Private and public payers are focusing on measuring and rewarding quality and efficiency in health care (Milgate and Cheng 2006
; Rosenthal et al. 2006
;). Such efforts include “pay-for-performance” (P4P) systems that reward measured performance, capitation systems that put providers at financial risk for high utilization, and tiered networks in which insurers use measured performance to assign providers to “preferred” status levels.
A key component of the design of such systems is the determination of whose performance to measure and reward. Typically, patients have contact with multiple physicians and institutions (Pham et al. 2007
). For example, surgical outcomes could be affected by the surgeon, the surgical team (surgeons, nurses, and anesthetists), and the institution where the surgery was performed. Therefore, determining rules regarding attribution of outcomes to providers creates major challenges in payment system design. Failure to accurately identify the provider or providers with the greatest influence on outcomes could adversely affect the credibility and impact of the payment system, and it could make providers accountable for decisions outside their control. Similarly, the validity of provider profiles, which are being developed for quality assessment and improvement (Bodenheimer 1999
) using more rigorous methods (Huang et al. 2005
; Shahian et al. 2005
; Zheng et al. 2006
;), also depends on whether they identify the providers with the greatest ability to affect the outcome the policy maker is trying to influence.
In principle, outcomes could be measured and rewarded for any or all of the providers or types of providers involved in a patient's care. However, doing so would present a variety of challenges. Data may not be available across all providers and only limited case mix adjusters may be available to control for differences in the patient populations. Even when data are available, statistical power to differentiate outcome variations associated with different physicians and facilities is often limited by sample size. Likewise, providers treating atypical patient populations could face substantial financial risks under a prospective payment system (PPS).
Owing in part to these difficulties, decisions about whom to measure and reward have generally not been based on empirical analyses of the relative impact of particular providers (or of different types of providers) on outcomes. Rather, decisions about which provider (or type of provider) to attribute responsibility have been based on factors such as convenience (e.g., measurement at the institutional level due to easier availability of data or large sample sizes), prospective assignment of patients to a designated “gatekeeper” physician (Rosenthal et al. 2006
), or arbitrary retrospective rules such as attributing responsibility to all providers with a minimum level of patient contact or to the single provider with the most patient contact during the year (Dudley and Rosenthal 2006
; Milgate and Cheng 2006
;). The Medicare Payment Advisory Commission (MedPAC) as well as physicians and their professional societies have expressed concern regarding the attribution methods in use today (American College of Cardiology 2006
; Milgate and Cheng 2006
; Sinsky 2007
; American Academy of Family Physicians 2008
This paper uses renal dialysis services to demonstrate a method for identifying the extent to which different types of providers influence variation in resource use and patient outcomes. Dialysis provides an excellent context for this study. Patients have ongoing relationships with both an institutional provider (the dialysis facility) and a physician (the nephrologist who manages dialysis-related services). Multiple nephrologists practice within most dialysis facilities, and most nephrologists practice in multiple facilities. This double “cross-over” facilitates the statistical identification of physician and facility effects on outcomes. Because the vast majority of dialysis patients are insured by Medicare, available data include a large number of patients. Further, detailed clinical data are available to adjust for case mix. Finally, several clinical performance measures are well established and relatively well accepted.
Dialysis facilities have a financial incentive to increase the use of the services, primarily injectable medications, which are paid on a fee-for-service basis by Medicare. However, physicians, who generally do not profit from these services, are ultimately responsible for prescribing care. Given the recent controversy about appropriate anemia management in dialysis facilities, these issues are particularly salient. Researchers have presumed that the organization is the decision making locus (e.g., Thamer et al. 2007
), while others have argued that institutional protocols are physician driven and modified by individual physicians in response to patient condition (Lazarus and Hakim 2007
Although outcomes and resource utilization may depend on both the dialysis facility and the nephrologist, public reporting of performance measures (Dialysis Facility Compare; http://www.cms.hhs.gov/DialysisFacilityCompare/
), quality improvement initiatives (e.g., http://www.esrdnetworks.org
), P4P proposals (Milgate and Cheng 2006
), and the development of an expanded case mix–adjusted dialysis PPS as required by the Medicare Prescription Drug, Improvement and Modernization Act of 2003 (Pub. L. 108-173) all use the dialysis facility as the locus of measurement and/or reward. Not only does this implicitly attribute responsibility to the facility for the practices of nonemployee physicians but also failure to report at the physician level provides no guidance to patients regarding choice of physician, and failure to provide physician incentives may forego opportunities to improve care.
Two prior studies are particularly relevant. Krein et al. (2002)
developed an empirical basis for deciding which provider level to profile (facility, professional group, or physician) in the context of diabetes care in the Veterans Administration (VA) system. They found that for outcome and resource use measures, variation at the facility level is dramatically higher than that at the physician level. Physician variation was substantial only for narrow process measures (ordering of specific laboratory tests), and the provider group explained relatively little variation in any measure. However, their study was limited to 13 facilities in one VA region.
A second prior study investigated the relative variation of resource use in U.S. dialysis facilities across four levels: facilities, nephrologists, patients, and time (different months for a given patient) (Turenne et al. 2008
). The analysis of four levels of variation created computational limitations which required a sampling strategy that limited the analysis to a 4 percent random sample of facilities and distinguished provider-level effects only through multiple physicians practicing within a facility (and not from physicians practicing in multiple facilities). Although this study also found that the variation across facilities exceeded across physicians, the physician-level variation was relatively more important than that found by Krein, with financially significant variation in resource use across both facilities and physicians.
The current study extends this previous research in several significant ways. First, by aggregating data across multiple months for each patient, this study uses data from almost all physician-facility pairs and both types of “cross-over” between physicians and facilities. Second, the prior dialysis study only examined resource utilization (costs per dialysis session for a set of services, primarily injectable medications and laboratory tests). The current study uses the same utilization measure but adds two outcome measures (achieving treatment targets for dose of dialysis and anemia management). Third, this study uses slightly more recent data (2004) than the prior dialysis study (2003).