Predicting the perioperative transfusion risk of proximal femoral antirotation nailing (PFNA) for elderly patients with intertrochanteric fractures: a new predictive nomogram.

Abstract

Background: Proximal femoral antirotation nailing (PFNA) for treating elderly patients with intertrochanteric fractures (EIFs) is often associated with substantial hidden blood loss. Perioperative blood transfusion to restore the lost blood has no effect on postoperative mortality and it increases the risk of postoperative infection. The goal of this study was to develop and validate a nomogram for predicting the risk of perioperative transfusion and intervening ahead of time to reduce the risk in EIF patients receiving PFNA.

Methods: This study retrospectively examined and collected risk factors associated with transfusion in EIF patients treated with PFNA. Random forest with least absolute shrinkage and selection operator (LASSO) regression analysis was used to select characteristic variables and construct nomograms with the screening variables. The predictive model's discriminatory efficacy and calibration efficacy were assessed by receiver operating characteristic (ROC) curves, C-index, and calibration curves, respectively. Clinical usefulness was assessed by decision curve analysis (DCA).

Results: The final nomogram consisted of five predictor variables: lower preoperative haemoglobin (HGB), age, preoperative urea, preoperative albumin, and surgical position. The nomogram showed good discriminatory and calibration efficacy with an area under the curve (AUC) value of 0.865 and a calibration curve highly approximating the ideal curve. In internal validation, the C-index of the model was calculated to be 0.823, indicating that the model exhibited superior predictive power.

Conclusions: The nomogram constructed from preoperative HGB, age, urea, albumin, and surgical position can be used to predict more accurately the risk of perioperative transfusion in EIF patients treated with PFNA. Validation of the accuracy of this predictive model requires multicenter, prospective, and larger populations.