Interpretable artificial intelligence for predicting blood transfusion after surgery for femoral shaft fractures: A retrospective analysis

Authors

• Rameshbabu Manyam, Department of Biostatistics and Bioinformatics, Rollins School of Public Health, Emory University, Atlanta, GA, USA.
• Puneet Gupta, Department of Anesthesiology and Critical Care Medicine, The George Washington University School of Medicine and Health Sciences, Washington, DC, USA.
• Pengfei Lou, Department of Biostatistics and Bioinformatics, Rollins School of Public Health, Emory University, Atlanta, GA, USA.
• Rui Guo, Department of Anesthesiology and Critical Care Medicine, The George Washington University School of Medicine and Health Sciences, Washington, DC, USA.
• Andrew Carro, Department of Anesthesiology and Critical Care Medicine, The George Washington University School of Medicine and Health Sciences, Washington, DC, USA.
• Eric R. Heinz, Department of Anesthesiology and Critical Care Medicine, The George Washington University School of Medicine and Health Sciences, Washington, DC, USA.

Document Type

Journal Article

Publication Date

3-1-2026

Journal

Indian journal of anaesthesia

Volume

70

Issue

3

DOI

10.4103/ija.ija_810_25

Keywords

Artificial intelligence; SHAP; SHapley Additive exPlanations; XGBoost; blood transfusion; blood urea nitrogen; body mass index; creatinine; femoral shaft fracture; gradient boosting; machine learning; white blood cell count

Abstract

BACKGROUND AND AIMS: Blood transfusions are often necessary in the surgical repair of orthopaedic fractures. However, these transfusions are associated with significant morbidity. The purpose of this study was to assess whether artificial intelligence (AI) based models can be utilised to predict blood transfusions after surgery for femoral shaft fractures and to identify the most important preoperative risk factors using interpretable AI methods. METHODS: This retrospective study utilised 2015-2020 data of the femoral shaft fracture patient population drawn from the National Surgical Quality Improvement Program database. Five AI based models were developed using patient clinical information for predicting blood transfusions within 72 hours of surgery. SHapley Additive exPlanations were performed to visualise and interpret the risk factors that contributed to the model's performance. RESULTS: A total of 1720 patients were included, of which 570 (33.1%) required a blood transfusion within 72 hours of femoral shaft fracture surgery. The Extreme Gradient Boosting model demonstrated the best predictive performance with an area under the receiving operating characteristic curve of 0.81 and a Brier score of 0.02. The most important risk factors for prediction were pre-operative haematocrit, age, platelet count, preoperative blood urea nitrogen, body mass index, preoperative white blood cell count, and preoperative creatinine. CONCLUSION: This study developed and internally validated an interpretable AI model for predicting blood transfusions in an isolated population of femoral shaft fracture patients with good performance. Interpretable AI based models may support anaesthesiologists and orthopaedic surgeons in perioperative risk stratification, management, and patient education but require external validation before clinical translation.

APA Citation

Manyam, Rameshbabu; Gupta, Puneet; Lou, Pengfei; Guo, Rui; Carro, Andrew; and Heinz, Eric R., "Interpretable artificial intelligence for predicting blood transfusion after surgery for femoral shaft fractures: A retrospective analysis" (2026). GW Authored Works. Paper 8953.
https://hsrc.himmelfarb.gwu.edu/gwhpubs/8953

Department

Anesthesiology and Critical Care Medicine