The Effectiveness of a Machine Learning Model in Predicting Blood Transfusion Probability in Bipolar Hemiarthroplasty Hip Replacement Surgery

Abstract

Purpose: To verify a machine learning-based prediction model for blood transfusion risk in patients undergoing bipolar hemiarthroplasty and to determine whether there are significant differences between the accuracy results of this verification and the original model.

Methods: A retrospective study using purposive sampling was designed to gather 136 samples with the inclusion criterion of undergoing bipolar hemiarthroplasty for femoral neck fractures at the author’s institution between January 1, 2021, and June 30, 2024. The research instruments included (1) a machine learning-based prediction model for blood transfusion probability (smskbl.streamlit.app), which was constructed using 232 femoral neck fracture samples undergoing bipolar hemiarthroplasty at the author’s institution from 2015 to 2020, and (2) a research questionnaire created by the researcher, including six items: one on demographic data, four on medical health conditions, and one on actual blood transfusion during surgery.

Results: The prediction model accuracy was 89%, compared with that of the original model (80%). The comparison of the accuracy results was not statistically significant (Z = 0.424, p > 0.05). In the blood transfusion group, the precision was 0.70, recall was 0.73, and F1-score was 0.72, whereas the group that did not receive blood transfusion had a precision of 0.94, recall of 0.93, and an F1-score of 0.93. The area under the curve was 0.83.

Conclusions: The blood transfusion prediction model demonstrated good performance in predicting transfusion risk. The model provides confidence in its risk prediction outcome and can be used to perform optimal risk management in preparation for bipolar hemiarthroplasty.