Plast Reconstr Surg. 2026 Feb 2. doi: 10.1097/PRS.0000000000012869. Online ahead of print.
ABSTRACT
INTRODUCTION: Targeted Muscle Reinnervation (TMR) can prevent and treat neuropathic pain in amputees, but the degree of success varies. This study developed a Machine Learning (ML) model to predict the likelihood of sustained pain mitigation following primary and secondary TMR based on patient characteristics.
METHODS: Patients who underwent TMR at a tertiary care center (2017-2024) were included. Patients were categorized as achieving good or poor pain outcomes based on predefined criteria: ≥3/10-point reduction (Numeric Rating Scale) for secondary TMR, or pain scores ≤3/10 for ≥3 months for primary TMR. Three ML architectures (lasso logistic regression, random forest classifier, and relevance vector machine (RVM)) were tested. Model performance was evaluated using area under the receiver operating characteristic (AUROC) curve; feature importance was quantified using Shapley additive explanations (SHAP).
RESULTS: In total, 77 primary TMR and 101 secondary TMR patients were included (median follow-up: 2.0 years). The RVM model achieved test prediction accuracy scores of 0.74±0.12 for both primary and secondary TMR, with AUROC scores of 0.78±0.13 and 0.80±0.05, respectively. For primary TMR, pre-operative opioid use, male sex, and history of depression showed strong negative impacts. For secondary TMR, pre-operative smoking, elevated pain scores, and history of anxiety were strong negative predictors. The model significantly outperformed traditional statistical approaches.
DISCUSSION: This novel custom ML model achieved strong predictive capability for TMR outcomes, demonstrating proof of concept of a practical tool for surgical planning and patient selection. The identification of several key modifiable risk factors suggests opportunities for pre-operative optimization to improve surgical outcomes.
PMID:41628603 | DOI:10.1097/PRS.0000000000012869