The effect of a Medicaid primary care case management program on the high-cost recipient population in one state.

Managed care delivery systems are the predominant form of coverage for Americans with private health insurance in the United States. Estimates of total managed care enrollment vary, depending upon the definitions used. InterStudy reports that as of July 2000 over 80 million Americans are enrolled in a Health Maintenance Organization (HMO), while Preferred Provider Organization (PPO) enrollment may have reached as high as 133 million lives (Interstudy 2001a, 2001b). This implies that about 90 percent of Americans with coverage are enrolled in some form of managed care.

The rapid growth in Medicaid expenditures, especially during the 1980s, prompted many states to evaluate the potential for managed care delivery systems to control that growth. States took a variety of approaches to managed care for the Medicaid population, ranging from mandatory enrollment of all Medicaid recipients in HMOs to the least restrictive Primary Care Case Management (PCCM) programs. Managed care has now become the dominant delivery system for the Medicaid population. According to the most recent figures, 56 percent of all Medicaid recipients are in some type of managed care program. Of the over 18 million Medicaid managed care participants, one quarter (4.6 million) were enrolled in a PCCM program in 2000 (HCFA 2000a, 2000b).

PCCM programs are designed to control utilization of services by providing physicians with a positive incentive to act as care managers, but avoid the incentives of some common risk sharing arrangements that might limit access to necessary care. A PCCM program typically locks-in recipients to a single Primary Care Provider (PCP) who manages medical care by providing authorization for hospital and specialty physician services. In addition to discounted fee-for-service reimbursement, PCPs are paid nominal monthly case management fees ($3 PMPM in Georgia). The PCP is expected to ensure proper utilization of services but the PCP income is not subject to a withhold or capitation that would put them at financial risk for excessive utilization. Since the PCCM program has only minimal effect on fees paid, any savings generated by the program must arise through reduced utilization.

High-cost Medicaid recipients are often included in Medicaid managed care programs, including PCCM programs. On the one hand, the skewed distribution of health care expenditures makes this population an excellent target for cost containment efforts. Small reductions in the volume of services provided to this group can generate substantial savings for public payers. On the other hand, it is possible that some of the care delivered to the general Medicaid population is provided to the "worried well" and that reductions in utilization should target these services, leaving the high-cost recipient unaffected. In addition, policy makers have expressed concern that managed care plans that reduce utilization may impose an undue burden on individuals with special needs and may compromise the quality of their care. Therefore, it is important to evaluate the effects of any programmatic changes on this population separately from the effects on the total population.

In this study, we evaluate the effect of a PCCM program on utilization and spending for a small group of very high-cost patients. When compared with the general Medicaid population, we found that the PCCM program had minimal impact on costs and no impact on utilization for this population, suggesting that this unrestrictive form of managed care poses no undue burden for this population. On the contrary, the PCCM program had a substantially greater effect on both cost and utilization for the general population, contributing to a further concentration of, expenditures in the high-cost population. These results suggest that non-restrictive managed care plans in general, and PCCM programs in particular may have little impact on the care provided to the high-cost population. Other programs, such as targeted disease management or individualized case management programs might better serve the needs of this group and achieve greater cost savings. Additional research on the effect of other types of managed care programs is necessary to identify the best mechanism for controlling utilization while ensuring access for this population. At the same time, the results suggest that savings attributed to PCCM may understate the full impact of the program on the vast majority of the participants in the program if such calculations are made without excluding the very high cost participants.

BACKGROUND AND LITERATURE REVIEW

Medicaid Managed Care and Primary Care Case Management

The rapid movement of Medicaid enrollees into one of several forms of managed care that occurred throughout the decade of the nineties appears to have stabilized. The latest managed care enrollment figures indicate that managed care growth between 1999 and 2000 was insignificant for the Medicaid population. PCCM programs have been one mechanism for implementing managed care, typically chosen in states with large, rural populations. PCCM programs are seen to have an advantage in rural areas because they focus on expanding access to services while avoiding the potential for providers to forgo quality or withhold services in response to capitated payments. In addition, traditional safety net providers can continue to serve the Medicaid population under a PCCM program.

There have been several published analyses of the effects of Medicaid managed care programs on cost, quality and access. Holahan et al. (1998) reviewed the experience of 13 states that had moved towards managed care in an effort to improve access and control costs. The authors found that even under the most restrictive managed care arrangements, projected savings to the states were only at 5 to 10 percent relative to fee-for-service, and that most savings would be generated through reductions in utilization. The study suggests that PCCM programs are a transitory step to more 1Lraditional managed care arrangements and the authors predict a gradual phase out of PCCM programs. The enrollment figures in Table 1 indicate that although most of the growth in managed care has been in other, more traditional managed care programs, the predicted reduction or elimination of PCCM enrollment has not occurred. While overall enrollment in Medicaid managed care plans grew rapidly and then stabilized, the number of Medicaid enrollees in PCCM programs has also grown, although not as rapidly as in other types of plans.

Slifkin et al. (1998) found that representatives from states implementing PCCM programs expressed a greater concern with ensuring access, while states that moved toward capitated delivery systems appeared to be more focused on cost containment. Felt-Lisk et al. (1999) did a follow-up study of Medicaid managed care in rural areas and found that providers experience increased competition for Medicaid patients in response to managed care programs, which enhances choice of providers for the Medicaid enrollees. The results highlight the benefit of enhanced choice between providers for the Medicaid population and may explain the persistence of the PCCM programs, despite their predicted demise.

High-Cost and Special Needs Populations

Policy makers at the state and federal levels share the concern about the effects of managed care on vulnerable Medicaid populations. The Secretary of Health and Human Services issued a report in November 2000 documenting the need to implement safeguards for individuals with special health care needs enrolled in Medicaid managed care programs. The Balanced Budget Act of 1997 (BBA) specifically directed the department to "conduct a study concerning the safeguards that may be needed to ensure that the health care needs of individuals with special health care needs ... enrolled with Medicaid managed care organizations are adequately met" (Section 4705(c)(2) of the BBA of 1997). The report generated in response to this provision specifically excludes an analysis of the effect of PCCM programs on the special-needs populations. However, that report noted the sparse information about the populations with special needs and made specific recommendations to states, including the need to:

* Ensure that these patients retain access to experienced providers,

* Ensure that care for this population is coordinated and continuous,

* Utilize appropriate reimbursement mechanisms for this population, and

* Facilitate additional research targeted at identifying the effects of various delivery systems on meeting the needs of this population (Shalala 2000).

PCCM programs do not give providers an incentive to under-ulilize services or restrict access to specialty care for vulnerable populations. This lack of risk sharing makes PCCM programs less suspect when implemented for broad populations, and is most likely the rationale for the exclusion of PCCM programs from the above mentioned study. However, the high concentration of expenditures within a small portion of the population implies that any managed care program, including PCCM programs, will be most successful in controlling costs if the utilization of high-cost enrollees is managed.

For any insured population, the distribution of health care expenditures is highly skewed (Berk & Monheit, 2001). The skewed distribution of health care expenditures implies that a small portion of the population consumes a large share of the program dollars and utilizes the most services. Berk & Monheit evaluate the distribution of health care expenditures in a series of publications (I 988, 1992, 2001) and show that the concentration of expenditures in the top five percent has remained constant over many decades of spending. This skewed distribution of expenditures implies that cost containment efforts targeting the highest cost enrollees have the greatest potential to generate savings. However, the authors find that the resources used by the bottom half of the population has remained constant over time at about three percent of total health care resources. The authors postulate that managed care might be expected to generate a more uniform distribution of expenditures because of an enhanced access to routine care for the healthy, and a reduction in resource intensity for the high-users. However, their findings do not support that hypothesis. Instead, they suggest that managed care has not appreciably altered the allocation of health care dollars within the population.

Morris Weinberger et al. (1996) evaluated the effect of enhanced primary care access on hospital readmission rates for a sample of chronically ill veterans. Patients with specific high-cost, chronic conditions from nine Veterans Affairs Medical Centers were randomly assigned to a study group with intensive primary care intervention or to receive routine care upon discharge. Patients who received the intensive primary care intervention had significantly higher readmission rates and more days of hospitalization than did those in the control group. These results suggest that access to primary care may not reduce costs for the chronically ill, and are consistent with the findings of Berk & Monheit cited above.

Taken together, these studies suggest that savings generated through managed care in general and through expanded primary care access specifically may represent a reduction in care provided to the "worried well" rather than a reduction in resource intensity for the high-users. We tested this hypothesis with respect to a primary care case management program for the Medicaid population in one state.

Georgia and Primary Care Case Management

Georgia implemented its PCCM program called GBHC pursuant to authorization obtained starting October 1, 1993. Enrollment proceeded gradually, starting in 1994 and continuing until February 1998 at which time all 159 counties in Georgia were included in the GBHC program. The stated goals of the program were to improve access to medical care, enhance continuity of care through creation of a medical home, and reduce unnecessary utilization of medical services. In Georgia, Medicaid recipients are mandated to enroll in the PCCM program called Georgia Better Health Care (GBHC) unless they reside in a nursing home, personal care home, mental health hospital or other domiciliary facility, or if they are eligible for short term Medicaid benefits (migrant workers, Right-from-the-Start Medicaid mothers). Medicaid recipients who failed to enroll in the program at the time it was implemented in their counties were auto-assigned to primary care providers. The gradual phase in of the program provides a natural experiment to test the effects of GBHC, as during that phase-in period enrollees and non-enrolled individuals vary only by county of residence.

This study is an analysis of the impact of Georgia Better Health Care as one example of a PCCM program on cost and utilization for a small subset of the population that has been identified as disproportionately high-cost enrollees. As mentioned, program savings are generated by reductions in utilization through the care management of the PCP, since there are only slight discounts associated with GBHC. If the PCCM program reduces utilization proportionately for high- and low-cost enrollees, the per-person savings generated among the high-cost enrollees will be substantially greater than that generated in the general population enrolled in the program. On the other hand, if the PCCM has no effect on utilization for the very high-cost population, then total reported savings may under estimate the actual effect on the low-cost enrollees in the system. This result would support the hypothesis that the high-cost patients are not adversely affected by this system, and that savings are achieved through a reduction in services to those with less resource intense conditions. The effect of GBHC on each population is assessed through bi-variate analysis of the distribution of health care expenditures for the entire eligible population, and through multivariate regression of the impact of GBHC on cost and utilization of physician services for the highest cost enrollees. Logistic analysis is used to estimate the probability of any emergency or hospital utilization as a function of enrollment in the GBHC program.

DESCRIPTION OF THE DATA

Data for this study are taken from a summary file extracted from claims paid for all Medicaid recipients in Georgia who are eligible for GBHC. The claims analyzed covered the period from program inception in 1994 through the third quarter of 1998, and data were extracted six months after the last date of service to allow for claims lag. Variables available for analysis include aggregate expenditures, expenditures for in- and out-patient services, physician services, and pharmaceuticals, and utilization data for each GBHC eligible participant. In addition, the data captured includes the category of eligibility, age of each recipient, county of residence, and case management fees paid for each claimant if actually participating in GBHC. Unfortunately, the data do not include any diagnostic or procedure codes that would enable researchers to identify chronically ill individuals by condition. Age is typically correlated with utilization and category of eligibility captures the fact that some individuals are medically needy and therefore eligible for GBHC. This study analyzes the effect of participation in GBHC on expenditures and on utilization, and the availability of some demographic information allows for control for some of the known determinants of utilization (age, sex, category of service).

Georgia implemented its PCCM program gradually, starting prior to the quarters studied and completing statewide implementation during the first quarter of 1998. Data from 1/94 to 9/94 (three quarters) were utilized to identify the high-cost population. Individuals were considered "high-cost" if total expenditures for each of these three quarters were at least one standard deviation above the mean. In addition, the high-cost definition excluded any individual who resided in a nursing home at any point in the study, all dual-eligible participants (Medicaid and Medicare) and any individual who was not eligible for GBHC for the entire time period studied (4th quarter 1994 though 3rd quarter 1998). This definition of high-cost claimants is very restrictive, resulting in a relatively small sample of 1,047 individuals. These individuals have consistently high expenditures in every calendar year and were not classified as high cost due to a temporary condition or trauma. All individuals classified as high-cost had significant expenditures for care in each calendar year of the study.

Table 2 demonstrates enrollment in GBHC for the quarters analyzed for the entire GBHC eligible population and for the high-cost sample. Individuals across the state were progressively enrolled into the GBHC program based on county of residence, and as this Table demonstrates, the high-cost individuals enrolled at the same rate as did the remaining GBHC eligible population. Statewide implementation was completed during the first quarter of 1998. After that time, any individuals not enrolled with a GBHC provider had obtained an exemption and were systematically different than program participants.

ANALYSIS OF GBHC IMPACT

Concentration of Expenditures

After identifying the high-cost sample, claims from October 1994 through September 1998 were analyzed to evaluate the effect of GBHC on per-person expenditures for the high-cost population compared other GBHC eligible aid-categories. Many individuals are eligible for Medicaid for a limited time, in particular women and infants with eligibility tied to a specific pregnancy. There may be a substantial difference in the propensity to utilize services for those who are short-term eligible and those who are continuously eligible over the study period. For example, lack of familiarity with GBHC procedures could systematically alter utilization and expenditures for those with brief periods of eligibility when compared to the high-cost, continuously eligible study group. In addition, GBHC should have little effect on utilization of services for pregnant women. Therefore, the control group of GBHC eligible individuals was restricted to those for whom data were available for all 16 quarters of the study period. This restriction resulted in a control group of 235,954 individuals who were in the aid categories eligible for GBHC during the four years studied.

The per-person quarterly expenditures for the high-cost sample and for all continuously GBHC eligible Medicaid recipients (control group) are shown in Table 3. The ratios shown in the last column compare the mean per-person expenditures for the high-cost group with the control group. It appears that as a greater proportion of the eligible population enrolled in GBHC, the concentration of expenditures among the high-cost sample increased rather than decreased. The high-cost sample represents only 0.4 percent of the continuously eligible population for all 16 quarters analyzed, but expenditures as a share of total for the continuously eligible population increased from 6.4 at the beginning of the GBHC program to 8.3 at the completion of the phase-in of the PCCM program.

Effect of GBHC on Expenditures

The mean per-person expenditures for individuals enrolled with a GBHC provider and those not enrolled are provided by quarter for the high-cost sample and for the sample of continuously enrolled GBHC eligible individuals in Table 4. The quarters of rolling enrollment are used. After that time, any individuals not enrolled in the program (less than 10 percent in each quarter) had received approval to remain outside the GBHC program and are therefore systematically different from enrollees. From Table 4 it is clear that the savings associated with GBHC are small for both groups but more consistent in the control group than in the high-cost sample.

It is difficult to assess the effect of GBHC on level of expenditures over time without adjusting for medical care cost inflation and seasonal variations in expenditures. However, a fixed effects model that controls for quarter of estimation should allow for an estimate of the impact of GBHC on costs. Therefore, the following model was estimated for the high-cost and the non-high-cost control group:

(1) LTE = [[beta].sub.1](Quarter)+ [[beta].sub.2](age) + [[beta].sub.3](GBHC) + [epsilon].

LTE is the log of total expenditures, setting total expenditures to 0.01 for all individuals with zero quarterly expenditures. The estimate of [[beta].sub.3] in each regression demonstrates the relative effect of GBHC on quarterly costs. Due to system limitations, the regression was performed on a random sample of 20 percent of the observations from the 16 quarter study period for the control group.

The full results of each regression can be found in Appendix 1. Key results are shown in Table 5. Age is included as a categorical variable, with the omitted age category adults 45 and older. It is interesting to note that the effect of age on expenditures differs between the control group and the high-cost group. For the control group, age is inversely related to expenditures, as expected. However, for the high-cost sample, all age groups had higher expenditures compared to the oldest group. This unusual result requires additional study.

The use of log of expenditures implies that the coefficients reflect the percent change associated with that variable. Therefore, the results indicate that the effect of GBHC on reducing costs for the non-high-cost group (-0.197) is almost twice the effect for the high-cost group (-0.102) although GBHC appears to reduce total expenditures for both groups. These results are consistent with the univariate analysis.

Utilization of Services

One of the primary goals of the GBHC program is to reduce excess utilization of emergency room services through increased access to primary care services provided in a physician's office. Therefore, it is important to evaluate the effect of GBHC on utilization of emergency room services and physician visits for high-cost and all other participants. If GBHC is controlling utilization of emergency room services equally for all participants, then the ratio of high-cost group utilization to' other long-term GBHC eligible group utilization, will be constant during and after the phase-in of GBHC. On the other hand, if GBHC is more effective in reducing excess utilization of emergency rooms for the control group, then the difference will be increasing over the study period.

Similarly, the ratio of total physician visits and inpatient days for the high-cost group and the control group of continuously eligible individuals demonstrates the relative effectiveness of the GBHC program in controlling physician visits and hospitalizations for the two groups. Table 6 demonstrates that for all types of utilization considered in this study, utilization for the control group dropped substantially during the phase-in of GBHC and remained constant once GBHC was fully implemented, but that the ratio between the two groups is increasing during the study period. This implies that GBHC had a much smaller or no effect on controlling utilization for the high-cost population, while utilization was reduced for the control group of continuously GBHC eligible Medicaid participants. The drop in utilization noted overall through the program is not indicative of access problems for those with chronic conditions, but rather indicates reduction in utilization for those who are not chronically ill.

Regression Analysis 

In order to quantify the effects of GBHC on utilization for the high-cost group, the probability of having an emergency room visit or any inpatient stay was estimated using a fixed effects logit model. This approach allows control for systematic variation in utilization by quarter that would relate to seasonality rather than to the GBHC program. In addition, the number of total physician visits (specialty and primary combined) was also estimated as a function of age and GBHC participation using a fixed effects model.

(2) Probability(ERV > 0) = [[beta].sub.1](Quarter) + [[beta].sub.2](age) + [[beta].sub.3](GBHC) + [epsilon].

(3) Probability(IPD > 0) = [[beta].sub.1](Quarter) + [[beta].sub.2](age) + [[beta].sub.3](GBHC) + [epsilon].

(4) #MDvisits = [[beta].sub.1](Quarter) + [[beta].sub.2](age) + [[beta].sub.3](GBHC) + [epsilon].

The full results of each regression are available upon request. Key results are shown in Table 7. (1)

The effect of GBHC on the probability of utilizing the emergency room (-0.108) and of having any inpatient days (-0.0476) is negative and significant for the non-high cost sample, and is either positive or insignificant for the high-cost sample. The effect of GBHC on total MD visits is positive and significant for the high-cost sample (0.379) and negative and significant for the control group (-0.10). These results provide additional support to the univariate analysis. GBHC is reducing utilization for the long-term low cost participants, but the only measurable impact of the program on utilization for the high-cost participants is positive.

DISCUSSION

An ongoing concern at the federal level is that managed care organizations serving the Medicaid population would disproportionately burden participants with special health care needs. Primary care case management programs do not have the same problematic incentives associated with capitated programs, and this study provides evidence that individuals who rely on Medicaid to meet substantial long-term health care needs do not experience restricted access to care as a result of the case management program. It is clear that the PCCM program in one state has not substantially hindered the high-cost participants identified for this study from utilizing services. On the contrary, the ability of case managers to influence utilization appears to be limited for this subset of the population, although overall the case management program reduced utilization for the control group. This finding is consistent with the findings of Monheit and Berk that managed care did not decrease the relative expenditures for the high-cost segment of the population. It is also consistent with the findings of Weinberger et al. that enhanced access to primary care case management for individuals with significant health problems did not significantly decrease costs or utilization.

If policy makers hope that through case management programs such as the one studied, patients with high utilization might be discouraged from consuming care with a low marginal benefit, there is little evidence from this study that this is occurring. In fact, resource utilization is more heavily concentrated among high-cost participants as a result of the PCCM. While the PCCM accounts for about a 10 percent decline in per-person expenditures in a fixed-effects model for high-cost participants (95 percent confidence interval -0.04 to -0.16), it accounts for about a 19 percent decline using the same model for the control group (95 percent confidence interval -0.17 to -0.22). Thus the effect of GBHC on expenditures for high-cost participants is substantially smaller than for the control group.

Limitations and Direction for Additional Research

This study presents a preliminary analysis of the differential effect of a case management program on high-cost and non-high cost long term participants. The sample selected for this study is representative of very high-cost users. It is possible that the health status of the high-cost sample deteriorated over time, resulting in higher utilization for this group. Although it is not possible to obtain health status information for this particular population, additional studies that control for deteriorating health status over time or use a cross sectional approach would be valuable. In addition, a more refined analysis should be undertaken using a data set that includes procedure and diagnostic codes. It is possible that the impact of any case management program on utilization and costs varies by diagnosis. If the PCCM program has been unsuccessful in affecting utilization for specific high-cost diagnoses, it would indicate that targeted disease management programs might be a better choice for particular high-cost long-term enrollees.

APPENDIX 1

PANEL A: ANALYSIS OF VARIANCE

                      REGRESSION: HIGH-COST
                            INDIVIDUALS

                              Sum of        Mean
Source              DF        Squares      Square

Model                 16       290.408      18.15
Error              12957     25915.341       2.00
C Total            12973     26205.749
Root MSE                                     1.41
Dep. Mean                                    8.51
C.V.                                        16.62
F value                                      9.08
Prob. > F                                    0.00
[R.sup.2]                                    0.01
Adj. [R.sup.2]                               0.01

                      REGRESSION: NON-HIGH COST
                            CONTROL SAMPLE

                              Sum of          Mean
Source              DF        Squares        Square

Model                 17       551450.53    32438.27
Error             657348     13528958          20.58
C Total           657365     14080409
Root MSE                                        4.54
Dep. Mean                                       2.32
C.V.                                          195.73
F value                                      1576.1
Prob. > F                                       0.0001
[R.sup.2]                                       0.039
Adj. [R.sup.2]                                  0.039

PANEL B: PARAMETER ESTIMATES

                      REGRESSION: HIGH-COST
                          INDIVIDUALS

            Parameter      Std.      T for [H.sub.0]:
Variable    Estimate      Error        Parameter = 0

INTERCEP      8.36         0.05          159.06
GBHC         -0.10         0.03           -3.54
Q5            0.16         0.07            2.46
Q6            0.13         0.07            1.93
Q7            0.04         0.07            0.67
Q8            0.11         0.07            1.60
Q9            0.01         0.07            0.09
Q10           0.07         0.07            1.03
Q11          -0.01         0.06           -0.16
Q12           0.03         0.06            0.53
Q13           0.01         0.06            0.09
Q14           0.04         0.06            0.71
Q15           0.08         0.06            1.31
Q16           0.13         0.06            1.98
INFANT         N/A
CHILD         0.07         0.04            1.71
YNGADULT      0.05         0.05            0.86
ADULT         0.28         0.03           10.07

                     REGRESSION: NON-HIGH-COST
                          CONTROL SAMPLE

            Parameter      Std.      T for [H.sub.0]:
Variable    Estimate      Error        Parameter = 0

INTERCEP      4.51         0.03           143.10
GBHC         -0.20         0.01           -15.36
Q5            0.61         0.03            20.23
Q6            0.79         0.03            26.58
Q7            0.37         0.03            12.57
Q8            0.36         0.03            12.22
Q9            0.55         0.03            18.55
Q10          -0.20         0.03            -5.96
Q11          -0.57         0.03           -17.27
Q12           0.08         0.03             2.75
Q13           0.24         0.03             8.44
Q14           0.29         0.03            10.15
Q15           0.18         0.03             6.20
Q16           0.14         0.03             4.91
INFANT       -1.765        0.028            0.000
CHILD        -3.08         0.02          -130.00
YNGADULT     -1.90         0.04           -51.77
ADULT        -1.38         0.03           -48.50
TABLE 1

MEDICAID MANAGED CARE AND PCCM ENROLLMENT

                                                          PCCM as
         Total       Managed                              Share of
        Medicaid       Care      % Managed      PCCM      Managed
Year   Population   Enrollment     Care      Enrollment     Care

2000   33,690,364   18,786,137    55.76%     4,635,409     24.67%
1999   31,940,188   17,756,603    55.59%     4,274,456     24.07%
1998   30,896,635   16,573,996    53.64%     4,003,421     24.15%
1997   32,092,380   15,345,502    47.82%     4,337,486     28.27%
1996   33,241,147   13,330,119    40.10%     4,016,773     30.13%
1995   36,200,000   11,619,929    32.10%     3,628,449     31.23%
1994   33,634,000   7,794,250     23.17%     2,385,157     30.60%

Source: Adapted from HCFA, 2001.
TABLE 2

PERCENT ENROLLED IN GBHC THROUGH PHASE-IN PERIOD AND BEYOND

            All GBHC
            Eligible     High-cost
             Groups       Sample

Q4, 1994      13.9%        12.6%
Q1, 1995      19.6%        16.4%
Q2, 1995      22.8%        19.0%
Q3, 1995      22.7%        18.7%
Q4, 1995      27.9%        24.6%
Q1, 1996      28.4%        24.8%
Q2, 1996      35.5%        33.1%
Q3, 1996      47.6%        47.0%
Q4, 1996      55.5%        57.3%
Q1, 1997      64.8%        63.8%
Q2, 1997      73.6%        71.2%
Q3, 1997      77.1%        74.0%
Q4, 1997      86.5%        81.5%
Q1, 1998      95.6%        91.4%
Q2, 1998      95.7%        90.7%
Q3, 1998      91.7%        90.4%
TABLE 3

AVERAGE PER-PERSON EXPENDITURES AND
RATIO OF HIGH-COST TO NON-HIGH-COST EXPENDITURES

                 HIGH-COST GROUP
                   (N = 1,047)

Quarter      Mean       Median     Std. Dev.

Q4, 1994    $7,544      $5,476        6,350
Q1, 1995    $7,770      $5,354        7,021
Q2, 1995    $7,297      $5,103        6,965
Q3, 1995    $7,657      $5,464        7,062
Q4, 1995    $7,329      $5,301        6,479
Q1, 1996    $7,973      $5,511        8,358
Q2, 1996    $7,694      $5,304        7,962
Q3, 1996    $7,793      $5,487        7,201
Q4, 1996    $7,617      $5,510        6,488
Q1, 1997    $8,126      $5,586       11,656
Q2, 1997    $8,206      $5,962        8,659
Q3, 1997    $8,113      $6,071        6,904
Q4, 1997    $7,456      $5,020        7,204
Q1, 1998    $7,952      $5,546        7,711
Q2, 1998    $7,966      $5,459        7,270
Q3, 1998    $9,195      $6,143        8,307

                        CONTROL GROUP
                        (N = 235,954)
                                                 Ratio of
Quarter       Mean       Median     Std. Dev.      Means

Q4, 1994     $490        $101         1,818        15.39
Q1, 1995     $482        $108         1,661        16.12
Q2, 1995     $427         $87         1,559        17.09
Q3, 1995     $442         $86         1,617        17.31
Q4, 1995     $444         $96         1,603        16.52
Q1, 1996     $455         $94         1,636        17.53
Q2, 1996     $424         $78         1,622        18.14
Q3, 1996     $425         $75         1,609        18.32
Q4, 1996     $435         $82         1,542        17.51
Q1, 1997     $451         $88         1,576        18.02
Q2, 1997     $453         $81         1,655        18.13
Q3, 1997     $445         $78         1,629        18.22
Q4, 1997     $417         $78         1,682        17.87
Q1, 1998     $452         $88         1,983        17.61
Q2, 1998     $442         $74         1,818        18.04
Q3, 1998     $454         $75         1,969        20.25
TABLE 4

AVERAGE EXPENDITURES FOR ENROLLED AND
NOT YET ENROLLED GBHC ELIGIBLE POPULATIONS
DURING GBHC PHASE-IN

                        CONTROL
                         GROUP

                                     Per-person
                                    Savings from
            Non-GBHC      GBHC          GBHC

Q4, 1994     $503        $412           $91 **
Q1, 1995     $494        $432           $62 **
Q2, 1995     $440        $383           $57 **
Q3, 1995     $457        $394           $63 **
Q4, 1995     $459        $403           $56 **
Q1, 1996     $471        $414           $57 **
Q2, 1996     $431        $412           $19 **
Q3, 1996     $435        $414           $21 **
Q4, 1996     $437        $434            $3
Q1, 1997     $470        $441           $29 **
Q2, 1997     $476        $444           $32 **
Q3, 1997     $456        $442           $14 *
Q4, 1997     $430        $415           $15
Q1, 1998     $420        $453          ($33)
Q2, 1998     $447        $441            $6
Q3, 1998     $318        $466         ($148)

                       HIGH-COST
                      STUDY GROUP

                                     Per-person
                                      Savings
                                     (increase)
            Non-GBHC      GBHC       from GBHC

Q4, 1994     $7,403      $8,520       ($1,117)
Q1, 1995     $7,571      $8,780       ($1,209)
Q2, 1995     $7,440      $6,687          $753
Q3, 1995     $7,695      $7,491          $205
Q4, 1995     $7,579      $6,565        $1,014 **
Q1, 1996     $8,150      $7,435          $715
Q2, 1996     $7,828      $7,422          $406
Q3, 1996     $7,863      $7,713          $150
Q4, 1996     $7,463      $7,732         ($269)
Q1, 1997     $8,081      $8,151          ($70)
Q2, 1997     $8,566      $8,059          $507
Q3, 1997     $9,209      $7,729        $1,481 **
Q4, 1997     $9,271      $7,044        $2,227 **
Q1, 1998    $11,149      $7,651        $3,498
Q2, 1998    $11,369      $7,618        $3,751
Q3, 1998    $13,034      $8,785        $4,249

** Savings significant at the [alpha] = 0.05 level.

* Savings significant at the [alpha] = 0.10 level.
TABLE 5

EFFECT OF GBHC ON LOG OF TOTAL QUARTERLY EXPENDITURES

           REGRESSION: HIGH-COST INDIVIDUALS
              DEPENDENT VARIABLE: LOGTTX

                Parameter        Std.
  Variable       Estimate       Error       Prob. > |T|

INTERCEP          8.365         0.053          0.000
GBHC             -0.102         0.029          0.000
INFANT             n/a
CHILD
(1 to 18)         0.070         0.041          0.088
YNGADULT
(19 to 24)        0.047         0.055          0.393
ADULT
(25 to 44)        0.276         0.027          0.000

       REGRESSION: NON-HIGH-COST CONTROL SAMPLE
             DEPENDENT VARIABLE: LOGTTX

                Parameter        Std.
  Variable       Estimate       Error       Prob. > |T|

INTERCEP           4.511        0.032          0.000
GBHC              -0.197        0.013          0.000
INFANT            -1.765        0.028          0.000
CHILD             -3.081        0.024          0.000
YNGADULT          -1.905        0.037          0.000
ADULT             -1.384        0.029          0.000
TABLE 6

UTILIZATION PER 1,000 BY SERVICE TYPE AND RATIO OF UTILIZATION:
CONTROL GROUP TO HIGH-COST STUDY GROUP

                         ER VISITS PER 1,000

               Control         High-cost
                Group            Group
            (N = 235,954)     (N = 1,047)         Ratio

Q4, 94           195              629              3.2
Q1, 95           196              545              2.8
Q2, 95           185              496              2.7
Q3, 95           182              527              2.9
Q4, 95           183              474              2.6
Q1, 96           146              465              3.2
Q2, 96           137              509              3.7
Q3, 96           125              474              3.8
Q4, 96           130              439              3.4
Q1, 97           122              523              4.3
Q2, 97           126              557              4.4
Q3, 97           121              591              4.9
Q4, 97           117              508              4.3
Q1, 98           118              509              4.3
Q2, 98           118              526              4.5
Q3, 98           113              611              5.4

                       MD VISITS PER 1,000

               Control         High-cost
                Group            Group
            (N = 235,954)     (N = 1,047)         Ratio

Q4, 94          1,504            4,837             3.2
Q1, 95          1,593            5,144             3.2
Q2, 95          1,331            4,397             3.3
Q3, 95          1,310            5,060             3.9
Q4, 95          1,438            4,589             3.2
Q1, 96          1,377            4,876             3.5
Q2, 96          1,204            4,570             3.8
Q3, 96          1,173            4,368             3.7
Q4, 96          1,258            4,341             3.4
Q1, 97          1,269            4,521             3.6
Q2, 97          1,181            4,321             3.7
Q3, 97          1,139            4,518             4
Q4, 97          1,150            4,188             3.6
Q1, 98          1,221            4,598             3.8
Q2, 98          1,092            4,575             4.2
Q3, 98          1,070            5,444             5.1

                      INPATIENT DAYS PER 1,000

               Control         High-cost
                Group            Group
            (N = 235,954)     (N = 1,047)         Ratio

Q4, 94           195             2,609            13.4
Q1, 95           161             2,440            15.2
Q2, 95           125             1,855            14.9
Q3, 95           124             2,562            20.6
Q4, 95           127             2,330            18.3
Q1, 96           125             2,232            17.9
Q2, 96           102             1,706            16.8
Q3, 96           102             1,528            15
Q4, 96           111             1,714            15.4
Q1, 97           112             2,138            19.1
Q2, 97           110             1,987            18
Q3, 97           103             1,744            16.9
Q4, 97           101             1,699            16.9
Q1, 98           112             1,964            17.6
Q2, 98           103             1,807            17.6
Q3, 98           108             2,619            24.1

All differences between high-cost and control group are significant
at the [alpha] = 0.01 level
TABLE 7

EFFECT OF GBHC ON UTILIZATION OF SERVICES

                                      Parameter    Std.
             Variable                 Estimate     Error    Prob. > |T|

Logistic Regression--Any ER Visits:
  High Cost Sample: GBHC Effect        0.0981     0.0463      0.0343
  Non High Cost Sample: GBHC Effect   -0.1081     0.00868     0.0001

Logistic Regression--Any IP days:
  High Cost Sample: GBHC Effect        0.0576     0.0503      0.2521
  Non High Cost Sample: GBHC Effect   -0.0475     0.0187      0.011

Regression--Total MD Visits:
  High Cost Sample: GBHC Effect        0.379      0.1295      0.0035
  Non High Cost Sample: GBHC Effect   -0.100      0.0063      0.0001

REFERENCES

Berk, M.L., A.C. Monheit, & M.M. Hagan (1988) "How the U.S. Spent Its Health Care Dollar: 1929-1980," Health Affairs, 7(3):46-60.

Berk, M.L. & A.C. Monheit (1992) "The Concentration of Health Expenditures: An Update," Health Affairs, 12(4):145-149.

Berk, M.L., & A.C. Monheit (2001) "The Concentration of Health Care Expenditures, Revisited," Health Affairs, 20(2):9-18.

Felt-Lisk, S., S. Hoag, P. Silberman & R. Slifkin (1999) "Medicaid Managed Care in Rural Areas: A Ten-State Follow-up Study," Health Affairs, 18(2):238-246.

Health Care Financing Administration (2000a) 2000 Enrollment Fact Sheet, US Department of Health and Human Services, Health Care Financing Administration, www.hcfa.gov/medicaid/omc2000.htm, accessed June 2001.

Health Care Financing Administration (2000b) 2000 Medicaid Managed Care Enrollment Report, US Department of Health and Human Services, Health Care Financing Administration, www.hcfa.gov/medicaici/plansum0.pdf, accessed June 2001.

Health Care Financing Administration (2001) Medicaid Statistics and Data, www.hcfa.gov/medicaid/trends1.htm, www.hcfa.gov/medicaid/trends00.pdf, www.hcfa.gov/ medicaid/typenrol.pdf, www.hcfa.gov/medicaid/plansum4.pdf, www.hcfa.gov/medicaid/ plansum7.pdf, www.hcfa.gov/medicaid/plansum8.pdf, www.hcfa.gov/medicaid/plansum9.pdf, www.hcfa.gov/medicaid/plansum0.pdf, accessed June 2001.

Holahan, J., S. Zuckerman, A. Evans & S. Rangarajan (1998) "Medicaid Managed Care in Thirteen States," Health Affairs, 17(3):43-63.

InterStudy (2001a) HMO Enrollment Continues to Decrease, (May 7), www.hmodata.com/Pdf/ir111pr.pdf, accessed June 2001.

InterStudy (200lb) InterStudy's PPO Database, Directory and Performance Report Now Available, (February 26), www.hmodata.com/pdf/PPO10PressRelease(11_00).pdf, accessed June 2001.

Miller, W., A. Sumner, S. McLaren, J. Johnson, J. Newman & R. Curry (2000): "Georgia Better Health Care 1999-2000 Program Evaluation Final Report," Institute of Health Administration, Georgia State University, (June 30).

Shalala, D. (2000): Report to Congress: Safeguards for Individuals with Special Health Care Needs Enrolled in Medicaid Managed Care, (November).

Slifkin, R., S. Hoag, P. Silberman, S. Felt-Lisk & B. Popkin (1998) "Medicaid Managed Care Programs in Rural Areas: A Fifty State Overview," Health Affairs, 17(6):217-227.

Weinberger, M. & E. Oddone (1996): "Does Increased Access to Primary Care Reduce Hospital Readmissions?" New England Journal of Medicine, 334 (22):1441-1448.

(1) As a control for sample size effects, the regression for the non-high cost group was repeated on a sub-sample of approximately equal size to the high-cost group. The effect of GBHC on utilization of ER and physician visits remained negative and significant, while the impact on the probability of inpatient days was negative but not significant.

Patricia G. Ketsche
Georgia State University (USA)

William J. Miller
Georgia State University (USA)

Susan A. McLaren
Georgia State University (USA)

Address for correspondence: Patricia G. Ketsche, Institute of Health Administration, Georgia State University, University Plaza, Atlanta, Ga 30303, USA, [e-mail].