawal, whether or not they also had a diagnosis of schizophrenia. The maximum proportion of variance accounted for was nearly 14 percent, with personality disorder and withdrawal each accounting for around 6 percent.

These results must be interpreted in light of certain assumptions and limitations. First, we took no account of indirect costs (9), such as costs to relatives and other caregivers and charitable and nongovernmental organizations. It is unclear whether the numerous assumptions made in our calculation of costs were adequate. Although we have reasonable confidence in the reliability of the Life Skills Profile, we do not know how reliable the diagnostic information was. Despite these factors, this study found clear and interpretable relationships between client characteristics and costs, accounting for a reasonable proportion of the cost variation across both hospital and community locations and without recourse to the exclusion of outliers.

Dr. Trauer is senior lecturer in the department of psychological medicine at Monash University in Victoria, Australia. Mr. Duckmanton was formerly senior psychologist at the North Eastern Metropolitan Psychiatric Service in Melbourne. He is currently coordinator of utilization management and quality improvement at Mental Health Services West in Portland, Oregon. Dr. Chiu is associate professor in the department of psychiatry at Melbourne University. Send correspondence to Dr. Trauer at the department of physiological medicine at Monash Medical Centre, 246 Clayton Road, Clayton 3168, Victoria, Australia (e-mail, [email protected]. edu.au). Steven S. Sharfstein, M.D., is editor of this column.

Enlarge table