Polishing the Crystal Ball: Predicting Prosthetic Usage after Lower-Limb Amputation

Home > Articles > Polishing the Crystal Ball: Predicting Prosthetic Usage after Lower-Limb Amputation
By Phil Stevens, MEd, CPO, FAAOP
hands around crystal ball

In 2009, a research team from Leeds, England, published a systematic literature review reporting on preamputation predictors of walking ability after prosthetic fitting.1 The undertaking was an immense one and reviewed the findings of 57 publications with variable study methodologies and definitions of success with a prosthesis. Ultimately, the authors condensed the observed trends to the following: Predictors of good walking ability include sound cognition and fitness, the ability to stand and balance on the sound side limb, and good preoperative mobility. Predictors of poor walking ability include delays between surgery and rehabilitation and problems with the residual limb, such as protracted limb pain and delayed healing. In general, younger patients with unilateral and distal amputations are also predictive of better walking ability.1

Yet, as the authors declared, "[It] is difficult to accurately predict mobility following rehabilitation with a prosthetic limb."1 This article reviews two recent contributions to this body of evidence. The first attempted to create and validate a set of preprosthetic variables that might predict prosthetic usage at four, eight, and 12 months post discharge from rehabilitation.2 The second attempted to determine the threshold of preprosthetic exercise capacity that might predict successful prosthetic use after transfemoral amputation due to peripheral vascular disease (PVD).3


The first study is based on the simple premise that if you examine a large group of individuals who have lower-limb amputations, some of whom have continued to use their prostheses and others who have not, across a range of variables, some of those variables will be more predictive of disuse than others and may help identify those individuals who are more likely to abandon the use of their prostheses.2 The authors began by identifying a range of potential variables that were categorized into intrinsic predictor variables, amputation predictor variables, and functional predictor variables. Some of these variables are shown in Table 1.

Table 1

For this study, successful prosthetic use was defined as the "use of the prosthesis for locomotor activities (e.g., transfers, standing, and walking) on one or more weekdays." By contrast, disuse was the failure to use a prosthesis for locomotor activities on any days. In the first phase of the study, 135 community-dwelling individuals with a major lower-limb amputation (transtibial or higher) functioning at Medicare Functional Classification Levels K1-K4 were contacted to determine their current and retrospective use of their prostheses. Those who were no longer using a prosthesis were asked to recall when they had stopped using it. Prior interviews, conducted four months after discharge and every two months thereafter for the first year post amputation, provided additional data on prosthetic use.

Evaluating demographic variables between users and nonusers began to illustrate which factors might prove predictive of prosthetic use. Age at the time of amputation, for example, appeared to be a poor predictor, with the mean age of users and nonusers reported at 55.1 and 58.3 years respectively. Gender also appeared relatively consistent between the two groups as male subjects constituted 79 percent and 71 percent of the user and nonuser cohorts respectively.

Other variables began to suggest predictive capacity. For example, type 2 diabetes was present in 46 percent of the nonusers, but only 37 percent of the users. Peripheral arterial disease was present in 61 percent of the nonusers and only 47 percent of the users. Traumatic amputation etiology was observed in 31 percent of the users, compared to 22 percent of the nonusers. Transtibial amputations were observed among 83 percent of the user cohort, but only 61 percent of the nonuser cohort. By contrast, transfemoral amputations were only seen in 21 percent of the users, but 68 percent of the nonusers. Bilateral major lower-limb amputations were seen in 9 percent of the users, compared to 34 percent of the nonusers. This data is summarized in Table 2.

Table 2

Forty potential variables, as well as their correlation to prosthetic nonuse over time, were ultimately examined by the authors. Using the statistical technique of backward-stepwise logistic regression modeling, the most significant predictors of nonuse at four, six, eight, and 12 months were identified. At four and six months post discharge, the five variables predictive of nonuse were:

  • An amputation level proximal to transtibial
  • The use of mobility aids at the time of discharge
  • Dependent outdoor walking on concrete at the time of discharge (i.e., required assistance or unable to perform)
  • Not having a diagnosis of type 2 diabetes
  • 19 or more comorbidities

At eight months post discharge, this list reduced to a proximal amputation level along with the use of mobility aids and dependent outdoor walking at the time of discharge. The predictors of nonuse one year post discharge were a proximal amputation level, mobility use at discharge, and a delay of more than 160 days to prosthesis fitting.


With their prediction rules established, the authors conducted the second phase of the study in which they attempted to validate these rules prospectively on a new cohort of individuals with lower-limb amputations. The study followed 66 subjects, 55 of whom ultimately remained prosthetic users. Attrition from the user's cohort increased over time with eight nonusers at four and six months, ten at eight months, and 11 at one year post discharge.

The earlier prediction rules were generally found to be fairly accurate. For example, the probability of nonuse at four months among the entire cohort was 12 percent. However, this probability leapt to 86 percent among subjects presenting with four of the five predictors at discharge. Similarly, the probability of nonuse at eight months post discharge was 15 percent among the entire cohort, but 86 percent among those presenting with the three predictors of nonuse by this stage of rehabilitation. By one year post discharge, predictive capacity began to wane, with the presence of two of the three predictors increasing the probability of nonuse from 17 percent to 36 percent.

However, to appreciate the clinical utility of these predictive observations, additional insights are required. To be truly valuable, a set of predictions needs to be able to determine who's at risk for nonuse without falsely classifying users as likely nonusers. For example, in the four-month post-discharge window, had the bar been lowered such that subjects need only present with one of the five predictive characteristics, 100 percent of the nonusers would have been identified as likely nonusers at discharge. However, for about 85 percent of those it would have been a false prediction. Requiring the presence of two of the five predictors would have still identified about 90 percent of the nonusers, but incorrectly classified about 30 percent of them. Using a threshold of three of the five predictors would have only caught 50 percent of the nonusers, falsely classifying 14 percent. The threshold of four of the five predictors had no false negatives (no eventual users predicted to be nonusers) but only detected 38 percent of the nonusers. So while the predictive variable set identified some of eventual nonusers, it failed to identify almost two-thirds of them.

A similar pattern was seen during the eight-month post-discharge window. The presence of one of the three variables correctly identified about 90 percent of the users, but incorrectly classified half of the users as nonusers. Requiring two variables decreased the false classification rate to 18 percent but only captured 70 percent of the nonusers. Requiring all three variables eliminated all of the false classifications but only identified 30 percent of the nonusers.

So while a consideration of these predictive rules at the time of discharge may help identify a minority cohort of those individuals at a high risk for nonuse, and do so with a relatively low risk of falsely classifying eventual users as nonusers, the majority of nonusers are unlikely to be predicted by these variables observed at discharge. The crystal ball of prosthetic use remains hazy.


The next article under consideration comes from Slovenia, where the government-funded healthcare system will cover the cost of a transfemoral prosthesis or a wheelchair, but not both. Within that system, there is a need to forecast whether or not an individual patient will be able to use his or her prosthesis as a primary means of mobility. Also, recognizing that the energy cost of ambulation with a transfemoral prosthesis, particularly among individuals with PVD, is quite high and may place these individuals at risk for sudden cardiac events, there is additional need to determine if prosthetic ambulation for daily mobility constitutes a significant risk to the patient. The intent behind this effort was to create a screening method that might predict whether individuals with transfemoral amputations due to PVD would be capable of sustained safe ambulation with prostheses.3

The authors used a convenience sample of the 101 subjects who went through their rehabilitation programs and were ultimately able to complete a six-minute walk test (6MWT) without the need to stop and rest during the test. At the time of admission, each subject performed an exercise test in which he or she was seated in front of a handheld wheel ergometer. As part of the test, ECG electrodes were attached to monitor each patient's well-being. The patients began handcycling at a resistance equivalent to ten watts of power, maintaining a cranking rate of 50 to 55 rotations per minute.

This was sustained for two minutes, followed by a one-minute rest period. At each successive two-minute exercise period, the workload was increased by ten watts up to the preset maximum of 50 watts. This continued until the patient reached his or her submaximal heart rate or when a contraindication for further exercise testing occurred. This workload was recorded for each individual. In addition, oxygen uptake was calculated according to the American College of Sports Medicine formula for arm exercise using each individual's demonstrated exercise power and body mass.

The 6MWT was performed during the final three days before each patient's discharge from rehabilitation. Each individual was asked to walk at a comfortable speed without a resting period. Metabolic parameters (including direct measurements of oxygen uptake), ECG, and heart rate were monitored during the test.

The subjects, with an average age of 70, had all undergone transfemoral amputations due to PVD. The average time from amputation to admission to the rehabilitation program was 74 days, at which time all subjects were ambulant using crutches or a walker. Three-quarters of the subjects presented with high blood pressure and half had comorbid diabetes. Rehabilitation lasted, on average, 44 days.

Performance on the 6MWT ranged from 22 to 203 meters with an average of 90 meters. As might be reasonably expected, walking distance increased with increasing exercise power capacity at admission (Table 3). The shortest average distances were observed among those with a maximum exercise power of 20 watts (55m) and the longest average distances among those who reached the maximum exercise power of 50 watts (130m).

Table 3

However, the relationship of greatest interest to the study authors was the oxygen uptake (VO2 max) experienced by the four subgroups during the arm exercise and the subsequent 6MWT-specifically, the relationship between which handwheel ergometer workload corresponded most closely to the workload experienced during the 6MWT.

Statistical analysis confirmed what was visually evident in simple scatter plots of the VO2 max during both events. Oxygen uptake values were most similar among those participants with a maximum exercise power of 30 watts. In other words, the two tasks of hand-wheel ergometry and the 6MWT test were similarly strenuous for this group. For those who only reached 20 watts, the oxygen uptake during the 6MWT was comparatively higher than during the ergometry task, suggesting that walking with a prosthesis is a strenuous activity for this group. By contrast, for those capable of 40 to 50 watts of exercise power, the oxygen uptake during the 6MWT was much lower, suggesting that for them ambulation with a prosthesis represented a mild to moderate exertion.

The authors concluded that successful prosthetic ambulation as a primary means of mobility is more likely among those individuals who are capable of a workload of 30 watts with an arm ergometer. When this level of fitness is present, prosthetic fitting should be initiated. For those who lack this fitness, prosthetic fitting may be reasonably delayed in favor of additional therapy toward improved general fitness.


Forecasting which patients will successfully transition to sustained prosthetic use remains challenging. Even when 40 potential predictors were documented in the preprosthetic period, the resultant prediction rules were unable to predict the majority of those individuals who became nonusers in the first year after discharge from rehabilitation programs.2 The concept of exercise screening to establish whether an individual has the cardiac capacity for prosthetic ambulation has its merits, yet may exclude those individuals who are initially only capable of limited walking but may progress to more sustained daily use. However, both strategies allow for the identification of those at risk. Ultimately, the role of this crystal ball may not be to determine prosthetic candidacy, but to provide insight into which patients may benefit from additional interventions for a successful outcome.

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


  1. Sansam, K., V. Neuman, R. O'Connor, and B. Bhakta. 2009. Predicting walking ability following lower limb amputation: A systematic review of the literature. Journal of Rehabilitation Medicine 41 (8):593-603.
  2. Roffman, C. E., J. Buchanan, and G. T. Allison. 2014. Predictors of nonuse of prostheses by people with lower limb amputation after discharge from rehabilitation: Development and validation of clinical prediction rules. Journal of Physiotherapy 60 (4):224-31.
  3. Erjavec, T., G. Vidmar, and H. Burger. 2014. Exercise testing as a screening measure for ability to walk with a prosthesis after transfemoral amputation due to peripheral vascular disease. Disability and Rehabilitation 36 (14):1148-55.