Printer Friendly


The overall health maintenance organization (HMO) market has been expanding in the US during the past two decades. The estimated enrollment grew from 9.1 million enrollees in 1980 to 52.5 million in 1996 (Interstudy 1997). The Medicare program, that insures 38 million of the nation's retirees, however, has been an exception to this growth phenomenon until recently. As of 1998, there were approximately 6.8 million Medicare enrollees in HMOs, compared to only 1.7 million enrollees 10 years ago. Of the 6 million managed care enrollees, approximately 90 percent are under capitated risk contracts, while the remainder are cost reimbursed contracts (Health Care Financing Administration [HCFA] 2000). This means approximately one out of every six Medicare beneficiaries is enrolled in an HMO. Although Medicare's HMO enrollment rate still remains lower than the overall market average, it has been growing vigorously at a rate of more than 25 percent per year during the last five years.

Since the Medicare program began contracting with HMOs, it has relied on an administrative capitation rate setting method known as the Adjusted Average Per Capita Cost (AAPCC). This method paid risk HMOs an amount equivalent to 95 percent of the estimated five-year moving-average amount of FFS healthcare expenditures for beneficiaries with similar characteristics in each county. This capitation method was based solely on the county's FFS sector experience.(1)

A growing proportion of prepaid patients in an area may influence the behaviors of healthcare providers. One can hypothesize that it creates an incentive for healthcare providers to increase service utilization in the FFS sector. However, the spillover effect alternatively theorizes that HMOs reduce the health expenditure in the FFS sector. The exact mechanisms of the spillover are not well understood, however, the following have been proposed. As an HMO promotes a cost-conscious style of healthcare practice among its providers, those physicians who have contracts with HMOs are likely to treat all of their patients more cost consciously. In particular, the risk HMOs that bear all financial risk in providing healthcare under a capitation system have strong incentives to promote cost-conscious medicine among their contracting physicians. According to recent data, 64 percent of physicians in the US had contracts with HMOs, and 83 percent had contracts with more broadly defined managed care organizations (Emmons & Simon 1996). If the physicians who treat both FFS and HMO patients apply the practice styles of managed care to FFS patients, as the hypothesis contends, a reduction of per capita healthcare costs in the FFS sector will follow.

The practice-style spillover is but one of several ways in which HMOs can influence healthcare expenditure in the FFS sector. Chernew (1995) suggests that managed care may fundamentally alter the healthcare environment in which the FFS sector operates, and consequently affect the FFS sector expenditures. For example, the managed care sector may change the hospital market structure and the availability of expensive technologies, which could lead to a cost reduction in the FFS sector. Others point to a competitive response from the FFS sector as the cause of reduction in healthcare costs. When managed care enters a market, the FFS sector may realize that it has to behave competitively lest it lose patients to the HMO sector. This would force the FFS sector to become more cost conscious, and offer more restrictive benefit plans at lower premiums. Despite differences in purported mechanisms, these hypotheses would predict that, other things being equal, HMOs will decrease the area's FFS sector healthcare expenditures. Additionally, these mechanisms seem to imply that an area with a high HMO enrollment rate will have relatively low FFS sector expenditures, controlling for other factors. This study examines the spillover experience by the Medicare program in the US, where HMOs' entry into FFS dominated healthcare markets may have influenced the behavior of the FFS sector. The US Medicare experience with the spillover phenomenon should be an instructive model for other countries when experimenting with capitated healthcare payment systems.

Interest in the effects of HMOs on healthcare expenditures is not new. Numerous studies have examined the effect of HMOs on overall healthcare expenditures using various data, costs, and utilization measures. An early study by Chiswick (1976) found that hospital admission rates were lower in areas with HMOs. However, as shown in Table 1, many studies on the subject have found conflicting evidence.

(subject population)

 Affirmative Inconclusive or Negative

Dowd 1986 McLaughlin 1984
(Inpatient) (Inpatient)

McLaughlin 1987 Hay & Leahy 1984
(Inpatient) (Inpatient)

Welch 1991 Newhouse, Schwartz, et al 1985
(Total) (Inpatient)

Robinson 1991 Feldman, Dowd, et al. 1986
(Inpatient) (Inpatient)

Clement, et al.(a) 1994 Luft, Maerki, et al 1986
(Total) (Total)

Baker(a) 1994 Rossiter(a) 1988
(Total) (Total)

Zwanziger, et al. 1994 McLaughlin 1988
(Inpatient) (Inpatient)

Rodgers and Smith(a) 1995 Morgan, et al 1997
(Total) (Inpatient)

(a) Studies that used FFS sector cost as a dependent variable.

There is a limit to the conclusions one can draw from most of the prior research. Because these papers examined changes in overall healthcare expenditures and utilization, rather those of the FFS sector, we cannot tell whether a reduction in an area-wide healthcare expenditures is due to spillover, or due to reduced utilization in the HMO sector. However, several authors began to examine the FFS sector costs in isolation, in order to test the spillover hypothesis. A study by Clement et al. (1992) is one such example. They sampled a total of two thousand Medicare FFS beneficiaries residing in 48 market areas with sufficient HMO enrollment between 1985 and 1989. Their analysis found that a 10 percentage point increase in capitated HMO enrollment is associated with up to a rive percentage point reduction in Medicare FFS sector per capita expenditure. More recently, others investigated the spillover effects in Medicare using nationwide data. Baker (1993) and Rodgers & Smith (1995) are such studies. Using national data between 1986 and 1990 Baker estimated the effect of the total HMO market share (risk and cost combined) as the key variable in his study. The results indicate that a 10 percentage point increase in the HMO share of Medicare enrollment is associated with 1.8 percent decrease in Medicare expenditure for Hospital Insurance and 0.7 percent decrease in Supplemental Medical Insurance.(2) Based on such a relationship, the author suggested increasing HMO enrollment to contain healthcare expenditures in the FFS sector of Medicare. Similarly, Rodgers & Smith (1995) advocated taking advantage of the spillover phenomenon to contain FFS sector costs in Medicare. In this paper, we go beyond the basic approaches taken by earlier researchers, and investigate several additional aspects of the spillover hypothesis.


An ideal spillover study would use total FFS market data. However, the fragmented market structure and propriety nature of private sector data in the US make it difficult to construct an accurate and comprehensive national FFS sector expenditure data base. Medicare (the Federal medical insurance for the 38 million retirees), on the other hand, maintains accurate nation-wide health expenditure data available to the public. Consequently, we investigate this relationship using Medicare sector data despite the limited scope.

For this study, we used the following data. The average Medicare FFS sector per capita cost was obtained from the AAPCC Masterfile for all US counties between 1988 and 1993 (source: Office of Actuary, HCFA). The Medicare HMO enrollment figures were obtained from the Medicare HMO Enrollment file to construct Medicare HMO penetration rates (source: Office of Managed Care, HCFA). In order to measure spillover effects from non-Medicare HMOs, we used InterStudy HMO enrollment data (InterStudy 1997) The Area Resource File (ARF) supplied other relevant county-level socioeconomic variables.

During the period under examination, a major payment reform in Medicare was phased in. The physician payment methodology based on the Resource-Based Relative Value Scale (RBRVS) began to revise physician payments by procedures and physician specialties starting in 1992 (Hsiao 1992). A relevant aspect of the reform was that it adjusted a national physician fee schedule across geographic areas called localities. To control for differential impacts of the RBRVS reform across these areas, a locality-level fee schedule impact variable was included (Office of the Actuary HCFA).


A long run implication of the spillover hypothesis is that areas with high HMO penetration rates will eventually exhibit lower FFS sector expenditures. To test this, we examined the association between FFS sector costs and HMO penetration rates using cross-sectional data. Using the ordinary least squares (OLS) regression method, the effects of Medicare HMO penetration rates on the FFS sector per capita expenditures are estimated. Included in the regression are factors expected to influence healthcare utilization such as income and availability of healthcare, e.g., number of hospital beds and physicians per resident. Regressions were run separately for each year.

Table 2 reports estimates from the cross-sectional study using 1993 data. The cross-sectional estimates for each of the six years were highly similar, and demonstrate a positive association between FFS costs and both the capitated HMO and cost-reimbursed HMO market penetration. All variables were significant at the 90 percent confidence level or higher. The risk variables' coefficients were consistently larger than those of the cost contract penetration variables, and the magnitudes in Medicare Part B were consistently larger than Part A in each year.


Dependent Variables Part A Cost Part B Cost Total Cost

Cost 93 0.3796(***) 0.3267(***) 0.3473(***)
 (4.323) (3.622) (4.152)
Risk 93 0.5250(***) 1.1582(***) 0.7647(***)
 (12.969) (27.855) (19.830)
Bed 93 -0.0012 -0.0051(***) -0.0026(*)
 (-0.691) (-2.764) (-1.516)
MD 93 0.0470(***) 0.0301(***) 0.0411(***)
 (14.602) (9.102) (13.406)
Income 93 0.0028(***) 0.0106(***) 0.0057(***)
 (3.172) (11.474) (6.608)
Intercept 7.5927 6.928 8.0116
Adj. [R.sup.2] 0.2476 0.3807 0.3165

(a) Dependent variable is log transformed. Regressions are weighted by
county Medicare population. The t-statistics (in parentheses) follow
the normal distribution:

(***) p [is less than or equal to] 0.01 (2.326)

(**) p [is less than or equal to] 0.05 (1.645)

(*) p [is less than or equal to] 0.10 (1.282)

The results show that high HMO penetration markets have higher FFS Medicare average costs, rather than lower as predicted by the spillover hypothesis. This is contrary to the long run predictions of the spillover hypothesis, and even seems to support the cost shifting hypothesis instead. This positive relationship, however, can be consistent with spillover hypothesis under the following cases. Studies have shown that a high Medicare capitation rate is an important predictor of capitated risk HMO enrollment in Medicare (Adamache & Rossiter 1985; Porell & Wallack 1990). This means it is the high capitation rate that attracts risk contract HMOs to selectively enter high FFS sector cost areas. As long as FFS sector costs were initially higher in these markets, despite any FFS cost reducing effects of spillover, the FFS sector costs could still remain higher than in other areas with little or no risk HMO presence. In other words, the selective location effect may outweigh any possible effect of spillover in the short run. If the selective market entry is, indeed, a prevalent phenomenon, the cross-sectional study design will not allow measurement of a spillover effect. For further investigation, we turned to a fixed effect model that takes full advantage of our panel data.


It is impossible to identify and obtain all relevant variables that cause healthcare costs to differ in each county. Recognizing this problem, we used a fixed-effects regression model. The model controls for unique characteristics of the local market area with county-specific qualitative (dummy) variables. Also, we explicitly controlled for factors expected to influence area healthcare utilization, i.e., availability of physicians and hospital beds per population, and average income. A set of year-specific intercepts was employed to control for changes in overall Medicare program cost during the study period.

Tables 3-5 report estimation results from the full data analyses. The risk penetration and cost penetration rates measure current year effects of spillover generated by the respective types of HMOs, while the lagged penetration variables estimate delayed effects. Thus, for example, the sum of the current and the lagged effects represents the total effect from HMO penetration rates. We also included non-Medicare HMO penetration rates in our regressions to control for spillover effects from other HMOs. The regression results from Part A show that all Medicare HMO penetration rates are positively associated with the FFS sector inpatient cost per capita, although the magnitudes are small. This seems to suggest that the cost-shifting effect may outweigh the spillover effect for inpatient care. This is consistent with a Medicare inpatient cost study by Morgan et al. (1997), which found that Medicare inpatient cost is positively correlated with the HMO enrollment rate in southern Florida. The fact that Medicare beneficiaries are allowed to disenroll from their HMOs on a monthly basis, and are allowed back into FFS might be an important factor behind the observed phenomenon.(3) Non-Medicare HMOs were not observed to affect Medicare average expenditure.


Dependent Part A Part A Part A
Variable Per Capita Per Capita Per Capita
 Cost Cost Cost

Cost -0.0993(*) -0.1035(*) -0.1264(*)
 (-1.314) (-1.373) (-1.640)
Lag Cost 0.2708(***) 0.2411(***)
 (4.471) (4.019)
Risk -0.0244
Lag Risk 0.0809(***)
Other HMO 0.0001 -0.0001 0.0001(*)
 (0.778) (-0.881) (1.307)
Lag HMO -0.0001(***) -0.0006(***)
 (-4.127) (-4.002)
Income 0.0018(***) 0.0019(***) 0.0018(***)
 (2.586) (2.694) (2.549)
MD -0.0251(***) -0.0262(***) -0.2603(***)
 (-3.866) (-4.047) (-4.013)
Bed -0.0013(**) -0.0013(*) -0.0014(**)
 (-1.648) (-1.614) (-1.733)
Intercept -0.1467 -0.1461 -0.1460
F-statistics 7623.74 9519.68 10862.70
Adj. [R.sup.2] 0.8611 0.8610 0.8608

Dependent Part A Part A
Variable Per Capita Per Capita
 Cost Cost


Lag Cost

Risk -0.0069 0.0428(**)
 (-0.211) (1.794)
Lag Risk 0.0620(***)
Other HMO 0.0001 0.0001
 (0.749) (1.081)
Lag HMO -0.0006(***)
Income 0.0017(***) 0.0018(***)
 (2.369) (2.496)
MD -0.2574(***) -0.0266(***)
 (-3.968) (-4.110)
Bed -0.0015(**) -0.0014(**)
 (-1.852) (-1.765)
Intercept -0.1463 -0.1462
F-statistics 9509.81 10863.61
Adj. [R.sup.]2 0.8609 0.8608

(a) Dependent variable is log transformed. Regressions
weighted by inverse of error variance. Models include year
dummies and individual county dummies. The t-statistics (in
parenthesis) follow the normal distribution:

(***) p [is less than or equal to] 0.01 (2.326)

(**) p [is less than or equal to] 0.05 (1.645)

(*)p [is less than or equal to] 0.10 (1.282)


Dependent Part B Part B Part B
Variable Per Capita Per Capita Per Capita
 Cost Cost Cost

Cost -0.3493(***) -0.1862(***) -0.1970(***)
 (-5.190) (-2.675) (-2.858)

Lag Cost -0.0168 0.0624
 (-0.313) (1.128)
Risk -0.5637(***)
Lag Risk -0.1075(***)
Other HMO 0.0002(***) 0.0001 0.0001
 (2.487) (0.131) (0.147)
HMO lag 0.0008(***) 0.0005(***)
 (5.746) (3.781)
RBRVS 0.0025(***) 0.0052(***) 0.0052(***)
 (3.618) (7.119) (7.109)
Income 0.0018(***) 0.0022(***) 0.0022(***)
 (2.794) (3.383) (3.349)
MD 0.0238(***) 0.0244(***) 0.0244(***)
 (4.127) (4.086) (4.094)
Bed 0.0029(***) 0.0035(***) 0.0034(***)
 (3.914) (4.573) (4.546)
Intercept -0.1206 -0.1226 -0.1226
F-statistics 4443.00 4948.19 5566.58
Adj. [R.sup.2] 0.7990 0.7836 0.7836

Dependent Part B Part B
Variable Per Capita Per Capita
 Cost Cost


Lag Cost

Risk -0.5504(***) -0.6484(***)
 (-18.816) (-30.325)
Lag Risk -0.1123(***)
Other HMO 0.0002(**) 0.0002(***)
 (2.264) (2.443)
HMO lag 0.0008(***)
RBRVS 0.0027(***) 0.0031(***)
 (3.842) (4.414)
Income 0.0016(***) 0.0015(***)
 (2.538) (2.345)
MD 0.0220(***) 0.0236(***)
 (3.813) (4.096)
Bed 0.0026(***) 0.0025(***)
 (3.633) (3.453)
Intercept -0.1208 -0.1210
F-statistics 5414.88 6077.90
Adj. [R.sup.2] 0.7985 0.7981

(a) Dependent variable is log transformed. Regressions
weighted by inverse of error variance. Models include year
dummies and individual county dummies. The t-statistics (in
parenthesis) follow the normal distribution:

(***) p [is less than or equal to] 0.01 (2.326)

(**) p [is less than or equal to] 0.05 (1.645)

(*) p [is less than or equal to] 0.10 (1.282)


Dependent Total Total Total
Variable Per Capita Per Capita Per Capita
 Cost Cost Cost

Cost -0.2203(***) -0.1608(***) -0.1906(***)
 (-4.399) (-3.195) (-3.828)
Lag Cost 0.1450(***) 0.1717(***)
 (3.618) (4.289)
Risk -0.2090(***)
Lag Risk -0.0328(**)
Other HMO -0.0001 -0.0001(*) -0.0001
 (-0.159) (-1.308) (-1.251)
HMO lag 0.0001 0.0001
 (1.010) (0.130)
RBRVS -0.0102(***) -0.0093(***) -0.0093(***)
 (-19.293) (-17.578) (-17.609)
Income 0.0007(*) 0.0009(**) 0.0008(**)
 (1.524) (1.873) (1.720)
MD -0.0081(*) -0.0080(**) -0.0078(**)
 (-1.892) (-1.850) (-1.816)
Bed 0.0002 0.0004 0.0004
 (0.436) (0.848) (0.725)
Intercept -0.1380 -0.1387 -0.1386
F-statistics 12460.60 14931.20 16771.85
Adj. [R.sup.2] 0.9177 0.9162 0.9161

Dependent Total Total
Variable Per Capita Per Capita
 Cost Cost


Lag Cost

Risk -0.1929(***) -0.2317(***)
 (-8.860) (-14.600)
Lag Risk -0.0458(***)
Other HMO -0.0001 -0.0001
 (-0.329) (-0.326)
HMO lag 0.0001
RBRVS -0.0102(***) -0.0100(***)
 (-19.196) (-19.015)
Income 0.0005 0.0005
 (1.162) (1.057)
MD -0.0093(***) -0.0087(**)
 (-2.187) (-2.037)
Bed 0.0001 0.0001
 (0.076) (0.023)
Intercept -0.1380 -0.1381
F-statistics 15181.04 17069.51
Adj. [R.sup.2] 0.9174 0.9174

(a) Dependent variable is log transformed. Regressions weighted by
inverse of error variance. Models include year dummies and individual
county dummies. The t-statistics (in parenthesis) follow the normal

(***) p [is less than or equal to] 0.01 (2.326)

(**) p [is less than or equal to] 0.05 (1.645)

(*) p [is less than or equal to] 0.10(1.282)

On the other hand, we found that Medicare Part B costs are negatively associated with HMO penetration variables (Table 4). The risk contract variable shows particularly strong effects here; i.e., its (current and lagged) combined estimates closely cluster between values of -0.641 and -0.660. Should we assume that the relationship is causal, the range corresponds to approximately a 6.61 percent to 6.88 percent reduction in Part B expenditure for a 10 percentage point increase in the risk contracting rate.(4) The cost contract variables report smaller coefficients that range between -0.150 and -0.212, or assuming causality, a reduction of 1.39 percent to 1.91 percent per 10 percentage point increase in the cost HMO contracting rate.

When the total expenditures (Part A and B combined) are examined, we find a small but consistent negative effect from the cost contracts (range from -0.079 to -0.193). In the case of risk contracts, we find a sizeable effect of spillover with the parameter values ranging from -0.230 to -0.240. They are translated into approximately a 2.05 to 2.13 percent reduction in FFS cost per one percentage point increase in risk contracting rate. Here again, the private sector HMO enrollment rates have no statistical significance in relation to the Medicare FFS sector expenditure variable.

Medicare has annual FFS sector expenditures in excess of $190 billion. This means that even a small cost reducing effect of HMOs translates into an enormous amount of savings should we assume a structurally stable and causal relationship in spillover. Before agreeing with that assumption, it seemed prudent to further examine the nature of this relationship.


While most Medicare beneficiaries lire in a county with one or more Medicare HMO plans, a majority of enrollments are heavily concentrated in a few states such as California, Florida, New York, and Arizona. As such, estimating a single parameter across high and low HMO penetration areas seems to be a restrictive approach. It is very plausible that magnitudes of the relationship may vary depending on the degrees of HMO penetration. To this end, the data were stratified into several groups with various thresholds at 2.5 percent, 5 percent and 7.5 percent Medicare HMO penetration rates and analyzed using the fixed effect model. Tables 6 and 7 report our regression results for areas with risk penetration rates above and below 2.5 percent as the threshold. We find the spillover coefficients generally show larger magnitudes in low penetration areas than in high penetration areas--for a given percentage point increase in HMO enrollment, the areas with low HMO penetration rates exhibit a larger reduction in FFS sector per capita costs. The inverse relationship holds essentially unchanged in other sets of stratified regressions using alternative thresholds; the relationship is stronger in markets with lower HMO penetration rates.


Dependent Part A Part A Part B
Variable Per Capita Per Capita Per Capita
 Cost Cost Cost

Risk 0.2796(***) 0.2217(***) -0.4251(***)
 (2.611) (2.635) (-4.301)
Lag Risk -0.0677 -0.1527(**)
 (-0.875) (-2.154)
RBRVS 0.0001
Income -0.0071(**) -0.0076(***) 0.0006
 (-2.181) (-2.378) (0.232)
MD -0.0076 -0.0074 0.050(**)
 (-0.243) (-0.239) (1.761)
Bed 0.0052 0.0044 -0.0341(***)
 (0.592) (0.507) (-4.175)
Intercept -0.1471 -0.1476 -0.1204
F-statistics 664.19 759.26 346.06
Adj. [R.sup.e] 0.8998 0.8998 0.8401

Dependent Part B Medicare Medicare
Variable Per Capita Total Total
 Cost Per Capita Per Capita
 Cost Cost

Risk -0.5534(***) -0.0495 -0.1415(***)
 (-6.992) (-0.794) (-2.830)
Lag Risk -0.1095(***)
RBRVS 0.0003 -0.0061(***) -0.0059(***)
 (0.139) (-4.423) (-4.272)
Income 0.0005 -0.0044(***) -0.0053(***)
 (-0.166) (-2.363) (-2.849)
MD 0.0510(**) 0.0124 0.0128
 (1.777) (0.688) (0.708)
Bed -0.0360(***) -0.0074(*) -0.0087(**)
 (-4.423) (-1.436) (-1.706)
Intercept -0.12 -0.1387 -0.1395
F-statistics 386.33 1349.14 1504.16
Adj. [R.sup.e] 0.8401 0.9536 0.9532

(a) Dependent variable is log transformed. Regressions weighted by
inverse of error variance. Models include year dummies and individual
county dummies. The t-statistics (in parenthesis) follow the normal

(***) p [is less than or equal to] 0.01 (2.326)

(**) p [is less than or equal to] 0.05 (1.645)

(*)p [is less than or equal to] 0.10(1.282)


 Dependent Part A Part A Part B
 Variable Per Capita Per Capita Per Capita
 Cost Cost Cost

Risk -0.0292 0.0498(**) -0.5721(***)
 (-0.834) (1.955) (-18.346)
Lag Risk 0.0898(***) -0.0967(***)
 (3.282) (-3.942)
RBRVS 0.0035(***)
Income 0.0025(***) 0.0025(***) 0.0017(***)
 (3.368) (3.486) (2.586)
MD -0.0220(***) -0.0236(***) 0.0215(***)
 (-3.307) (-3.555) (3.631)
Bed -0.0017(**) -0.0016(**) 0.0031(***)
 (2.130) (-2.006) (4.191)
Intercept -0.1460 -0.1459 -0.1202
F-statistics 8839.75 10092.60 5067.10
Adj. [R.sup.2] 0.8580 0.8579 0.7958

 Dependent Part B Medicare Medicare
 Variable Per Capita Total Total
 Cost Per Capita Per Capita
 Cost Cost

Risk -0.6560(***) -0.2042(***) -0.2294(***)
 (-28.790) (-8.735) (-13.432)
Lag Risk -0.0289(*)
RBRVS 0.0039(***) -0.0104(***) -0.0103(***)
 (5.125) (-18.045) (-17.994)
Income 0.0016(***) 0.0009(**) 0.0009(**)
 (2.515) (1.842) (1.814)
MD 0.0232(***) -0.0078(**) -0.0073(**)
 (3.929) (-1.770) (-1.659)
Bed 0.0029(***) 0.0001 -0.0001
 (4.043) (0.043) (-0.016)
Intercept -0.1203 -0.1378 -0.1378
F-statistics 5691.47 13936.23 15675.97
Adj. [R.sup.2] 0.7955 0.9147 0.9146

(a) Dependent variable is log transformed. Regressions
weighted by inverse of error variance. Models include
year dummies and individual county dummies. The t-statistics
(in parenthesis) follow the normal distribution:

(***) p [is less than or equal to] 0.01 (2.326)

(**) p [is less than or equal to] 0.05 (1.645)

(*) p [is less than or equal to] 0.10 (1.282)

The inverse relationship with respect to the HMO penetration rates is a surprising finding. It indicates that the phenomenon peaks at an early phase of HMO penetration, but diminishes quickly thereafter, due to a rapidly declining marginal effect. This seems to challenge the practice style spillover notion as the mechanism behind this relationship. If the Medicare FFS sector cost reduction were due to the spread of managed care style medical practice promoted by the Medicare HMOs, the spillover effects would likely grow stronger with an increase in HMO enrollment. However, the fact that we observe stronger spillover effect in areas with very low penetration rates, e.g., 5 percent or even 2.5 percent, seems to mitigate against the practice style spillover mechanism as a cause of this phenomenon. On the other hand, the competition hypothesis may not be inconsistent with this relationship if the FFS sector reacts competitively in anticipation of the growing significance of HMOs in the marketplace.


It is important to recognize that the HMO markets with similar HMO penetration rates could be in different stages of maturity. One would wish to analyze a long time series of data to examine spillover effects over time. Limited data, however, made it impossible to observe such phenomenon over a long time period. Instead, we recognize that the HMO markets may be at different stages of maturity, and selected states that are characterized by a long history of HMO presence and high rates of enrollment.

Incidentally, the states in the Western Census region seemed to be a good model of mature HMO markets. These include California, Oregon, Washington, and Hawaii, all of which have a long history and significant HMO presence. The regression results of these states are shown in Table 8. Few observations in the data made most of the estimates insignificant, but the Western region reported positive coefficients, while the test of the country (Table 9) continued to show some spillover relationship. If the HMO sector influences FFS sector expenditures through some mechanism, there could be a time period beyond which the HMO may hot further affect the FFS sector cost. The absence of a spillover relationship in mature HMO markets confirms this expectation, and raises a question about the sustainability of this phenomenon. This suggests that spillover may be a transitory phenomenon which occurs in earlier stages of HMO development. This point will be revisited in our conclusion section, below.


 Dependent Part A Part A Part B
 Variable Per Capita Per Capita Per Capita
 Cost Cost Cost

Risk 0.0304 -0.1440 0.3666(**)
 (0.178) (-1.017) (2.053)
Lag Risk -0.1686(**) -0.070
 (-1.823) (-0.741)
RBRVS -0.0080(*)
Income -0.0058(**) -0.0076(***) 0.0055(**)
 (-1.911) (-2.613) (1.696)
MD -0.0279 -0.0309 0.2030(***)
 (-0.818) (-0.905) (5.824)
Bed 0.0027 0.0026 -0.0011
 (1.139) (1.091) (-0.459)
Intercept -0.1518 -0.1527 -0.0892
F-statistics 807.42 918.37 242.616
Adj. [R.sup.2] 0.9213 0.9210 0.8161

 Dependent Part B Medicare Medicare
 Variable Per Capita Total Total
 Cost Per Capita Per Capita
 Cost Cost

Risk 0.2904(**) 0.1897(**) 0.0284
 (1.990) (1.717) (0.313)
Lag Risk -0.1500(***)
RBRVS -0.0087(*) 0.0043(*) 0.0030
 (-1.618) (1.299) (0.900)
Income 0.0048(*) -0.0021 -0.0034(**)
 (1.559) (-1.061) (-1.767)
MD 0.2015(***) 0.0646(***) 0.0613(***)
 (5.793) (2.995) (2.836)
Bed -0.0011 0.0010 0.0009
 (-0.475) (0.694) (0.639)
Intercept -0.0894 -0.1288 -0.1294
F-statistics 306.87 1304.34 1452.03
Adj. [R.sup.2] 0.8162 0.9551 0.9547

(a) Dependent variable is log transformed. Regressions weighted
by inverse of error variance. Models include year dummies and
individual county dummies. The t-statistics (in parenthesis)
follow the normal distribution:

(***) p [is less than or equal to] 0.01 (2.326)

(**) p [is less than or equal to] 0.05 (1.645)

(*) p [is less than or equal to] 0.10(1.282)


 Dependent Part A Part A Part B
 Variable Per Capita Per Capita Per Capita
 Cost Cost Cost

Risk -0.0268 0.0501(**) -0.5495(***)
 (-0.798) (2.065) (-18.560)
Lag Risk 0.08847(***) -0.1181(***)
 (3.303) (-4.967)
RBRVS 0.0028(***)
Income 0.0023(***) 0.0024(***) 0.0012(**)
 (3.188) (3.315) (1.917)
MD -0.0265(***) -0.0280(***) 0.0171(***)
 (-4.000) (-4.236) (2.934)
Bed -0.0024(***) -0.0022(***) 0.0030(***)
 (-2.734) (-2.589) (3.860)
Intercept -0.1460 -0.1458 -0.1223
F-statistics 87771.02 10014.01 5215.16
Adj. [R.sub.2] 0.8566 0.8565 0.7998

Dependent Part B Medicare Medicare
Variable Per Capita Total Total
 Cost Per Capita Per Capita
 Cost Cost

Risk -0.6509(***) -0.1985(***) -0.2320(***)
 (-30.303) (-8.911) (-14.371)
Lag Risk -0.0390(**)
RBRVS 0.0033(***) -0.0104(***) -0.0103(***)
 (4.676) (-19.245) (-19.121)
Income 0.0012(**) 0.0005 0.0005
 (1.812) (1.049) (1.004)
MD 0.0190(***) -0.0114(***) -0.0107(***)
 (3.277) (-2.600) (-2.456)
Bed 0.0028(***) -0.0002 -0.0003
 (3.639) (-0.482) (-0.580)
Intercept -0.1226 -0.1388 -0.1389
F-statistics 5852.17 14062.49 15814.62
Adj. [R.sub.2] 0.7994 0.9151 0.9151

(a) Dependent variable is log transformed. Regressions weighted
by inverse of error variance. Models include year dummies and
individual county dummies. The t-statistics (in parenthesis)
follow the normal distribution:

(***) p [is less than or equal to] 0.01 (2.326)

(**) p [is less than or equal to] 0.05 (1.645)

(*) p [is less than or equal to] 0.10(1.282)


The notion that HMOs influence physician practice style in the FFS sector has a very potent cost containment implication in the United States. Our study analyzed Medicare program data and found evidence that supports this view. However, several questions and issues remain unanswered. The following points discuss the limitations in our study.

Primarily, the scope of the study is limited to the national Medicare program experience. Although it accounts for about 20 percent of the US healthcare expenditures, there are many private payers in the US health insurance market. As the observed relationship is based solely on the Medicare experience, it is unclear whether findings from Medicare are generalizable to the test of the US health insurance market. Secondly, unavailability of data for the overall managed care activity variables weakens confidence in our estimates. It seems reasonable that it is the overall managed care industry that contributes to the spillover phenomenon. However, the lack of appropriate measures of managed care activity required us to focus only on the HMO sector as the source of spillover. It is not clear to what extent the broader managed care sector contributes to the overall spillover phenomenon, and how much our estimates are inflated by the omission of such a variable.

This article examined if high HMO penetration areas have lower FFS expenditure using cross-sectional information, under the assumption that, at least in the long run, areas with high HMO penetration rates may show lower FFS sector expenditures. We found no such evidence in the cross-sectional data. If anything, our FFS cost variable showed a strong positive correlation with HMO enrollment rates. We interpret this as collaborating evidence that Medicare HMOs tend to enter high FFS sector markets.

Alternatively, using full panel data, our fixed effect regression models measured spillover effects in Medicare. The results showed some spillover effects from capitated contracting, but not from cost contracting in Medicare. It is notable that the spillover effect was stronger in ambulatory care (Part B) than in inpatient care (Part A). This seems consistent with the studies that found HMOs tend to enroll a healthier population than the average FFS sector population, possibly leaving in the FFS sector a population with higher likelihood of hospital admission (Brown & Bergeron 1993; Riley et al. 1994; Morgan et al. 1997). We found the capitated HMO contracting rate is negatively associated with FFS expenditure. Our estimates ranged from a 2.05 to a 2.13 percent reduction in FFS expenditure per 10 percentage point growth in HMO enrollment.

Since we wanted to find out if this phenomenon was enduring and stable, we first compared magnitudes of the spillover across high and low HMO penetration areas. Contrary to our expectations, the results showed a stronger effect in lower HMO penetration areas than in higher HMO enrollment areas. This meant that the phenomenon is subject to a rapidly declining marginal effect, even at low penetration rates.

Second, recognizing that the HMO markets are at different stages of maturity, we comparatively estimated magnitudes of spillover, singling out the Western census region which has a long history and well-established HMO sector. Surprisingly, the estimates were insignificant and positive, while regressions from the rest of the country confirmd the presence of a spillover relationship; we did not find evidence of a spillover relationship in the mature markets.

Once the HMO sector in an area reaches a certain maturity level, further changes in the significance of the HMO sector may not influence the FFS sector cost any longer. In other words, the spillover effect may be a transitory phenomenon.

This study analyzed Medicare data and found that capitated enrollment is associated with lower FFS sector cost, particularly in the ambulatory care setting. The spillover hypothesis represents a potent policy for cost containment, provided that a stable and causal relationship exists. However, further analysis suggested that this may be a transitory phenomenon, whose effects dissipate over time with rising HMO enrollments. Until we clearly understand the nature and the mechanism of the spillover phenomenon, we recommend, for the time being, that healthcare policymakers in the US and abroad refrain from developing a proactive cost containment policy based on assumptions regarding this phenomenon. A microlevel study may be required to improve our understanding of the actual nature of the spillover effect, and thus avoid incurring any unintended consequences of implementing such a policy. Medicare has physician level data for both HMO and FFS sector enrollees starting with the year 2000. We are optimistic that further research using the latest data will shed valuable light on this issue.

(1) A change in the payment policy made in the Balanced Budget Act (BBA) of 1997 changed the Medicare HMO payment system by blending county-specific rates with the national rate. However, the Medicare capitation amount continues to be set using the FFS sector average cost.

(2) Commonly known as Part A benefit, the Hospital Insurance fund reimburses inpatient hospital, skilled nursing facility, home health, and hospice services. Medicare Part B, or Supplemental Medical Insurance, on the other hand, reimburses for physician services, diagnostic tests, radiology, pathology, various therapies, and medical supplies. We will refer to them as Part A and Part B hereafter.

(3) This insightful point was raised by an anonymous referee of the journal. The Medicare program has begun to phase in an annual lock-in enrollment period to deal with this problem.

(4) To calculate the spillover effect in models using the log transformation, use the following formula: exp[(risk rate coefficient + lag risk coefficient) x %change] -1. Smearing's estimate correction (Duan 1983) was calculated, but the parameter results remain virtually unaffected.


Adamache, K. & L. Rossiter (1985) The Entry of HMOs into the Medicare Market: Implications for TEFRA Mandate, Princeton, NJ: Mathematica Policy Research, Inc.

Baker, L.C. (1993) Can Managed Care Control Health Care Costs: Evidence from the Medicare Experience, Princeton University and Robert Wood Johnson Foundation (November).

Brown, R.S. & J. Bergeron (1993) The Medicare Risk Program for HMOs: Final Summary Report on Findings from the Evaluation, Princeton, NJ: Mathematica Policy Research, Inc. (February).

Chernew, M. (1995) "The Impact of Non-IPA HMOs on the Number of Hospitals and Hospital Capacity," Inquiry 32(2):143-154.

Chiswick, B.R. (1976) An Analysis of SMSA Differences in Occupancy Rates, Admission Rates, and Bed Rates, Explorations in Economic Research, 3:326-378.

Clement, D.G., P. Gleason & R. Brown (1992) The Effects of Risk Contract HMO Market Penetration on Medicare Fee-for-Service Costs, Princeton, NJ: Medical College of Virginia and Mathematica Policy Research, Inc.

Dowd, B. (1987) "HMOs and Twin Cities Admission Rates," Health Services Research, 21:177-188.

Duan, N. (1983) "Smearing Estimate: A Nonparametric Retransformation Method," Journal of the American Statistical Association, 78(3):605-610.

Emmons, D. & C. Simon (1996) "Managed Care: Evolving Contractual Arrangements, Socioeconomic Characteristics of Medical Practice 1996," Studies on the Socioeconomic Environment of Medicine, American Medical Association, 15-25.

Feldman, R. & B. Dowd (1986) "The Competitive Impact of Health Maintenance Organizations on Hospital Finances: An Exploratory Study," Journal of Health Politics, Policy, and Law, 10(4):675-698.

Hay, J.W. & M.J. Leahy (1984) "Competition Among Health Plans: Some Preliminary Evidence," Southern Economic Journal, 1:831-846.

Health Care Financing Administration (2000) Profiles of Medicare--Medicare 2000:35 Years of Improving Americans' Health and Security, Baltimore, MD: Health Care Financing Administration (July).

Hsiao, W.C., P. Braun, E.R. Becker, D.L. Dunn, N. Kelly, N. Causino, M.D. McCabe & E. Rodriguez (1992) "An Overview of the Development and Refinement of the Resource-Based Relative Value Scale," Medical Care, 30(11):NS1-NS12.

InterStudy Publications (1997) InterStudy Competitive Edge, Minneapolis, MN: InterStudy Publications.

Luft, H.S., S.C. Maerki & J.B. Trauner (1986) The Competitive Effects of Health Maintenance Organizations: Another Evidence from Hawaii, Rochester, and Minneapolis/St. Paul, Journal of Health Politics, Policy and Law, 10:625-658.

Morgan, R.O., B.A. Viring, C.A. DeVito & N.A. Persily (1997) "The Medicare-HMO Revolving Door--The Healthy Go In and the Sick Go Out," The New England Journal of Medicine, 37:3 169-175.

McLaughlin, C.G. (1987) "HMO Growth and Hospital Expenses and Use: A Simultaneous Equation Approach," Health Services Research, 22:183-202.

McLaughlin, C.G. (1988) "The Effects of HMOs on Overall Hospital Expenses: Is Anything Left After Correcting for Simultaneity and Selectivity?" Health Services Research, 421-441.

McLaughlin, C.G., J.C. Merrill & A.J. Freed (1984) "The Impact of HMO Growth on Hospital Costs and Utilization," Advances in Health Economics and Health Service Research, Greenwich, CT: JAI Press, 57-93.

Newhouse, J.P., A.P. Schwartz & C. Witsberger (1985) "Are Fee-for-Service Costs Increasing Faster than HMO Costs?" Medical Care, 23(8):960-966.

Porell, F. & S. Wallack (1990) "Medicare Risk Contracting: Determinants of Market Entry," Health Care Financing Review, 12:75-85.

Riley, G., C. Tutor, Y. Chiang & M. Ingber (1996) "Health Status of Medicare Enrollees in HMOs and FFS in 1994," Health Care Financing Review, 17(4):65-76.

Robinson, J.C. (1992) "HMO Market Penetration and Hospital Cost Inflation in California," Journal of the American Medical Association, 266:2719-2723.

Rodgers, J. & K.E. Smith (1995) Do Medicare HMOs Reduce Fee-for-Service Costs? Report to GHAA, Health Policy Economics Group, Price Waterhouse.

Rossiter, L.F. (1988) National Medicare Competition Evaluation: Impacts on the Fee-For-Service Sector, Final Analysis Report, Report to HCFA, Williamson Institute for Health Studies, Medical College of Virginia.

Welch, W.P. (1991) HMO Market Share and its Effect on Local Medicare Costs, Washington, DC: The Urban Institute.

Zwanzinger, J., G. Melnick & A. Bamezai (1994) "Costs and Price Competition in California Hospitals, 1980-1990," Health Affairs, 13(4):118-126.

The author wishes to thank Phil Cotterill, Jerry Riley, Al Peden, and most of all Mel Ingber for their encouragement and helpful comments.

The author was with the Office of Strategic Planning, US Health Care Financing Administration when this paper was written. The opinions expressed are those of the author and do not necessarily reflect those of the US Health Care Financing Administration.

Address for correspondence: Jay P. Bae, Department of Economics, Georgia State University, University Plaza, Atlanta, GA 30302-4039 USA,
Jay P. Bae
Georgia State University (USA)
COPYRIGHT 2001 isRHFM Ltd. Towson, MD. All rights reserved.
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2001, Gale Group. All rights reserved. Gale Group is a Thomson Corporation Company.

Article Details
Printer friendly Cite/link Email Feedback
Title Annotation:health maintenance organization
Author:Bae, Jay P.
Publication:Research in Healthcare Financial Management
Article Type:Statistical Data Included
Geographic Code:1USA
Date:Jan 1, 2001

Related Articles
HMO ripoff.
Study: HMO Pullouts Will Hurt Only a Few Medicare Markets.
Prior authorization of pharmaceuticals: health policy in search of evaluation. (2001 Best Clinical Paper Award Recipient).
Are factor substitutions in HMO industry operations cost saving?
Medicare HMO exodus: system correction or prediction? (Health Policy Update).
Healthcare cost differences in the 1990s: the influence of metropolitan area marketplace dynamics.

Terms of use | Copyright © 2018 Farlex, Inc. | Feedback | For webmasters