Development and external validation of an immediate-postoperative prediction model for high hidden blood loss after PFNA in older intertrochanteric-fracture patients

Abstract Background Hidden blood loss (HBL) is a frequent and clinically important complication after proximal femoral nail antirotation (PFNA) for intertrochanteric fractures. A practical model that combines perioperative predictors and is externally validated may improve individualized blood management. Methods This retrospective prediction model study included a development cohort of 186 older patients who underwent PFNA and an independent external validation cohort of 84 patients from Xianning First People's Hospital between 2021 and 2023. High HBL was defined as HBL > 650 mL based on an X-tile-derived cutoff. A main perioperative model was developed using predictors available before and during surgery, with least absolute shrinkage and selection operator screening and multivariable logistic regression. Internal validation was performed using 1000 bootstrap resamples. Model performance was assessed using discrimination, calibration, Brier score, decision curve analysis, and external validation. A supplementary preoperative model using only preoperatively available variables was also constructed. Results High HBL occurred in 99 of 186 patients (53.2%) in the development cohort and 27 of 84 patients (32.1%) in the external validation cohort. The final perioperative model included preoperative anticoagulant use, preoperative albumin, and operation time. In the development cohort, the model achieved an apparent AUC of 0.715 (95% CI, 0.640–0.790) and a bootstrap optimism-corrected AUC of 0.698. In the external validation cohort, the AUC was 0.894 (95% CI, 0.826–0.963), with a Brier score of 0.165, calibration intercept of -1.806, and calibration slope of 1.026. Because the validation cohort had a lower event rate, an exploratory intercept-only recalibration was performed; this reduced the mean predicted risk from 53.4% to 32.1%, improved the Brier score from 0.165 to 0.121, and preserved the AUC at 0.894. The supplementary preoperative model showed moderate discrimination, with an external AUC of 0.839. Conclusions A simple perioperative model incorporating anticoagulant use, preoperative albumin, and operation time showed good external discrimination for high HBL after PFNA. The model may support individualized perioperative blood management, although local recalibration and multicenter prospective validation are required before broad clinical implementation.