Risk Prediction Equations for Lower-limb Amputation: The Changing Face of Healthcare

Content provided by The O&P EDGE
Current Issue - Free Subscription - Free eNewsletter - Advertise

Earlier this year, as part of the presidential primary election cycle, a candidate was confronted with a position he had taken on a healthcare issue some 20 years earlier. His challenger observed that the candidate’s position on the issue had since reversed, and snidely inquired, “What has changed?” I found myself immersed in the related question, “In healthcare, what hasn’t changed over the past 20 years?”

Among the changes, today's healthcare system is increasingly informed by large datasets, which are used to predict the likely outcomes of a given pharmaceutical or surgical intervention. A rather ambitious effort in this realm was recently undertaken in the United Kingdom to determine the relative risk of amputation over a ten-year period among individuals with diabetes.1 In addition to weighting various characteristics and presentations, the result is an online tool, accessible at qdiabetes.org/amputation-blindness/index.php, where anyone with diabetes can calculate his or her individual risk of amputation and vision loss.

grahic of bar and pie charts

Analyzing this effort not only gives us some insights into our patients with diabetes, but also allows for contemplation of just how much might be learned from the aggregation of large datasets in our modern healthcare environment and how a similar analytical process might be applied to predict the outcomes of O&P interventions.

The Data

The authors of the study drew from two large datasets. The first was the QResearchR database from the United Kingdom, which includes clinical and demographic data from over 1,000 general practices, covering more than 20 million patients. This resource provides primary care data such as demographics, diagnoses, prescriptions, and laboratory results. Roughly three-quarters of the practices from the QResearch database were randomly assigned to a subdataset, composed of nearly half a million patients 25-84 years old with diabetes, that would be used to derive the risk prediction equations.

Data from the remaining practices in the QResearch database provided an additional 142,419 patients who met the same criteria (25-84 years old, with diabetes) that would be used to validate the risk equations. A second database, the Clinical Practice Research Datalink (CPRD) represents a similar resource that uses a different clinical computer system and provided data about an additional 200,000 patients with diabetes to be used as part of the validation effort.

In deriving the equations, a number of predictor variables were considered. These included the following:

  • Age
  • Ethnic group
  • Type of diabetes (type 1 or type 2)
  • Number of years since diagnosis of diabetes (< 1, 1-3, 4-6, 7-10, ≥ 11 years)
  • Smoking status (nonsmoker, ex-smoker, light smoker (1-9 cigarettes/day), moderate smoker (10-19 cigarettes/day), heavy smoker (≥ 20 cigarettes/day)
  • Townsend Deprivation Score (a measure of economic affluence and deprivation)
  • Glycated hemoglobin (HbA1c)
  • Systolic blood pressure
  • Body mass index (BMI)
  • Total serum cholesterol/high-density lipoprotein (HDL) cholesterol ratio

In addition to these largely continuous variables, the presence of a number of comorbid conditions was recorded, including:

  • Atrial fibrillation
  • Congestive cardiac failure
  • Cardiovascular disease
  • Treated hypertension
  • Peripheral vascular disease (PVD)
  • Chronic renal disease
  • Rheumatoid arthritis (RA)

As would be reasonably expected, the authors of this investigation considered the prevalence of these variables as observed between the three subdatasets (i.e., the derivation cohort and the two cohorts used to validate the resultant equations). The similarities between the two groups were astounding.the prevalence values for the different variables were within percentage points of one another with respect to nearly every consideration.


Risk Variables in Predicting Lower-limb Amputation

HbA1c: When the body processes sugar, glucose in the blood- stream naturally attaches itself to hemoglobin, the protein in red blood cells that transports oxygen throughout the body. When this occurs, the hemoglobin protein becomes glycated and will remain that way for the duration of the red blood cell’s life. HbA1c is a measure of the amount of glycated hemoglobin in the body and is proportional to the total amount of sugar that has been in the system. Because red blood cells survive in the human body for two to three months, measuring HbA1c provides a reasonable assessment of the blood glucose levels over that time period. This is generally considered more informative than measuring blood glucose levels, which only indicate the amount of sugar in the system at the time of the test.

HbA1c is measured in millimole/mole (mmol/mol) . Normal values for this index are below 42mmol/mol. Values between 42 and 47 are considered prediabetic, while values of 48 or higher are indicative of diabetes. Among 1,000 people with diabetes who reported hypoglycemic events, only 9 percent had HbA1c values below 42. By contrast, 30 percent reported HbA1c values between 42 and 53, 28 percent between 53 and 64, 14 percent between 64 and 75, with the remaining 19 percent reporting HbA1c values in excess of 75.1 HbA1c values can be lowered through exercise, diet, insulin, and pharmaceuticals.

  1. Lipska, K. J., E. M. Warton, E. S. Huang, H. H. Moffet, S. E. Inzucchi, H. M. Krumholz, and A. J. Karter. 2013. HbA1c and risk of severe hypoglycemia in type 2 diabetes: The diabetes and aging study. Diabetes Care 36 (11):3535-42.

Systolic Blood Pressure: Blood pressure rises and falls with each heartbeat. Accordingly, blood pressure is generally reported with two numbers. The first, the systolic pressure, is the higher of the two and represents the pressure in the arteries when the heart contracts. The second, the diastolic pressure, is the smaller number and represents the pressure in the arteries between heartbeats when the heart muscle is resting and the heart is refilling with blood. Of the two, the systolic pressure is monitored more closely as a risk factor for cardiovascular disease. Systolic blood pressure generally increases with age as the large arteries become stiffer due to plaque buildup.

Reported as millimeters of mercury (mmHg), normal systolic blood pressure is below 120. Values between 120 and 139 are considered prehypertensive. High blood pressure (hypertension) is defined as systolic pressures between 140-159, with pressures above 160 considered stage 2 hypertension. Values above 180 indicate a hypertensive crisis where emergency care is indicated.

Townsend Deprivation Score: The Townsend Deprivation Score is primarily used in analyzing populations according to four variables: unemployment, car ownership, homeownership, and household overcrowding. Scores range from -7 (most affluent) to 11 (most deprived). In the study in question, for every increase in the Townsend Deprivation Score of five units (i.e., greater deprivation) there was an associated increased hazard ratio of 1.1 for women and 1.3 for men.

All three cohorts were predominantly men, with the men averaging 60 years old and the women averaging 62 years old. Individuals with type 2 diabetes outnumbered those with type 1 diabetes by a ratio of roughly 95 percent to 5 percent. Across all three cohorts, just over 50 percent had diabetes that was diagnosed within the past year with the remainder presenting at one to three years post-diagnosis (≈ 16 percent), four to six years post-diagnosis (≈ 9 percent), seven to ten years post-diagnosis (≈ 8 percent), and more than ten years post-diagnosis (≈ 12 percent). Nonsmokers and ex-smokers were predominant, especially among women, with light, moderate, and heavy smokers at approximately 8 percent, 4 percent, and 4 percent, respectively.

Average BMI was only slightly higher for women (31) than men (30) and systolic blood pressure averaged 139mmHg for both genders across all three cohorts. The most frequently encountered comorbid condition was treated hyperglycemia, found more frequently among women (39 percent) than men (33 percent). This was followed by cardiovascular disease, observed more frequently among men (22 percent) than women (16 percent).

The Results

Within the equation derivation cohort of roughly 455,000 subjects, just under 5,000 amputations occurred. This amounted to an amputation rate of 1.34 per 1,000 person-years among women and 2.36 per 1,000 person-years among men. Remarkably, this rate was nearly uniform across all three cohorts. When variables were considered against observed amputations, a final model emerged for women that included age, systolic blood pressure, HbA1c, deprivation, duration of diabetes, smoking status, ethnicity, RA, congestive heart failure, PVD, and chronic renal disease. The same variables appeared in the model among men, with the additions of both type 1 and type 2 diabetes and the presence of atrial fibrillation. Neither BMI nor cholesterol was associated with an increased risk of amputation.

The relationships were what you might expect. Increasing durations of diabetes were associated with an increased risk of amputation, with individuals who were more than ten years removed from their diabetes diagnosis were almost 3.5 times more likely to experience an amputation. Smoking also increased the risk of amputation, with women who smoked heavily at nearly twice the risk for amputation and men who smoked heavily at 1.25 times the risk. Among comorbid conditions, PVD carried the greatest additional risk, with hazard ratios of 4.26 and 3.16 among women and men, respectively. This was followed by chronic renal disease, with hazard ratios of 2.7 and 2.3, respectively.

The hazard ratios associated with age increased, as might be expected, in a roughly linear fashion. The risk of amputation was reduced until about 40 years of age, after which it began to increase, reaching just over 2 and just under 2 for men and women, respectively, by the age of 80. HbA1c values appeared to be quite predictive, with values of 120 associated with a threefold increased risk among both genders. Systolic blood pressure was another strong predictor, with a linear increase beginning at 130mmHg and increasing to a threefold risk for lower-limb amputation at a value of 250mmHg.


The final equations are striking for a number of reasons. First of all, they appear to be fairly accurate. When the validative populations were divided into tenths according to their relative predictive risks for amputation at study entry and observed amputation rates ten years later, the pairing was close.

Additionally, the equations underscore the reality that healthcare is complex. There is no single variable that overwhelms the others in its ability to predict a lower-limb amputation. Rather, amputation risk is a function of numerous overlapping variables and considerations. Using the online predictive tool, a range of scenarios and predictive risks can be calculated.

For example, consider a 54-year-old woman who does not smoke and received a diagnosis of diabetes in the past year. Assuming an HbA1c count of 46 and a systolic blood pressure of 122mmHg, her relative risk of amputation in the next ten years is less than 0.5 percent. Such information could dramatically reduce potential anxiety for a patient who has heard that diabetes can lead to amputation.

By contrast, consider a 68-year-old man who is a moderate smoker and was diagnosed with diabetes five years ago. Assuming a comorbid diagnosis of PVD and ongoing medication for hypertension, with an HbA1c of 64 and a systolic blood pressure of 150mmHg, his risk of lower-limb amputation is just under 10 percent. Given this elevated risk, preventive steps in the form of regular screenings, medications, and lifestyle changes might be reasonably implemented to reduce the risk of eventual amputation.

These scenarios highlight another striking outcome of this herculean effort. As complex as a given individual’s risk for lower-limb amputation may be, with enough data, the modern medical community is beginning to understand the relative relationships and contributions of each of the different variables and considerations.

The Bigger Picture

Looking beyond the immediate implications of diabetes and the associated relative risks of amputation, the process that was used to generate the predictive equations could eventually be used to justify prosthetic care if adequately large datasets could be assembled. Variables such as age, comorbid health conditions, vital capacity, measures of single-leg standing balance, aerobic capacity, muscle strength, and amputation level might be used to predict a patient’s likelihood of attaining community ambulation with a prosthesis or benefiting from certain components.

As the battle over healthcare spending continues, it is important for those in prosthetic rehabilitation to recognize that competing modalities in healthcare are collecting such databases to leverage the value of their services. The more prosthetic rehabilitation professionals can ensure a high likelihood of success with a given intervention for a given patient, the greater the likelihood of reimbursement for its procurement. Within this mindset, the argument for comprehensive evaluations and outcomes-based assessments is not to secure reimbursement on the individual case under consideration, but to ensure continued reimbursement for cases that will be seen in the years ahead.

Phil Stevens, MEd, CPO, FAAOP, is in clinical practice with Hanger Clinic, Salt Lake City. He can be reached at .


  1. Hippesley-Cox, J., and C. Coupland. 2015. Development and validation of risk prediction equations to estimate future risk of blindness and lower limb amputation in patients with diabetes: Cohort study. The BMJ 351:h5441.

Bookmark and Share