Abstract
Objective:
This study aimed to create a novel prediction model (AMPREDICT MoRe) that predicts death and re-amputation after dysvascular amputation, which overcomes prior implementation barriers by using only predictors that are readily available in the electronic health record (EHR).
Methods:
This was a retrospective cohort study of 9 221 patients with incident unilateral transmetatarsal, transtibial, or transfemoral amputation secondary to diabetes and or peripheral arterial disease identified in the Veterans Affairs Corporate Data Warehouse between 1 October 2015 and 30 September 2021. The prediction model evaluated factors falling into several key domains: prior revascularisation; amputation level; demographics; comorbidities; mental health; health behaviours; laboratory values; and medications. The primary outcome included four categories: (i) no death and no re-amputation (ND/NR); (ii) no death and re-amputation (ND/R); (iii) death and no re-amputation (D/NR); and (iv) death and re-amputation (D/R). Multinomial logistic regression was used to fit one year post-incident amputation risk prediction models. Variable selection was performed using LASSO (least absolute shrinkage and selection operator), a machine learning methodology. Model development was performed using a randomly selected 80% of the data, and the final model was externally validated using the remaining 20% of subjects.
Results:
The final prediction model included 23 predictors. The following outcome distribution was observed in the development sample: ND/NR, n = 4 254 (57.7%); ND/R, n = 1 690 (22.9%); D/NR, n = 1 056 (14.3%); and D/R, n = 376 (5.1%). The overall discrimination of the model was moderately strong (M index 0.70), but a deeper look at the c indices indicated that the model had better ability to predict death than re-amputation (ND/NR vs. ND/R, 0.64; ND/NR vs. D/NR, 0.78; grouped ND vs. D, 0.79 and NR vs. R, 0.67). The model was best at distinguishing individuals with no negative outcomes vs. both negative outcomes (ND/NR vs. D/R, 0.82).
Conclusion:
The AMPREDICT MoRe model has been successfully developed and validated, and can be applied at the time of amputation level decision making. Since all predictors are available in the EHR, a future decision support tool will not require patient interview.
Keywords: Amputation, Death, Dysvascular, Mortality, Prediction, Re-amputation
INTRODUCTION
The absolute number of amputations performed in the United States Department of Veterans Affairs (VA) increased by more than 40% from 2008 to 2018,1 and the overall healthcare burden of amputation is projected to rise with the increasing prevalence of diabetes and peripheral arterial disease (PAD).2 Amputation has a profound impact on mobility3,4 and quality of life.5 Moreover, 48% of individuals undergoing diabetes related amputations will be re-admitted to hospital in the subsequent six month period, predominantly for residual limb complications and re-amputation.6 An initial more proximal amputation may mitigate these re-amputation risks, but higher level amputation is associated with increased risk of mobility loss, with adverse effects on social participation7 and quality of life.8 Therefore, amputation level choice must balance these tradeoffs.9
Multiple factors are potentially relevant for this decision, including clinical and laboratory assessments, limb perfusion, revascularisation opportunities, comorbidities, and baseline mobility status. Prediction models and decision support tools are being used increasingly to assist providers make complex clinical decisions. The previously developed AMPREDICT mortality,10 re-amputation,11 and mobility12 models predict one year outcomes at each lower extremity major amputation level, in addition to the transmetatarsal amputation level. These prediction models have subsequently been translated into the AMPREDICT Decision Support Tool for point of care access.13
While this decision support tool has been used internationally (https://www.ampdecide.org), the original AMPREDICT models included predictor variables that are not readily available in the electronic health record (EHR) and required manual entry, making implementation more challenging. The aim of the current study was to create a novel prediction model (AMPREDICT MoRe) that predicts four death and re-amputation combinations and overcomes prior implementation barriers by using only predictors that are readily available in the EHR.
METHODS
Study design
This was a retrospective cohort study of patients who underwent an incident lower extremity amputation at the transmetatarsal (TM), transtibial (TT), or transfemoral (TF) level.
Data source
The VA Corporate Data Warehouse (CDW) is a repository comprising data from multiple VA clinical and administrative systems, including inpatient and outpatient data as well as demographic information. The CDW was used to identify patients aged ≥ 30 years undergoing their first diabetes and or PAD related amputation at the TM, TT, or TF level between 1 October 2015 and 30 September 2021, as determined by Current Procedural Terminology (CPT) and International Classification of Diseases, Ninth and Tenth Revision (ICD-9 and ICD-10) procedure codes. Patients were excluded if they had had a prior amputation at the TM level or higher, were undergoing a bilateral amputation, or an amputation due to trauma. Patients were also excluded if they had a diagnosis of paraplegia, quadriplegia, spinal cord injury, or dementia, or a body mass index (BMI) of < 15 or > 52 kg/m2. Local ethics board approval was received for this study.
Defining incident amputation level
Once a qualifying amputation code was identified, a five year look back was performed to identify any prior amputation at TM level or higher. The presence of any diagnostic or procedure code related to prior amputation or its treatment resulted in exclusion of those subjects using similar methods to previous studies.10,11 For guillotine procedures at the TT and TF level, it was assumed that a closure procedure would be performed within three weeks; therefore, a search forward three weeks for the next procedure code was performed to classify the incident level. If the next subsequent procedure was more than three weeks after the initial guillotine procedure or there was no subsequent procedure, an error in the initial coding was presumed and the guillotine code was recoded as a definitive amputation. If no subsequent codes were identified, ankle disarticulation procedures were classified as TT amputations and knee disarticulations were classified as TF amputations.
Classifying laterality for incident and subsequent amputations
Laterality was determined by ICD-10 procedure codes or CPT code modifiers and was classified as either left, right, bilateral, or unknown based on all procedures recorded on the same day. If there were both left and right procedures recorded on the same day, the procedure was classified as bilateral. Laterality was considered unknown only if none of the recorded procedures on the same day had laterality.
Candidate predictor variables
The prediction model evaluated factors available within the EHR that fell into several key domains: prior revascularisation; amputation level; demographics; comorbidities; mental health; health behaviours; laboratory values; and medications (Supplementary Table S1). The selection of candidate predictors was based on prior literature evidence and clinical expert opinion.10–12,14–29 All predictors were identified through the CDW using corresponding ICD-9 and −10 codes and CPT codes, where appropriate, and Structured Query Language (SQL) for labs and medications. All 35 candidate predictors preceded the incident amputation, which was considered time zero for the prediction model.
Outcomes
The primary outcomes of death and re-amputation were combined into the following four categories: (i) no death and no re-amputation (ND/NR); (ii) no death and re-amputation (ND/R); (iii) death and no re-amputation (D/NR); and (iv) death and re-amputation (D/R). Both death and re-amputation were counted if they occurred within twelve months of the incident amputation. The CDW’s Vital Status File was used to determine the date of death among those who died.30 Re-amputation was identified by an amputation procedure code that occurred 1 – 365 days after the definitive incident amputation (or 22 – 365 days after guillotine or disarticulation procedures to allow for revisions within three weeks) and was ipsilateral to the incident procedure, had unknown laterality, or was bilateral at the same or higher level than the incident amputation. Three categories of re-amputation were identified and were all included in the definition of re-amputation: soft tissue revisions for TT and TF; re-amputations at the same level for TT and TF; and definitive amputation at a higher level (e.g., TM to TT, or TT to TF). Because TM amputations do not have codes for soft tissue revision or re-amputation at the same level, all TM codes after an incident TM amputation were coded as a subsequent TM amputation and classified as a re-amputation at the same level. It is understood that re-amputation and death could only be counted if the re-amputation preceded death.
Model development and validation
To explore general patterns in the data, the distributions of candidate predictors were assessed following data cleaning. Categorical variables were tabulated, while continuous predictors were summarised with quartiles and measures of central tendency. Each predictor was also evaluated in a bivariable manner with the four category death and re-amputation outcomes.
Multinomial logistic regression was used to fit one year post-incident amputation risk prediction models. Interactions of dialysis, diabetes (any), PAD, peripheral neuropathy, retinopathy, proteinuria, microvascular disease (combination of retinopathy, proteinuria, and peripheral neuropathy), and ipsilateral revascularisation with all amputation levels and age with only TF amputation level were considered, as the level of incident amputation may have different effects on the risks of death and re-amputation.
Variable selection was performed using LASSO (least absolute shrinkage and selection operator), a machine learning methodology that identifies models which optimise future predictive performance. This leads to more parsimonious models, tuned to reduce prediction error by minimising overfitting.31 Tenfold cross validation was used to determine the optimal tuning parameter using a standard criterion, which proposes the sparsest model with prediction error within one standard deviation of the minimum.32 Further, the more flexible ungrouped LASSO model was chosen to allow different sets of variables to be leveraged for predicting death vs. re-amputation if necessary.
Data were missing in < 5% in a limited number of variables. Multiple imputation by chained equations (MICE) was implemented to handle these missing data. Variable selection using LASSO was performed concurrently with multiple imputation by stacking the imputed data sets and weighting each observation by its proportion of non-missing variables.33 Additionally, to allow for a variety of possible non-linear relationships with the outcome, fractional polynomials were included for continuous predictors, age, BMI, creatinine, albumin, blood urea nitrogen, platelets, leukocytes, haematocrit, and haemoglobin.34 Model development was performed using a randomly selected 80% of the data, and the final model selected by LASSO was externally validated using the remaining 20%. To assess the predictive ability of the model, a discrimination plot was evaluated in parallel with various performance measures including pairwise c indices for each pair of outcome categories, grouped c indices for death and re-amputation alone, and the polytomous discrimination index (PDI), which compares sets instead of pairs of outcomes.35 The PDI for random performance with a four category outcome is 0.25, compared with 0.5 for c indices; therefore, lower indices are common compared with c indices commonly reported from binary outcome prediction models since the scale is wider. The M index, which averages all pairwise c indices, is also provided.
All statistical analyses were carried out using R statistical software (R Foundation for Statistical Computing, Vienna, Austria) with LASSO multinomial regression performed using the glmnet package.31,36
Model application
Two hypothetical patient vignettes are presented that could represent real patients based on one of the author’s thirty years of clinical experience in the pre-operative evaluation of patients requiring amputation. One patient is undergoing amputation secondary to long standing type II diabetes and its associated comorbidities; and the other with PAD and extensive smoking history, with its associated comorbidities.
RESULTS
Summary of sample
Among the consecutive 14 143 persons with an amputation identified because of PAD and or diabetes, 4 922 (34.8%) were excluded, leaving 9 221 incident unilateral amputations eligible for the analysis, including 3 116 (33.8%) TM, 4 031 (43.7%) TT, and 2 074 (22.5%) TF (Fig. 1; Table 1). The distribution of outcomes and candidate predictors by development (n = 7 376) and validation (n = 1 845) samples are given in Table 2. The distribution of candidate predictors was similar between the development and validation samples. Statistically significant differences (p < .050) were observed for sex and opioid use, although the differences were relatively small and probably not clinically significant.
Figure 1.
Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) flowchart depicting total number of patients with an amputation because of peripheral arterial disease and or diabetes acquired by the Veterans Affairs Corporate Data Warehouse and total number excluded to achieve final prediction model cohort. BMI = body mass index.
Table 1.
Distribution of candidate predictors by death and re-amputation category for eligible subjects (n = 9 221).
| Variable | No death/no re-amputation (n = 5 271) | No death/re-amputation (n = 2 134) | Death/no re-amputation (n = 1 352) | Death/re-amputation (n = 464) |
|---|---|---|---|---|
| Amputation level | ||||
| Transmetatarsal | 1 643 (31.2) | 997 (46.7) | 276 (20.4) | 200 (43.1) |
| Transtibial | 2 346 (44.5) | 920 (43.1) | 576 (42.6) | 189 (40.7) |
| Transfemoral | 1 282 (24.3) | 217 (10.2) | 500 (37.0) | 75 (16.2) |
| Demographics | ||||
| Age – y* | 67.0 (11.0) | 67.0 (11.0) | 71.0 (10.0) | 70.0 (12.0) |
| Sex | ||||
| Male | 5 173 (98.1) | 2 099 (98.4) | 1 339 (99.0) | 455 (98.1) |
| Female | 98 (1.9) | 35 (1.6) | 13 (1.0) | 9 (1.9) |
| Ethnicity | ||||
| White, non-Hispanic | 3 258 (64.9) | 1 256 (61.2) | 863 (67.2) | 267 (59.9) |
| Black | 1 299 (25.9) | 605 (29.5) | 302 (23.5) | 132 (29.6) |
| Hispanic | 342 (6.8) | 143 (7.0) | 89 (6.9) | 36 (8.1) |
| Other | 124 (2.5) | 47 (2.3) | 31 (2.4) | 11 (2.5) |
| Marital status | ||||
| Married | 2 052 (40.7) | 825 (40.6) | 537 (42.0) | 177 (39.9) |
| Single/never married | 686 (13.6) | 297 (14.6) | 160 (12.5) | 53 (11.9) |
| Separated/divorced | 1 875 (37.2) | 754 (37.1) | 442 (34.6) | 182 (41.0) |
| Widowed | 432 (8.6) | 158 (7.8) | 139 (10.9) | 32 (7.2) |
| Urban/rural classification | ||||
| Rural | 1 669 (31.7) | 633 (29.7) | 424 (31.4) | 134 (28.9) |
| Urban | 3 600 (68.3) | 1 501 (70.3) | 928 (68.6) | 330 (71.1) |
| Geographic region | ||||
| Continental | 911 (17.3) | 388 (18.2) | 229 (16.9) | 90 (19.4) |
| Midwest | 1 084 (20.6) | 417 (19.5) | 285 (21.1) | 78 (16.8) |
| North Atlantic | 1 137 (21.6) | 461 (21.6) | 289 (21.4) | 124 (26.7) |
| Pacific | 974 (18.5) | 418 (19.6) | 219 (16.2) | 84 (18.1) |
| Southeast | 1 165 (22.1) | 450 (21.1) | 330 (24.4) | 88 (19.0) |
| BMI – kg/m2 | 28.2 (8.7) | 27.5 (8.3) | 26.1 (9.2) | 26.7 (8.1) |
| Comorbidities | ||||
| Asthma | 462 (8.8) | 161 (7.5) | 116 (8.6) | 50 (10.8) |
| Kidney dialysis | 464 (8.8) | 264 (12.4) | 337 (24.9) | 143 (30.8) |
| Coronary atherosclerosis | 2 588 (49.1) | 1 147 (53.7) | 908 (67.2) | 319 (68.8) |
| Diabetes | 4 682 (88.8) | 1 875 (87.9) | 1 165 (86.2) | 405 (87.3) |
| Peripheral arterial disease | 2 674 (50.7) | 1 261 (59.1) | 841 (62.2) | 307 (66.2) |
| COPD | 1 967 (37.3) | 792 (37.1) | 687 (50.8) | 233 (50.2) |
| Chronic liver disease | 303 (5.7) | 135 (6.3) | 127 (9.4) | 40 (8.6) |
| Heart failure | 1 919 (36.4) | 835 (39.1) | 861 (63.7) | 299 (64.4) |
| Myocardial infarction | 232 (4.4) | 137 (6.4) | 148 (10.9) | 53 (11.4) |
| Peripheral neuropathy | 3 378 (64.1) | 1 333 (62.5) | 801 (59.2) | 283 (61.0) |
| Stroke | 1 028 (19.5) | 479 (22.4) | 413 (30.5) | 133 (28.7) |
| Retinopathy | 2 838 (53.8) | 1 217 (57.0) | 760 (56.2) | 283 (61.0) |
| Proteinuria | 764 (14.5) | 340 (15.9) | 270 (20.0) | 88 (19.0) |
| Microvascular disease | 488 (9.3) | 221 (10.4) | 155 (11.5) | 49 (10.6) |
| Mental health | ||||
| Anxiety | 1 622 (30.8) | 605 (28.4) | 380 (28.1) | 120 (25.9) |
| Depression | ||||
| No depression | 2 648 (50.2) | 1 126 (52.8) | 757 (56.0) | 247 (53.2) |
| Depressive disorder/dysthymia | 1 019 (19.3) | 403 (18.9) | 259 (19.2) | 88 (19.0) |
| Major depressive disorder | 1 604 (30.4) | 605 (28.4) | 336 (24.9) | 129 (27.8) |
| Mild cognitive impairment | 212 (4.0) | 57 (2.7) | 74 (5.5) | 27 (5.8) |
| Health behaviours | ||||
| AUDIT-C category† | ||||
| Mild | 4 766 (91.0) | 1 937 (92.0) | 1 231 (92.0) | 436 (94.2) |
| Moderate | 355 (6.8) | 123 (5.8) | 74 (5.5) | 24 (5.2) |
| Severe | 118 (2.3) | 46 (2.2) | 33 (2.5) | 3 (0.6) |
| Smoking status | ||||
| Never | 1 896 (36.8) | 733 (35.7) | 484 (36.9) | 170 (37.6) |
| Former | 1 522 (29.6) | 601 (29.2) | 418 (31.9) | 152 (33.6) |
| Current | 1 730 (33.6) | 722 (35.1) | 410 (31.2) | 130 (28.8) |
| Opioid use disorder | 403 (7.6) | 135 (6.3) | 78 (5.8) | 35 (7.5) |
| Laboratory values | ||||
| Creatinine – mg/dL | 1.1 (0.7) | 1.2 (0.9) | 1.5 (2.5) | 1.7 (3.7) |
| Albumin – g/dL | 3.2 (1.1) | 3.1 (1.0) | 2.8 (1.0) | 2.9 (1.0) |
| Blood urea nitrogen – mg/dL | 19.0 (15.0) | 19.0 (16.0) | 27.0 (25.0) | 28.0 (27.0) |
| Platelets – 1 000/μL | 284 (146) | 289 (153) | 260 (144) | 261 (148) |
| Leukocytes – 1 000/μL | 9.2 (4.8) | 10.0 (5.4) | 10.0 (5.6) | 10.6 (6.2) |
| Haematocrit – % | 32.9 (9.1) | 31.9 (9.0) | 30.3 (8.0) | 30.4 (8.6) |
| Haemoglobin – g/dL | 10.7 (3.1) | 10.4 (3.1) | 9.8 (2.7) | 9.7 (2.6) |
| Anticoagulants | ||||
| Dabigatran | 65 (1.2) | 19 (0.9) | 10 (0.7) | 2 (0.4) |
| Low molecular weight heparin | 180 (3.4) | 75 (3.5) | 48 (3.6) | 12 (2.6) |
| Factor Xa inhibitors | 400 (7.6) | 176 (8.2) | 133 (9.8) | 45 (9.7) |
| Warfarin | 431 (8.2) | 235 (11.0) | 181 (13.4) | 59 (12.7) |
| Antiplatelet agents | 1 816 (34.5) | 792 (37.1) | 521 (38.5) | 186 (40.1) |
| Revascularisation | ||||
| Any ipsilateral revascularisation | 1 654 (31.4) | 863 (40.4) | 497 (36.8) | 192 (41.4) |
Data are presented as n (%) or mean (standard deviation). BMI = body mass index; COPD = chronic obstructive pulmonary disease; AUDIT-C = Alcohol Use Disorders Identification Test–Consumption.
Unless otherwise specified, all diagnoses represent any diagnosis prior to the date of incident amputation.
At the time of incident amputation.
Most recent prior to the date of incident amputation (mild: women, < 3, men, < 4; moderate: women, ≥ 3 and < 8, men, ≥ 4 and < 8; severe: ≥ 8 for both men and women). When multiple scores were recorded on the same date and time, the higher score was used.
Table 2.
Outcome and candidate predictors for AMPREDICT MoRE one year death and re-amputation full prediction model, stratified by development (n = 7 376) and validation (n = 1 845) samples.
| Variable | Development sample (n = 7 376) | Validation sample (n = 1 845) | Total (n = 9221) |
|---|---|---|---|
| Outcome | |||
| No death/no re-amputation | 4 254 (57.7) | 1 017 (55.1) | 5 271 (57.2) |
| No death/re-amputation | 1 690 (22.9) | 444 (24.1) | 2 134 (23.1) |
| Death/no re-amputation | 1 056 (14.3) | 296 (16.0) | 1 352 (14.7) |
| Death/re-amputation | 376 (5.1) | 88 (4.8) | 464 (5.0) |
| Amputation level | |||
| Transmetatarsal | 2 496 (33.8) | 620 (33.6) | 3 116 (33.8) |
| Transtibial | 3 214 (43.6) | 817 (44.3) | 4 031 (43.7) |
| Transfemoral | 1 666 (22.6) | 408 (22.1) | 2 074 (22.5) |
| Demographics | |||
| Age – y* | 68.0 (10.0) | 67.0 (10.0) | 68.0 (10.0) |
| Sex | |||
| Male | 7 264 (98.5) | 1 802 (97.7) | 9 066 (98.3) |
| Female | 112 (1.5) | 43 (2.3) | 155 (1.7) |
| Ethnicity | |||
| White, non-Hispanic | 4 518 (64.2) | 1 126 (63.8) | 5 644 (64.1) |
| Black | 1 869 (26.5) | 469 (26.6) | 2 338 (26.6) |
| Hispanic | 482 (6.8) | 128 (7.3) | 610 (6.9) |
| Other | 171 (2.4) | 42 (2.4) | 213 (2.4) |
| Marital status | |||
| Married | 2 875 (40.9) | 716 (40.5) | 3 591 (40.8) |
| Single/never married | 939 (13.4) | 257 (14.5) | 1 196 (13.6) |
| Separated/divorced | 2 612 (37.1) | 641 (36.3) | 3 252 (37.0) |
| Widowed | 607 (8.6) | 154 (8.7) | 761 (8.6) |
| Urban/rural classification | |||
| Rural | 2 283 (31.0) | 577 (31.3) | 2 860 (31.0) |
| Urban | 5 091 (69.0) | 1 268 (68.7) | 6 359 (69.0) |
| Geographic region | |||
| Continental | 1 279 (17.3) | 339 (18.4) | 1 618 (17.5) |
| Midwest | 1 489 (20.2) | 375 (20.3) | 1 864 (20.2) |
| North Atlantic | 1 627 (22.1) | 384 (20.8) | 2 011 (21.8) |
| Pacific | 1 370 (18.6) | 325 (17.6) | 1 695 (18.4) |
| Southeast | 1611 (21.8) | 422 (22.9) | 2 033 (22.0) |
| BMI – kg/m2 | 27.7 (8.7) | 27.5 (8.6) | 27.7 (8.7) |
| Comorbidities | |||
| Asthma | 634 (8.6) | 155 (8.4) | 789 (8.6) |
| Kidney dialysis | 945 (12.8) | 263 (14.3) | 1 208 (13.1) |
| Coronary atherosclerosis | 3 952 (53.6) | 1 010 (54.7) | 4 962 (53.8) |
| Diabetes | 6 505 (88.2) | 1 622 (87.9) | 8 127 (88.1) |
| Peripheral arterial disease | 4 059 (55.0) | 1 024 (55.5) | 5 083 (55.1) |
| COPD | 2 912 (39.5) | 767 (41.6) | 3 679 (39.9) |
| Chronic liver disease | 484 (6.6) | 121 (6.6) | 605 (6.6) |
| Heart failure | 3 103 (42.1) | 811 (44.0) | 3 914 (42.4) |
| Myocardial infarction | 446 (6.0) | 124 (6.7) | 570 (6.2) |
| Peripheral neuropathy | 4 632 (62.8) | 1 163 (63.0) | 5 795 (62.8) |
| Stroke | 1 643 (22.3) | 410 (22.2) | 2 053 (22.3) |
| Retinopathy | 4 084 (55.4) | 1 014 (55.0) | 5 098 (55.3) |
| Proteinuria | 1 153 (15.6) | 309 (16.7) | 1 462 (15.9) |
| Microvascular disease | 723 (9.8) | 190 (10.3) | 913 (9.9) |
| Mental health | |||
| Anxiety | 2 165 (29.4) | 562 (30.5) | 2 727 (29.6) |
| Depression | |||
| No depression | 3 844 (52.1) | 934 (50.6) | 4 778 (51.8) |
| Depressive disorder/dysthymia | 1 402 (19.0) | 367 (19.9) | 1 769 (19.2) |
| Major depressive disorder | 2 130 (28.9) | 544 (29.5) | 2 674 (29.0) |
| Mild cognitive impairment | 299 (4.1) | 71 (3.8) | 370 (4.0) |
| Health behaviours | |||
| AUDIT-C category† | |||
| Mild | 6 694 (91.5) | 1 676 (91.7) | 8 370 (91.5) |
| Moderate | 466 (6.4) | 110 (6.0) | 576 (6.3) |
| Severe | 159 (2.2) | 41 (2.2) | 200 (2.2) |
| Smoking status | |||
| Never | 2 636 (36.7) | 647 (36.1) | 3 283 (36.6) |
| Former | 2 152 (30.0) | 541 (30.2) | 2 693 (30.0) |
| Current | 2 389 (33.3) | 603 (33.7) | 2 992 (33.4) |
| Opioid use disorder | 547 (7.4) | 104 (5.6) | 651 (7.1) |
| Laboratory values | |||
| Creatinine – mg/dL | 1.2 (0.9) | 1.2 (1.0) | 1.2 (1.0) |
| Albumin – g/dL | 3.1 (1.0) | 3.1 (1.0) | 3.1 (1.0) |
| Blood urea nitrogen – mg/dL | 20.0 (17.0) | 20.0 (18.0) | 20.0 (17.0) |
| Platelets – 1 000/μL | 280 (149) | 279 (148) | 280 (148) |
| Leukocytes – 1 000/μL | 9.6 (5.3) | 9.4 (4.8) | 9.6 (5.2) |
| Haematocrit – % | 32.1 (8.9) | 32.1 (9.1) | 32.1 (8.9) |
| Haemoglobin – g/dL | 10.5 (3.0) | 10.5 (3.2) | 10.5 (3.0) |
| Anticoagulants | |||
| Dabigatran | 78 (1.1) | 18 (1.0) | 96 (1.0) |
| Low molecular weight heparin | 263 (3.6) | 52 (2.8) | 315 (3.4) |
| Factor Xa inhibitors | 621 (8.4) | 133 (7.2) | 754 (8.2) |
| Warfarin | 744 (10.1) | 162 (8.8) | 906 (9.8) |
| Antiplatelet agents | 2 682 (36.4) | 633 (34.3) | 3 315 (36.0) |
| Revascularisation | |||
| Any ipsilateral revascularisation | 2 590 (35.1) | 616 (33.4) | 3 206 (34.8) |
Data are presented as n (%) or mean (standard deviation). BMI = body mass index; COPD = chronic obstructive pulmonary disease; AUDIT-C = Alcohol Use Disorders Identification Test–Consumption.
Unless otherwise specified, all diagnoses represent any diagnosis prior to the date of incident amputation.
At the time of incident amputation.
Most recent prior to the date of incident amputation (mild: women, < 3, men, < 4; moderate: women, ≥ 3 and < 8, men, ≥ 4 and < 8; severe: ≥ 8 for both men and women). When multiple scores were recorded on the same date and time, the higher score was used.
Summary of outcomes distribution
In the development sample, the following outcome distribution was observed: ND/NR, n = 4 254 (57.7%); ND/R, n = 1 690 (22.9%); D/NR, n = 1 056 (14.3%); and D/R, n = 376 (5.1%) (Table 1). The outcome distributions were similar in both the development and validation samples.
Risk prediction model development
The final prediction model included 23 predictors. Estimated coefficients of predictors for each four category death and re-amputation outcome level are provided in Supplementary Table S2. Since the outcome has more than two levels, the individual coefficients do not simply represent log odds ratios as they would for binary logistic regression, and their interpretability is more difficult, particularly for quantitative predictors that are non-linearly modelled. Generally, the more positive the coefficients, the greater the associated probability is for those specific outcome levels; and the more negative they are, the lower the associated probability. The continuous predictors that are modelled non-linearly (age, BMI, platelets, leukocytes, haemoglobin, blood urea nitrogen, and albumin) are illustrated graphically in Supplementary Figure S1.
Model validation
The discrimination summaries computed from the validation set are shown in Table 3. The overall discrimination is moderately strong (M index 0.70) and shows a 44% chance of correctly identifying a patient’s true outcome among a randomly selected set of four patients from each outcome category (PDI 0.44), but a deeper look at the c indices indicates that the model has better ability to predict death than re-amputation (ND/NR vs. ND/R, 0.64; ND/NR vs. D/NR, 0.78; grouped ND vs. D, 0.79 and NR vs. R, 0.67). Furthermore, the model was best at distinguishing individuals from each end of the outcome spectrum, i.e., no negative outcomes vs. both negative outcomes (ND/NR vs. D/R, 0.82).
Table 3.
Quantitative discrimination measures for the final model.
| Discrimination measure | Model estimate* |
|---|---|
| Polytomous discrimination index | 0.44 |
| M index | 0.70 |
| Pairwise c indices | |
| No death/no re-amputation vs. no death/re-amputation | 0.64 |
| No death/no re-amputation vs. death/no re-amputation | 0.78 |
| No death/no re-amputation vs. death/re-amputation | 0.82 |
| No death/re-amputation vs. death/no re-amputation | 0.74 |
| No death/re-amputation vs. death/re-amputation | 0.62 |
| Death/no re-amputation vs. death/re-amputation | 0.58 |
| Grouped c indices | |
| No death vs. death | 0.79 |
| No re-amputation vs. re-amputation | 0.67 |
Model estimates are calculated from the validation sample.
The discrimination plot for the final model is shown in Figure 2. The modelled predicted probabilities for each of the four outcome levels, stratified by the observed outcome, are shown as grouped boxplots. If the boxplot of the observed outcome has the highest probability in the corresponding predicted outcome category, then the model is properly discriminating. The greatest separation in boxplots, and therefore the greatest discriminative ability, is shown for individuals with neither death nor re-amputation relative to both death and re-amputation.
Figure 2.
Discrimination plot of predictions for the final model. Distributions of predicted probabilities for each combined death and re-amputation outcome category are stratified by the actual observed category. Red lines show the prevalence of each outcome category. ND = no death; NR = no re-amputation; R = re-amputation; D = death.
Calibration slopes were computed using a logistic regression on the validation set and were found to be > 1.0, indicating model underfitting (ND vs. D, 1.67 vs. NR vs. R, 1.36). This indicates that patients with extremely high risk of death or re-amputation have underestimated predictions.
Model application
The AMPREDICT MoRe model can be used at the time a patient has been identified who requires lower extremity amputation for complications of PAD and or diabetes. Table 4 describes two hypothetical patients and their clinical backgrounds at the time of amputation. In patient #1, because of the extent of disease in the foot, a TM amputation is not an option. If he undergoes a TT amputation, the predicted mortality risk is significant but much lower than a TF amputation (27% vs. 42%). In contrast, a TT amputation has a greater re-amputation risk (29% vs. 18%). Patient #2, by contrast, exhibits the manifestations of chronic arterial occlusive disease in the absence of diabetes. He has had many revascularisation procedures. He shares the many comorbidities of patients with long term smoking and abnormal serum lipids, including cardiovascular disease, cerebrovascular disease, and chronic obstructive pulmonary disease. Like the first patient, the only amputation level options available are TT or TF. The one year mortality risks are significant, but essentially the same at either amputation level (21% and 24%). The risks of failure of healing and requiring re-amputation are 36% at the TT level compared with 22% at the TF level. These risks are also displayed in Figure 3.
Table 4.
Example clinical scenarios for two hypothetical patients requiring amputation.
| Predictor | Patient #1 | Patient #2 |
|---|---|---|
| Age at time of amputation – y | 70 | 70 |
| BMI – kg/m2 | 30 | 22 |
| Marital status | Married | Married |
| Heart failure | No | Yes |
| Kidney dialysis | Yes | No |
| Diabetes | Yes | No |
| Coronary atherosclerosis | No | Yes |
| Peripheral arterial disease | No | Yes |
| Chronic liver disease | No | No |
| Myocardial infarction, within last 6 mo | No | Yes |
| Peripheral neuropathy | Yes | No |
| Stroke | No | Yes |
| Retinopathy | Yes | No |
| Blood urea nitrogen – mg/dL | 35 | 10 |
| Platelets – 1 000/μL | 220 | 381 |
| Leukocytes – 1 000/μL | 16.3 | 7.5 |
| Proteinuria | Yes | No |
| Haemoglobin – g/dL | 10.9 | 13 |
| Albumin – g/dL | 1.9 | 4.2 |
| Warfarin | No | Yes |
| Ipsilateral revascularisation | No | Yes |
| Mild cognitive impairment | No | No |
| Risk of outcomes – % | ||
| No death/no re-amputation | ||
| Transmetatarsal | 43 | 45 |
| Transtibial | 51 | 49 |
| Transfemoral | 46 | 60 |
| No death/re-amputation | ||
| Transmetatarsal | 34 | 34 |
| Transtibial | 22 | 30 |
| Transfemoral | 12 | 15 |
| Death/no re-amputation | ||
| Transmetatarsal | 15 | 14 |
| Transtibial | 20 | 14 |
| Transfemoral | 35 | 17 |
| Death/re-amputation | ||
| Transmetatarsal | 7 | 7 |
| Transtibial | 7 | 7 |
| Transfemoral | 8 | 7 |
BMI = body mass index.
Figure 3.
Predicted risk of hypothetical clinical scenarios: (A) patient 1; and (B) patient 2. TF = transfemoral; TT = transtibial; TM = transmetatarsal.
DISCUSSION
The primary purpose of this study was to develop and validate a patient specific multivariable model to predict a four category outcome of twelve month risk of re-amputation and or death for individuals undergoing their first unilateral amputation secondary to diabetes and or PAD. The overall performance of the model was moderately strong for a four category outcome (M index 0.70 in the validation sample). It discriminated best among those who did and did not die within the first year after amputation. The model was less effective at discriminating among those who did and did not require a re-amputation. This led to more modest performance (pairwise c index 0.67); however, this remains clinically meaningful and can be used to support clinical decision making regarding the most appropriate amputation level for a patient or the decision not to amputate in favour of palliative care. The more limited performance characteristics for re-amputation prediction may be partly explained by the competing risk of death and re-amputation.
The global vascular guidelines emphasise the importance of the amputation level decision as well as its reliance on clinical experience owing to the absence of any diagnostic tests that can reliably predict outcome.37 Amputation level decisions can impact quality of life as well as healthcare costs.38,39 TM amputations are associated with 60% readmission rates, longer lengths of hospital stay, and increased incidence of stroke.40 They also have up to a 47% non-healing rate, with 35% requiring higher level amputation, both of which adversely affect healthcare utilisation and patient functional mobility.41 Rates of TM amputation have been increasing despite these risks.11,42 Although the motivation behind these decisions is unclear, the literature suggests that patient prioritisation of preserving as much limb length as possible and preservation of mobility6,37,41 are important factors. There are clear tradeoffs between each amputation level. Balancing the risks of healing failure, the need for re-amputation, and potential mobility in a population with a high one year mortality rate is extremely challenging.43
A recent systematic review evaluating the accuracy of surgeon clinical outcome prediction in a variety of surgical conditions concluded that surgeons are generally good at predicting peri-operative risk, but poorer at predicting longer term outcomes. This review recommended the use of decision support tools to augment clinical decision making.44 More specifically, in the context of amputation outcome prediction, surgeons overpredicted death and amputation revision in the first month post-amputation.45 In addition to the challenges surgeons face in predicting outcome, many patients undergoing vascular surgery procedures46 and amputation surgery38,47 do not feel that they are offered the opportunity to participate in emotionally difficult amputation level decisions and desire greater involvement.48
As computing power increases, the ability to use advanced machine learning techniques to identify and quantify personalised risk through prediction models has improved.49 However, implementation of these models requires careful thought and planning to include experienced teams with diverse scientific, clinical, and technical backgrounds to ensure maximum accuracy and utility as well as efficiency.13,50–53
The model described here has several strengths as well as potential limitations. A different dataset than the original AMPREDICT mortality and re-amputation models was used. The former models used data from the VA Surgical Quality Improvement Program (VASQIP) database, which is not directly linked to the EHR and contains some variables that require patient interview and potentially a physical assessment. Like the prior published AMPREDICT models,11,12,42,54 this model carefully ensured that the amputations were incident amputations to avoid the heterogeneity typically seen in cross sectional studies where it is unknown whether the predictors preceded the amputation and whether they were first amputations vs. re-amputations. Furthermore, a careful selection of predictors was performed, as many of these predictors are difficult for the individual surgeon to consider, especially in aggregate, which the AMPREDICT MoRe model achieves. One area where too much granularity was avoided was in prior revascularisation. Attempts were made to categorise open, endovascular, stents, balloons, and inflow and outflow procedures; however, in some cases there were large amounts of missing data, and in others the coding was less than clear, which could have led to misclassification. Frailty indices were also considered; however, two of the indices (JEN and VA) were unavailable in the CDW, and while CAN (Care Assessment Need) scores were available, there were large numbers of missing scores. Given the large number of comorbidities considered, many of which are used by these frailty indices, it was felt that this domain was captured well. It was fortunate that a dataset of nearly 10 000 incident amputations was available. The results should only be applied to individuals who have undergone their first unilateral amputation at the TM level or higher due to diabetes and or PAD. The number of women in both the development and validation samples was small due to the nature of the US veteran population. Hence, they did not contribute significantly to the risk predictions, although sex, race, and geography were purposefully not included as predictors to avoid disparate amputation level decisions based on these factors. While the performance characteristics for predicting death were very strong, the prediction of re-amputation was acceptable, albeit less robust. One reason may be that the prediction is made prior to surgery; therefore, it does not consider operative or post-operative factors, which may also modify the risk of healing. However, that is the intent of these models. Identifying higher risk patients in the preoperative period may influence amputation level decision making or it may help focus post-operative care on a multidisciplinary approach involving vascular surgery, endocrinology, nutrition, and rehabilitation so that limb perfusion, diabetes management, and residual limb care can be optimised. The use of this model outside of a VA population should be considered with some level of caution, and it is recommended that it is externally validated in non-veteran populations in the USA and internationally.
Conclusion
The AMPREDICT MoRe model has been successfully developed and validated and can be applied at the time of amputation level decision making using predictors available in the medical record. This model will soon be translated to a point of care decision support tool (i.e., risk calculator), available both externally for all users and within the VA EHR. Such a tool will accomplish the following: (i) augment provider clinical expertise to formulate and quantify key outcomes of individual patients, therefore assisting them in developing treatment and disposition plans; (ii) provide prognostic evidence at the time of the amputation decision to facilitate patient and surgeon shared decision making around the most appropriate level; and (iii) assist in setting appropriate patient outcome expectations.
Supplementary Material
WHAT THIS PAPER ADDS.
This paper describes a novel prediction model (AMPREDICT MoRe) that predicts death and re-amputation among individuals facing their first amputation because of diabetes and or dysvascular disease. This model overcomes prior implementation barriers by using only predictors that are readily available in the electronic health record. The model can be used at the time of amputation level selection.
ACKNOWLEDGEMENT
This material is based upon work supported by the US Department of Veterans Affairs, Office of Research and Development, Rehabilitation Research and Development [grant no. RX003690-01A1].
APPENDIX A. SUPPLEMENTARY DATA
Supplementary data to this article can be found online at https://doi.org/10.1016/j.ejvs.2025.02.016.
Footnotes
CONFLICTS OF INTEREST
None.
REFERENCES
- 1.Cai M, Xie Y, Bowe B, Gibson AK, Zayed MA, Li T, et al. Temporal trends in incidence rates of lower extremity amputation and associated risk factors among patients using Veterans Health Administration services from 2008 to 2018. JAMA Netw Open 2021;4:e2033953. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Dillingham TR, Pezzin LE, MacKenzie EJ. Limb amputation and limb deficiency: epidemiology and recent trends in the United States. South Med J 2002;95:875–83. [DOI] [PubMed] [Google Scholar]
- 3.Czerniecki JM, Turner AP, Williams RM, Hakimi KN, Norvell DC. Mobility changes in individuals with dysvascular amputation from the presurgical period to 12 months postamputation. Arch Phys Med Rehabil 2012;93:1766–73. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Columbo JA, Nolan BW, Stucke RS, Rzucidlo EM, Walker KL, Powell RJ, et al. Below-knee amputation failure and poor functional outcomes are higher than predicted in contemporary practice. Vasc Endovascular Surg 2016;50:554–8. [DOI] [PubMed] [Google Scholar]
- 5.Suckow BD, Goodney PP, Nolan BW, Veeraswamy RK, Gallagher P, Cronenwett JL, et al. Domains that determine quality of life in vascular amputees. Ann Vasc Surg 2015;29:722–30. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Ratliff HT, Shibuya N, Jupiter DC. Minor vs. major leg amputation in adults with diabetes: six-month readmissions, reamputations, and complications. J Diabetes Complications 2021;35:107886. [DOI] [PubMed] [Google Scholar]
- 7.Roepke AM, Williams RM, Turner AP, Henderson AW, Norvell DC, Henson H, et al. A longitudinal study of social participation after dysvascular lower extremity amputation. Am J Phys Med Rehabil 2017;96:741–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Almasri J, Adusumalli J, Asi N, Lakis S, Alsawas M, Prokop LJ, et al. A systematic review and meta-analysis of revascularization outcomes of infrainguinal chronic limb-threatening ischemia. J Vasc Surg 2019;69(6S):126S–36S. [DOI] [PubMed] [Google Scholar]
- 9.Preece RA, Dilaver N, Waldron C-A, Pallmann P, Thomas-Jones E, Gwilym BL, et al. A systematic review and narrative synthesis of risk prediction tools used to estimate mortality, morbidity, and other outcomes following major lower limb amputation. Eur J Vasc Endovasc Surg 2021;62:127–35. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Norvell DC, Thompson ML, Boyko EJ, Landry G, Littman AJ, Henderson WG, et al. Mortality prediction following non-traumatic amputation of the lower extremity. Br J Surg 2019;106: 879–88. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Czerniecki JM, Thompson ML, Littman AJ, Boyko EJ, Landry GJ, Henderson WG, et al. Predicting reamputation risk in patients undergoing lower extremity amputation due to the complications of peripheral artery disease and/or diabetes. Br J Surg 2019;106: 1026–34. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Czerniecki JM, Turner AP, Williams RM, Thompson ML, Landry G, Hakimi K, et al. The development and validation of the AMPREDICT model for predicting mobility outcome after dysvascular lower extremity amputation. J Vasc Surg 2017;65:162–71.e3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Norvell DC, Sukow BD, Webster JB, Landry G, Henderson AW, Twine CP, et al. The development and usability of the AMPREDICT Decision Support Tool: a mixed methods study. Eur J Vasc Endovasc Surg 2021;62:304–11. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Brier MJ, Williams RM, Turner AP, Henderson AW, Roepke AM, Norvell DC, et al. Quality of relationships with caregivers, depression, and life satisfaction after dysvascular lower extremity amputation. Arch Phys Med Rehabil 2018;99:452–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Acar E, Kacira BK. Predictors of lower extremity amputation and reamputation associated with the diabetic foot. J Foot Ankle Surg 2017;56:1218–22. [DOI] [PubMed] [Google Scholar]
- 16.Beaulieu RJ, Grimm JC, Lyu H, Abularrage CJ, Perler BA. Rates and predictors of readmission after minor lower extremity amputations. J Vasc Surg 2015;62:101–5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Blume P, Salonga C, Garbalosa J, Pierre-Paul D, Key J, Gahtan V, et al. Predictors for the healing of transmetatarsal amputations: retrospective study of 91 amputations. Vascular 2007;15:126–33. [DOI] [PubMed] [Google Scholar]
- 18.Curran T, Zhang JQ, Lo RC, Fokkema M, McCallum JC, Buck DB, et al. Risk factors and indications for readmission after lower extremity amputation in the American College of Surgeons National Surgical Quality Improvement Program. J Vasc Surg 2014;60:1315–24. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Font-Jimenez I, Llaurado-Serra M, Pallares-Marti A, Garcia-Hedrera F. [Psycho-social factors involved in amputation. Systematic review of the literature]. Aten Primaria 2016;48:207–10 (in Spanish). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Fortington LV, Geertzen JH, van Netten JJ, Postema K, Rommers GM, Dijkstra PU. Short and long term mortality rates after a lower limb amputation. Eur J Vasc Endovasc Surg 2013;46:124–31. [DOI] [PubMed] [Google Scholar]
- 21.Hasanadka R, McLafferty RB, Moore CJ, Hood DB, Ramsey DE, Hodgson KJ. Predictors of wound complications following major amputation for critical limb ischemia. J Vasc Surg 2011;54:1374–82. [DOI] [PubMed] [Google Scholar]
- 22.Hoffstad O, Mitra N, Walsh J, Margolis DJ. Diabetes, lower-extremity amputation, and death. Diabetes Care 2015;38:1852–7. [DOI] [PubMed] [Google Scholar]
- 23.Kono Y, Muder RR. Identifying the incidence of and risk factors for reamputation among patients who underwent foot amputation. Ann Vasc Surg 2012;26:1120–6. [DOI] [PubMed] [Google Scholar]
- 24.Nelson MT, Greenblatt DY, Soma G, Rajimanickam V, Greenberg CC, Kent KC. Preoperative factors predict mortality after major lower-extremity amputation. Surgery 2012;152:685–96. [DOI] [PubMed] [Google Scholar]
- 25.Phair J, DeCarlo C, Scher L, Koleilat I, Shariff S, Lipsitz EC, et al. Risk factors for unplanned readmission and stump complications after major lower extremity amputation. J Vasc Surg 2018;67:848–56. [DOI] [PubMed] [Google Scholar]
- 26.Robbins JM, Strauss G, Aron D, Long J, Kuba J, Kaplan Y. Mortality rates and diabetic foot ulcers: is it time to communicate mortality risk to patients with diabetic foot ulceration? J Am Podiatr Med Assoc 2008;98:489–93. [DOI] [PubMed] [Google Scholar]
- 27.Swaminathan A, Vemulapalli S, Patel MR, Jones WS. Lower extremity amputation in peripheral artery disease: improving patient outcomes. Vasc Health Risk Manag 2014;10:417–24. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.van Netten JJ, Fortington LV, Hinchliffe RJ, Hijmans JM. Early post-operative mortality after major lower limb amputation: a systematic review of population and regional based studies. Eur J Vasc Endovasc Surg 2016;51:248–57. [DOI] [PubMed] [Google Scholar]
- 29.Yip VS, Teo NB, Johnstone R, Robertson AG, Robertson JH, Welch GH, et al. An analysis of risk factors associated with failure of below knee amputations. World J Surg 2006;30:1081–7. [DOI] [PubMed] [Google Scholar]
- 30.Sohn MW, Arnold N, Maynard C, Hynes DM. Accuracy and completeness of mortality data in the Department of Veterans Affairs. Popul Health Metr 2006;4:2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Friedman J, Hastie T, Tibshirani R. Regularization paths for generalized linear models via coordinate descent. J Stat Softw 2010;33:1–22. [PMC free article] [PubMed] [Google Scholar]
- 32.Hastie T, Tibshirani R, Friedman JH. The Elements of Statistical Learning. Data Mining, Inference, and Prediction , Vol 2. New York, NY: Springer; 2009. [Google Scholar]
- 33.Wood AM, White IR, Royston P. How should variable selection be performed with multiply imputed data? Stat Med 2008;27:3227–46. [DOI] [PubMed] [Google Scholar]
- 34.Royston P, Sauerbrei W. Multivariable Model-building. A Pragmatic Approach to Regression Analysis Based on Fractional Polynomials for Modelling Continuous Variables. Chichester, UK: John Wiley & Sons; 2008. [Google Scholar]
- 35.Van Calster B, Vergouwe Y, Looman CW, Van Belle V, Timmerman D, Steyerberg EW. Assessing the discriminative ability of risk models for more than two outcome categories. Eur J Epidemiol 2012;27:761–70. [DOI] [PubMed] [Google Scholar]
- 36.The R Core Team. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; 2013. [Google Scholar]
- 37.Conte MS, Bradbury AW, Kolh P, White JV, Dick F, Fitridge R, et al. Global vascular guidelines on the management of chronic limb-threatening ischemia. Eur J Vasc Endovasc Surg 2019;58(1S): S1–109.e33. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Columbo JA, Davies L, Kang R, Barnes JA, Leinweber KA, Suckow BD, et al. Patient experience of recovery after major leg amputation for arterial disease. Vasc Endovascular Surg 2018;52:262–8. [DOI] [PubMed] [Google Scholar]
- 39.Franklin H, Rajan M, Tseng CL, Pogach L, Sinha A, Mph Md. Cost of lower-limb amputation in U.S. veterans with diabetes using health services data in fiscal years 2004 and 2010. J Rehabil Res Dev 2014;51:1325–30. [DOI] [PubMed] [Google Scholar]
- 40.Zambetti BR, Stiles ZE, Gupta PK, Stickley SM, Brahmbhatt R, Rohrer MJ, et al. Present-day analysis of early failure after forefoot amputation. Surgery 2020;168:904–8. [DOI] [PubMed] [Google Scholar]
- 41.Landry GJ, Silverman DA, Liem TK, Mitchell EL, Moneta GL. Predictors of healing and functional outcome following transmetatarsal amputations. Arch Surg 2011;146:1005–9. [DOI] [PubMed] [Google Scholar]
- 42.Norvell DC, Czerniecki JM. Risks and risk factors for ipsilateral re-amputation in the first year following first major unilateral dysvascular amputation. Eur J Vasc Endovasc Surg 2020;60:614–21. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43.Goodney PP, Tarulli M, Faerber AE, Schanzer A, Zwolak RM. Fifteen-year trends in lower limb amputation, revascularization, and preventive measures among Medicare patients. JAMA Surg 2015;150:84–6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44.Dilaver NM, Gwilym BL, Preece R, Twine CP, Bosanquet DC. Systematic review and narrative synthesis of surgeons’ perception of postoperative outcomes and risk. BJS Open 2020;4:16–26. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 45.Gwilym BL, Pallmann P, Waldron C-A, Thomas-Jones E, Milosevic S, et al. , Vascular and Endovascular Research Network (VERN) and the PERCEIVE Study Group. Long-term risk prediction after major lower limb amputation: 1-year results of the PERCEIVE study. BJS Open 2024;8:zrad135. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 46.Corriere MA, Avise JA, Peterson LA, Stafford JM, Easterling D, Boone DS Sr, et al. Exploring patient involvement in decision making for vascular procedures. J Vasc Surg 2015;62:1032–9.e2. [DOI] [PubMed] [Google Scholar]
- 47.Leonard C, Sayre G, Williams S, Henderson A, Norvell D, Turner AP, et al. Perceived shared decision-making among patients undergoing lower-limb amputation and their care teams: a qualitative study. Prosthet Orthot Int 2023;47:379–86. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 48.Henderson AW, Turner AP, Leonard C, Sayre G, Suckow B, Williams SL, et al. Mortality conversations between male veterans and their providers prior to dysvascular lower extremity amputation. Ann Vasc Surg 2023;92:313–22. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 49.Ankolekar A, Dekker A, Fijten R, Berlanga A. The benefits and challenges of using patient decision aids to support shared decision making in health care. JCO Clin Cancer Inform 2018;2:1–10. [DOI] [PubMed] [Google Scholar]
- 50.Ngiam KY, Khor IW. Big data and machine learning algorithms for health-care delivery. Lancet Oncol 2019;20:e262–73. [DOI] [PubMed] [Google Scholar]
- 51.Obermeyer Z, Emanuel EJ. Predicting the future – big data, machine learning, and clinical medicine. N Engl J Med 2016;375:1216–19. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 52.Poldrack RA, Huckins G, Varoquaux G. Establishment of best practices for evidence for prediction: a review. JAMA Psychiatry 2020;77:534–40. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 53.French MA, Roemmich RT, Daley K, Beier M, Penttinen S, Raghavan P, et al. Precision rehabilitation: optimizing function, adding value to health care. Arch Phys Med Rehabil 2022;103: 1233–9. [DOI] [PubMed] [Google Scholar]
- 54.Norvell DC, Thompson ML, Baraff A, Biggs WT, Henderson AW, Moore KP, et al. AMPREDICT PROsthetics—predicting prosthesis mobility to aid in prosthetic prescription and rehabilitation planning. Arch Phys Med Rehabil 2023;104:523–32. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.