Geographic Variation of Lower-Extremity Major Amputation in Individuals With and Without Diabetes in the Medicare Population

As part of ongoing work with the Dartmouth Atlas of Health Care project, we analyzed claims data during hospitalizations for all nontraumatic lower-limb major amputations in individuals with and without diabetes enrolled in Medicare from 1996 through 1997 (6).

Numerators for nontraumatic amputations were selected from the MEDPAR file for 1996–1997. Numerators were calculated from hospital discharges using diabetes code (ICD-9-CM, code 250.x) and the highest-level amputation during the hospitalization. Major amputations were defined as transtibial (84.15) or transfemoral (84.17).

Denominators were calculated using the Medicare Denominator file (HISKEW) for 1996–1997. Denominators for individuals with diabetes were estimated by multiplying the regional prevalence of diabetes by the regional Medicare population. The regional prevalence of diabetes was based on HEDIS 3.0 using Medicare Part B claims data for 1995–1996. Patients with diabetes were defined by having a diagnosis code (ICD-9-DM 357.2x, 362.x, 648.0x, 250.xx, or 366.41) listed on at least two outpatient visits or one inpatient stay evaluation and management code (9,10). Denominators for individuals without diabetes were the remaining Medicare beneficiaries. To ensure that subjects counted in the numerator corresponded to the denominator, enrollees younger than 65 years of age or older than 99 years of age were excluded. Enrollees in risk-bearing health management organizations (∼11% for 1996) were excluded from both the numerator and denominator (11).

The unit of analysis was the hospital referral region (HRR). The 306 HRRs represent health care markets for their respective tertiary medical centers. Individual regions were created by the zip codes in which a plurality of patients sought care. The average Medicare population for an HRR was 92,600; the smallest was 14,900 enrollees (6). Previous authors studied 11 surgical procedures and found 90% of the enrollees resided in the same HRR as the surgical center (7). Mapping of HRR boundaries by zip code was created using zip code boundary files from Geographic Data Technology (Lebanon, NH).

An indirect method was used to adjust procedure rates for sex, race (black, nonblack), and age (65–69, 70–74, 75–79, 80–84, 85+). To assess regional variation, the ratio from the highest to lowest HRR rates was calculated (7). HRRs with rates based on 10 or fewer procedures were dropped from the analysis to avoid unstable estimates of regional variation (7). For individuals without diabetes, two HRRs were dropped for major amputation.

Minor amputations (i.e., toe, ray resection, transmetatarsal, and Chopart’s) were dropped from this analysis. We found 52,285 minor amputations in our analysis of 1996–1997 Part A data. A separate analysis of 1997 Part B data found 53,705 minor amputations. It seems that most minor amputations missed in Part A data were performed in individuals without diabetes. This is supported in our analyses, because 35 HRRs had to be dropped from the measure of extremal ratio and 80 HRRs were found to have rates >25% below the national average for minor amputation in individuals without diabetes. There will be future limitations in using Part B Medicare data for minor amputation in individuals with diabetes. Claims are limited to one diagnostic field and will likely yield low sensitivity for the diabetes diagnosis because minor amputations are performed for multiple diagnostic reasons (i.e., gangrene, osteomyelitis, septic arthritis, diabetes, etc.).

In the Medicare population during 1996 and 1997, a total of 44,599 and 39,111 major amputations were performed in individuals with and without diabetes, respectively. A total of 6,037,804 Medicare patients with diabetes were identified in 1996; there were 50,416,125 Medicare beneficiaries without the diabetes code in 1996. This resulted in a diabetes prevalence of 12% for the Medicare population in 1996. Patients with diabetes were 10 times more likely to undergo major amputation. The adjusted rate of major amputation for individuals with diabetes was 3.83 per 1,000 (95% CI 3.60–4.06) compared with 0.38 per 1,000 (95% CI 0.354–0.406; P < 0.0000) individuals without diabetes. The overall rate of major amputation was 3.69 per 1,000 patients with diabetes. For patients without diabetes, the overall rate was 0.39 per 1,000. Diabetes-related amputations accounted for 53% of major amputations.

Higher rates of major amputation for individuals without diabetes were observed in a wide path across the south, extending from North Carolina to Texas, and high rates were noted in the Mid-Atlantic states. However, in individuals with diabetes, the regions with the highest rates were more scattered, with pockets located in California, Texas, New Mexico, the upper Midwest, and the Mid-Atlantic states (Figs. 1 and Fig. 2).

Variation (the ratio from highest to lowest rate) across HRRs for individuals with diabetes was 8.6-fold compared with 6.7-fold in individuals without diabetes for major amputations (Table 1). The Pearson correlation coefficient was calculated for major amputation for individuals with diabetes to major amputation for individuals without diabetes and was found to be 0.56 (adjusted R² = 0.31).

Major amputation rates for individuals with diabetes exhibited 8.6-fold variation compared with 6.7-fold variation for major amputations for individuals without diabetes. Major amputation rates exhibited a high degree of variation when compared with 11 other procedures in the Medicare population over the same time period using the same methods. Only lower-extremity revascularization, carotid endarterectomy, back surgery, and radical prostatectomy exhibited higher variation than amputation in individuals with diabetes (7). In The Netherlands, rates of diabetes-related amputation for the 27 health care regions varied fourfold, whereas nondiabetes-related amputation varied only twofold (8). Our study found higher rates of variability than that observed in The Netherlands but a similar pattern of higher variability in individuals with diabetes as compared with individuals without diabetes.

Certain regions exhibited higher rates of major amputation in the Medicare population. For individuals without diabetes, higher rates were generally observed in the Southern and Atlantic states. However, for individuals with diabetes, higher rates of major amputation were much more varied.

Geographic variations in surgical care have been ascribed to population characteristics, underlying surgical capacity and lack of physician agreement on diagnostic or treatment modalities (12). We attempted to control for differences in the population characteristics by adjusting for age, sex, and race. Severity of disease and comorbid conditions may also affect rates of amputation. Previous authors using the same data set and methods found no change in rates of other procedures after the additional adjustment for comorbid conditions using modified Charlson scores (13). Other important population characteristics, such as education and socioeconomic factors, that may influence self-care and access to care (14,15,16) were not controlled because those data were not available.

Amputation rates may also be affected by lack of access to medical care. Common barriers include physical barriers of geography or transportation as well as financial and cultural issues. We have no evidence for geographic variation of these types of barriers.

Increased surgical capacity has also been shown to increase procedure rates (17). We have no data on geographic distribution of surgeons or providers of preventive foot care. For this reason, the HRR was selected as the unit of analysis in the present study because it is likely that most tertiary care facilities have a multidisciplinary foot clinic.

One of the more intriguing explanations for the differences in variability across regions involves regional differences in diagnosis and management, which develop when physicians do not agree on the best way to diagnose or manage a particular problem (6,7,18). This theory suggests that in the face of controversy and uncertainty, physicians develop a local culture or regional “practice style.” For example, the rates of internal fixation of hip fracture exhibit little geographic variation, reflecting a straightforward diagnostic process and essentially one treatment option. In fact, the performance of internal fixation for hip fracture closely approximates the incidence of hip fracture in each population (7). On the other hand, radical prostatectomy for prostate cancer exhibits high regional variation, reflecting the plethora of opinions on diagnostic and treatment strategies (7). We suspect that performance of major amputation, with its dire implications for disability and mortality, would be used only in extreme situations and would thus exhibit low regional variation. The correlation of 0.56 (adjusted R² = 0.31) suggests that within each region, major amputation decisions for diabetic and nondiabetic populations are influenced in a similar fashion.

The risk of amputation may be affected by management of risk factors leading to amputation. Preventive measures provided by multidisciplinary foot care teams implementing risk assessment, therapeutic foot wear, and patient education have demonstrated a 50–85% decreased risk of ulcer and amputation (19). Management of foot ulcers remains controversial, but effective management should include adequate assessment and treatment of infection, the use of debridement, and off-loading of weight during ambulation to increase the rate of ulcer healing and minimize amputation (20). Although there is significant disagreement regarding the most effective way to provide preventive care and ulcer management, there is currently no evidence for geographic variations in preventive foot care or ulcer management.

Peripheral vascular disease is the major cause of amputation in nondiabetic individuals and contributes to approximately half of all amputations in individuals with diabetes (21). Controversies regarding the appropriate assessment and management of peripheral vascular disease also exist. Although some centers of excellence have reported a decrease in amputation rates after aggressive surgical revascularization (22,23,24), population-based studies have found no change in amputation rates associated with a tripling of the vascular procedure rate (25,26). Interestingly, large geographical variations for lower-extremity vascular bypass procedure have also been described (6).

Our analysis has several potential limitations. Hispanic ethnicity is associated with lower rates of amputation in individuals with diabetes as compared with African-Americans or whites (16). We did not adjust for Hispanic ethnicity; we adjusted only for race by African-American status, which might overstate the amputation rate in regions with large Hispanic populations. Second, there may be some misclassification in ascertaining diabetes in the numerator and denominator. The sensitivity for the diabetes diagnosis in Medicare claims data has been reported to be 84% (95% CI 81–86%) when compared with chart audit (27). This same study reported that coding of major procedures had high sensitivity, ranging from 88 to 100%. A separate study reported that the diabetes diagnosis based on claims data in an ambulatory setting was sensitive (>70%), specific (>98%), and reliable (κ > 0.80) (28). That study used a method very similar to the present study. Also, estimates of the diabetes denominator in the present study compare quite favorably with other published studies (29).

Our analysis provides a glimmer of hope in reducing amputation rates. If there truly are differences in prevention and management that lead to amputation, these could be identified and adopted by other areas. If the rest of the country performed amputations at the 10th percentile rates of major amputation, there would be 6,299 fewer major amputations per year in Medicare patients with diabetes.

In summary, lower-limb amputation in individuals with and without diabetes exhibits high regional variation in the Medicare population, even after adjusting for the patient characteristics of age, sex, and race. We propose that this variation may be due to systematic differences in preventive care and treatment decision-making. Future work should explore regional variation in preventive care and identify the discretionary aspects of decision-making leading to amputation. Lastly, the observation that amputation rates are low in some areas suggests that care practices in those regions might be adopted on a widespread basis. This potential decrease in the amputation rate for the entire population should encourage public health officials, providers, and patients that there are indeed ways of decreasing the amputation risk for all diabetic individuals.

This work was supported in part by the Robert Wood Johnson Foundation and the Dartmouth Atlas of Health Care. G.E.R. was supported as a Career Scientist from the Veterans’ Administration Health Services Research and Development Program (grant no. RCS 98353). G.E.R. and J.A.M. were also partially supported by the Rehabilitation Research and Development Center for Excellence in Amputation, Prosthetics, and Limb Loss Prevention of the Veterans’ Administration (VA) Puget Sound Health Care System under the VA Rehabilitation Research and Development Program Grant no. A0806C.

We thank Sally Sharp and Janice Drexler for their help in data acquisition and John Birkmeyer and Jack Wennberg for their thoughtful suggestions and help with data acquisition.