Prediction of post-surgical complications in hand reconstruction using machine learning

Abstract

Background: Hand reconstruction is a complex surgical procedure in which various postoperative complications may arise, such as infections, flap necrosis, and joint stiffness. The prediction of these complications has traditionally relied on the surgeon’s experience and conventional clinical models. However, artificial intelligence (AI), particularly machine learning, has proven to be an effective tool for analyzing large volumes of clinical data and enhancing predictive capabilities in various medical fields.

Methods: A retrospective study was conducted with a sample of 200 patients who underwent hand reconstruction, using exclusively clinical record data. Three machine learning models were evaluated: XGBoost, Random Forest, and an artificial neural network. A data preprocessing pipeline, feature selection, and cross-validation were applied to optimize model performance. Predictive capability was assessed using the ROC curve and the area under the curve (AUC).

Results: XGBoost achieved the best performance with an AUC of 0.88, followed by Random Forest (AUC = 0.88) and the artificial neural network (AUC = 0.86). The most relevant variables for predicting complications included patient age, comorbidities such as diabetes mellitus, type of injury, and surgery duration.

Conclusions: AI models proved to be useful tools for predicting postoperative complications in hand reconstruction, surpassing the accuracy of conventional methods. In particular, XGBoost demonstrated the highest predictive capacity. These findings suggest that machine learning could optimize surgical planning and clinical decision-making, although further studies are needed to validate its applicability across different populations.