This article analyzes the cost-effectiveness of inpatient substance abuse treatment, using readmission rates as the outcome measure. Our goal is to identify the characteristics of inpatient treatment programs that yield the most benefit at the least cost. Despite the widespread use of inpatient treatment, there has been a paucity of cost-effectiveness studies. The Institute of Medicine's review of drug abuse treatment literature found no studies of the cost-effectiveness of inpatient care (Gerstein and Harwood 1990). Two literature reviews found that the cost and effectiveness of different modes of alcoholism treatment are not correlated; however, the authors did not find the evidence persuasive enough to make recommendations about funding or treatment decisions (Finney and Monahan 1996; Holder et al. 1991).
The cost of inpatient treatment increases with its duration. There is conflicting evidence on how the length of treatment affects outcomes. Some observational studies have found that longer stays are associated with better outcomes in therapeutic communities (Bleiberg et al. 1994), halfway houses (Moos, Pettit, and Gruber 1995), and hospitals (Welte et al. 1981). Observational studies may suffer from selection bias, however. When the length of stay (LOS) is not randomly assigned, it is likely to be a function of the same patient characteristics that affect outcome.
Randomized clinical trials are designed to avoid this selection bias. Random assignment to a longer inpatient stay has not resulted in better outcomes for patients being treated for alcoholism (Mattick and Jarvis 1994) or drug abuse (McCusker, Vickers-Lahti, Stoddard, et al. 1995). This finding may not be definitive. Many LOS trials have few subjects and limited statistical power. Moreover, the results of trials may not apply to patients with the most severe disorders. Trials usually exclude these patients as it is regarded as unethical to enroll them in a protocol where they might be assigned a short LOS.
The cost of treatment also depends on the intensity of staffing. It is uncertain if intensively staffed programs yield better outcomes. A comparison of two inpatient alcohol treatment programs, one with 40 percent fewer staff, found no significant difference in effectiveness (Stinson et al. 1979). However, an evaluation of residential programs for adolescent drug abusers suggested that higher staffing levels were associated with better outcomes (Friedman and Glickman 1987).
There is little information on the effect of longer stays and more intensive levels of staffing on the cost of inpatient treatment. Also lacking is an analysis of whether the extra cost is justified by additional effectiveness.
The Department of Veterans Affairs (VA) relies heavily on inpatient programs to treat veterans with substance abuse disorders. In the 1994 fiscal year, VA provided 1.42 million days of inpatient substance abuse treatment (Piette, Baisden, and Moos 1995) at a cost of some $468 million. Until 1996, VA eligibility rules encouraged treatment in the inpatient setting. Most veterans qualify for care based on their income; low-income veterans were eligible for free outpatient care only if it was in preparation for a hospital stay or needed to prevent one.
VA's inpatient substance abuse programs have been studied using readmission as a measure of effectiveness (Peterson, Swindle, Phibbs, et al. 1994). Programs that performed better than expected had longer intended treatment duration, used assessment interviews involving family or friends, treated more patients on a compulsory basis, and had fewer early discharges and higher rates of participation in aftercare. The study used readmission after 180 days as the outcome. Only small changes resulted when the follow-up period was changed to 30, 60, 90, or 365 days.
We expand on this earlier study by considering costs and cost-effectiveness and by employing random effects regression.
DATA
Data on the design of treatment programs were obtained by mailed survey to all administrators of VA inpatient treatment programs in October 1990 (Peterson, Swindle, Phibbs, et al. 1994). The survey gathered information on the design of the program, such as the intended LOS and methods used in treatment, as well as a count of the number and type of direct treatment staff. We obtained detailed cost and utilization data for the preceding year, the period October 1, 1989 to September 30, 1990. Information on patients was obtained from the Patient Treatment File, the VA database of hospital discharges. The VA discharge file includes a unique patient identifier, patient demographics, diagnoses, and LOS. We obtained data on program cost and staffing from the Cost Distribution Report, the cost-accounting system used by VA medical centers. We divided the total cost from this report by the total days of inpatient care from the discharge file to find the average cost per day of care for each program. The components of the average daily cost of treatment are presented in Table 1. Using the assumption that all patients incur costs at the program's mean daily rate, we multiplied each patient's LOS by daily cost to find the total cost of treatment. The VA cost report reported research costs of $10.01 per patient day and education costs that averaged $17.42 per day. These costs were excluded from our analysis.
The VA cost report may suffer from some inaccuracies (Swindle, Beattie, and Barnett 1996). We created an alternative estimate of the cost of treatment staff. The number of each type of staff reported in the program survey was multiplied by the national average salary and benefits cost obtained from the VA summary expense journal, the Computerized Accounting for Local Management. We substituted this estimate of the cost of each program's staff to create an alternative measure of treatment cost.
We studied treatment provided by 98 programs that could be matched to the discharge file and cost report. These programs treated 38,683 unique patients during the year ending September 30, 1990. We examined the cost and effectiveness of the first treatment received by each patient during the study year. When a patient received more than one treatment during the year, we included only this first treatment as the index treatment for our analysis.
We did not include 77 of the 175 VA inpatient programs. These were excluded because the VA databases did not always allow us to distinguish the cost and utilization of individual programs when several alternative programs operated in a single medical center. The excluded programs were larger, less intensively staffed, and had longer intended LOS (Peterson, Swindle, Phibbs, et al. 1994). There was no difference in patient characteristics, as measured by the severity of illness index developed for VA substance abuse patients (Phibbs, Swindle, and Recine 1997).
METHODS
Cost-effectiveness analysis requires a single measure of outcome. Our only information on patients came from the discharge database. Given this limitation, [TABULAR DATA FOR TABLE 1 OMITTED] we defined a treatment as effective if the patient was not readmitted to any VA hospital within the United States for medical detoxification, substance abuse rehabilitation, or psychiatric care within 180 days of discharge from the index treatment. Using this definition, 75.0 percent of the treatments were effective. Data from non-VA programs were not available, so readmission to other facilities was not considered by our study.
Variables and their mean values are presented in Table 2. Medical and psychiatric conditions, and the substances abused by patients, are based on the diagnoses in the discharge file. Prior admissions represent the number of inpatient treatment episodes in the year before the index treatment. "High-income" means an income of more than twice the upper limit established by the VA eligibility test; in 1990, a single veteran with income in excess of $34,480 would have been considered high-income, as would a veteran with two dependents and an income in excess of $48,276.
We wished to find the patient and program level characteristics that explain the cost of treatment and the probability of readmission for further treatment. If we had used the program as the unit of observation, patient characteristics would have entered our model as a mean value for each program, resulting in a substantial loss of statistical power. A patient-level analysis, however, cannot make the standard assumption that the error terms are independent. When the error terms of patients in the same program are correlated, then standard models overstate the statistical significance of the regression coefficients.
Random-effects models account for the correlation of patients within programs. We had a continuous dependent variable, cost, and a dichotomous dependent variable, an indicator of whether the patient was readmitted within six months. Random-effects models can be used in both linear (Laird and Ware 1982) and logistic regression (Wong and Mason 1985). We used simple random-effects regressions, treating the intercept as a random variable whose variation is explained by program characteristics. We did not estimate any program-by-patient interaction terms.
We were interested in discovering the program characteristics that affect readmission rates while controlling for patient characteristics. One important patient characteristic is the number of times the patient was hospitalized in the previous year. This depends on the characteristics of both patient and program, but we wished to control for only the patient's contribution. To keep program factors out of this measure of patient severity, we excluded previous admissions to the program that provided the index treatment.
We considered the program-level factors previously found to predict rates of readmission (Peterson, Swindle, Phibbs, et al. 1994). Because our focus was on cost-effectiveness, we added factors associated with resource use, including the intensity of staffing and program size.
We wished to consider the effect of program-level factors that influence the LOS. We used the intended length of a completed treatment, according to the program director. We did not use the actual LOS because it would reflect patient-level characteristics as well as the design of the program.
We did not include early discharge or participation in aftercare in our analysis. These variables were excluded because of our concern that they are endogenous, that is, that they are correlated with the error term. The [TABULAR DATA FOR TABLE 2 OMITTED] unobserved patient attributes associated with retention in treatment are likely to be correlated with the likelihood that the patient avoided readmission. Inclusion of endogenous variables could bias our regression coefficients.
Table 3: Cost of Inpatient Substance Abuse Treatment VA Medical
Centers, 1990 Fiscal Year; Random-Effects Regression (N = 38,683
patients, 98 programs)
Coefficient p-Value
Intercept 1,303.56 .512
Program-Level Factors
Intended length of stay (days) 122.88 .000
Log program size -524.66 .002
Wage index 1,977.58 .003
Treatment staff per patient (FTE) 2,228.64 .000
Percent compulsory admissions 966.85 .013
[greater than] 50% family/friends assessment 248.67 .270
Patient-Level Factors
3 or more prior admissions -496.71 .000
2 prior admissions -229.79 .003
1 prior admission -95.12 .048
Age 13.41 .177
Age-squared -0.21 .036
Service-connected disability -66.16 .024
High income 96.19 .542
Non-veteran 44.71 .845
Not married 148.42 .000
African American 410.47 .000
Opiate diagnosis -281.21 .038
Marijuana 242.47 .003
Nicotine 384.87 .010
Amphetamine 26.60 .809
Schizophrenia -328.20 .000
Bipolar disorder 124.03 .210
Post-traumatic stress disorder 390.15 .000
Depression 396.29 .000
Other personality disorder 410.69 .071
Heart disease 78.35 .188
Arthritis 416.34 .000
Back problems 281.18 .000
Cancer 737.65 .000
Liver diagnoses 334.76 .001
HIV 361.28 .240
Alcohol withdrawal -637.47 .000Bleiberg, J. L., P. Devlin, J. Croan, and R. Briscoe. 1994. "Relationship Between Treatment Length and Outcome in a Therapeutic Community." International Journal of the Addictions 29 (6): 729-40.
Finney, J. W., and S.C. Monahan. 1996. "The Cost-Effectiveness of Treatment for Alcoholism: A Second Approximation." Journal of Studies on Alcohol 57 (3): 22943.
Friedman, A., and N. Glickman. 1987. "Residential Program Characteristics for Completion of Treatment by Adolescent Drug Abusers." Journal of Nervous and Mental Disease 175 (7): 419-24.
Gerstein, D. R., and H.J. Harwood (eds.). 1990. Treating Drug Problems: A Study of the Evolution, Effectiveness, and Financing of Public and Private Drug Treatment Systems. Washington, DC: National Academy Press.
Gfroerer, J. C., E. H. Adams, and M. Moien. 1988. "Drug Abuse Discharges from Non-federal Short-stay Hospitals." American Journal of Public Health 78 (12): 1559-62.
Holder, H., R. Longabaugh, W. R. Miller, and A. V. Rubonis. 1991. "The Cost Effectiveness of Treatment for Alcoholism: A First Approximation." Journal of Studies on Alcohol 52 (6): 517-40.
Kamlet, M. S. 1992. A Framework for Cost-Utility Analysis of Government Health Care Programs. Washington, D C: Office of Disease Prevention and Health Promotion, Public Health Service, U.S. Department of Health and Human Services.
Laird, N., and J. Ware. 1982. "Random-Effects Models for Longitudinal Data." Biometrics 38 (4): 963-74.
Mattick, R., and T. Jarvis. 1994. "Inpatient Setting and Long Duration for the Treatment of Alcohol Dependence?: Outpatient Is As Good." Drug and Alcohol Review 13 (2): 127-35.
McCusker, J., M. Vickers-Lahti, A. Stoddard, R. Hindin, C. Bigelow, M. Zorn, F. Garfield, R. Frost, C. Love, and B. Lewis. 1995. "The Effectiveness of Alternative Planned Durations of Residential Drug Abuse Treatment." American Journal of Public Health 85 (10): 1426-29.
Mechanic, D., M. Schlesinger, and D. D. McAlpine. 1995. "Management of Mental Health and Substance Abuse Services: State of the Art and Early Results." Milbank Quarterly 73 (1): 19-55.
Moos, R. H., B. Pettit, and V. Gruber. 1995. "Longer Episodes of Community Residential Care Reduce Substance Abuse Patients' Readmission Rates." Journal of Studies on Alcohol 56 (4): 433-43.
Peterson, K., R. Swindle, C. Phibbs, B. Recine, and R. Moos. 1994. "Determinants of Readmission Following Inpatient Substance Abuse Treatment: A National Study of VA Programs." Medical Care 32 (6): 535-50.
Phibbs, C. S., R. W. Swindle, and B. Recine. 1997. "Does Case Mix Matter for Substance Abuse Treatment? A Comparison of Observed and Case Mix-Adjusted Readmission Rates for Inpatient Substance Abuse Treatment in the Department of Veterans Affairs." Health Services Research 31 (6): 755-71.
Piette, J. D., K. L. Baisden, and R. H. Moos. 1995. Health Services for VA Substance Abuse Patients: Utilization for Fiscal Year 1994. Palo Alto, CA: U.S. Department of Veterans Affairs, Program Evaluation and Resource Center, HSR&D Center for Health Care Evaluation.
Stinson, D. J., W. G. Smith, I. Amidjaya, and J. M. Kaplan. 1979. "Systems of Care and Treatment Outcome for Alcoholic Patients." Archives of General Psychiatry 36 (5): 535-39.
Swindle, R. W., M. C. Beattie, and P. G. Barnett. 1996. "The Quality of Cost Data: A Caution from the VA Experience." Medical Care 34 (3): MS83-MS90.
Weisner, C., T. Greenfield, and R. Room. 1995. "Trends in the Treatment of Alcohol Problems in the U.S. General Population, 1979 through 1990." American Journal of Public Health 85 (1): 55-60.
Welte, J., G. Hynes, L. Sokolow, and J. Lyons. 1981. "Effect of Length of Stay in Inpatient Alcoholism Treatment on Outcome." Journal of Studies on Alcohol 42 (5): 483-91.
Wong, G., and W. Mason. 1985. "The Hierarchical Logistic Regression Model for Multilevel Analysis." Journal of the American Statistical Association 80 (391): 513-24.
Paul G. Barnett, Ph.D. is a health economist with the HSR&D Center for Health Care Evaluation and the Program Evaluation and Resource Center, Veterans Affairs Palo Alto Health Care System, Menlo Park, California, and the Department of Health Research and Policy, Stanford University School of Medicine. Ralph W. Swindle, Ph.D. is a research health scientist with the HSR&D Service, Roudebush VAMC, Indianapolis, IN, and the Department of Medicine, Indiana University School of Medicine and the Regenstrief Institute for Health Care. Address correspondence and requests for reprints to Paul G. Barnett, Ph.D., Health Economist, Center for Health Care Evaluation, Veterans Affairs Palo Alto Health Care System, 795 Willow Rd. (152 MPD), Menlo Park, CA 94025. This article, submitted to Health Services Research on July 23, 1996, was revised and accepted for publication on February 28, 1997.
Paul G. Barnett "Cost-effectiveness of inpatient substance abuse treatment". Health Services Research. FindArticles.com. 16 Aug, 2010. http://findarticles.com/p/articles/mi_m4149/is_n5_v32/ai_20575055/
COPYRIGHT 1997 American College of Healthcare Executives
COPYRIGHT 2004 Gale Group
COPYRIGHT 1997 American College of Healthcare Executives
COPYRIGHT 2004 Gale Group
Posts
No comments:
Post a Comment