Enrolling the elderly in prepaid plans raises a number of additional issues. In theory, health maintenance organizations may provide better continuity and coordination for care of chronic disease than the fee-for-service system [3]. Yet, some authors have expressed concern that health maintenance organizations may be insensitive to special needs of the elderly because their highly structured care systems may be difficult for elderly patients to use, creating nonfinancial barriers to care [4]. Under prepayment, physicians may respond to economic incentives to restrict services by seeing chronically ill patients less often [5]. Incentives to limit treatment may be particularly powerful among high-cost enrollees such as the elderly.

Previous studies of the effects of capitation have focused on either nonelderly poor or elderly nonpoor persons. None has studied populations that are both elderly and poor, which may be at particular risk for underservice in prepaid plans. Further, they suffer either from incomplete follow-up of study patients or from the fact that patients were not randomly assigned to prepaid and fee-for-service groups, introducing the substantial threat of selection bias. We describe the experience of noninstitutionalized elderly Medicaid beneficiaries who were randomly assigned to prepaid compared with fee-for-service Medicaid care.

Methods

Top Methods Results Discussion Author & Article Info References

The study was conducted as part of the Hennepin County Medicaid Demonstration Project, one of the Health Care Financing Administration-sponsored Medicaid Competition Demonstration sites. This site enrolled a broad range of Medicaid beneficiaries, including the elderly, and randomly assigned 35% of them to prepaid care. The remaining 65% continued to use fee-for-service providers participating in Medicaid. Once randomly assigned to the capitation group, beneficiaries were given an opportunity to choose among seven health plans. These included a closed-panel health maintenance organization, a county-sponsored network health maintenance organization that formed in response to the demonstration, and five independent practice association plans. The 8% of persons who did not voluntarily choose a plan were randomly assigned to one. Beneficiaries were required to remain in the plan for at least a year, unless they successfully appealed.

Because over 40% of the Twin Cities' population is enrolled in health maintenance organizations, it is likely that nearly all physicians caring for study patients had some patients in their practices for whom they were reimbursed on a capitation or reduced-fee basis, with risk-sharing through "withhold" arrangements in which part of their compensation was determined by their success in containing costs.

Almost the entire study sample was enrolled in both Medicare and Medicaid, and the Medicaid portion of care for the prepaid group was capitated as part of the demonstration. Under this capitation payment, Medicaid paid for the copayment and deductible portion of the Medicare program, as well as for services not covered by Medicare, such as drugs, dental care, and physical, speech, and occupational therapy. These Medicaid costs were fixed at 95% of estimated fee-for-service costs, which constituted about half of total health care expenditures for this population. Plan participation was voluntary, but all plans participating in the demonstration chose to enroll elderly patients.

We identified from Medicaid tapes all 1496 noninstitutionalized, aged ( 65 years) Medicaid beneficiaries in Hennepin County, Minnesota, and randomly selected 400 beneficiaries for an experimental (prepaid) group and 400 beneficiaries for a comparison group for evaluation. Sample members who identified themselves as hypertensive or diabetic at baseline were included in a predesigned substudy to assess physiologic outcomes. Sample sizes were chosen based on an of 0.05, a ß of 0.8, and estimates of the prevalence of hypertensive and diabetic persons in the population. The sample was designed to include enough hypertensive persons to detect a 10 mm Hg change in systolic and a 5 mm Hg change in diastolic blood pressure, enough diabetic persons to detect a 15% change in glycosylated hemoglobin, and a ±4 percentage point difference for dichotomous outcome variables for the entire study sample.

Overall, health status was assessed along the dimensions specified by the World Health Organization [6]. Patients rated their health as excellent, good, fair, or poor. Physical functioning was assessed with the nine-item battery used in the RAND Health Insurance Experiment [7]. Social functioning was measured using a modified five-item scale developed by Kane and colleagues [8]. Role function was measured with a two-item scale, and general health perceptions were measured with a four-item general health scale, both from the RAND Health Insurance Experiment [7]; Activities of Daily Living and Instrumental Activities of Daily Living were assessed with standard measures [9, 10]. Other than the physical functioning, Activities of Daily Living and Instrumental Activities of Daily Living measures, which are scored in terms of numbers of limitations, measures were scored such that a high score indicated better health.

Three domains of mental health status were measured: well-being, anxiety, and depression. Well-being was measured using items from the RAND Health Insurance Experiment [7]. Anxiety was measured with items from the Hopkins Symptom Checklist [11], and depression was measured with items from the Zung Depression Scale [12]. In all cases, a higher score indicated better health. Because pilot testing indicated that beneficiaries found the long mental health scales to be too intrusive, we used the three items from each scale with the highest published factor loadings.

We selected three physiologic indicators of health status: blood pressure control for hypertensive persons, glycosylated hemoglobin in diabetic persons, and visual acuity for the entire elderly population. These were chosen because they have been shown to be sensitive to changes in access to care [13, 14]. In the RAND Health Insurance Experiment, far visual acuity was better among low-income enrollees receiving free care [15], and we hypothesized that access to eye glasses or cataract surgery might differ among the fee-for-service and prepaid enrollees. Finally, we collected information regarding sociodemographic characteristics, access to care (usual source of care, delay and refusals of care, travel and waiting time), satisfaction with care (global satisfaction and satisfaction with provider and staff), and use of health services. Utilization data were available from client self-report at baseline and 1 year later and from the state Medicaid program and Part A Medicare claims for the demonstration year.

We used Medicare and Medicaid claims to measure inpatient use during the demonstration year. However, when we conducted an audit of medical records to validate a sample of the outpatient claims submitted to the state by the health plans, we found that they were incomplete, that there was substantial under- and over-reporting, and that the degree of accuracy varied by plan. Thus, in our analyses we use only self-reported outpatient use.

We interviewed sample members at baseline, which was the period between assignment to prepaid plans and 2 weeks after coverage started for experimental group patients. Control group interviews were conducted during a similar period. All patients were reinterviewed 1 year later, at which time we also interviewed proxy respondents when patients had died or were too ill to be interviewed. Methods for achieving high response rates are described, in part, by Bindman and colleagues [16].

Near and far visual acuity were measured for all patients using standard Snellen charts. Patients were instructed to use glasses if they routinely wore them.

After patients were interviewed for the evaluation, those who reported having hypertension or diabetes or both were visited by a physician or medical student. Standardized blood pressure measurements were obtained for all hypertensive persons; glycosylated hemoglobin levels were measured for all diabetic persons. These were repeated at the 1-year follow-up interview.

Data Analysis

We compared the distributions of variables between experimental and control populations for the baseline and follow-up periods using t-tests and chi-square techniques. In analyzing follow-up data, ordinary least-squares techniques were used to analyze continuous variables, whereas logistic regression was used for dichotomous variables. In all these analyses, the dependent variable was a health status measure at the follow-up interview. In addition, for health status measures that were continuous variables, we computed the difference between the value of the variable at baseline and follow-up and used the difference as the dependent variable. These results were similar to those in which the dependent variable was a health status measure at follow-up. For hospitalization and nursing home utilization data, we used tobit regression [17], a method for handling censored data (because of the large number of people with no admissions) as well as logistic regression. To minimize the loss of data, mean sample values were substituted for missing values of independent variables if the number of observations for which data on a specific variable were missing was less than 10%. Otherwise, the variable with missing data was not included in any analyses. In the regression models, we controlled for baseline values of sociodemographic characteristics, inpatient and outpatient use, general health status, physical function, activities of daily living, instrumental activities of daily living, social function, insurance, and length of time in the plan. For the health status variables, the regression-adjusted results were similar in magnitude and direction to the unadjusted findings, and the results of analyses using the dependent value at follow-up did not differ from those using change scores as the dependent variable. Thus, we report only unadjusted data. Regression adjustment did alter the magnitude of the utilization differences, however. Thus, unadjusted and regression-adjusted scores are presented for these data. For the logistic regression analyses, we report both the odds of having a visit in the prepaid group compared with fee-for-service group as well as the difference in the likelihood of a patient in the prepaid group having a visit compared with a patient in the fee-for-service group.

Finally, we calculated the average annual expenditures per person. For the fee-for-service group, this was the total of actual Medicare and Medicaid payments for the sample divided by the total number of beneficiaries. For the prepaid group, this was the total of capitation payments and estimated reinsurance payments divided by the total number of beneficiaries, plus the Medicare payments.

Results

Top Methods Results Discussion Author & Article Info References

We obtained second interviews for 387 control and 384 experimental group patients 1 year after the baseline interview, yielding 96% and 97% completion rates, respectively. Reasons for loss to follow-up appear in Table 1. Prepaid and fee-for-service groups did not differ significantly in any sociodemographic characteristics, baseline utilization Table 2, or baseline health status measures (Table 3). Patients were mostly female and white, and had, on average, three chronic conditions. Consistent with our expectations, the study sample reported significantly poorer health than did the overall Medicare population in the Twin Cities, based on survey data collected in 1989 (Wisner C. Personal communication). Sixty percent reported being in fair or poor health, in contrast to only 13% of the general Medicare population.

Use of services was lower in the prepaid group (Table 4). Based on logistic regression analyses, the difference in the likelihood of a patient in the prepaid group reporting an outpatient visit compared with a patient in the fee-for-service group was –16.6%(adjusted odds ratio, 0.44; CI, 0.29 to 0.74) and –21.2% for an emergency department visit (odds ratio, 0.40; CI, 0.25 to 0.63). Claims data indicated that, relative to the fee-for-service group, the difference in likelihood of hospitalization for the prepaid group was –11.2%(odds ratio, 0.55; CI, 0.32 to 0.94) and that considering all patients, length of stay for the prepaid group was 1.3 days shorter than for the fee-for-service group (CI, –0.06 to –7.78 days). The likelihood of being admitted to a nursing home did not change.

Despite these differences in use, beneficiaries' reports of access to or satisfaction with care did not differ. For example, 92% of prepaid and 94% of fee-for service patients were "very satisfied" or "satisfied" with their care (P > 0.2). Eighty-six percent of each group reported having a usual source of care, and 16.5% of prepaid and 18.6% of fee-for-service patients reported "at least some" difficulty getting emergency care (P > 0.2). The difference in average annual per-person expenditures made by Medicaid was $715 (CI, $103 to $1326), which was 27% lower for patients in the prepaid group. Medicare expenditures did not differ statistically between the two groups ($462; CI, -$1118 to $194).

Forty-four patients died during the 1-year follow-up period; 24 were in the fee-for-service group and 20 received capitated care (P > 0.2). Table 3 compares health outcomes of the prepaid and fee-for-service groups at baseline and follow-up. Blood pressure and glycosylated hemoglobin for hypertensive and diabetic persons, respectively, were similar in both groups, as were self-rated health, physical functioning, mental health and Activities of Daily Living and Instrumental Activities of Daily Living dependencies for the entire study population. Patients in the fee-for-service group reported slightly worse general health than patients in the prepaid group at follow-up, and these differences remained statistically significant after regression adjustment (0.4 points; CI, 0.06 to 0.72 points).

Discussion

Top Methods Results Discussion Author & Article Info References

The current health reform debate focuses on expanding access to care while maintaining quality and controlling costs. Capitation, managed care, and competition are seen by some as desired features of health care reform. This study is the first to report on use of services and quality of care from a randomized trial of capitated compared with fee-for-service Medicaid payment for elderly poor, a group at particular risk for adverse health outcomes as well as underservice. It reports on beneficiary outcomes aggregated across seven health plans of different types. States contemplating the use of capitation for elderly Medicaid beneficiaries, nearly all of whom are also Medicare beneficiaries, will only be able to capitate the Medicaid contribution for eligible patients. Thus, this demonstration represents a "real world" test of capitation for high-risk elderly Medicaid beneficiaries. Partially on the basis of this demonstration experience, health care for Medicaid recipients in Hennepin County is now being delivered via a capitation arrangement.

Although enrollees in the prepaid group used significantly less care, there was no evidence that they experienced poorer health during the study period, and beneficiary reports of access and satisfaction were comparable. Such findings are consistent with those of the National Medicare Competition Evaluation, in which Medicare enrollees in health maintenance organizations received equivalent or better quality of care for selected conditions and had comparable health outcomes [18-21] to those in fee-for-service Medicare. However, capitated and fee-for-service groups in those evaluations often differed in their characteristics, and follow-up data in some cases were incomplete. Studies of private employed groups that have addressed similar issues have generally not used a randomized design and therefore may suffer from selection bias because enrollees in prepaid plans may have differed in important ways from those who were cared for in the fee-for-service sector [22]. The only randomized trial to date examined the experience with capitation in the RAND Health Insurance Experiment [23]. This study found that the rate of hospital admissions for prepaid health plan enrollees was 40% lower than for fee-for-service patients. Two other studies from the same experiment [24, 25] reached differing conclusions regarding health outcomes. However, the health maintenance organization studied in the RAND Health Insurance Experiment was a staff model health maintenance organization with salaried physicians, which is not typical of most current prepaid plans.

Our results are also consistent with recent studies about outcomes of capitated care for poor, nonelderly populations. Carey and colleagues [26, 27] compared Medicaid enrollees receiving Aid to Families with Dependent Children (AFDC) in counties with capitated demonstration programs to AFDC populations in similar counties with traditional Medicaid fee-for-service care and found no difference in several aspects of process of care. Finally, Lurie and colleagues [28] found that health outcomes of chronically mentally ill Medicaid beneficiaries enrolled in capitated health plans did not differ statistically from those remaining under fee-for-service care.

Although neither sample sizes nor our previous agreements with the health plans permit plan-specific analyses, it is important to consider the reasons that use may have been lower for the prepaid group. Switching Medicaid beneficiaries from fee-for-service to capitation Medicaid financing might reduce their use because of changes in the financial incentives faced by their physicians, the application of managed-care techniques to control service use, or disruption in the continuity of care for beneficiaries that reduced use until new care patterns were established. Fortunately, because most patients did not change doctors, we can exclude disruption as the cause of lower use. Most of the physicians serving beneficiaries in our sample continued to receive fee-for-service payments from plans, often with discounts on fees and risk-sharing through a "withhold pool." Because the county-sponsored health maintenance organization was formed in response to the demonstration, this was the first exposure to capitation for many physicians practicing there. All of the plans used managed-care techniques such as prior authorization for surgery or physical therapy, concurrent review during hospitalization, or restricted formularies. Thus, it seems most likely that the observed reductions in services are caused largely by these efforts.

Several study limitations should be noted. First, because patients were followed for only a year, we do not know if adverse effects would have become evident over a longer period. However, the demonstration continued after our evaluation ended, and 101 prepaid and 111 fee-for-service group enrollees died in the first 3 years of the demonstration (P > 0.2). Although this is a crude measure of outcome, it is consistent with our other findings. Second, because only the Medicaid portion of expenditures was capitated, we cannot be certain that the findings would be similar if Medicare payments had also been capitated. Third, we made many comparisons between the prepaid and fee-for-service groups. The relatively few significant differences observed between the groups may have occurred on the basis of chance alone. Because of the large numbers of comparisons made and the consistent findings, it seems unlikely that use of additional measures would have altered the general conclusions of the study. Also, blood pressure and glycosylated hemoglobin levels vary from hour to hour, so measurements made at baseline and 1 year later are liable to substantial sampling error. Fourth, our outpatient use measures are based on client self-report because claims data proved inaccurate. However, we know of no reason that there would be differential self-reporting between the two groups that would influence our comparisons of use. Finally, most patients were enrolled in plans that also cared for privately insured populations, and we doubt that they treated study patients differently than their other capitated enrollees. Other Medicaid programs may enroll patients in prepaid plans serving only Medicaid patients. The evidence on whether others would fare as well in such plans is inconclusive, but the specter of the California Medicaid scandals in the 1970s is a reminder that the enrollment of Medicaid beneficiaries in such settings should be carefully monitored.