63/100

Moderate evidence

Cite
More
EmailLinkedInX

Admission model flagged 98% of diabetic foot amputations in a 771-patient Taiwan study

Study title: Admission-Time Prediction of In-Hospital Major Amputation in Diabetes-Related Foot Disease: Model Development and Temporal Validation Study

Evidence type
Observational study
Sample
n = 2,171
Sources
1 sourcePeer-reviewed
Published
2 min read

Study title: Admission-Time Prediction of In-Hospital Major Amputation in Diabetes-Related Foot Disease: Model Development and Temporal Validation Study

Researchers at a tertiary center in Taiwan built a model to estimate, at admission, which patients with diabetes-related foot disease would undergo major above-ankle amputation during the same hospital stay. It used 20 routinely collected variables and was trained on 1,400 patients admitted from 2015 to 2021.

Read the study abstract

Major above-ankle amputation is a devastating outcome of diabetes-related foot disease. The Wagner classification is routinely available but places many inpatients in the same grade; few prediction models for major amputation have undergone temporal validation or reported calibration. This study aimed to develop an admission-time model for in-hospital major amputation from routinely collected data and to validate it internally in a later period at the same center. Its intended role is to support prognostic commu…

55Gain side

How the two sides scoreProblem scores higher

60Problem side

The gain

In a temporal validation of 771 diabetic foot disease patients at a Taiwan tertiary center, a ridge logistic regression model scored an AUROC of 0.882 for in-hospital major amputation versus 0.804 for Wagner grade alone.

The problem

In the same 771-patient Taiwan temporal validation, the amputation model overpredicted absolute risk, with a mean predicted risk of 9.6% against 6.6% observed and an observed-to-expected ratio of 0.69.

A fixed formula sets both scores from the evidence and its sources. Reader votes don’t move them.

What the scores mean

G is the gain side, out of 100. It goes up when the gain is specific, when it measurably helped people and when the sources behind it are strong.

P is the problem side, out of 100. It goes up when the problem is specific, when it measurably hurt people and when the sources behind it are strong.

In brief

On 771 later patients (2022-2025), the model reached an AUROC of 0.882 versus 0.804 for Wagner grade alone. At the preset screening threshold it flagged 44.5% of patients, with sensitivity of 0.980 and negative predictive value of 0.998. No major amputation occurred among the 396 patients in the lowest risk band.

The model overpredicted absolute risk, and positive predictive value was only 14.6%. It was built to support prognosis talks and early specialist assessment, not to decide on amputation, and it was validated only at one center, so external and prospective testing are still needed.

Main points

  1. The development cohort had 1,400 patients with 126 major amputations (9.0%); the temporal test cohort had 771 patients with 51 (6.6%).
  2. Five algorithms were compared by nested cross-validation; ridge logistic regression had the lowest out-of-fold log loss and became the primary model.
  3. At the prespecified threshold, 343 patients (44.5%) were flagged, with sensitivity 0.980 and specificity 0.593.
  4. Positive predictive value was 14.6%, while negative predictive value was 0.998.
  5. None of the 396 patients (51.4%) in the lowest risk band had a major amputation; 22 of 54 (40.7%) in the highest band did.
  6. Random forest and gradient boosting reached higher AUROCs (0.915 and 0.912) in secondary comparisons.
  7. Adding segmental limb pressures did not clearly improve discrimination (change in AUROC -0.004).

The full story

The retrospective cohort study covered adults with a single hospitalization for diabetes-related foot disease at one tertiary center in Taiwan from 2015 through 2025. Patients were split by admission date, with 2015-2021 for development and 2022-2025 for testing, and operating thresholds were fixed before a single evaluation in the test cohort.

The authors chose the model for its transparency. Calibration was imperfect: the intercept was -0.47 and the slope 1.33. The model's best use is ruling out risk, since its high negative predictive value contrasts with a low positive predictive value of 14.6%.

The authors state the model must not be used to indicate amputation or withhold limb-salvage evaluation. Independent external validation, recalibration where needed and prospective evaluation are required before clinical use.

What this doesn’t fix

The model was validated internally at a single center, estimates amputation under one institution's care pathway, was not compared with multidimensional classifications such as Wound, Ischemia, and foot Infection, and whether acting on it improves outcomes is untested.

Sources

1 on file
  1. JMIR Medical Informaticsdoi:10.2196/110740
    Peer-reviewed2026-11-13

Talk it over

Three questions to start on, two more if you have time
The model scored an AUROC of 0.882 against 0.804 for Wagner grade alone, but predicted a mean risk of 9.6% against 6.6% observed. How do you weigh the better sorting against the overprediction?

Start with a question

Admission model flagged 98% of diabetic foot amputations in a 771-patient Taiwan studyhttps://truvace.com/item/admission-time-prediction-of-in-hospital-major-amputation-in-diabetes-related-fo

  1. The model scored an AUROC of 0.882 against 0.804 for Wagner grade alone, but predicted a mean risk of 9.6% against 6.6% observed. How do you weigh the better sorting against the overprediction?

    Answer in the thread
  2. No major amputation occurred among the 396 patients in the lowest risk band, while positive predictive value was only 14.6%. If you were a specialist at the Taiwan center, how would you use a flag from this model?

    Answer in the thread
  3. The model was validated only at one tertiary center in Taiwan, using later patients from 2022-2025. What would you want to see from other hospitals before trusting it?

    Answer in the thread
If you have more time
  • The authors say the model must not be used to indicate amputation or withhold limb-salvage evaluation. What safeguards in a hospital you know could keep it to that role?

    Answer in the thread
  • Random forest and gradient boosting reached higher AUROCs (0.915 and 0.912) but the authors chose ridge logistic regression for transparency. Which would you pick, and why?

    Answer in the thread

What readers say