Primary Care Physician Compensation Method in ...

Primary Care Physician Compensation Method in Medical Groups Does It Influence the Use and Cost of Health Services for Enrollees in Managed Care Organizations? Douglas A. Conrad, PhD; Charles Maynard, PhD; Allen Cheadle, PhD; Scott Ramsey, MD, PhD; Miriam Marcus-Smith, MHA; Howard Kirz, MD, MBA; Carolyn A. Madden, PhD; Diane Martin, PhD; Edward B. Perrin, PhD; Thomas Wickizer, PhD; Brenda Zierler, PhD; Austin Ross, MPH; Jay Noren, MD; Su-Ying Liang, PhC

Context.—Growth of at-risk managed care contracts between health plans and medical groups has been well documented, but less is known about the nature of financial incentives within those medical groups or their effects on health care utilization. Objective.—To test whether utilization and cost of health services per enrollee were influenced independently by the compensation method of the enrollee’s primary care physician. Design.—Survey of medical groups contracting with selected managed care health plans, linked to 1994 plan enrollment and utilization data for adult enrollees. Setting.—Medical groups, major managed care health plans, and their patients/ enrollees in the state of Washington. Study Participants.—Sixty medical groups in Washington, 865 primary care physicians (internal medicine, pediatrics, family practice, or general practice) from those groups and affiliated with 1 or more of 4 managed care health plans, and 200 931 adult plan enrollees. Intervention.—The effect of method of primary care physician’s compensation on the utilization and cost of health services was analyzed by weighted least squares and random effects regression. Main Outcome Measures.—Total visits, hospital days, and per member per year estimated costs. Results.—Compensation method was not significantly (P ..30) related to utilization and cost in any multivariate analyses. Patient age (P,.001), female gender (P,.001), and plan benefit level (P,.001) were significantly positively related to visits, hospital days, and per member per year costs. The primary care physician’s age was significantly negatively related (P,.001) to all 3 dependent measures. Conclusions.—Compensation method was not significantly related to use and cost of health services per person. Enrollee, physician, and health plan benefit factors were the prime determinants of utilization and cost of health services. JAMA. 1998;279:853-858

JAMA, March 18, 1998—Vol 279, No. 11

IN TODAY’S health care environment, physicians, their patients, health care leaders, and public policymakers are confronting the challenges of managed care. Financial incentives and externally imposed utilization management constraints are being applied to medical practice with unprecedented intensity in a climate of increased competition and cost consciousness. Despite professional and public concern regarding the impact of these economic pressures on the quality and efficiency of health care, there have been few quantitative studies of the impact of managed care on the cost and utilization of health services.1-3 There have been only 2 prior studies2,3 on the impact of financial incentives for physicians on the general utilization and cost of health services, and both of those focused on plan payment method rather than compensation by the medical group to the individual physician. Moreover, a recent descriptive study of large medical groups in California 4 suggested that placing medical groups at risk through health plan capitation might result in reduced utilization, but did not address indi-

From the Department of Health Services, University of Washington, Seattle. Reprints: Douglas A. Conrad, PhD, Department of Health Services, University of Washington, PO Box 357660, Seattle, WA 98195-7660 (e-mail: dconrad@u .washington.edu).

Physician Compensation Effects on Health Services Use and Cost—Conrad et al

©1998 American Medical Association. All rights reserved. Downloaded From: http://jama.jamanetwork.com/ on 02/25/2013

853

Table 1.—Factors Posited to Influence Per Capita Utilization and Cost of Health Services Enrollee characteristics • Age • Sex • Case mix classification (Ambulatory Care Group) resource consumption weight • Benefit level of the individual enrollee’s health plan Primary care physician characteristics • Age • Sex • Primary care specialty: family practice, internal medicine, pediatrics, or general practice Medical group characteristics • Group size: 3 to 5, 6 to 19, or $20 physicians • Group type: multispecialty or not • Extensiveness of utilization management protocols applied in or to the group Market area characteristics • Proxied by dummy variables for rural areas and urban areas .250 000 population Health plan payment method (measured by percentage of group’s revenues) • Full risk capitation (group at risk for utilization of primary care, specialty, hospital, and ancillary services utilization) • Professional capitation (at risk for primary care and specialty services only) • Primary care capitation (at risk for primary care services only) • Fee for service plus a withhold (withhold only returned if utilization of services below certain target amount) • Fee for service without risk (not subject to withhold or other form of risk account) Primary care physician compensation method (of the group to the typical primary care physician) • Salary only (compensation based on 100% straight salary with no incentives) • .50% Salary based, with other incentives accounting for rest of compensation • .50% Production based, with other incentives accounting for the remainder • Production only (compensation based solely on individual primary care physician’s production) • Other method (did not fit in other 4 categories)

vidual physician compensation methods within those groups. The latter mechanism, the method of physician compensation within medical groups, is arguably the most important financial incentive influencing the individual physician’s behavior. In this article, we estimate a model that examines the relationship between the method of primary care physician (PCP) compensation and the utilization and estimated cost of health services for adult enrollees of managed care organizations (MCOs). The study hypothesis posited that utilization and cost of health services per capita (per enrollee) would be greater, other things equal, among enrollees in the panel of PCPs who are compensated predominantly on the basis of their individual production, generally measured as fee-for-service equivalent services provided to patients (production based), as compared with those cared for by PCPs predominantly compensated by salary (salary based). In our study sample, no medical groups compensated PCPs on a capitation basis.

of 82%. Two single-specialty pediatric groups were excluded from the analysis of adult enrollees reported in this article. This resulted in a study sample of 200 931 adult enrollees in the panels of 865 PCPs within 60 medical groups. Individuals covered by public programs (eg, Medicare, Medicaid, Civilian Health and Medical Program of the Uniformed Services, and Veterans Affairs) were not included in the study. The medical groups were identified by the 4 participating MCOs. These MCOs included a large staff model-health maintenance organization (HMO); a major network-model HMO; a preferred provider organization, which selected PCPs based on their estimated efficiency in caring for a broad range of clinical conditions; and a groupmodel HMO. In all but the group-model HMO, each individual enrollee was assigned a PCP. The individual-level analysis included 200 931 privately insured members 18 years of age or older, continuously enrolled for the calendar year 1994, and who were in 1 of the 3 MCOs that assigned individual enrollees to PCPs.

METHODS Study Sample This study focused on PCPs in medical groups, which were defined as 3 or more physicians practicing in a common setting and sharing revenues and expenses. A survey of clinic practices with respect to physician compensation, health plan payment, as well as patient care management and information dissemination was mailed to 76 medical groups and completed by 62, for a final response rate

Dependent Variables Measures of cost and utilization included per member per year (PMPY) estimated costs of health services, total physician and outpatient visits PMPY, and total hospital days per enrollee per year. The PMPY measure was constructed by assigning resource-based relative value scale weights to each ambulatory service or procedure and then multiplying the resource-based relative value scale units by $46.23 (the average

854 JAMA, March 18, 1998—Vol 279, No. 11

dollar conversion factor for commercial health insurance plans in the state of Washington in 1994). Total ambulatory care visits included primary care and specialty referral visits to hospital outpatient clinics, physicians, and other health care providers. Hospital days were converted to dollar equivalents by first applying all-payer diagnosis related group weights to each hospital discharge, second dividing by the average length of stay for the particular discharge, and third multiplying by the dollar payment per hospital day among private, commercially insured patients in Washington in 1994. Dental and pharmacy claims were not included in the measures of cost and utilization. Independent Variables The independent variables used in evaluating the study hypothesis are listed in Table 1. Individual enrollee data were obtained from the MCO enrollment files. The benefit level measure was constructed by scoring each of 3 types of covered services, inpatient hospitalization, physician office or hospital outpatient care, and emergency department visits, as 0 (low), 1 (medium), or 2 (high), depending on whether the level of patient cost sharing was above average, average, or below average, respectively. Adding the 3 subscales produced a score ranging from 0 to 6, in order of increasing overall health plan share of payment. An ambulatory care group (ACG) relative resource consumption measure, which was based in 1994 on age, gender, and International Classification of Diseases, Ninth Revision (ICD-9)5 diagnosis codes, was assigned to each enrollee.6 The algorithm proceeds by first assigning groups according to diagnoses (ambulatory diagnostic groups) and then hierarchically subclassifying those groups into ACGs (aggregations of diagnostic groups with common levels of annual resource consumption). Earlier validation work by Weiner et al6 has established that ACGs are a powerful predictor of annual health care expenditures. The 1994 PCP characteristics were obtained from the 4 health plans, the Washington State Medical Association, the American Medical Association’s Physician Masterfile, and the AMA Directory of Medical Specialists. Age and gender were obtained, and self-reported specialty was defined as family practice, internal medicine, pediatrics, or general practice. Medical group characteristic variables were derived from the medical group practice survey that was sent to medical group administrators. The survey requested 1994 information about compensation method for the typical PCP, the number of all physicians by spe-



cialty, utilization management protocols, and the distribution of revenues from all health plans, not just the 4 MCOs. For each group, a utilization management score was calculated for fee-forservice contracts, as well as for at-risk managed care contracts. The score ranged from 0 to 9 with 1 point assigned for doing the following: (1) preauthorization for referrals to specialists, (2) preauthorization for hospital admissions, (3) preauthorization for outpatient procedures, (4) preauthorization for hospital admissions, (5) concurrent review of hospital stays, (6) information feedback to PCPs, (7) clinical guidelines, (8) case management for high-cost, catastrophic episodes of care, and (9) internal utilization review by physician peers. Group revenues from health plans were distributed among the different payment methods described in Table 1. An alternative definition of plan payment was implemented, and each of the MCOs provided 1994 information concerning how they paid the medical groups with which they contracted. Other variables of interest were characteristics of the geographic areas where medical groups were located. Since groups were concentrated in 2 large metropolitan areas with a smaller number located in 6 other counties throughout the state, market area characteristics were defined by a series of 8 dichotomous geographic variables. The key variable of interest, PCP compensation method, was determined by the medical group survey that contained several questions adapted from the Medical Group Management Association. We asked what percentage of the typical PCP’s compensation was derived from the following: (1) straight salary with no incentives, (2) individual capitation, (3) group capitation based on average per member per month “at risk” within the group, (4) individual production, (5) equal share of net income, (6) structured bonus or incentive, and (7) other. These responses were used to derive the 5 mutually exclusive categories of compensation (Table 1). Statistical Methods To assess the effect of PCP compensation method on the cost and utilization of health services, several different strategies were used. First, at the group level, measures of cost and utilization were examined by the 5 different compensation methods. Second, using ordinary least squares regression with the group as the unit of observation (N=60), the association between the 3 dependent variables and compensation method was examined, while controlling for average enrollee and medical group characteris-

tics. Given the relatively small number of observations, it was possible to control for only a limited number of independent variables. A third approach used a mixed-model analysis of variance in which the medical group and the individual PCP were modeled as random effects.7-10 In this model, SEs of the regression coefficients were adjusted to reflect the nesting of PCPs within medical groups and of individual enrollees within PCPs’ panels. Ordinary least squares and weighted least squares regression were used to validate the results of the mixed-model analysis of variance. In these analyses, the dependent variables were transformed to normalize the distributions of the residuals in the estimated regression equations; the natural logarithm of PMPY costs, hospital days, and total visits per year was used as the transformation. Since the 1994 ACG case-mix classification was derived principally from the diagnoses recorded for utilization in the same year, there could be a direct correspondence between this variable and measures of cost and utilization. Accordingly, the mixedmodel analysis of variance was run with and without the ACG weight. All variables were allowed to enter the model and the resulting regression coefficients and P values were reported. RESULTS Univariate Results Measures of cost and utilization, as well as characteristics of enrollees and physicians, are shown in Table 2. The average number of total visits was approximately 6.5. The average PMPY cost was equivalent to a per month cost of $105, a representative level of cost for that period. The average hospital days per 1000 were 280. The average probability of hospitalization for the year, 5.2%, was within the range of averages reported in the Medical Outcomes Study.11 The average age of enrollees was 40 years, 55% were women, and the average plan benefit level was slightly above 3 (on a 0-6 scale). Most enrollees in this sample (65%) were panel members of family practice physicians; the average age of these physicians was 45 years and 30% were women. The majority of the medical groups included family practice, pediatrics, and general internal medicine practices; 30% of groups were multispecialty. Almost half the groups had 3 to 5 physicians (48%), and the remainder was equally distributed between groups of 6 to 19 (26%) or 20 or more physicians (26%). Utilization management for both fee-forservice enrollees (5.5 of 9) and managed care enrollees (6.9 of 9) was extensive.

JAMA, March 18, 1998—Vol 279, No. 11

Table 2.—Enrollee and Physician Characteristics*

Variable No. of annual visits per enrollee Annual per member costs, $ Annual No. of hospital days per enrollee Annual probability per enrollee of hospitalization, % Enrollee age, y Women, % Case mix, score Plan benefit level Primary care physician age, y Female primary care physicians, % Pediatricians, % Internists, % Family physicians, % General practitioners, %

Adults Aged .17 y (n=200 931) 6.56 (8.09) 1268 (3896) 0.28 (4.15) 5.2 (22.2) 42 (13) 54.6 (49.8) 0.66 (0.91) 4.09 (1.09) 45 (6) 29.4 (45.6) 1.2 (10.8) 9.7 (29.6) 89.1 (31.2) ,0.1 (1.8)

*Data are presented as mean (SD).

Compensation methods were as follows: 18.3% of the groups (46% of PCPs, 90% of PCP enrollees) were on salaryonly compensation, 16.7% (17% of PCPs, 2% of enrollees) of the groups were compensated on the basis of greater than 50% base salary plus other methods, 16.7% of groups (16% of PCPs, 4% of enrollees) were on greater than 50% productionbased plus other methods, and 45% (20% of PCPs, 3% of enrollees) were production-based only. The 2 remaining groups (1% of PCPs, 1% of enrollees) compensated PCPs based on equal shares of net income and a mixed scheme, respectively. The differing compensation method percentages (by groups, PCPs, and enrollees) reflect primarily the large size of 1 of the salary-only groups (in terms of the number of PCPs and enrollees). For purposes of the analysis, the nature of health plan payments to the medical group was summarized in the dummy variable, at-risk, which is coded as 1, if the plan’s contract with the specific medical group was based on any 1 of 4 types of payment (as defined in Table 1): full-risk capitation, professional services capitation, primary care capitation, or fee-for-service plus withhold. Under these payment types, the medical group was effectively at risk for unanticipated variation in utilization. Fee-for-service contracts (whether discounted or not) without withhold were coded as 0 for the at-risk variable. Ninety-six percent of the individual enrollees were in health plans whose payment arrangements placed their PCP’s medical group at risk (ie, their at-risk variable equaled 1). All types of capitation (including full-risk, which accounted for less than 5% of average total revenues; professional services capitation for primary care and specialty referrals; and capitation for primary care only) comprised 21% of total health plan payments to medical groups in Washington in 1994.



855

Table 3 displays the unadjusted level of per enrollee cost and utilization by compensation method. The PMPY cost was lower for enrollees whose physicians received salary only. However, there was no apparent trend in the number of visits or hospital days. Multivariate Results Multivariate regression analysis at the level of the medical group revealed no statistically significant association between measures of compensation method and cost and utilization. Mixedmodel analysis of variance at the level of the individual enrollee was then used to further test the study hypothesis. The unit of analysis for all multivariate results presented in this article was the individual enrollee. For significant variables, and the at-risk health plan payment and compensation variables, the impact of a 10% increase in continuous variables or a unit change in categorical variables is displayed in Table 4. The ACG resource consumption weight was not included in these models because of its codeterminacy with utilization, but including the ACG weight did not alter the significance or sign of the estimated effects of compensation method. The coefficient of determination

(R2) for the final model (without inclusion of ACG weight) is presented for each dependent variable in Table 4. The impact of PCP compensation method on use and cost was statistically insignificant for all 3 dependent measures. Our most conservative estimates of power (for hospital days per enrollee) revealed b=.90 to detect (with a=.05) a difference of .018 hospital days per enrollee, which is a relatively small difference. The measure for health plan payments placing the medical group at risk was also statistically insignificant in the analyses of visits, hospital days, and PMPY estimated cost. Tests for interactions between compensation method and health plan payment and between compensation method and group size revealed no statistically significant interactions. The identical mixed analysis of variance model was estimated for 4 specific clinical conditions of relatively high frequency in our sample (asthma, congestive heart failure, hypertension, and pneumonia). The results were the same: compensation method was not significantly related to utilization or PMPY cost. In contrast, enrollee age, gender, plan benefit level, and physician age were significantly associated with all 3 dependent variables. For example, adjusted

Table 3.—Unadjusted Utilization and Cost by Method of Primary Care Physician Compensation Measures of Utilization and Cost Per Year, Mean (SD) Method of Primary Care Physician Compensation Salary only .50% Salary plus other

Cost Per Member, $ 1252 (2822) 1420 (4073)

No. of Visits 6.6 (8.2) 6.4 (7.0)

No. of Hospital Days Per 1000 280 (4309) 253 (1803)

.50% Production-based plus other Production-based only Other

1414 (3940) 1403 (4078) 1275 (2820)

6.9 (7.6) 6.4 (7.4) 7.4 (6.8)

301 (2843) 240 (1893) 115 (726)

for all other variables in the model, a 10% increase in enrollee age was associated with a 7.8% increase in PMPY cost and with smaller increases in visits and in hospital days per enrollee. Adjusting for other factors, women enrollees used more services than men; adjusted PMPY cost and visits per year were 77% and 4.7% greater, respectively, for women. Patient benefit level and physician age also had statistically significant, though smaller, impacts on use and cost. COMMENT In analysis at both the group and individual enrollee levels for all adult enrollees and medical groups in the study sample, the effect of PCP compensation method on the use and cost of services per enrollee was statistically insignificant. The principal drivers of use and cost were characteristics of individual enrollees and their PCPs and the level of health plan benefit coverage. Use and cost increased with age, female gender, and richness-of-plan benefit coverage. These findings are not surprising and are consistent with earlier studies of health services utilization.12,13 The finding that the enrollees of older PCPs had lower levels of use and cost might reflect the clinical maturity and efficiency of more experienced practitioners, a preference for less service-intensive practice styles, or underprovision of services. The significantly higher number of visits and hospital days per year of female enrollees is consistent with 1994 National Health Interview Survey age-adjusted estimates.14 The significant age and gender effects on utilization and cost highlight the necessity of age and sex adjustments in health plan payments.

Table 4.—Adjusted Utilization Differences for Key Variables* Per Year Per Member Costs Mean Effect (±1.96 SE), % +7.8‡ (±0.2) +77.1‡ (±0.1) +6.2‡ (±0.1) −1.8‡ (±1.0) −0.8 (±2.8) +0.8 (±2.4) −5.0 (±13.3) ... +0.7 (±1.8) −0.1 (±2.0) +0.8 (±1.6) +0.7 (±9.4)

Variable Enrollee age† Enrollee sex, female vs male Plan benefit level§ Physician age§ Utilization management of FFS enrollees§ Utilization management of managed care§ At risk Salary only\ .50% Salary plus other .50% Production based plus other Production based only Other

No. of Visits

P Value ,.001 ,.001 ,.001 ,.001 ..56 ..49 ,.45 ... ,.42 ,.95 ,.32 ,.89

Mean Effect (±1.96 SE), % +4.5‡ (±0.1) +4.7‡ (±0.1) 4.3‡ (±0.04) −1.5‡ (0.7) −0.4 (±1.5) +0.5 (±1.4) −2.4 (±8.4) ... +3.5 (±9.7) +1.4 (±11.5) +2.9 (±9.7) +11.6 (±61.8)

No. of Days in Hospital

P Value ,.001 ,.001 ,.001 ,.001 ..56 ..43 ,.57 ... ,.47 ,.81 ,.55 ,.71

Mean Effect (±1.96 SE), % +0.4‡ (±0.04) +1.2‡ (±0.02) +0.4‡ (±0.13) −0.2‡ (±0.1) −0.02 (±0.23) −0.1 (±0.3) −0.3 (±2.0) ... +0.4 (±1.8) −1.2 (±2.1) −0.4 (±1.8) +0.1 (±17.8)

P Value ,.001 ,.001 ,.001 ,.001 ..86 ..37 ..75 ... ,.67 ,.27 ,.64 ,.99

*Model R2 is 10.1% for per member costs, 6.7% for No. of hospital visits, and 0.1% for No. of days in the hospital. †Calculated for 10% change in the variable. ‡The estimated coefficient was significantly different from zero and P,.05. §Calculated for unit change in the variable. \Ellipses indicate data that are not available because of an omission of this category.

856 JAMA, March 18, 1998—Vol 279, No. 11



The multivariate results regarding the effect of PCP compensation method are not consistent with our original hypothesis, but are in accord with the statements of PCPs and administrators in key informant interviews.15 The PCPs generally reported that their treatment decisions were based on clinical indications rather than financial incentives. Physicians do respond to incentives and how they are compensated is perceived to affect their productivity, but not their treatment decisions for individual patients. Indeed, previous studies indicate a major effect of individual productionbased compensation on physician productivity.16,17 It is also possible that the statistically insignificant effect of PCP compensation method observed in this study was the result of offsetting effects. For example, PCPs compensated on a production basis compared with those on straight salary might do more primary- and specialty-type care themselves, but make fewer referrals to specialists. In contrast, our original hypothesis was that the financial incentive to the PCP to do more under production-based compensation would override any potential savings from reduced referrals. Our findings should be viewed as complementary, rather than contradictory, to the earlier studies.2,3 Those studies did find significant effects of plan payment incentives on health services use, whereas we find no significant effects of individual PCP compensation or health plan payment method (the at-risk variable). The insignificance of the health plan payment variable might reflect the limited variation in the at-risk variable in our sample. The fact that 96% of the enrollees were in panels of PCPs who were at risk implies that our choice of health plans, in effect, has eliminated the influence of health plan payment in this sample. This actually increases one’s confidence that we have isolated the effect (or noneffect, in this case) of individual physician compensation, which is the variable of interest in this study. This is an important consideration because the incentive effects of health plan payment and physician compensation are potentially significant independent influences on use and cost, and they might interact with one another.18 The absence of capitation compensation to individual PCPs within medical groups is not surprising. Medical groups exist, in part, to spread economic risk among physicians17; thus groups are unlikely to adopt compensation arrangements that place substantial risk for unanticipated variance in utilization on individual physicians. This is consistent with the physician and administrator

responses in our key informant interviews.15 Of course, at the group level, where the risk of unanticipated variance in volume of services per person is diversified across physicians, capitation payments by health plans are being accepted in groups as our data suggest. It is important to acknowledge the potential limitations of this study. First, the compensation of PCPs, not specialists, was the focus of this study, although we recognize the large role of specialists. The PCPs are experiencing declining choice in the specialists to whom they can refer their enrollees. Within those constraints, which are imposed at the level of the MCO (benefit design) or the group (intergroup agreements), the PCP will respond to his or her own incentives to manage care by purposeful referral to specialists. On a related point, because the health plan administrative data for this study did not allow one to distinguish reliably between primary care and referral care, we were unable directly to test hypotheses regarding the differential effects of compensation on primary care visits vs specialty referral care. However, the trade-off between these types of care is ultimately captured in the PMPY cost measure. Second, since this study examined Washington groups and 1994 quantitative data only, our results do not necessarily generalize to other environments or market conditions. While 1994 was a relatively stable year in terms of physician compensation arrangements in Washington, the somewhat more turbulent market conditions of 1996 to 1997 might produce different incentive effects of compensation on utilization and cost. We find no suggestion of this in the 1996 interview responses,15 but it is possible. It is also conceivable that the presence of relatively efficient PCPs in the preferred provider organization plan reduced the variation in use and cost within our sample, thus making it somewhat more difficult to detect compensationrelated differences. Moreover, the relatively high penetration of at-risk managed care in our sample may have affected the behavior of all physicians, thus narrowing the scope of potential compensation effects. Third, we view compensation method not as an externally imposed condition, but as a component of work that is chosen, in part, by the practitioner. This selfselection was suggested by the findings of the key informant interviews of physicians and group-practice administrators.15 Self-selection might impart a positive (nonconservative) bias to our estimate of the effect of production-based compensation on use and cost since phy-

JAMA, March 18, 1998—Vol 279, No. 11

sicians with higher production potential might gravitate to groups compensating on a production basis. Thus, the finding of negligible effect of compensation method is reinforced. Reestimating the model with instrumental variables, taking into account self-selection of compensation method, did not change the finding of no significant compensation effects. Finally, this study focused on PCP compensation within medical groups. It is possible that the nature of interactions among physicians within medical groups might attenuate the individual physician’s response to financial incentives. In contrast, individual physicians in solo or 2-person practices might react much more strongly to compensation incentives. Future analyses should examine the impact of financial incentives on physicians in solo or 2-person practices. This is especially important in light of the finding in a 1995 national survey that a large percentage of solo and small practice physicians seemed unfamiliar with the provisions of their managed care contracts,19 and therefore might be taking on excessive financial risk. Compensation method is 1 of many factors potentially influencing PCP behavior. When other factors are controlled for, PCP compensation method does not appear to have a significant effect on cost and use of health care services per person for managed care enrollees whose PCP practices within a medical group. Future research should examine the robustness of this finding in other settings and as the managed care marketplace evolves over time. Douglas A. Conrad, PhD, was principal investigator of the project on which this article was based and assumed the lead role in research design and writing of this article. Charles Maynard, PhD, assumed the lead role in data analysis for the project and took the lead in writing the “Methods” section for this article. Allen Cheadle, PhD, assisted in designing the survey of medical groups, in programming the mixed-model analyses of variance, and in the writing throughout the article. Scott Ramsey, MD, PhD, took the lead in developing case-mix measurement, assisted directly in designing the conceptual model for estimating cost and use effects, and participated in writing the “Methods” and “Comment” sections. Miriam Marcus-Smith, MHA, was project manager and took the lead in developing the medical group survey instrument and participated in writing both the introduction and “Comment” section. Howard Kirz, MD, MBA, assisted in writing the introduction and “Comment” section and played a major role in designing health plan data collection methods. Carolyn A. Madden, PhD, assisted in writing the econometric model for the utilization and cost analyses and in writing the “Results” section. Diane Martin, PhD, assisted in formulating measures of use and cost and in the specification of statistical models used in the article, as well as in writing the “Results” section. Edward B. Perrin, PhD, assisted in designing appropriate statistical tests and in writing the “Methods” and “Results” sections. Thomas Wickizer, PhD, played an important role in designing survey strategies



857

with the medical groups and in writing the protocol for collecting key informant interview and health plan data. Brenda Zierler, PhD, assisted in writing the “Results” section. Austin Ross, MPH, formulated the original protocol for selecting medical groups and assisted in writing the introduction. Jay

Noren, MD, assisted in design and conduct of the survey of medical groups and in the formulation and written description of the conceptual framework for the empirical tests. Su-Ying Liang, PhC, performed several of the statistical analyses using instrumental variables (as a check on the mixed-model analy-

sis of variance results) and helped in writing the “Results” section. The research on which this article is based was supported by a grant from the Robert Wood Johnson Foundation Health Care Financing and Organization Initiative.

8. Salonen JT, Kottke TE, Jacobs DR Jr, Hannan PJ. Analysis of community-based cardiovascular disease prevention studies: evaluation issues in the North Karelia Project and the Minnesota Heart Health Program. Int J Epidemiol. 1986;15: 176-182. 9. Whiting-O’Keefe Q, Simborg D. Choosing the correct unit of analysis in medical care experiments. Med Care. 1984;22:1101-1114. 10. Williams PT, Fortmann SP, Farquhar JW, Varady A, Mellen S. A comparison of statistical methods for evaluating risk factor changes in community-based studies: an example from the Stanford Three-Community Study. J Chronic Dis. 1981; 34:565-571. 11. Greenfield S, Nelson EC, Zubkoff M, et al. Variations in resource utilization among medical specialties and systems of care. JAMA. 1992;267: 1624-1630. 12. Manning W, Leibowitz A, Goldberg G, Rogers W, Newhouse J. A controlled trial of the effect of a prepaid group practice on use of services. N Engl J Med. 1984;310:1505-1510. 13. Manning W, Newhouse JP, Duan N, et al. Health insurance and the demand for medical care: evidence from a randomized experiment. In: Health

Insurance Experiment Series. Rockville, Md: Department of Health and Human Services, RAND; February 1988:1-25. 14. Current Estimates From the National Health Interview Survey. 10th ed. Rockville, Md: US Dept of Health and Human Services; 1994. Vital and Health Statistics publication 95-1518. 15. Zierler B, Marcus-Smith M, Cheadle A. Effect of compensation method on the behavior of primary care physicians in managed care organizations: evidence from interviews with physicians and medical leaders in Washington State. Am J Managed Care. 1998;4:209-220. 16. Gaynor M, Pauly MV. Compensation and productive efficiency in partnerships: evidence from medical group practice. J Political Econ. 1990;98:544-573. 17. Gaynor M, Gertler P. Moral hazard and risk spreading in partnerships. Rand J Econ. 1995;26: 591-613. 18. Hillman AL, Welch W, Pauly M. Contractual arrangements between HMOs and primary care physicians: three-tiered HMOs and risk pools. Med Care. 1992;30:136-148. 19. Simon CJ, Emmons DW. Physician earnings at risk: an examination of capitated contracts. Health Aff (Millwood). 1997;16:120-126.

References 1. Conrad D, Wickizer T, Maynard C, et al. Managing care, incentives, and information: an exploratory look inside the “black box” of hospital efficiency. Health Serv Res. 1996;31:235-260. 2. Hillman AL, Pauly MV, Kerstein JJ. How do financial incentives affect physicians’ clinical decisions and the financial performance of health maintenance organizations? N Engl J Med. 1989;321:86-92. 3. Stearns SC, Wolfe BL, Kindig DA. Physician responses to fee-for-service and capitation payment. Inquiry. 1992;29:416-425. 4. Robinson JC, Casalino LP. The growth of medical groups paid through capitation in California. N Engl J Med. 1995;333:1684-1687. 5. World Health Organization. International Classification of Diseases, Ninth Revision (ICD-9). Geneva, Switzerland: World Health Organization; 1977. 6. Weiner JP, Starfield BH, Steinwachs DM, Mumford LM. Development and application of a population-oriented measure of ambulatory care case mix. Med Care. 1991;29:452-472. 7. Koepsell TD, Martin DC, Diehr PH, et al. Data analysis and sample size issues in evaluations of community-based health promotion and disease prevention programs: a mixed model analysis of variance approach. J Clin Epidemiol. 1991;44:701-713.

858 JAMA, March 18, 1998—Vol 279, No. 11

