Incidence of Lower-Extremity Amputation in American Indians: The Strong Heart Study

The SHS was initiated in 1988 to investigate cardiovascular disease and its risk factors in American Indians (11). The design, methods, and laboratory techniques of the SHS have been previously reported (12,13). The SHS cohort consists of 4,549 participants aged 45–74 years from 13 American-Indian communities in Oklahoma, the Dakotas, and Arizona who were seen at the baseline examination conducted between 1989 and 1992. The second and third examinations were conducted from 1993 to 1995 and 1997 to 1999, respectively. Within the study population, nonparticipants were similar to participants in age, BMI, and self-reported frequency of diabetes and hypertension (13).

At each of the three SHS examinations, trained examiners collected amputation data from direct observation of both legs. These data included the presence/absence of portions of lower limbs and the cause of missing extremities. Missing extremities were coded by study personnel as toe/foot, below-the-knee amputation (BKA), or above-the-knee amputation (AKA). Participants were asked to attribute each missing extremity to a specific factor. These attributions were coded as diabetes (LEA), trauma, congenital, other, and unknown.

Participants were categorized as having diabetes if they had a fasting glucose ≥126 mg/dl (14), reported use of hypoglycemic medications, or reported that they had been told by a health care professional that they had diabetes. Diabetes duration among participants who reported having previously been diagnosed with diabetes was assessed by questionnaire at the baseline examination.

SHS methods and protocols have been described (11,12). Age, diabetes, smoking, hypertension, loss of protective sensation, and male sex have been associated with increased LEA risk in other studies (4,5,9–11,15). Participants were considered hypertensive if they had a systolic blood pressure ≥140 mmHg, a diastolic blood pressure ≥90 mmHg, or if they were taking antihypertensive medication. Smoking history was determined by questionnaire and coded as ever versus never. At-risk drinking was defined as report of either ≥5 drinks on one occasion in the past month or ≥14 drinks in a typical week (1,16). Cholesterol, triglycerides, and fasting glucose were determined by enzymatic methods using a Hitachi chemistry analyzer and consistent standardized reagents (Boehringer Mannheim Diagnostics, Indianapolis, IN). Fasting insulin and fibrinogen were measured by established methods (17,18). Microalbuminuria was defined as a ratio of urinary albumin (in milligrams per milliliter) to creatinine (grams per milliliter) of 30–299 mg/g and macroalbuminuria as a ratio of ≥300 mg/g.

An ankle-brachial index (ABI) of <0.90 (“low”) is 95% sensitive and 99% specific for angiographically documented peripheral arterial disease (19,20) and has been used in previous work (21) in the SHS. Data from the SHS show that an ABI >1.40 (“high”) is associated with a similar level of cardiovascular disease and mortality risk as low ABI (22). Baseline ABI was used to identify three groups of individuals: those with low ABI (<0.90), those in the normal range of ABI (0.90 ≤ ABI ≤ 1.40), and those with high ABI (>1.40). Monofilament data and other measures of neuropathy were not collected at the baseline examination, and information on history of foot ulcers was unavailable.

χ² and Student’s t tests were used to examine differences in proportions and means between categorical and continuous variables, respectively. Because the purpose of this report is to understand risk factors for a first amputation, comparisons focused on differences in baseline risk factors between participants with diabetes who did and did not experience a first (incident) LEA during the follow-up period. We report selected supplemental data on recurrent amputations among individuals with incident LEA. We were also interested in the effects of combinations of LEA risk factors, particularly duration of diabetes, HbA_1c, ABI, and renal function. Accordingly, we created dummy variables for these factors and examined their age-adjusted joint effects on LEA risk. We then used the logistic regression model to identify risk factors that were independently associated with increased odds of incident LEA. Variables chosen for inclusion in the logistic regression models were based on findings from univariate analysis and previous studies.

Following examination of main effects in the logistic models, we explored the interaction of elevated HbA_1c and renal dysfunction to test the hypothesis that this combination of risk factors acted synergistically to increase LEA risk. Despite our interest in other key interactions, sparse cells prevented testing of the interactions between ABI × HbA_1c and ABI × renal function in the logistic models. In some analyses, we examined LEA risk factors separately in relation to minor (toe/foot) and major (BKA+AKA) amputation. We employed the Hosmer-Lemeshow test to evaluate the goodness of fit of the models and examined receiver-operating characteristic curves to assess how well the models predicted an incident LEA.

Of the 4,549 participants in the SHS cohort, 1,974 (43.4%) had diabetes and no LEA at baseline. This group constitutes the analysis sample in this report. Of these 1,974, 87 (4.4% of individuals with diabetes; 6% of men and 3.5% of women) experienced an incident LEA during a mean of 7.9 years of follow-up (Table 1). Among both men and women, amputation of toes was most common, followed by BKA and AKA. Compared with women, a slightly larger proportion of men had incident LEA above the knee: 14 vs. 11%. The difference in proportions was somewhat larger for BKA: 35% in men vs. 25% in women. Conversely, among women with incident LEA, amputation of toes was somewhat more common than among men. A total of 159 anatomical sites were amputated among the 87 individuals with incident LEA, yielding ∼1.8 amputated sites per individual. However, men with incident LEA had more amputated sites than women. The 43 men with incident LEA had 85 sites, yielding ∼2.0 sites per man, whereas the 44 women had 74 sites, yielding ∼1.7 sites per woman. Twenty-five participants with incident LEA (29% of all LEAs) had more than one amputation during follow-up. Among participants with incident LEA at the second examination, eight individuals had ipsilaterial LEA at the third exam. Two participants had an incident LEA at exam 2 and a contralateral LEA at exam 3, seven participants had a bilateral LEA ascertained at exam 2, and nine participants had a bilateral LEA ascertained at exam 3. One participant had both ipsilateral and contralateral LEAs between exams 2 and 3.

Table 2 compares the baseline characteristics of SHS participants with and without incident LEA during follow-up. Participants with diabetes who experienced incident LEA were more likely to be male; to be from the Arizona center; to have longer duration of diabetes; to have renal disease, abnormal ABI, and/or higher systolic blood pressure; and to lack a high school education (P < 0.01, for all). Compared with participants with normal albuminuria, risk of LEA was increased 5.6 and 5.4 times among those with micro- and macroalbuminuria, respectively. Participants with high ABI had an LEA risk 2.4 times that of those with normal ABI, and the risk was marginally elevated among those with low ABI. Baseline BMI was lower in participants who experienced an incident LEA compared with those who did not.

Figure 1 shows the age-adjusted effects of HbA_1c and renal function on odds of incident LEA. In general, as HbA_1c increased and renal function diminished, risk of LEA increased, with the combination of high HbA_1c and poor renal function having stronger effects than either characteristic alone. Similarly, the combination of unfavorable ABI and increasing HbA_1c was also associated with substantially increased age-adjusted odds of incident LEA (Fig. 2), and the combination of abnormal ABI and diminishing renal function was also associated with increased risk of LEA (Fig. 3). Of note in these analyses is the unexpectedly lower risk of LEA in some categories of macroalbuminuria. Twenty-two percent of participants with baseline macroalbuminuria (compared with 11% in the microalbuminuria and 8% in the normal groups) had died by the second SHS examination, when incident LEA was first ascertained.

Logistic regression models are based on data for 81 participants with an LEA and 1,799 participants without an LEA who had complete data for all covariates of interest. Table 3 shows that men were twice as likely as women to experience an incident LEA and having a high school education reduced the risk by >50%. Increasing diabetes duration and HbA_1c both influenced risk of incident LEA. The effect of poor glycemic control on risk of LEA was considerable: compared with participants with HbA_1c <6.5%, LEA risk was increased two- and threefold among participants with HbA_1c between 6.5 and 9.5% and among those with HbA_1c >9.5%, respectively. Microalbuminuria (odds ratio [OR] 2.67, 95% CI 1.48–4.84) was associated with increased LEA risk, as was a high baseline ABI (1.80, 1.02–3.17). The Hosmer-Lemeshow test suggested that the model fit the data well (P = 0.88), and inspection of a receiver-operating characteristic curve indicated that in this sample the model correctly identified LEA cases 80% of the time. We observed no significant interactions between pairs of risk factors and LEA risk in the adjusted models. Although stratification of our models according to major/minor LEA reduced statistical power considerably, these analyses did not result in meaningful differences in the results with respect to the direction or magnitude of point estimates.

We also examined the association of baseline risk factors with risk of multiple (more than one) incident LEA. We defined participants as having multiple LEAs if they were missing portions of extremities on both legs or feet at a single SHS examination or if they had missing extremities at the second examination, followed by additional LEAs at the third examination. Forty-four percent of participants with more than one LEA had macroalbuminuria at baseline compared with 25% of those with one LEA and 18% among those without LEA (P < 0.0001).

This report presents a systematic analysis of incident LEA in a diverse, population-based sample of American Indians with diabetes and indicates that the 8-year cumulative incidence of a first LEA was 4.4%. The adjusted risk of LEA in men was twice that of women, and increasing diabetes duration, poor glycemic control, low educational attainment, renal disease, and ABI >1.40 were all significantly associated with increased LEA risk.

In our study, the adjusted risk of LEA associated with male sex was 2.06, a consistent finding in at least two previous prospective studies of incident LEA in American Indians. In the Oklahoma Indian Diabetes Study (5) (n = 875), men’s risk of incident LEA was twice that of women’s, and in a study of 4,399 Pima Indians (6), the rate of LEA in men was 2.6 times higher than in women, adjusted for age and diabetes duration. However, the observation that men are at higher risk of LEA is not a universal finding. In a national study of 20-year risk of LEA, sex did not predict incident LEA (15). One reason for the absence of a sex effect in the latter study may be due to the fact that it included a mix of individuals with and without diabetes, potentially masking an important sex effect in studies exclusively of individuals with diabetes.

Several factors might explain the persistent observation that men with diabetes experience more LEAs than women with diabetes. First, this observation may be due to anatomical differences predisposing to diabetic neuropathy, a known LEA risk factor. Insensate neuropathy likely reflects severe underlying nerve damage and is a known risk factor for LEA (23). A twofold sex difference in insensate neuropathy reported in one study (24) is consistent with twofold sex differences in LEA risk in the current report and other LEA studies in American Indians with diabetes (5,6). The SHS did not collect measures of sensory neuropathy at the baseline examination that would have permitted direct examination of whether men had more neuropathy than women. However, a number of studies support the key role of neuropathy as a risk factor for LEA, including one report (25) that showed that South Asians with diabetes have about one-quarter the risk of LEA of Europeans with diabetes, a difference attributable to lower rates of neuropathy and peripheral arterial disease.

It is also possible that men experience more minor trauma to the foot that ultimately results in LEA. Sex differences in type of employment rather than biological differences between men and women may be one way that Indian men are exposed to potentially high-risk foot trauma. Unfortunately the SHS does not have information that permits testing of this hypothesis. A third explanation might be related to the tendency for men to have less contact with the health care system or different attitudes toward self-care than women. A small study of the influence of sex among individuals with diabetes and severe foot lesions indicated that women were active in self-care and preventive care and actively sought information. In contrast, men more often sought care only for acute problems, had a passive attitude toward foot care, and frequently sought complementary care from their wives (26). Although the latter study was not conducted in American Indians, it is possible that attitudes and beliefs about foot care among Indian men, rather than sex per se, may contribute to the persistent association between male sex and LEA risk.

Peripheral arterial disease, defined in various studies as absent or diminished lower-extremity pulses (ABI <0.90 or <0.80), has been shown to predict incident LEA in individuals with diabetes (5,24,26). Our models showed an elevated (OR 1.50, 95% CI 0.66–3.40) but nonsignificant risk associated with ABI <0.90. Among participants with an LEA, 9.2% were in the low ABI group, compared with 6% in the no LEA group. However, there were only eight LEAs in the low ABI group, and this small number limited precision in the low ABI point estimate for LEA risk. Despite the absence of statistical significance, both the magnitude and direction of the point estimate are consistent with previous studies.

The association between high ABI and LEA persisted in multivariable analyses, conferring an adjusted LEA risk of 1.80 (1.02–3.17). Our finding that high ABI predicted LEA risk is consistent with a study showing that radiographically documented medial arterial calcification predicted risk of LEA in Pima Indians (6). A high prevalence of elevated ABI has been reported in the SHS and other studies of American Indians (22,27). The significant association between high ABI and LEA is likely due to the larger number (n = 20) of LEAs in the high ABI group, and the fact that high ABI was more than twice as common among those with an LEA compared with those without (22.9 vs. 11.1%).

One observation relevant to our findings is that among individuals with diabetes, medial arterial calcification is often present in individuals with both neuropathy and poor renal function. In previous work in the SHS, elevated ABI was associated with albuminuria, and both independently predicted mortality (22). In the current report, both ABI >1.40 and albuminuria independently predicted incident LEA. An interesting finding is this work relates to the protective role of education in risk of LEA. In the SHS cohort as a whole, 52% of participants had completed high school, but this proportion differs according to diabetes status: 46% of participants with diabetes completed a high school education compared with 54% of participants without diabetes. Part of the diabetes-associated difference in educational attainment in the cohort is an age effect, with older participants less likely to have completed high school. However, the effect of education persisted in multivariable models that adjusted for age and other LEA risk factors, with a high school education being associated with a reduction in LEA risk of >50%.

The importance of education in reducing risk of LEA was shown in a previous study (15) of black and white Americans in the National Health and Nutrition Examination Survey (NHANES) Epidemiologic Follow-up Study. This study showed that although there were substantially higher age-adjusted rates of LEA in black, compared with white, Americans, this excess risk was eliminated when the presence or absence of a high school education was considered, along with age, diabetes, smoking, and hypertension. Of particular interest in these two reports are the nearly identical adjusted risk estimates associated with having a high school education: in the NHANES Follow-up Study, the adjusted risk estimate for LEA associated with having completed high school was 0.47, and the corresponding risk estimate in the Strong Heart Study was 0.46. Thus, completing high school and/or the complex set of favorable factors that is associated with completing high school reduces the risk of LEA by >50%.

Data concerning the importance of blood pressure as a predictor of LEA in American Indians are conflicting. In the current report, systolic blood pressure was found to be an important predictor of LEA, results that are consistent with findings for men in the Oklahoma Indian Diabetes Study (5) but inconsistent with findings from Pima Indians (6).

Our finding of no association between smoking and LEA risk is consistent with findings from a study of Pima Indians (6). However, smoking is a known risk factor for LEA in other large prospective studies (15). A likely explanation for this finding is that smokers in the SHS cohort smoke fewer cigarettes per day than the national average, with use as low a six cigarettes per day among women in the Arizona center (28). Further, long duration of diabetes among SHS cohort members could overwhelm a weak association between smoking and LEA.

Our data show differences in the cumulative incidence of LEA across the three SHS centers, with the highest rates in Arizona (6.2%) and similar lower incidence in Oklahoma (3.3%) and the Dakotas (2.5%). Although regional differences in physician practice patterns and patient preferences may explain some of the geographic differences in LEA among individuals in the SHS, a more likely explanation for these differences is that participants in the Arizona center had longer duration of diabetes at baseline (13.6 vs. 9.0 and 10.1 years in the Dakotas and Oklahoma, respectively).

Our study has several important limitations. First, information on neuropathy, history of ulcers, or description of foot deformity was not available. If these data had been available at baseline, we expect that we would have observed significant relationships between these factors and incident LEA. To what extent our parameter estimates for risk factors such as sex and HbA_1c would have been influenced by inclusion of these factors in the model is difficult to estimate. However, if we had been able to exclude people with a history of foot ulcers from the analysis, it is likely that the number of incident LEAs would have been reduced, and this would have reduced the power of the study. Although we ascertained LEA among participants at each exam cycle by direct examination of the legs and feet, our protocol did not include the use of hospitalization records. We may therefore have missed some LEAs that occurred among participants who died between exams. Addition of these events may have resulted in increased power, which would have added precision to some LEA risk estimates with wide CIs that included 1.0, rendering some relationships statistically significant. Finally, our models were based on data from 81 of the 87 individuals with incident LEA and 1,799 of the 1,887 individuals without LEA who had complete covariate data. Empirical comparison of available risk factor data between the six LEA participants with incomplete covariate data and the 88 non-LEA participants with incomplete data suggested less favorable profiles for those with incident LEA for several important variables, including diabetes duration (19.9 vs. 13.5 years), high ABI (50 vs. 34%), low ABI (16 vs. 7%), and HbA_1c (10.9 vs. 8.6%). Although small numbers precluded reliable comparisons, these data suggest that the strength and precision of the associations we report for some LEA risk factors may be underestimated.

Duration of diabetes and baseline HbA_1c both appear to contribute important information on LEA risk among individuals with diabetes. Our data show increased risk of LEA associated with worse glycemic control, adjusted for diabetes duration, and the reverse was also true. Despite the importance of diabetes duration in our models of LEA risk, this factor is not one for which interventions can be effectively designed. However, our data suggest that it is possible to identify American Indians at particularly high risk of LEA. LEA prevention efforts may therefore be especially effective in this group. Thus, public health resources for LEA prevention should be focused on designing and implementing effective interventions for modifiable risk factors such as glycemic control, blood pressure, renal disease, and peripheral vascular disease.

This study was supported by grants U01-HL-41642, U01-HL-41652, and U01-HL-41654 from the National Heart, Lung, and Blood Institute.

The authors acknowledge the assistance and cooperation of the AkChin Tohono O’odham (Papago)/Pima, Apache, Caddo, Cheyenne River Sioux, Comanche, DE, Spirit Lake Community, Fort Sill Apache, Gila River Pima/Maricopa, Kiowa, Oglala Sioux, Salt River Pima/Maricopa, and Wichita Indian communities, without whose support this study would not have been possible. The authors also thank the Indian Health Service hospitals and clinics at each center and Betty Jarvis, RN; Tauqeer Ali, MD, PhD; and Marcia O’Leary, RN, directors of the SHS clinics and their staffs.