|
MARKETWATCHPhysician Practice Size And Variations In Treatments And Outcomes: Evidence From Medicare Patients With AMI
Little is known about the relationships between physician practice size and patient treatments or outcomes. We examined whether the practice size of attending physicians was related to within-hospital differences in care for Medicare patients with acute myocardial infarction (AMI). We found that patients treated by solo physicians were less likely to receive cardiac catheterization and angioplasty within a day of admission and more likely to die than other patients in the same hospital, even after a number of patient and physician characteristics were taken into account. These differences suggest that solo practitioners are less likely to follow guidelines calling for quick use of angioplasty.
ALTHOUGH THE SIZE of medical groups has grown over time, solo physicians remain prevalent in the United States.1 The size of a physician practice might influence the delivery of care, yet little information about the relationship between practice size and care is available. In this paper we examine whether practice size is associated with the care of heart attack patients covered by traditional Medicare. Physician practice size could influence care in a number of ways. Larger practices might be more readily able to adopt helpful infrastructure such as information technology (IT).2 They also might more readily be able to investigate and implement new guidelines, protocols, and other care-improving processes.3 They might use different governance techniques than smaller practices and might place different financial incentives on physicians.4 Physicians in larger practices might have easier access to clinical information and consultations through their peers and support staff.5 Higher caseloads in larger practices might allow physicians to narrow their focus to certain types of patients. Physicians in larger practices might have different referral networks, including their practice colleagues, or differential access to outside consultants. Larger practices might also have more influence within the hospitals that their physicians use, so their physicians might find it easier to get timely treatments for their patients.6 A number of studies have looked at costs of care in larger practices, but existing evidence linking practice size with patient care is sparse.7 Two relatively small studies reported that larger practices were associated with greater resource use for hypertensive patients.8 Another study reported that larger practices provide shorter office visits.9 The Medical Outcomes Study indicated that patients of larger multispecialty practices had lower rates of hospitalization, office visits, tests, and prescription drug use.10
Data source. We began with a cohort of all traditional, fee-for-service (FFS) Medicare patients with a new occurrence of acute myocardial infarction (AMI, or heart attack) in 1999.11 Examining only patients with traditional Medicare minimizes the likelihood that differences in insurance coverage would confound relationships between practice size and care. For each patient, the initial ("index") hospitalization was identified, and claims after the index admission, including transfers and rehospitalizations, were tracked for one year. Death dates were identified from Social Security Administration death records. The entire cohort consists of 233,756 patients. Assignment of physician information. For each patient, we identified the Medicare Unique Physician Identification Number (UPIN) of the attending physician recorded on the claim from the index admission. We were able to identify a unique "index attending" UPIN for 99.8 percent percent of the cohort patients. Although a number of physicians might contribute to the care of the patient, in claims data the attending is the clinician who is primarily responsible for the care of the patients from the beginning of the hospital episode.12 Some prior research on the relationship between physician characteristics, such as specialty, and care for AMI patients has also focused on the index attending physician.13 The UPIN was used to link physician characteristics from the American Medical Association’s (AMA’s) Physicians’ Professional Database and practice size for groups of three and up from the AMA’s Medical Group Census. The practice size of the index attending physician might not be the same as the practice size of other physicians who might have influenced the care of the patient. The AMA data might also be incomplete or inaccurate in some cases. Thus, our measure of practice size is likely to contain some measurement error. Measurement error in practice size increases the risk that we will understate the true size of any relationship between practice size and treatment patterns and outcomes. Like many studies, however, the validity of our results does not require that measured practice size be perfectly accurate, as long as it is reasonably correlated with the true practice size. Study sample definition. From the full sample, we extracted patients of physicians whose primary employment is indicated to be a non-HMO (health maintenance organization) "office-based practice," excluding about 114,000 patients of physicians in other types of practices or with missing AMA employment information. We then excluded a small number of patients for whom a resident or fellow was listed as the index attending or whose index attending had missing physician demographic information. This left 116,671 patients for analysis. The observable demographic characteristics and comorbidities of these patients are similar to those in the full cohort.14 Variables. We assigned patients to one of six categories based on the practice size of their index attending physician. We developed measures of receipt of cardiac catheterization, percutaneous transluminal coronary angioplasty (PTCA), and coronary artery bypass graft (CABG) for each patient. We measured receipt of cardiac catheterization, PTCA, and CABG within one, seven, or thirty days of initial admission. We did not analyze one-day CABG further because this was received by only 1.7 percent of patients. We measured mortality within thirty and ninety days of the index admission. We created a number of patient control variables to account for severity of illness and other risk factors. These include sex, race (black or nonblack), rural (that is, non–metropolitan statistical area, or MSA) county of residence, age, comorbid conditions at admission, and patients who had admissions for cardiovascular disease one to two years before the index admission. We measured a number of physician characteristics in addition to practice size, including sex, age, medical school graduation year, and specialty. We coded a measure of whether the physician reported being primarily employed by a medical school. We coded the location of the medical school attended and of the most recent residency program. Statistical analysis. We present descriptive statistics on treatment and outcome measures and on patient and physician characteristics, according to practice size categories, testing differences across categories using chi-square statistics. We estimated multivariate regression models in which the dependent variable is one of our treatment or outcome measures, and the main independent variables of interest are indicators for the practice size categories. The models control for patient characteristics, including a full set of interactions among age, sex, and race categories, and physician characteristics, including interactions between sex and age. The models also included an indicator for each hospital at which each patient was initially admitted. These control for variations in technological capabilities, staffing, ownership, quality, or other hospital characteristics; demographic and health characteristics that might be common to all patients treated in a given hospital; and market-level factors such as the prevalence of managed care and competition between physicians. With these hospital controls included, our results indicate the extent of differences in treatment patterns and outcomes among patients receiving care at the same hospitals. We estimated our models using conditional logistic regression, which appropriately accommodates the inclusion of large numbers of indicator variables for individual hospitals in models with dichotomous dependent variables.15 We present odds ratios and derive relative risks around sample mean treatment and outcome rates. We used chi-square statistics to determine the statistical significance of the individual coefficients and odds ratios associated with each of the practice size measures. We also performed chi-square tests of the hypotheses that the coefficients on all of the practice size variables were jointly equal to 0. This approach accounts for clustering of patients within hospitals. We were unable to estimate conditional logit models that simultaneously accounted for both hospital-level clustering and the possibility of clustering among patients treated by the same doctor. On average, each physician treated 3.2 patients in the data set, and 71 percent of physicians treated three or fewer patients, which suggests that the degree of bias would be small. To further examine this, we relied on other (linear probability) models and found that accounting for physician-level clustering had little impact on the size or significance of the results.16
Sample characteristics. Of the 116,671 patients in our analysis data set, 39 percent had an index attending physician from a solo practice; 26 percent, from a practice of 2–5 physicians; 11 percent, from a practice of 6–9; 10 percent, from a practice of 10–19; 7 percent, from a practice of 20–49; and 7 percent, from a practice of 50 or more.
We found large, statistically significant differences in unadjusted average treatment patterns and outcomes across practice size categories (Exhibit 1
Many patient and physician characteristics also vary by practice size (Exhibits 2  and 3). Patients of solo physicians appear to be less healthy in many of the measures. They also were more likely to have physicians that were female (except those of the largest practices), over age fifty-five, in internal medicine, and with non-U.S. medical training.
Regression results. The upper portion of Exhibit 4 reports results from conditional logit regressions assessing the relationships between practice size and treatments. These models control for the physician and patient characteristics shown in Exhibits 2 and 3, and also for the hospital of the index admission. Values shown are odds ratios for the indicated practice sizes, compared with solo practice.17
Patients of non–solo practice physicians were statistically significantly more likely than patients of solo practitioners to receive catheterization within one day of admission and also significantly more likely to receive PTCA within one day. Physicians in the mid-size practices (10–49) had the highest one-day catheterization rates, and physicians in the largest practices (50+) had the highest one-day PTCA rates, although differences across the larger five size categories tended to be statistically insignificant.
The lower portion of Exhibit 4 The relationships between the index attending physician’s practice size and catheterization and PTCA rates were strongest within one day of admission. Over time, differences diminished in size and statistical significance, although all of the odds ratios remained greater than 1, and some remained statistically significant at thirty days. Patterns for CABG are less clear. Physicians in practices of 2–5 had much higher CABG rates than solo-practice physicians at seven and thirty days. However, solo physicians otherwise did not tend to differ from other practice sizes.
Exhibit 5
One of the most striking differences between the characteristics of solo-practice doctors and those in larger practices is in the prevalence of physicians who graduated from non-U.S. medical schools. Our regression models controlled for this characteristic, but to more fully explore these relationships, we reestimated models using just those physicians who had graduated from a U.S. or Canadian medical school. Results are consistent with those shown in Exhibits 2 and 3 (data not shown). For example, the odds ratios for practices of 2–5 physicians relative to solo practice were 1.11 for both one-day cardiac catheterization and one-day PTCA (p < .01 and p < .05, respectively).18
Exhibit 6
Traditional Medicare patients with AMI whose hospital care was primarily directed by solo-practice physicians had significantly different treatment pattern than patients treated by physicians from larger practices. Specifically, solo physicians’ patients were less likely to receive catheterization and PTCA quickly following their initial admission. Although the magnitude of the absolute differences was modest, the relative differences were more notable. For example, non-solo physicians’ patients had one-day catheterization rates 10–12 percent higher and one-day PTCA rates 10–26 percent higher. Results also indicated lower mortality among patients of non-solo physicians, although the statistical strength depended on the chosen time frame and sample. Among patients treated by non-solo physicians, there was not significant evidence for differences in care or outcomes across practice sizes. Some models yielded differences in treatment patterns between practices with 2–5 physicians and those with 10–49 or 50 or more physicians, but patterns were not consistent across the different treatments studied. These differences between the non-solo practice sizes usually were not statistically significant. Possible explanations. Three types of explanations could be associated with these findings. First, differences in the organizational structures, infrastructure, opportunities for communication, or other features could cause differences in treatments and outcomes. Our research was not designed to identify specific mechanisms that might be at work. However, to the extent that causal mechanisms are an explanation for our findings, our results suggest that whichever mechanisms are at work vary noticeably between solo and non-solo practices but less so between non-solo practices of different sizes. One factor that might differ significantly between solo and other practices, but not among larger practices, is the ease of information sharing and consultation. A second potential explanation is that the variations reflect underlying differences in either the types of physicians in solo practices or the types of patients treated by solo physicians relative to others. Our regression models controlled for a range of physician and patient characteristics, which should reduce the possibility that our results are due to this type of variation. Our results also indicate that differences in the treatments and outcomes of patients treated by solo and non-solo physicians exist even within the same hospital. However, these statistical adjustments cannot rule out the possibility that still other, unmeasured differences between solo and non-solo physicians and patients play a role in the findings. For example, physicians who choose to be in solo practice might, on average, have different approaches to practice or different preferences about interactions with other physicians that ultimately influence the ways in which they provide care. Similarly, patients of solo practitioners might have different views about treatment. For example, if such patients were less inclined toward catheterization or PTCA, our results might simply indicate differences in patients’ preferences. There are many different ways in which preferences could vary across patients, and we cannot rule them all out. However, we believe that differences in preferences about whether or not to seek treatments are unlikely to explain our results. We found that the primary difference between patients of solo and non-solo physicians was whether they received catheterization or PTCA within one day, and much less in whether or not they eventually received it. A third potential explanation for at least part of our results stems from the potential for error in our measure of practice size. This could play a role particularly in the finding of relatively few differences across the larger practice sizes, where measurement error could be more of an issue. The AMA data themselves, particularly those from the Medical Group Census, are subject to some measurement errors. In addition, measures of practice size might not fully capture the relevant information in large practices with, for example, multiple locations or many different specialties. Measurement error in practice size could attenuate the observable relationships to the point where we cannot detect differences among the larger practice sizes with these data. Study limitations. This study has limited ability to identify variations in quality, although some results provide suggestive evidence that could provide a foundation for future study. In some models, mortality rates appear lower in non-solo than in solo practices. In addition, guidelines for the treatment of AMI patients published in 1996 called for PTCA to be applied in a timely fashion, and a 1999 update called for PTCA within ninety minutes of admission for most patients.20 Although claims data do not provide sufficient detail to clearly evaluate adherence to these guidelines (for example, time to PTCA in minutes), results suggest that adherence to even the 1996 guidelines, while low in general, was lower still among those with solo attending physicians.21 Because we studied only 1999, we could not observe whether there have been more recent changes in treatment patterns. Differences might narrow over time. For example, if the explanation is simply that solo-practice physicians were slower than others to adopt new guidelines for treatment, perhaps they later adopted the guidelines, and differences narrowed. To the extent that other explanations are important, differences might have remained unchanged or even widened. Further research. Further research is needed to determine what creates these differences and whether they are found in settings beyond hospital care for AMI. If research establishes that physician practice size has widespread influence on practice patterns, policymakers could improve quality through a variety of means. Efforts to improve the dissemination of new information or increase guideline adherence might consider whether approaches could be effectively tailored to physicians in different types of practices or settings.22 Practice size might also interact with more generalized efforts to improve quality of care. For example, if larger practices are more successful at meeting pay-for-performance (P4P) initiative targets, these initiatives might promote the formation of larger groups.23 Public reporting of quality data might also stimulate such changes. Further research is needed to understand whether such reorganization would improve patient care. If guideline adherence is driven by access to infrastructure, referral networks, accountability, information sharing, or other features that are not found in solo practices, then such reorganization might be able to improve care. Alternatively, if our findings resulted from underlying differences in physician or patient characteristics, simply encouraging the formation of larger practices might have little or no impact on quality.
Jonathan Ketcham (ketcham{at}asu.edu) is an assistant professor in the School of Health Management and Policy, W.P. Carey School of Business, Arizona State University, in Tempe. Laurence Baker is an associate professor in the Department of Health Research and Policy, Stanford University, in Stanford, California. Donna MacIsaac is a data analyst in that department at Stanford. This project was supported by the Robert Wood Johnson Foundation Scholars in Health Policy Research Program at the University of California, Berkeley, and the University of California, San Francisco (Ketcham), and the Agency for Healthcare Research and Quality (Baker and MacIsaac).
This article has been cited by other articles:
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Top
Study Data And Methods
Study Results
