مهرنوش غظنفري هرندي

The rapid growth of COVID-19 has posed significant challenges to health care systems, and hospitals in particular, due to the increasing number of cases in each institution. In view of the fact that the demand for hospital beds is highly dependent on the length of stay of patients with COVID-19, the discovery of early biomarkers of the length of hospital stay may rationalize allocation of hospital resources from access to the emergency department and facilitate appropriate treatment and improvement of patient care. Therefore, the development of a statistical model to predict the length of stay at hospital for patients with COVID-19 is essential. The aim of this study was to determine the length of stay of COVID-19 patients in the hospital with statistical tools, data mining techniques, and machine learning, in regard to the imputation of missing data and feature selection. Thus, five regression algorithms-Random Forest Regression, Support Vector Regression, XGBoost Regression, Gradient Boosting Regression, and Artificial Neural Network Regression-were compared based on the eva‎luation criteria. In addition, to fill in the missing data, four methods were used: Predictive mean matching (PMM), Predictive Mean Matching with distance-aided donor selection (Midastouch), Classification and Regression Tree (CART), and Random Forest. According to the amount of error calculated for each method, the Random Forest method Was selected as the best way to fill in the missing data. Among the four methods of Lasso Regression, Ridge Regression, Elastic net Regression and Recursive Feature Elimination (RFE) that were used to feature selection, the Recursive Feature Elimination (RFE) method was selected to choose the most effective features. The case of this study is the data of 659 patients with COVID-19 collected from Imam Khomeini Hospital in Tehran from mid-March 2021 to early May 2021. Among the models introduced for prediction, Gradient Boosting Regression had the best performance in terms of eva‎luation criteria to predict the length of stay of patients with COVID-19 at the hospital, which according to the results of this model, age, fever, other symptoms, headache, dyspnea and hemoptysis were effective variables in predicting the length of hospital stay of coronary patients.

كوويد-19 , مدت زمان بستري , پيش بيني , يادگيري ماشين , رگرسيون , داده هاي گمشده

COVID-19 , length of stay , prediction , machine learning , regression , missing data

Mehrnoush Ghazanfariharandi

Rassoul Noorossana