A108-12 Machine Learning For Prognosis Prediction In Hospitalized Patients With COVID-19

Abstract Rationale Coronavirus disease 2019 (COVID-19) emerged as a global pandemic in 2019 and has since gradually subsided. However, the prognosis of hospitalized vulnerable populations remains a significant concern. With advancements in artificial intelligence and machine learning, their applications in clinical practice have become increasingly widespread. Methods This retrospective study collected data from hospitalized COVID-19 patients from May 2022 to April 2023, including information on age, gender, Charlson Comorbidity Index (CCI), quick Sequential Organ Failure Assessment score (qSOFA), vaccination status, laboratory data, and mortality in 30 days. Machine learning algorithms including decision trees, random forest, XGBoost, support vector machine and logistic regression, were employed to develop predictive models by Python. The performance of each model was evaluated based on metrics such as the area under the receiver operating characteristic curve (AUC) and average precision (AP). The SHapley Additive exPlanations (SHAP) value was applied to identify important predictive factors. Results Among the 548 hospitalized patients diagnosed with COVID-19, there were 451 survivors and 97 mortalities. In terms of performance metrics, the support vector machine model achieved the highest AUC (0.916), while the random forest model demonstrated the highest AP (0.741). Among all predictive models, the qSOFA consistently exhibited the greatest SHAP value, indicating that it was the most influential predictor. Additionally, CCI, age, vaccine doses, absolute neutrophil count, and hemoglobin also played significant roles in outcome prediction. Conclusion Machine learning models not only facilitate the prediction of COVID-19 prognosis but also offer novel perspectives beyond traditional statistical approaches in evaluating predictive indicators. This abstract is funded by: None