The application of machine learning algorithms in predicting the length of stay following femoral neck fracture

• DOI: 10.1016/j.ijmedinf.2021.104572
• Publication Date: 2021-09-13T15:36:08Z
• Authors: Hao Zhong; Bingpu Wang; Dawei Wang; Zirui Liu; Cong Xing; Yu Wu; Qiang Gao; Shibo Zhu; Haodong Qu; Zeyu Jia; Zhigang Qu; Guangzhi Ning; Shiqing Feng
• Keywords: Stepwise regression

Femoral neck fracture is a frequent cause of hospitalization, and length stay an important marker hospital cost quality care provided. As extension traditional statistical methods, machine learning provides the possibility accurately predicting stay. The aim this paper to retrospectively identify predictive factors (LOS) predict postoperative LOS by using algorithms. Based on admission perioperative data patients, linear regression was used analyze LOS. Multiple models were developed, performance different compared. Stepwise showed that preoperative calcium level (P = 0.017) lymphocyte percentage 0.007), in addition intraoperative bleeding (p 0.041), glucose sodium chloride infusion after surgery 0.019), Charlson Comorbidity Index 0.007) BMI 0.031), significant predictors best performing model principal component (PCR) with optimal MAE (1.525) proportion prediction error within 3 days 90.91%. Excessive intravenous surgery, hypocalcemia, high percentages lymphocytes, excessive bleeding, lower higher CCI scores related prolonged regression. Machine could This information allows administrators plan reasonable resource allocation fulfill demand, leading direct improvement more use scarce resources.