This study has measured performance across all US FFS Medicare physician-hospital networks for both racial disparities and process quality of ambulatory diabetes care. We measure performance at the level of groups of physicians whose work patterns suggest they form a virtual network around a local hospital. These networks have been proposed as a nexus for quality improvement and reporting. We note large variation in the magnitude of disparities across networks that are not associated with the level of network quality. This finding challenges the view that racial disparities in care are principally caused by whites receiving high-quality care, while blacks receive low-quality care in the same network. Rather, in the context of ambulatory care for diabetes whites and blacks receive a similar level of quality within networks. When racial disparities within network are present, they are just as likely to be evident in high-quality networks as in those with lower quality. We also found that the network-level patient characteristics explained only 2% of the variation in network level disparities, suggesting that role of patient level behaviors such as adherence, while important, is unlikely to be the leading explanation for variation in network performance.
Many studies have focused on disparities in expensive procedures and hospital care for chronic disease care, but few focus on the upstream practices in primary care outside the Veteran’s Administration system health system and managed care.10,22,23
It is important to look outside managed care and the VA because patterns may be different in the absence of an identifiable system with the capacity to implement quality improvement. The most comprehensive study in the FFS setting was conducted by RAND and included 12 communities in the United States on 439 measures in its Quality Assessment Tools to determine which individual characteristics predicted receiving low quality care.24
In contrast to our finding of a 7% absolute lower receipt of recommended care, the RAND study did not find significant differences in recommended care by race. The difference between our overall results and those from the RAND study may be explained by our focus on a more limited set of measures. Some evidence for this hypothesis comes from a secondary finding in the RAND study of greater disparities arising in a limited set of 33 quality measures. As well, the RAND study included just 12 communities, and so may not have reflected the true degree of disparities across the US, yet both studies demonstrate that the gap from disparities is small relative to the gap in quality.
For the purpose of measuring and reporting disparities in ambulatory care, our analysis focuses on groups of physicians caring for a population of patients rather than individual physicians. Prior investigators have made 2 key observations regarding efforts to report individual physician performance: (1) for reliable reporting, minimum sample sizes must be 50 to 100 patients25
; and (2) it is possible to game the system by avoiding small numbers (1–3) of patients.26
Regarding racial disparities, other investigators have also pointed out that a small group of physicians provide most minority primary care, often in less supportive environments.7,27
Yet individuals receive care simultaneously from many physicians and specialties.28
Consequently, reporting performance at the individual primary care physician level may not only have a disproportionally negative impact on providers who care for minorities but also create greater incentives to avoid caring for these patients. Reporting quality and disparities at the network level, as we have done, therefore offers several advantages. Only 1% of physicians are in networks with fewer than 100 patients in contrast to the 66% of individual physicians who would have fewer than 100 Medicare patients. 8
A network-based approach to measure performance reduces the incentive to avoid high-risk patients because the outcomes for patients who receive care across the network are included whether individual providers attempt to limit access or not. The administrative burden and cost of public reporting would be markedly reduced (eg, from 400,000 physicians to 5000 physician-hospital networks). Finally, reporting at the physician-network level acknowledges that most patients receive care from multiple providers, including primary care and specialist physicians, and it is the joint performance of these multiple providers, which affects care not the performance of a given physician. Therefore, with information on network performance, patients could switch to better networks. Such reporting is aligned with recent Institute of Medicine reports calling for performance measures that promote such shared accountability.29
Our aim is not merely to document the disparities, but to provide a mechanism by which information needed to motivate improvement can be generated. Reporting network-specific performance can facilitate collaborative medical and public health interventions for disparity reduction, a method employed successfully in Centers for Disease Control demonstration projects, and new initiatives such as the Robert Wood Johnson Aligning Forces for Quality Program.30–33
Furthermore, reporting performance at the network level will almost certainly become wide-spread, now that health care reform legislation will require Medicare to implement Accountable Care Organizations as a national voluntary program by 2012. Our data also inform how policies to improve network performance for minorities should be structured. One approach is to target within network disparities directly. Another approach is to intervene among lower quality providers since blacks are more likely to be treated in these settings.6,7,9,13,34
Because we also find that blacks are more likely to receive care from low-quality networks, we were able to ask which of these approaches would be most effective at reducing the national disparity between blacks and whites.
To explore the benefits of each approach, we performed a simulation to compare the gains from eliminating within network disparities completely in all 1616 networks that serve blacks, to one that improved quality to a benchmark of 85% of recommended testing in the 500 largest minority-serving networks (without regard to addressing disparities). The absolute national disparity would be reduced from 7% to 2% by eliminating disparities within every network, and would improve the quality of care received by blacks from 70% to 75% of recommended care. In contrast, the targeted intervention in 500 networks would raise the quality of care for blacks nationally to 81% of recommended care. There are no data with which to compare the costs of the 2 strategies but targeting low-quality minority provider groups as we simulated would involve over a 1000 fewer intervention sites and achieve greater testing rates.
Our study is not without its limitations. First, our analysis is restricted to beneficiaries who have had contact with the health care system by having at least 1 ambulatory care visit and blacks were more likely not to have any visits in a 2-year period. It is unlikely that these individuals are receiving recommended testing, so our estimates may underestimate racial disparities. Second, the use of Medicare claims data forces us to restrict the analysis to whether recommended testing occurred, not whether treatment goals were attained. These measures are important processes of care as reflected by their inclusion in the HEDIS measure set. Our choice of process measures might still be questioned. Although some studies have shown that measures used to assess hospital process quality have little relationship to mortality (in pneumonia and congestive heart failure), other studies have shown a positive relationship.35–37
Harman et al have shown that improvement on process measures in a health plan has a positive relationship to health in diabetes.38
Third, it is possible that the claims data do not capture all of the care patients received through incomplete billing, but there is little reason that this problem would differentially affect blacks as compared with nonblacks. Fourth, we were limited to single race categories that do not capture multiethnicity (Hispanic) classifications.18
There may be different patterns for each subgroup that we could not assess. This limitation underscores the importance of collecting better data on race and ethnicity. Finally, we used a 20% Medicare sample which resulted in many networks having a small sample size, a problem that would be ameliorated by using 100% Medicare data but highlights that even when reporting at the group level there will likely be limitations in reporting data for uncommon diseases or smaller subgroups of the population.
In summary, it is now possible to report racial disparities and the quality of care received by minority patients at the level of physician-hospital networks that provide ambulatory care to a population of diabetes patients. While these networks are “virtual” in the same way that states and regions are not directly responsible for care delivery, the measurement of disparities at a more disaggregated level provides patients with more relevant information on the performance of their care network, and it provides policy-makers with more specific guidance on where to focus efforts to improve minority health. Efforts to improve the quality of minority healthcare will have to focus on both eliminating within-network disparities as well as improving the performance of lower performing networks, which are more likely to care for black patients. Regardless of the specific policies that are chosen to improve networks, we have demonstrated that 1 key ingredient to these policies—the ability to measure and monitor performance for FFS ambulatory diabetes care—is now possible.