 |
 |
 |
|
| home | CME | help | feedback | subscriptions | archive |
|
© 1999 The Journal of Bone and Joint Surgery, Inc. Geographic Variations in the Rates of Operative Procedures Involving the Shoulder, Including Total Shoulder Replacement, Humeral Head Replacement, and Rotator Cuff Repair*   , NEW YORK, N.Y.Investigation performed at New York Orthopaedic Hospital and International Center for Health Outcomes and Innovation Research, Columbia-Presbyterian Medical Center, New York City
Background: Although geographic variations in the rates of orthopaedic procedures have been well documented, considerable controversy remains regarding the factors that drive these variations, particularly the role of the availability of orthopaedic surgeons. Moreover, little attention has been specifically focused on variations in the rates of commonly performed shoulder procedures. Methods: The current study documents state-to-state variations in the rates of total shoulder replacement, humeral head replacement, and rotator cuff repair and examines factors that might account for these variations. The regional incidences of these three procedures were analyzed with use of the Health Care Financing Administration Medicare database (MEDPAR, 1992). The rates were age-adjusted, and variations were measured with use of high:low ratios, variation coefficients, and systematic components of variation. Potential causes of variation were analyzed with use of Spearman and partial correlations as well as with Poisson regression. Results: Rates for the three procedures that were studied varied from one state to another by as much as tenfold. Humeral head replacement had the lowest rate of variation according to all three measures. All three procedures were performed less often in states that were more densely populated. With the numbers available for study, no consistent, significant relationship was found between the density of orthopaedists and shoulder surgeons and the rates of any procedure. Conclusions: The striking variations that were noted for these commonly performed procedures showed that there is a clear need for well designed clinical research to further define the factors that account for the variations and to examine the effectiveness and appropriate indications for the procedures.
Epidemiological studies have documented widespread and persistent variations in the utilization of both orthopaedic and nonorthopaedic procedures, such as those for the treatment of acute myocardial infarction8,21 as well as hysterectomy, prostatectomy, and coronary artery bypass grafting2,9,12,13,22,27. With regard to orthopaedics, geographic variations have been found for several types of operative intervention, including procedures on the lumbar spine1,15,18 and the cervical spine4, total hip and total knee arthroplasty19,28, and arthroscopy. However, virtually no variation has been found for other procedures, such as those for the treatment of fractures about the hip3,5-7,10,11,25. It is commonly believed that professional uncertainty is an important factor in these variations. This uncertainty is largely the result of incomplete scientific evidence concerning the so-called appropriateness of alternative treatments for some clinical conditions. For example, there is considerable controversy about the benefits and risks of operative compared with nonoperative treatment of herniated lumbar discs, and widespread variations have been documented15,18. In contrast, for fractures about the hip, internal fixation is widely accepted as providing the best outcome, and low rates of variation have been noted10,11. In addition to professional uncertainty, other factors, including differences in the availability and practice style of physicians and differences in the characteristics of patients, may determine the rates of utilization of particular procedures. Keller et al. suggested that variations in the incidences of several major orthopaedic procedures may be explained at least partially by differences in the number of orthopaedic surgeons among regions11. However, Peterson et al., in a 1992 study of variations in the utilization of hip and knee replacements in the Medicare population, were unable to confirm an association between the orthopaedic workforce and the volumes of such procedures19. Those authors found a strong inverse correlation between the rates of these procedures and the population density in each state (r = 0.5 for total hip replacement and r = 0.46 for total knee replacement). Thus, the factors that may account for widespread variation in the utilization of medical care remain to be further elucidated. As far as we know, no studies have examined whether there are regional variations in the rates of operative procedures on the shoulder. Furthermore, the relationship between factors related to the delivery of care (such as variations in the physician and orthopaedic-surgeon workforces) and factors related to the consumption of health care (such as the characteristics and preferences of the patients) has not been addressed with regard to operative procedures involving the shoulder. Despite ongoing innovations in the operative treatment of shoulder pain, uncertainty persists about when to use nonoperative treatment and when to intervene operatively. Moreover, patients who have shoulder problems differ with regard to their attitudes toward risk and their preferences for outcomes and, therefore, with regard to their decisions about appropriate treatment. These observations suggest that there may well be major variations in the rates of operative procedures involving the shoulder. The current study had two objectives. First, we sought to determine whether there are significant variations in the utilization of common shoulder procedures, including total shoulder replacement, humeral head replacement, and rotator cuff repair. Second, if such variations were found, we wanted to analyze certain factors that may contribute to them; we were especially interested in possible relationships between the rates of shoulder procedures and the distributions of orthopaedic surgeons, of orthopaedic surgeons with special expertise in operative treatment of the shoulder, and of primary-care physicians.
Sources of Data
Calculation of Rates The relationship between the age-standardized rates for each procedure in each state and various patient-related and physician-related factors then were examined.
Variables No truly valid measure of practice specialization relating to the shoulder is available for each state. Special training in the operative treatment of the shoulder may be included in various fellowships, including traditional shoulder fellowships, sports-medicine fellowships, and certain upper-extremity fellowships. Furthermore, many orthopaedic surgeons maintain an interest in operative treatment of the shoulder, and may subspecialize in it, without having had formal fellowship training. Such interest may be reflected in subscription to the Journal of Shoulder and Elbow Surgery; the rates of subscription to this publication (which were obtained courtesy of the Journal of Shoulder and Elbow Surgery) correlated highly with the distribution of members of the American Shoulder and Elbow Surgeons (obtained courtesy of that association) (r = 0.94). Thus, we used the rates of subscription to the Journal of Shoulder and Elbow Surgery as a proxy for specialization in the treatment of the shoulder. Primary-care physicians were defined as all physicians in family practice, general practice, and internal medicine combined. Because of our nation's increased focus on primary care and gatekeeping, we investigated whether the density of primary-care providers was correlated with the rates of procedures.
Correlations
Poisson Regression
Quantification of Variation Coefficient of variation: The coefficient of variation is a method of standardizing variation across procedures. These coefficients were calculated on the basis of the standard deviation divided by the mean for each of the three procedures. Systematic component of variation: The systematic component of variation has been used in the literature on variations in rates to calculate the amount of variation that is not random16,17,23,26. The systematic component of variation is equal to the estimate of total variance minus the random component of variation; that is, the value represents the amount of nonrandom variation that remains unexplained. A large systematic component of variation implies that many theoretically measurable factors have not yet been taken into account. The pioneers of this concept were McPherson and Wennberg16,17, and we used their method to perform our analysis. We calculated the systematic component of variation for each procedure, using directly age-standardized rates to quantify the amount of unexplained variation not related to age.
Variations in the Rates of Procedures
Rotator cuff repair had the largest coefficient of variation (0.477), and total shoulder replacement had the largest high:low ratio with regard to rates of procedures (10.36), whereas humeral head replacement had both the lowest coefficient (0.301) and the lowest high:low ratio (4.48) (Table II). Thus, the rate of humeral head replacement was less variable than the rates of the other procedures, according to all analyses.
We observed the same trend with regard to the relative amount of systematic variation for each procedure. When systematic components of variation were calculated according to age-adjusted rates, rotator cuff repair had the largest component of nonrandom (that is, area-related) variation (0.342) and humeral head replacement had the smallest (0.045) (Table II). A smaller systematic component of variation implies that age accounts for more of the variation in rates, whereas a larger one suggests that there is variation exceeding that which is expected on the basis of age alone. The physician workforce varied widely among states as well, not only in terms of absolute numbers but also when population density was taken into account. Of the fifty-one states included in our analysis, the District of Columbia had the highest population density (9649.5 people per square mile), followed by New Jersey (1049.9), and Alaska had the lowest rate (1.0) (Table III). In terms of overall physician density, the District of Columbia had the highest rate (53.44 per 1000 Medicare beneficiaries), followed by Maryland (31.65), and Mississippi had the lowest rate (10.17). With regard to the density of orthopaedic surgeons, Alaska had the highest rate (1.64 per 1000 Medicare beneficiaries), followed by Colorado (0.68), and Ohio had the lowest rate (0.05). In terms of the density of shoulder specialists, Wyoming had the highest rate (0.618 per 1000 Medicare beneficiaries), followed by Alaska (0.107), and North Dakota had the lowest rate (0.020). With regard to the density of primary-care physicians, Alaska had the highest rate (11.00 per 1000 Medicare beneficiaries), followed by Delaware (9.11), and Mississippi had the lowest rate (2.88).
Population Density
Workforce Regression estimates for the three procedures were derived with use of the generalized estimating equation method14 (Table V). The estimated relative risk is the exponential of the regression coefficient. The interpretation of this coefficient is that, for a one-unit increase in the predictor, with all other predictors held constant, the rate of procedures increases by a factor of relative risk. With regard to age-groups, relative risk represents the factor by which the rate of the procedure increases in a given age-group relative to the reference age-group of sixty-five to sixty-nine years. A relative risk of less than one suggests that higher values of the factor are associated with lower rates of procedures, whereas a relative risk of more than one suggests that higher values are associated with higher rates of procedures. Lower and upper limits of a 95 percent confidence interval also were calculated for relative risk. A confidence interval that does not include the number one indicates that the factor is significant at the 5 percent level. The age-groups with a higher proportion of men tended to have higher rates of procedures; these differences were significant with regard to humeral head replacement and rotator cuff repair (Table V). More sparsely populated states had higher rates of all three procedures. In addition, all of the older age-groups had significantly more humeral head replacements than did the reference group that was sixty-five to sixty-nine years old (p = 0.05). Rotator cuff repairs were positively associated with the density of primary-care physicians, and the oldest age-group had significantly fewer procedures than did the youngest age-group (p = 0.05). Other than population density, none of the factors were predictive of the rate of total shoulder replacement except age; the group that was seventy to seventy-four years old and the group that was seventy-five to seventy-nine years old had higher rates of procedures than did the group that was sixty-five to sixty-nine years old. The Poisson analysis did not confirm a significant association between the densities of primary-care physicians, orthopaedic surgeons, and shoulder specialists and the rates of procedures (Table V).
These results are based on correlation between the rates of procedures and the density of physicians, with both factors expressed as the number per Medicare beneficiary. For comparison, we examined the same correlations with use of the density of physicians (primary-care physicians, orthopaedic surgeons, and shoulder specialists) expressed as the number of physicians per resident in each state. Perhaps the availability of physicians to the Medicare population is determined more by the number of physicians per state resident than by the number per Medicare beneficiary. When physician density was expressed in this way, partial correlation revealed a significant correlation between the densities of orthopaedic surgeons and shoulder specialists and the rate of rotator cuff repair only (r = 0.39 and r = 0.29, respectively). However, the results of the Poisson multivariate analysis were essentially unchanged regardless of which way the analysis was done; population density showed a strong negative correlation with the rates of procedures, and there was no real correlation between physician density at any level of specialization and the rates of procedures.
Three general conclusions are supported by this study. First, there was significant variation in the rates of each of the three procedures, with humeral head replacement having the least variation. Second, population density showed a strong, significant negative relationship with the rates of all three procedures. Third, the physician workforce had a limited role in driving the rates of any of the procedures. Humeral head replacement is used to treat fractures of the proximal part of the humerus that are not amenable to primary fixation and to treat arthritis of the glenohumeral joint, and there is a roughly even distribution between these two major indications. Many authors have reported little variability in the rates of trauma-related procedures, such as those for fractures about the hip10, perhaps because there is less professional uncertainty regarding the proper management of these patients. Of the three procedures that were considered in the current study, humeral head replacement was by far the one most driven by acute trauma, which provides a possible explanation for the lower rates of variation associated with this procedure. Nevertheless, we suspect that there are differences in practice patterns among orthopaedic surgeons in different states with regard to the indications for humeral head replacement, even after trauma. It would be interesting to study whether low rates of humeral head replacement were offset by higher rates of open reduction and internal fixation of these fractures, but our database did not allow this analysis. Both correlation and multiple-regression analysis revealed a strong and consistent negative relationship between population density and the rates of all three procedures. This finding is similar to that reported by Peterson et al. in their study of total hip and total knee arthroplasty19. Database review cannot directly explain this finding, but there are several possible explanations. Population density may serve as a proxy for differences in other patient characteristics, such as lifestyle and occupation, that are not reflected by the available data. Residents of states with a low population density (for example, North Dakota, Idaho, Wyoming, and Montana) may have a higher prevalence of arthritis, overuse syndromes, and trauma as a result of the more strenuous, labor-intensive lifestyle in these rural areas, which have a large agricultural workforce. Alternatively, people who work in such occupations may have less tolerance for functional disability and incapacity because of the demands of their work and therefore may seek operative treatment more readily. In any event, the interesting and robust inverse relationship between rates of procedures and population density merits additional study. We were surprised to find relatively weak relationships between the availability of physicians with various levels of specialization and the rates of the procedures after controlling for age and population density. Pearson correlations between the densities of orthopaedic surgeons and shoulder specialists and the rate of total shoulder replacement were only moderate. The positive correlation between total shoulder replacement and the density of shoulder surgeons that is suggested by the Pearson correlation makes intuitive sense. Total shoulder replacement is a technically challenging procedure that has undergone considerable incremental innovations over the last few years. This procedure has only recently begun to be performed outside of large medical centers, and it still is likely to be predominantly the domain of specialized shoulder surgeons. Therefore, we expected that the availability of shoulder surgeons might drive the rate of total shoulder replacement. However, the Poisson multivariate analysis, which simultaneously adjusted for gender, age, population density, and workforce specialization, did not confirm these findings. Generally, we believe that the Poisson analysis more closely reflects reality in that it adjusts for a wider range of possibly confounding variables. Furthermore, the Pearson analysis does not weight observations according to sample size as the Poisson regression does; thus, it is more subject to the effect of large outliers. Our interpretation of the Poisson regression analysis is that there is no real relationship between the availability of physicians at any level of specialization and total shoulder replacement, after adjusting for a wider range of factors, such as population density, age, gender, and all of the physician densities. Instead, we found a weak negative relationship (relative risk = 0.91) between the density of orthopaedic surgeons and the rate of humeral head replacement as well as a stronger positive relationship (relative risk = 1.97) between the density of primary-care physicians and the rate of rotator cuff repair (Table V). Because these relationships are not detected with use of the Pearson model, are small in magnitude and barely reach the level of significance with use of the Poisson model, and are somewhat counterintuitive, their true importance, if any, is unclear. Our data suggest that there is little or no correlation between the density of either general orthopaedic surgeons or shoulder specialists and the rates of these procedures. Our results highlight some important points about variations among rates of procedures and their relationship to population and workforce variables. However, limitations of this study leave room for future investigations in these areas. First, use of only MEDPAR data limited our study to an older, Medicare population. Total shoulder replacement and humeral head replacement usually are performed in these age-groups, but rotator cuff tears often occur in a younger population. In addition, lifestyles and practice styles may vary in such a way that, in some states, patients may seek and receive treatment earlier (that is, before the age of sixty-five years old), whereas in other states, a systematic delay before treatment is sought may cause more patients to be counted as part of the Medicare population. These limitations can be addressed in future studies that include data on a wider range of age-groups. A second limitation of the current study is that shoulder specialists were accounted for with use of a proxy measure (subscription to the Journal of Shoulder and Elbow Surgery). Although this measure correlated strongly with the locations of members of the American Shoulder and Elbow Surgeons, not all physicians who consider themselves specialists in this area may have been included with use of this measure. Small differences in the density of shoulder specialists may have a major effect on our findings. A more direct measure of actual and self-described specialists in each state should be developed for use in future studies. Moreover, because of the available data, we used the state as the basic unit of analysis. Patients may cross state borders to obtain care, which introduces some error into the analysis. Our analysis did not account for this factor. This problem can be largely avoided by use of a hospital service area as the unit of analysis, as was done by Wennberg et al.27 and other investigators11, in numerous population-based studies. Finally, although we examined several important variables, many are left for future investigators to explore. The effects of payer mix, race, socioeconomic status (including occupational and lifestyle preferences), and surgical training are only some of the other factors that may contribute to variations in the rates of these operative procedures. In summary, we documented significant geographic variations in the rates of common shoulder procedures. In contrast to previous investigators, we failed to find a significant relationship between the availability of surgeons and the rates of these procedures; however, we found a strong negative relationship between the density of a given state population and the rates of operative procedures performed in that state. Variations in the utilization of health care have been well described for numerous areas of clinical intervention. These variations clearly have a profound effect on health-care costs and, more importantly, on the outcomes for patients, who often take some risk to achieve some apparent benefit. Patients who have an orthopaedic procedure take risks related to anesthesia and to the operation itself in order to obtain an improvement in functional status and quality of life. Technological innovations in orthopaedics have provided new opportunities for improving the quality of life of our aging population, but there is evidence that intervention may not be appropriately distributed. If the rates of shoulder procedures vary by as much as tenfold, as the current study suggests that they do, it is likely that we are either providing too little or too much intervention, or perhaps both. Although research based on a large database can reveal variations in the utilization of operative procedures, it cannot help to determine which rate is appropriate. The answer obviously depends on the prevalence of disease, patient preference, and the indications for operative intervention. Our findings suggest important areas that warrant additional research. First, we need to understand whether variations in the rates of shoulder procedures are indeed occupationally driven and whether these variations are the result of differences in the prevalence of disease or in patient preference, or both. Second, our finding of a particularly high variation in the rates of total shoulder replacement and rotator cuff repair suggests that additional high-quality clinical research is needed to determine the appropriate indications for operative intervention as well as the benefits, risks, and costs of different types of operative intervention. There is clearly a need for more clinical research to determine the effectiveness and appropriateness of these commonly performed procedures. Database review is a powerful tool that can help to determine how such research should be focused.
*No benefits in any form have been received or will be received from a commercial party related directly or indirectly to the subject of this article. No funds were received in support of this study.
  CiteULike  Connotea  Del.icio.us  Facebook  Technorati  Twitter What's this?
This article has been cited by other articles:
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Top
Abstract
Introduction
| home | CME | help | feedback | subscriptions | archive | site index |
|
Stanford University Libraries' HighWire Press (TM) assists in the publication of JBJS Online. |