Development and validation of a nomogram predicting the risk of necrosis in fingertip replantation: a single-center retrospective study

Background Necrosis remains an important complication after fingertip replantation. We aimed to develop and internally validate a model for estimating necrosis risk using clinical, procedural, and early-postoperative information. Methods We retrospectively analyzed 1,344 patients who underwent replantation of 1,497 fingertips at a single center between August 2018 and August 2023. The patient was the unit of analysis. Patients were divided by outcome-stratified random sampling (seed 2222) into a derivation cohort ( n = 942; 70%) and a split-sample internal validation cohort ( n = 402; 30%). Candidate predictors were reduced by 10-fold cross-validated least absolute shrinkage and selection operator regression. A multivariable logistic regression model was specified using six predictors available during the early postoperative course. Discrimination, calibration, Brier score, and decision-curve analysis were evaluated. Results Necrosis occurred in 115 patients (12.2%) in the derivation cohort and 48 (11.9%) in the internal validation cohort. The model retained surgeon experience, platelet count (PLT), neutrophil percentage (NEU%), D-dimer, hypertension, and vascular crisis. The area under the receiver operating characteristic curve was 0.953 (95% CI, 0.935–0.968) in the derivation cohort and 0.974 (95% CI, 0.958–0.987) in the internal validation cohort. Corresponding Brier scores were 0.051 and 0.042. In the internal validation cohort, the calibration intercept was −0.338 and the calibration slope was 0.759; the Hosmer–Lemeshow test yielded P = 0.729. Decision-curve analysis suggested potential net benefit across clinically plausible threshold probabilities, but external validation is required. Conclusion This split-sample internally validated nomogram estimates necrosis risk after fingertip replantation once early-postoperative laboratory results and vascular-crisis status are available. The observed discrimination should be interpreted alongside calibration shrinkage, possible overfitting, and the absence of external validation.