Lasso-Based Machine Learning Algorithm for Predicting Postoperative Lung Complications in Elderly: A Single-Center Retrospective Study from China

Abstract

Background: The predictive effect of systemic inflammatory factors on postoperative pulmonary complications in elderly patients remains unclear. In addition, machine learning models are rarely used in prediction models for elderly patients. Patients and Methods: We retrospectively evaluated elderly patients who underwent general anesthesia during a 6-year period. Eligible patients were randomly assigned in a 7:3 ratio to the development group and validation group. The Least logistic absolute shrinkage and selection operator (LASSO) regression model and multiple logistic regression analysis were used to select the optimal feature. The discrimination, calibration and net reclassification improvement (NRI) of the final model were compared with “the Assess Respiratory Risk in Surgical Patients in Catalonia” (ARISCAT) model. Results: Of the 9775 patients analyzed, 8.31% developed PPCs. The final model included age, preoperative SpO2, ANS (the Albumin/NLR Score), operation time, and red blood cells (RBC) transfusion. The concordance index (C-index) values of the model for the development cohort and the validation cohort were 0.740 and 0.748, respectively. The P values of the Hosmer–Lemeshow test in two cohorts were insignificant. Our model outperformed ARISCAT model, with C-index (0.740 VS 0.717, P = 0.003) and NRI (0.117, P < 0.001). Conclusion: Based on LASSO machine learning algorithm, we constructed a prediction model superior to ARISCAT model in predicting the risk of PPCs. Clinicians could utilize these predictors to optimize prospective and preventive interventions in this patient population.