Objectives: To study how payment for specialty services affects the rates of performance of invasive procedures by physicians in a number of specialties.
Study Design: Observational study.
Patients and Methods: Administrative data from 1996-1997 and 1997-1998 from 3 large health maintenance organizations (HMOs) in the Midwestern and western United States were used to study variations in procedure rates associated with different methods of paying for cardiology, gastroenterology, ophthalmology, orthopedic, and ear, nose and throat services within each HMO. The age-, sex-, and comorbidity adjusted probabilities of undergoing selected, potentially discretionary procedures, were compared within each plan by payment method.
Results: After adjustment, rates under fee-for-service payment tended to be higher than those under capitation or salary payment, whereas there was no clear pattern for salary versus capitation payment. Even within a single specialty in a single plan, however, rates did not always follow the same pattern for different procedures.
Conclusions: The payment method for specialty services used by these 3 health plans was variably associated with how likely patients were to undergo a variety of invasive procedures. The effects of contract payment methods for specialty services on health care costs, quality, and outcomes should be further studied, but such studies will challenge the capabilities of health plan data systems.
(Am J Manag Care. 2004;10:229-237)
Financial incentives are clearly among the more powerful motivators of human behavior, and there is no reason to believe that physicians are somehow immune. Indeed, part of the rationale for staff- and group-model health maintenance organizations (HMOs) is that salaried physicians should provide less expensive care in the absence of fee-for-service (FFS) incentives. However, the literature on the effects of financial incentives on physician practice style and costs of care has been mixed. Most studies of physician payment mechanisms and incentives have focused on payment of primary care providers1 "because of the central gatekeeper role."2 One study found no association of primary care physician compensation method in medical groups with utilization or costs of care.3 Another study found that capitation payment to clinics resulted in lower costs and salary payment of physicians resulted in higher costs4; a third suggested a modest decrease in primary care costs when primary care practitioners in an independent practice association (IPA) were subject to a substantial withhold and eligible for a bonus when expenditure targets were met or exceeded.5
Few published studies of the effects of HMO financial incentives have focused on specialist physicians, despite the facts that direct charges for specialty care are at least as high as those for primary care6,7 and that primary care physicians comprise a minority of physicians in the United States.8 It seems at least as important to focus on what happens after specialty referrals are made as on incentives affecting primary care physicians? referral patterns. Researchers have noted that HMOs "should consider the incentives facing affiliated specialists in order to achieve reductions in the amount of hospital resources used for hospitalized patients" and that "the role of specialists might be scrutinized if savings from different payment mechanisms are to be more substantial."9 In one study, a variable in a survey of HMOs indicating the presence of any specialist risk incentives was not significant in predicting hospitalization rates, annual visits per enrollee, or HMO breakeven status.10
Historically, use of capitation for paying specialists lagged behind use of capitation for paying primary care physicians, with specialists primarily getting FFS payments, often with a discounted fee scale.11,12 Through the mid-1990s, there appears to have been a trend toward capitation payment. Only 13% of HMOs paid specialists partly or fully by capitation in a 1987 survey.13 This figure rose to 25% in a 1990 survey,14 and in a 1994 survey, 69% of staff/group-model HMOs versus 47% of network/IPA-model HMOs paid for at least some specialty services by capitation.15 Whether the "managed care backlash" has reversed this trend is not certain, although a study in 1999 found that only 14% of health plans used capitation to pay specialists16 and the proportion of all physicians with capitation contracts was reported to have decreased slightly between 1996 and 1999.17
Few studies have actually compared specialist performances under different payment systems in the same plan or setting. A study of cataract surgery among Medicare beneficiaries receiving care in a large HMO in Southern California found lower rates among both staffmodel and IPA-model enrollees than among Medicare beneficiaries receiving FFS care in the same geographic region.18 However, ophthalmologists in the IPA-model component received discounted FFS payments and the referring primary care physicians were capitated and at risk for specialty services, so the lower rates of cataract surgery in these IPA-model HMO patients may have been driven by effective gatekeeping, nonfinancial incentives, or other, unmeasured factors. Another study found lower rates of elective sterilization when gynecologists were shifted from FFS to capitation payment.19
A variety of factors could affect observed activities of specialists, including both type and amount of compensation, other incentives, and differences in the populations of patients served. This paper reports on whether payment method (ie, salary, capitation, FFS) for specialty services was associated with different rates of performance of invasive procedures for 5 different specialties in 3 large HMOs. The study took advantage of payment variation for each studied specialty within each plan and compared results both across specialties and across HMOs. We hypothesized that procedure rates would be highest under FFS payment, intermediate under salary, and lowest under capitation.
We obtained encounter and claims data from 3 large, well-established HMOs located in the Midwest, the Rocky Mountain region, and the Pacific Northwest. All 3 HMOs were mixed-model plans with large staff/groupmodel components as well as other structures such as contracts with IPAs, employment of primary care providers with providers of specialty care under contract, and so forth. Each HMO used at least 2 different methods of paying for services of at least 2 of the studied medical specialties. In 2 of the HMOs (subsequently referred to as HMOs A and B), the payment variation was geographically based—some areas had salaried primary care physicians and most specialty services were provided by salaried specialists, while other areas used FFS (the term "fee-for-service" is used to denote any type of payment contingent upon performing a service, such as a negotiated fee schedule payment) and/or capitation (population-based capitation payment) payment for the same specialty services, with primary care providers predominantly or exclusively salaried. In the third HMO (HMO C), payment for specialty services was by either capitation or FFS and was not geographically based, but rather was determined by contractual agreements between providers of specialty services and the primary care clinics for enrollees in specific insurance product lines. Primary care providers all were paid under capitation arrangements in HMO C. Analyses used data from 1996-1997 for HMOs A and B and, because of data system changes, from 1997-1998 for HMO C.
We chose 5 specialties for study, based on the presence of payment variation within 1 or more of the HMOs for the specialty and the existence of common, reasonably expensive procedures performed exclusively by these specialists, the performance of which seemed likely to allow for significant provider discretion. Specialties and procedures studied are shown in Table 1. For HMOs A and B, where we analyzed cardiology procedures, we also gathered data for coronary artery bypass graft (CABG) surgery as a potential substitute for percutaneous translumenal coronary angioplasty/stenting/ atherectomy (PTCA). HMO A also had variation in payment for cardiothoracic surgery services, including CABG, whose geographic basis exactly matched the variation in payment for cardiology services. HMO B had no variation in payment for cardiothoracic surgery services but we analyzed variation in use of these services according to the manner in which cardiology services were paid for to look at overall revascularization rates according to payment for cardiology services. Procedures were identified by Current Procedural Terminology, 4th Edition (CPT-4)20 and International Classification of Diseases, Ninth Revision (ICD-9) procedure codes in the encounter and claims data. In one plan, information from claims and encounter databases had to be supplemented with information from operating room and special procedure laboratory logs because professional claims were not always generated; this source accounted for 5% to 18% of the studied procedures in this plan, except for ear, nose, and throat surgery (otorhinolaryngology), where 80% of the procedures were identified in this fashion.
To calculate population-based procedure rates, it was necessary to classify every HMO member into 1 of the 3 specialty payment groups (FFS, capitation, or salary), not just those members who received a procedure. This was possible in HMO C because the type of contract a member had determined whether specialty services would be paid for by FFS or capitation if such services were needed. However, no such information was available on individual members for the 2 HMOs with geographically based variation. Therefore, for HMOs A and B, members were classified according to the predominant specialty payment method at the primary care clinic to which they were assigned. For primary care clinics where more than 80% of the specialty procedures were paid for by 1 method, all of the patients assigned to that clinic were classified as belonging to that specialty payment group. Any clinic with fewer than 80% of studied procedures paid for by a single method was dropped from analyses for that procedure.
We attempted to learn how plan payment for procedures translated into physician payment but the response rate to our survey of the medical groups contracting with the health plans was too low for meaningful analysis.
We studied the odds of undergoing a procedure at least once. Standardized procedure rates were determined by using the age and sex distribution of the combined populations of the 3 HMOs for direct standardization. We computed Mantel-Haenszel odds ratios for procedure performance, comparing different payment methods to compare rates and adjust for any age and sex differences in the populations at risk. Then, we constructed multivariate logistic models for procedure performance. We adjusted for case mix by using dummy variables representing comorbidity categories for a Charlson-type index21 constructed from International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnostic and procedure codes and CPT-4 procedure codes, based on previous work by Deyo et al22 and Romano et al23 (see Appendix for details). Models also controlled for time at risk for a procedure under a payment method (whether due to plan enrollment for only part of the observation period or changing eligibility from one payment method to another over the course of the study) by using log (exposure time) as an offset term,24 Medicaid status, and disability status (persons receiving Medicare because of disability). Analyses were conducted with SAS version 6 software (SAS Institute, Cary, NC) and Stata 7.0 (Stata Corporation, College Station, Tex), which was used to adjust standard errors for clustering of observations within clinics.
This study was approved by the institutional review boards of the University of Washington and HealthPartners.
Table 2 shows procedure rates according to the payment method for specialty services in all 3 HMOs, standardized to the combined enrollment of the 3 HMOs. Rates under a payment method often were comparable across the health plans, but varied by as much as threefold to fourfold in some cases.
Table 3 presents pairwise comparisons of procedure rates by payment method as adjusted odds ratios from logistic regression models, controlling for age, sex, comorbidities, and exposure time to payment method. In HMO A, FFS payment was associated with significantly higher procedure rates than salary and capitation, and capitation was associated with higher rates of the studied orthopedic procedures than salary payment. The largest payment-associated differences in this study were seen here for cardiology services PTCA), with approximately threefold to fivefold higher rates under FFS versus salary payment. Coronary artery bypass graft rates also were twice as high under FFS versus capitation payment. As the population at risk for cardiology services under FFS payment was the same population at risk for cardiothoracic surgery under FFS payment, the higher rate of CABG surgery in this population means that the higher rate of CABG surgery was in addition to, not instead of, the much higher rate of PTCA.
Results were somewhat less clear-cut in HMO B. Procedure rates were similar or higher under FFS versus salary payment; rates under FFS payment also tended to be higher than those under capitation payment, but fewer of these differences were statistically significant. PTCA rates differed dramatically depending on whether or not there had been a diagnosis of myocardial infarction (MI); this effect modification was not present in HMO A, nor was it found for cardiac catheterization in either HMO. Patients with a diagnosis of MI had substantially greater odds of undergoing PTCA with FFS or capitation payment than with salary payment to cardiologists, although rates did not vary significantly by payment for patients without a diagnosis of MI. CABG surgery was paid for under FFS arrangements throughout all of HMO B, but CABG rates were higher in areas where cardiology services were paid for under FFS and capitation arrangements than where there were salaried cardiologists. Therefore, analogous to the situation in HMO A, for persons with a diagnosis of MI the higher CABG rates were in addition to higher PTCA rates rather than substituting for PTCA.
In HMO C, we were only able to compare FFS and capitation payment for 2 specialties. However, for all 4 procedures studied, the FFS rates were significantly higher than the rates under capitation payment.
We examined how controlling for comorbidity and exposure time to payment method affected our estimates of paymentassociated differences compared with age-sex adjustment alone (data not shown). In most cases, comorbidity adjustment made little difference. However, striking changes were found with comorbidity adjustment for cardiology services in HMO A. The odds ratios comparing FFS with salary payment increased markedly for cardiac catheterization, from 2.08 to 2.89, and even more markedly for PTCA, from 2.85 to 5.06. The FFS versus capitation odds ratio for CABG surgery also increased substantially, from 1.46 to 1.97. These increases with comorbidity adjustment are consistent with epidemiologic data showing that cardiovascular disease prevalence is actually lower in the area with the higher procedure rates. In HMO B, the odds ratio for CABG services comparing FFS with salary payment of cardiologists increased from 1.19 to 2.40. Part of the change resulted from differences in the prevalence of comorbid conditions, primarily cardiovascular disease, and the rest was due to differences in time exposed to cardiologists receiving FFS versus salary payment. This exposure time effect also was seen for hip and knee joint replacement surgery in HMO B. The odds ratio comparing FFS with salary payment increased from 0.92 to 1.52, with almost all of this change resulting from adjustment for the fact that patients eligible to see FFS orthopedists had substantially shorter exposure times (median 8 months) than patients eligible to see salaried orthopedists (median 24 months).
Our findings do not confirm our hypothesis in its entirety. After adjustment, there was a moderately consistent association between FFS payment and higher procedure rates relative to salary payment, but rates under capitation tended to be similar to or higher than those under salary payment, contrary to our hypothesis. However, findings were not always consistent within a plan or for the same specialties across different plans. A fair question to ask is whether our findings represent signal or noise. We would strongly answer this question, "Yes, both." We believe that the trend for higher procedure rates under FFS payment is probably signal and that the truly eye-popping difference in rates of invasive cardiology procedures for FFS versus salary payment in HMO A is not a random occurrence. However, the lack of consistency even within the same groups of specialists for different procedures (eg, comparisons of FFS vs salary for joint replacement surgery and knee arthroscopy in HMO B) indicates that factors other than payment also are in play.
As this was an observational study, taking advantage of existing payment variations, contract payment provisions cannot be viewed as strictly exogenous to groups and physicians; endogenous factors associated with type of payment may well also be associated with group practice patterns. HMO market share, number of HMOs in a market, and number of physician groups in a market all affect the probability that a health plan will negotiate capitated rather than FFS contracts.25 Large groups are more likely than small groups or solo practitioners to have the organizational willingness and ability to negotiate and manage capitated contracts.26 Discussions with personnel involved in negotiating contracts for the HMOs indicated that, when a decision to contract for services was made, important factors included previous experience with the group, a group?s position as a sole provider of services in an area versus being one of many providers, and the group?s willingness to consider capitation contracts. Physician choice of whether to commence and continue employment in an HMO also cannot be viewed as exogenous,27-29 and salaried HMO physicians are likely different from those in private practice (eg, they may place a higher value on free time, which would tend to decrease discretionary procedure performance).
An important factor we were unable to observe was how contracted specialists were actually paid. We knew how the HMOs paid for the services, but often these payments were made to "middle tier" organizations2 and not directly to the specialists. There is some evidence that groups with a higher proportion of income from various types of FFS arrangements use lower proportions of base salary for primary care physician compensation and are more likely to relate physician income to measures of productivity,30 supporting the supposition that group payment is related to physician payment. We were unable to achieve a useful response rate to our survey of middle-tier organizations to ascertain financial and nonfinancial provider incentives. However, when a plan is negotiating a contract, our findings are likely to be relevant: plans cannot dictate how middle-tier organizations pay their providers, but they can decide how to structure contracts for services. Beyond basic payment structure, contract provisions such as bonuses, withholds, and risk corridors can alter incentives. We obtained extracts of contracts between the HMOs and specialists, but for confidentiality reasons, the plans would not release information on the size of payments (eg, resource-based relative value scale conversion factors, actual capitation rates). Thus, we could not assess the size of incentives, which is probably more important than simple presence or absence of an incentive. No study of which we are aware analyzing performance under different payment methods has studied actual size of incentive payments.
We also sought to evaluate whether differential referral of patients to specialists by primary care providers was associated with payment differences, but were prevented from doing so by data limitations. None of the HMOs had databases that tracked historical referrals. We investigated using any visit to a specialist as a surrogate for a referral but found that this, too, could not be supported by the historical health plan data systems. One study of a change from salary/productivity to partial capitation payment reported minimal changes in specialist visits,31 suggesting that differential referral according to payment may not be a major issue.
Another potential source of error is the differential coding incentives experienced by physicians. FFS payment provides a clear incentive to maximize coding; this incentive is weaker under capitation and weaker yet for salaried physicians, which would bias results toward higher rates under FFS payment. However, as detailed above, we made extensive efforts to capture all procedures. Because we studied procedures, not intensity/ cost measures, our findings should not be biased by upcoding.
Why might physicians in groups paid by FFS contracts tend to have higher procedure rates than those paid under salary and capitation arrangements? FFS payment allows physicians to maximize income by increasing service volume, an incentive absent under salary and capitation payment. It is conceivable that the HMOs could monitor salaried physicians? procedure rates and modify the behavior of those performing what are perceived to be excessive numbers of procedures, although the literature does not suggest that it is easy to change physicians? practice behaviors, nor was there a strong sentiment among physician leaders in these organizations that they have had substantial success in such efforts. The "culture" of an HMO could act to modify the practice patterns of a physician who, upon starting employment with the HMO, had substantially different practice patterns from those of his or her peers. Another explanation could relate to staffing levels in HMOs. If specialty access is functionally rationed by having fewer providers per capita than in the community, it is likely these specialists will perform fewer procedures per enrollee.
Capitation agreements evolve over time and, because high costs in one year can be used to negotiate better terms in the following year, the ability of capitation contracts to limit costs (and procedures) may be more modest than many believe. Anticipating low costs and high payment under a capitation contract could lead to more discretionary procedure performance to avoid lower payment when the contract is renegotiated, also potentially driving up rates under capitation. Most of the specialists whose services were paid for under capitation contracts were presumably working in environments with a mix of capitation and FFS payment for services. It takes a substantial amount of effort for a physician to attend to differences in individual patients? payment, so even if physicians had no qualms about treating patients differently based on payment, they might find it difficult in practice to do so. Anecdotally, we were told that some groups specifically kept information about individual patients? insurance and payment from providers; others issued monthly reports detailing utilization and income for individual providers according to the insurance plans of the patients they saw. It seems more likely that physicians? practice styles could adjust over time to their overall practice environment, so that physicians in practices with mostly capitated patients might decrease utilization of discretionary procedures relative to what they would do in a mostly FFS environment.32,33 Unfortunately, we had no data to allow assessment of appropriateness of the procedures we studied.
As payment incentives in a contract will exert effects on preexisting procedure rates, observation of procedure rate changes when new payment incentives are put in place may be a better measure of the effects of the incentives than cross-sectional comparisons. However, this requires both observation of comparable "control" groups to adjust for secular trends and stable health plan data systems over several years surrounding contract changes (or data from a source other than the plan in the case of the first contract with a group).
As a methodologic note, the differences we observed between our multivariate models and the Mantel- Haenszel age- and sex-adjusted results highlight the need for adjustment for comorbidities, consistent with findings of others,34 and exposure time. Comorbidity adjustment was relatively unimportant for most of the specialties but was critically important for analyzing cardiovascular services. As payment innovation will tend to be more likely in locations of new growth, where enrollment is less established and less stable, our findings indicate that failure to adjust for exposure time could substantially bias findings. Underlying differences in the populations seen by the specialists in ways not accounted for by our comorbidity measures could confound our findings, but such differences have not been found to explain geographic variations nationally. 35 Where we were able to check our comorbidity measures with epidemiologic data on the underlying populations (for prevalence of cardiovascular disease in HMO A), the epidemiologic data were quite consistent with the results of our comorbidity adjustment.