In the course of the last 30 years, there have been several studies, initially by Wennberg and associates, that have demonstrated large differences in the rates of elective procedures across small geographic regions and across relatively large geographic areas (e.g., 306 hospital referral regions across the United States) [1–11]. Some of the procedures for which these differences have been found include tonsillectomy, hysterectomy, prostatectomy, carotid endarterectomy, hip replacement, and coronary artery bypass graft surgery.
For example, for cardiac procedures, Hannan and Kumar [11] found more than a three-fold variation in age/sex adjusted CABG rates and more than a two-fold variation in age/sex adjusted PCI rates in 12 regions in New York State. Also, Wennberg et al. found that there 82% of the variation (p < .001) in revascularization rates across twelve service areas in northern New England could be explained by differences in coronary angiography rates [9], and in another study Wennberg et al. found that in the same regions 43% of the variation in revascularization rates could be explained by differences in the number of catheterization laboratories per capita [10].
Our study differs from previous studies in that the regions that were identified (states) were considerably larger than regions that were examined in previous studies, and therefore less likely to exhibit large variations in utilization rates. Also, unlike some studies, we used HCUP data because it allowed us to identify all patients in a region receiving a procedure, not just Medicare patients.
Findings of our study were that there were large variations across states in age/sex adjusted PCI, CABG surgery and revascularization rates, although the variations were not as large as those found in other studies, all of which used considerably smaller geographical regions. We found variations of 1.83-fold for PCI, 1.54-fold for CABG surgery, and 1.54-fold for revascularization across the eleven states of interest.
In an attempt to explain differences in revascularization rates among states, we examined the relationship between revascularization rates and a few factors: the concentration of relevant specialists, the utilization of diagnostic catheterizations, proxies for the concentration of coronary artery disease, and socio-economic status. When these factors were tested in a linear regression model with statewide revascularization rate as the dependent variable, the only significant independent predictors were catheterization rate (positive correlation, p < 0.0001) and per cent white (p = 0.005). These two variables explained 94% of the variation in statewide revascularizaton rates, and cardiac catheterization rate explained 68% of the variation by itself.
High correlations between revascularization rates and cardiac catheterization rates have been found in earlier studies [9]. This could be related to the fact that states with sicker patients need to have more catheterizations and more revascularizations done. However, another possible explanation is that there are practice pattern differences among states in performing cardiac catheterization and revascularization, and that the states in which more patients are catheterized, more patients will be identified as needing revascularization. The latter hypothesis would appear to account for at least some of the differences found given the magnitude of the differences. In fact, one of the significant predictors of catheterization rate was the sum of the number of cardiac surgeons and interventional cardiologists per 100,000 population. Thus, the hypothesis that higher rates of procedures are associated with the "enthusiasm" and density of specialists who perform those procedures16 did appear to have some merit.
The significant relationship between white population and revascularization rate has been demonstrated in numerous studies conducted in smaller geographic regions than the regions examined in this study [20–27]. Our findings are disturbing because they suggest that minorities may have lower access to cardiac procedures after controlling as well as possible for need for these procedures, and that these access differences persist even across very large sub-populations of our country.
We also found a positive, although non-significant correlation between PCI and CABG surgery rates (R = 0.47, p = 0.14). However, when Oregon, which had a very low PCI rate, was removed, the correlation became statistically significant (R = 0.64, p = .046). Thus, there was not an observed tendency for substitution of procedures among states. These findings are similar to the findings of Kuhn et al., who reported that there was a .49 correlation between PCI and CABG surgery rates for Medicare patients across 305 Metropolitan Statistical Areas in 1988 [12]. These authors concluded that the rates were correlated because they were both highly correlated with the rates of cardiac catheterization (with respective correlations of .64 and .72), which was also the case in our study.
We also found that the proportion of patients who were revascularized who underwent PCI was quite variable across states. For patients with multiple vessels attempted, the risk-adjusted proportion who underwent PCI ranged from 10.4% in Oregon to 29.0% in Iowa. Both Oregon and Iowa had ratios that were more than 25% different than the eleven state mean.
There are a few caveats to the study. First, it is possible that differences in revascularization rates among states could be related to the differences in coronary heart disease, and therefore need for revascularization, rather than in practice pattern differences among states. Although we used age/sex adjusted acute myocardial infarction (AMI) admissions and coronary heart disease (CHD) admissions per 100,000 population to adjust for need for revascularization when examining the impact of race and catheterization rate on revascularization rate, they are both flawed when used for this purpose. For instance, many patients who need revascularization have not suffered AMIs. Also, many patients with CHD are not admitted to the hospital unless there is an intent to revascularize them, so hospitalizations for CHD are undercounts of the number of patients who need revascularization. They may also be overcounts because not all patients with CHD require revascularization. Also, the analyses in Table 3 that identified predictors of revascularization rate suffer from the use of ecological variables.
Second, identified differences between states in the choice of revascularization procedure were necessarily limited to inpatient data. Some PCIs are performed in an outpatient setting and the tendency for PCIs to be performed on an outpatient basis may vary between states. However, although many of the states studied do not have outpatient databases, it does not appear that many PCIs were performed outside of the inpatient setting in the year of our study (1999).
Third, we were limited to using administrative data in the study, and some data elements we would have liked to use were not available. In particular, it would have been desirable to have all data elements that are needed to determine the appropriateness of each procedure for purposes of the part of the study relating to choice of procedure, including the number of diseased coronary vessels. It is valuable to be able to distinguish between two-vessel disease and three-vessel disease because most of the latter group undergo CABG surgery. Nevertheless, one would expect that in regions as large as entire states, this type of bias would be minimized.
Thus, in conclusion, we found large inter-state differences in the rates of revascularization and in the tendency to choose revascularization procedures. As noted earlier, differences in regional procedure rates have been reported in several older studies and in some recent studies. Our study differs in that the regions we used (all patients in entire states) are considerably larger than regions used in earlier studies. Despite this fact we still found that there were substantial differences in the use of cardiovascular procedures. Other findings of the study suggest that these differences are due in part to access related to socio-economic status and to practice pattern differences. We look forward to other studies that examine variations in procedure choice and rates and to the development of databases that are capable of arriving at more definitive explanations for differences of this magnitude.