| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ALL ISSUES | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Classic Articles |
The literature on the phenomenon that the cost of outpatient psychotherapy may be offset by savings in medical expenditures began with a West German study of persons who had psychoanalysis or psychoanalytic psychotherapy and whose use of hospitalization for a 5-year period was less than that of a control group.1 This study and the subsequent literature were reviewed by Jones and Vischi, who concluded that the effect of psychotherapy was to reduce use of medical services by about 20%.2 A meta-analysis of 15 controlled offset studies up to 1978 that included some reviewed by Jones and Vischi yielded an estimate of the cost-offset effect between 0% and 14%.3 The range of estimates reflects methodologic flaws in many studies.
A meta-analysis of controlled studies of the effect of "psychologically informed intervention" on patients following heart attack or facing surgery showed that patients provided with information—about their condition, what to expect, and how to further recovery—or who were given emotional support did better than control subjects on most outcome indicators.4 Thirteen of these experimental studies included days in hospital as an outcome indicator, and their combined results showed that psychologically treated patients were discharged about 2 days sooner than were persons not so treated. Devine and Cook, from a meta-analysis of 49 controlled experiments of the effects of psychoeducational interventions with surgical patients, reported 1.31 fewer hospital days for patients receiving mental health services than for patients provided only the usual medical management.5
Since our last review of the cost-offset literature in 1978, the number of controlled studies has increased to 58 suitable for meta-analysis.1,6–62 It is feasible now to study the variables associated with reduced medical utilization following mental health treatment. A second resource, the massive fee-for-service research data base derived from the health insurance claims files of the Blue Cross and Blue Shield Federal Employees Program (FEP), provides a complementary perspective for studying the same variables. When we use these two large sets of data, each with special strengths that may compensate for weaknesses in the other, we can attempt to answer the same questions from two distinctly different perspectives.
METHOD 1: META-ANALYSIS OF THE COST-OFFSET LITERATURE
Meta-analysis is a quantitative procedure for summarizing findings across studies.4, 63 It makes use of any of several summary statistics that convert diverse findings from individual studies to a common base that is free of scale.
To update our literature search we began with the comprehensive list of references provided by Jones.64 We called Medlars and Index Medicus searches for January 1979 through July 1982, reviewed Excerpta Medica from January 1979 to July 1982, and obtained Automated Subject Citation Alert and PsycSCAN searches for cost-offset topics and key authors. We also searched Dissertation Abstracts and obtained microfilms of relevant entries. Finally, we surveyed reports from published lists of grants and contracts of government agencies and checked usable studies through Citation Index from 1979 to 1982. By May 1983 we had located 58 cost-offset studies suitable for meta-analysis (see Table 1A, Table 1B, Table 1C, Table 1D). Of these, 27 were doctoral dissertations, unpublished government grant or contract reports, or reports from private industry. The relatively large portion of unpublished studies should alleviate the fear that meta-analysis of published studies may be biased by the generally positive results of studies that are published. Eighteen additional cost-offset studies were not included in the meta-analysis because the data provided were insufficient or the design was inadequate to assess the impact of mental health treatment on utilization of medical services.65–83
|
|
|
|
General Cost-Offset Effects
Table 1A, Table 1B, Table 1C, Table 1D displays the characteristics and findings of the 58 studies of effects of outpatient psychotherapy on subsequent medical care utilization. The outcomes of all of the studies ranged from a 72.4% increase to –181.6% (decrease) in use of medical services following psychotherapy. Eighty-five percent of all of these studies reported a decrease in medical utilization following psychotherapy.
Twenty-six of these studies were naturalistic, time-series studies that compared persons' medical care utilization before and after psychotherapy. Each person served as his or her own control. Some of the studies also used comparison groups of persons who did not have psychotherapy. These studies did not assign patients to treatment groups randomly. Of the 26 time-series studies, all but six were conducted in prepaid clinic settings. This subset of studies yielded an average effect size of –33.10% (95% confidence interval is –57% to –20%). The weight of the findings from these 26 studies might be thought impressive considering that such naturalistic studies avoid the confounding problems of Hawthorne effects.84 On the other hand, the studies are open to other challenges.
First, the meaning of results from most such time-series studies has been challenged because experimental and comparison groups were selected differently. The medical care utilization of experimental subjects was recorded on "relative time" before and after the time of first mental health treatment. But the utilization data of comparison subjects were collected before and after an arbitrarily selected date. We expect that utilization of medical services may rise before the individual's entry into mental health treatment as a function of the same sense of distress that eventuated in his or her seeking mental health care. Thus the pre-psychotherapy utilization of the experimental groups might represent a peak or near peak. The medical care utilization of the mental health-treated group would be expected to fall from its peak regardless of benefits from the psychotherapy, since what goes up, in statistics as in nature, must come down. In contrast, for the control group there would be no such expectation either for a rise or fall. Thus results favoring the experimental group over the control group might be explainable in terms of statistical regression to the mean.
Self-selection for psychotherapy is also frequently invoked as a reason to question the findings of naturalistic studies. Random assignment to treatments is a cornerstone of methods developed in the biological sciences. But since self-selection for psychotherapy might well be regarded as part of that treatment, new methods to provide a functional equivalent of random assignment are called for. In the meantime, rather than simply dismissing the results of such a large number of studies, one can view the potential biasing effects of self-selection as an empirical matter to be settled by data.
Thirty-two studies were experimental in design, assigning patients to treatment conditions either randomly or through some matching scheme. Of these, 22 experiments determined the effects of psychological intervention on patients hospitalized for medical crises, with patients assigned randomly to a group receiving relevant information, emotional support, or both, or to a comparison group receiving only the standard medical regimen.
Analyzing only these 22 studies that are not vulnerable to bias resulting from self-selection or misinterpretation of the phenomenon regarding regression to the mean, we find that on the average these modest psychological interventions reduced inpatient hospitalization approximately 1.5 days below the control groups' average of 8.7 days. This effect is in the same direction as, although slightly smaller than, our earlier finding of about 2 days on the basis of 13 studies.4
In a comparison of the outcome measures of these 22 experimental studies that used random assignment to treatments with the 26 time-series studies in which patients had selected psychotherapy, the studies using random assignment yielded an average percent change of –10.4%. The 26 studies relying on self-selection yielded an average percent change of –33.1%. The offset effect is smaller when self-selection is ruled out by random assignment, but it appears both under conditions of random assignment and with self-selection of treatment. Devine and Cook5 performed a similar test in their meta-analysis of cost-offset effects of mental health treatment among surgical patients and concluded that the method of subject assignment was not systematically related to the size of estimates of effect.
0utcome Indicators: Outpatient Versus Inpatient Medical Utilization
Of the 48 estimates of the effects of mental health treatment on outpatient medical utilization, only five came from experimental studies. Of the 71 estimates of the effect of mental health treatment on inpatient utilization 62 came from experimental studies. The question is hopelessly confounded with study methodology and must be approached in a different way.
Five studies20,23,46,55,57 provided data that permit an unconfounded examination of the effects of psychotherapy on inpatient as well as outpatient medical care utilization. In all but one, the reduction in inpatient medical utilization exceeded the reduction in outpatient utilization. The average change was –73.4% for inpatient utilization and –22.6% for outpatient utilization. Only one study20 was an exception to this pattern. If one assumes that these five studies were drawn from a population of studies for which it is hypothesized that there is a 0.50 probability of inpatient utilization being reduced more than outpatient utilization, then the four "successes" (inpatient reduction greater than outpatient) in five "trials" have a probability less than 0.10 of being equaled or exceeded under the hypothesis.
These five studies have strengths and weaknesses that are complementary. On balance they permit the conclusion that the offset effect is likely to be greater for inpatient medical care utilization than for outpatient utilization. As we shall see, analysis of insurance claims will strengthen this impression.
Age of Patients as a Mediating Factor in Cost-Offset Effects
Most of the cost-offset studies did not report findings by age of patient; we found only two cost-offset studies of older people that were suitable for meta-analysis.29,39 Neither of these dealt with outpatient psychotherapy, possibly reflecting a misleading bias that older patients do not profit from outpatient psychotherapy. There are, of course, many case reports and studies of positive benefits of mental health treatment for geriatric patients. For example, Godbole and Verinis67 compared the effects of two forms of psychotherapy in a study of 61 hospitalized patients and reported benefits for both treatment groups as assessed by improvement in rating forms completed by nursing staff and author/therapists.
National statistics show the same trend as the research literature. In 1980 persons age 65 years and older constituted 11% of the population and accounted for 29% of all health expenditures.85 Yet they received a disproportionately small portion (2% to 4%) of outpatient mental health services.86 These figures suggest underutilization of mental health services by this age group. Older people may be less likely than other age groups to be referred for mental health treatment, although their needs may be greater and benefits would seem to be significant.
Levitan and Kornfeld39 provided psychiatric consultation to 24 elderly patients hospitalized for fractured femur and compared their hospital stays with those of a comparison group of 26 patients hospitalized for the same reason without psychiatric intervention in the same months of the previous year in the same hospital. Length of stay for the intervention group was 12 days shorter than the mean of 42 days for the control group, and twice as many of the patients who had been provided consultation returned home rather than being discharged to a nursing home or other institution.
Hill29 studied 40 cataract surgery patients between the ages of 50 and 91 years. They were randomly assigned to a behavioral training group, a sensory information training group, a combined behavioral and sensory training group, or a comparison group that received no special preparation. We would not expect important differences in length of stay, since the mean hospital stay for all four groups of patients was only a little over 3 days. However, a second outcome variable—first venture from home after discharge—did show significant differences in the expected direction. The "combined" group ventured out soonest from home, and both other treatment groups ventured out sooner than the comparison group.
Since we could find only two studies that directly addressed the impact of age on the offset effect, we measured its impact indirectly through meta-analysis of the 23 studies that did report the mean age of subjects. In 15 inpatient studies the mean age of the patients was 48.14 years, and the correlation between the mean age listed in each study and the effect size was –0.44, indicating that older subjects benefit more. In four outpatient studies that used visits to the doctor as the outcome measure, the mean age of the patients was 30.53 years, and the correlation between mean age and effect size was –0.31. In four alcohol outpatient studies the mean age of the clients was 35.8 years, and the correlation between mean age and effect size was –0.78. Thus in three different settings with three different populations a consistent finding emerges: Older people tend to have greater offset effects following mental health treatment.
METHOD 2: ANALYSIS OF HEALTH INSURANCE CLAIMS FILES
The claims files of the Blue Cross and Blue Shield FEP from 1974 through 1978 contain the medical care charges for a national sample of 6.7 million federal employees, retirees, survivors, and family members. About 53% of all federal employees were insured by FEP during these years, providing the largest fee-for-service data base available. The procedures for transforming the claims files to research files are described elsewhere.87 About 1.5% of persons covered received some form of mental health services in any 1 year during the 5-year period, or about 3.9% during the 5 years. This proportion is consistent with other reports that 1% to 1.8% of general medical patients receive psychiatric treatment in a 1-year period.88,89
Previous work87 has shown a dose-response relationship for psychotherapy and medical care utilization, with a cost-offset effect becoming clear after about six psychotherapy visits. In the present study, therefore, we examined the medical utilization of a group of persons who had at least seven outpatient mental health treatment visits beginning in 1975 but no psychiatric inpatient claims at any time. We compared their medical care utilization with that of a randomly selected subset of persons who filed no mental health claims throughout the 5 years of the data base. Each person in both groups was drawn from a contract that was active from 1974 through 1978 and was required to have at least one medical claim of any size in 1975 to enter the study. The data thus represent persons who made at least minimal use of medical care services. About 19% of contracts filed no claims during the 5 years. To ensure that differences in death rates would not bias the results, each person over age 55 had to have at least one claim of any kind in 1978, the last year of the data base.
This method of comparison avoids capitalizing on statistical regression to the mean, since both groups were compared on calendar time and had the same requirement to enter the study, a medical claim in 1975. We were thus able to compare the medical care utilization of the two groups for 1 year before the year of the entry requirement and for 3 years following it, which is also the year in which each person in the treatment group began a first episode of outpatient psychotherapy with or without drugs.
RESULTS 2
Evidence of General Cost-Offset Effects
Figure 1 shows that in 1974, the year before the start of mental health treatment, the medical charges for the treatment group were markedly higher than those for the comparison group, a finding consistent with the literature that suggests excess morbidity from physical disease among the mentally ill90,91 and our earlier findings.87 The medical charges of both groups rose in 1975 in part as an artifact of selection—each person was required to have at least one medical claim in that year. The medical care charges of both groups then fell in 1976 and rose again at a slower rate from 1976 to 1978. Following mental health treatment, the medical care charges of the treatment group increased more slowly than the average inflation rate of 13.6% per year. In contrast, the charges of the comparison group increased faster than the inflation rate. If we adjust the means for 1975–1978 for the difference between the groups in 1974, the adjusted means of the treatment group were significantly lower than those of the comparison group during each of these 4 years (t = –3.21, –2.44, –2.69, and –3.77, respectively, P < 0.05).
|
Cost-Offset Effects in Claims Files: Outpatient Versus Inpatient Medical Utilization
Figure 2 compares the outpatient and inpatient medical care charges of the persons whose total medical charges were graphed in Figure 1. Outpatient charges include physician office visits, outpatient laboratory charges, and prescription drugs. Inpatient charges include all medical charges incurred while the patient was hospitalized, e.g., hospital bed, physician fees, and other charges billed separately during the hospitalization.
|
Cost-Offset Effects in Claims Files as Mediated by Patients' Age
An examination of the cost-offset effect for narrow age subsets is complicated by the necessarily small sizes of these groups and the high variances characteristic of medical claims data. Since most persons obtain medical care only occasionally, claims data consist mostly of zero entries. Claims generally range from a few dollars to several hundred dollars, with a few much larger entries. In small groups, a single person with extraordinarily high medical claims can increase the variance considerably and complicate the interpretation of differences among group means. We can avoid this problem by removing the extreme cases, defined as persons with total medical charges over $20,000 in a single year, from both the mental health and comparison groups.
Removing the extreme cases from both groups lowered the mean of each group by only a few dollars and reduced the size of both groups by only 0.4%. Thus variance and standard errors were minimized without altering the general form of the findings.
To emphasize the relative differences in medical care utilization of age subsets, Figure 3 displays differences between the mean inpatient medical charges of the treatment group and the comparison group for four age groups. Figure 4 presents the same differences for outpatient medical utilization. Negative differences (below the zero line) indicate that the treatment group had lower charges than the comparison group. A falling curve, whether above or below the zero difference line, indicates a cost-offset effect. Graphing differences in this way removes the inflation component, since it affects both groups equally.
|
|
Figure 4 shows that the differences in outpatient medical charges of all the age groups remained fairly constant over the 5 years and that the expenditures of the mental health group were higher in every year than those of the comparison group. The slight dips in the curve of the oldest age group reflect the fact that those over age 55 in the mental health treatment group had significantly lower outpatient charges in 1975 and 1977 (t = –4.31 and –1.99, respectively, P < 0.05).
These findings for fee-for-service health insurance subscribers are generally in accord with findings derived from our meta-analyses of studies done in organized medical care settings and hospitals using both experimental and time-series methods.
DISCUSSION
Retrospective analysis of health insurance claims data and meta-analyses of time-series studies and prospective controlled experimental studies converge to provide evidence of a general cost-offset effect following outpatient psychotherapy. The widespread and persistent evidence of reduced rate of increase of medical expense following mental health treatment argues for the inseparability of mind and body in health care, and it also argues specifically for the likelihood that mental health treatment may improve patients' ability to stay healthy enough to avoid hospital admission for physical illness.
The clearest cost-offset effect appears largely in the reduction of inpatient rather than outpatient costs. As we noted in an earlier study,87 inpatient charges account for 75% of total medical charges and substantial savings would have to result from reduced hospitalization. Older patients show larger cost-offset effects than younger ones. These findings could be surprising to anyone believing that mental health treatment is necessarily more effective for younger than older people. The findings could also be surprising if one had assumed that reduction of medical services associated with psychotherapy is a function of keeping "the worried well" from "cluttering outpatient services." We have presented more detailed evidence elsewhere to show that recipients of mental health services suffer more chronic disease and are physically sicker than people who do not use psychiatric services.3, 87,93 The effects of outpatient mental health treatment cannot be explained as simple substitution of one outpatient service for another.
Older people generally use more medical services and more expensive inpatient services, leaving more room for cost reductions. But other factors may also contribute. Many older people have special mental health needs following emotionally distressing events such as suffering physical disease; experiencing loss of friends, spouse, social status, or income; being victims of crime; or being forced to relocate. The 1975 Harris survey showed that 8% of the respondents 65 and older said they had no close person to talk to, compared with 5% of the respondents under 65.94 Older men and women often have multiple social problems and more than one chronic disease or disability. Yet on average they are seen for a shorter period of time by their doctors during outpatient visits.95 Older people may also be in jeopardy because their lives lack the structure of a daily work routine and the supportive social networks associated with employment. The older patient—even if voluble about physical symptoms or peeves—may not volunteer much about emotional distress to a much younger physician, who also may not inquire about such problems when examining an elderly patient. Such a situation is not promising for early detection of need for mental health intervention, nor is it optimal for active cooperation between patient and physician in the effective management of chronic illness that would minimize need for hospitalization.
In view of the needs of the older population, planned psychological intervention may have special advantages. Provision of mental health services to older people could serve to shore up flagging determination to follow medical advice and to stay healthy and socially engaged. Evidence from one study of patient education and support for hypertensive patients reported that the special program had a more positive influence on compliance among elderly than among young patients.96
In view of the evidence from the literature and from our studies of health insurance claims, underutilization of mental health services by the elderly may result in needless suffering among the elderly and needless cost to society.
References
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ALL ISSUES | SEARCH | TABLE OF CONTENTS |
