Development and external validation of a prediction model for digit replantation failure after traumatic amputations based on a prospective multicenter cohort

Abstract

Background: Failure of digit replantation after traumatic amputation is difficult to predict. The authors aimed to develop a prognostic model to better identify factors that better predict replantation failure following traumatic digit amputation.

Materials and methods: In this multicenter prospective cohort, the authors identified patients who had received digit replantation between 1 January 2015 and 1 January 2019. Univariable and multivariable analyses were performed successively to identify independently predictive factors for failure of replanted digit. To reduce overfitting, the Bayesian information criterion was used to reduce variables in the original model. Nomograms were created with the reduced model after model selection. This model was then internally validated with bootstrap resampling and further externally validated in validation cohort.

Results: Digit replantation was failed in 101 of 1062 (9.5%) digits and 146 of 1156 digits (12.6%) in the training and validation cohorts, respectively. The authors found that six independent prognostic variables were associated with digit replantation failure: age, mechanism of injury, ischemia duration, smoking status, amputation pattern (complete or incomplete), and surgeon's experience. The prediction model achieved good discrimination, with concordance indexes of 0.81 (95% CI: 0.76-0.85) and 0.70 (95% CI: 0.65-0.74) in predicting digit failure in the training and validation cohorts, respectively. Calibration curves were well-fitted for both training and validation cohorts.

Conclusions: The proposed prediction model effectively predicted the failure rate of digit replantation for individual digits of all patients. It could assist in selecting the most suitable surgical plan for the patient.

Figure 1

Study screening and enrollment.

Figure 2

Proposed model for estimating failure rate of digit replantation after traumatic digit amputation. To use the prediction, first, the surgeon determines the subscore of each variable by finding the patient’s value on the corresponding scale and then drawing a vertical line up to the Points scale to obtain the number of points for that variable. This is done for each variable. Second, the surgeon calculates the sum of the variable subscores. Finally, the surgeon finds the sum of the variable subscores on the Total Points scale, and then draws a vertical line (index line or isopleth) from the total points scale down to the lower line (failure rate) of the model to determine the predicted probability of failure.

Figure 3

Bootstrap analysis results to determine prediction accuracy of model for failure rate of digit replantation. Calibration curve of the model for (A) training and (B) validation cohorts in estimating failure rate. The prediction model was validated with the internal data from the training cohort, and then externally evaluated from data of the validation cohort. Apparent (or actual) accuracy, bias-corrected accuracy, and ideal accuracy are plotted. Analysis used 1000 bootstrap samples.