Prediction of Short or Long Length of Stay COVID-19 by Machine Learning

Abstract

Aim: The aim of this study is to utilize machine learning techniques to accurately predict the length of stay for Covid-19 patients, based on basic clinical parameters. Material and Methods: The study examined seven key variables, namely age, gender, length of hospitalization, c-reactive protein, ferritin, lymphocyte count, and the COVID-19 Reporting and Data System (CORADS), in a cohort of 118 adult patients who were admitted to the hospital with a diagnosis of Covid-19 during the period of November 2020 to January 2021. The data set is partitioned into a training and validation set comprising 80% of the data and a test set comprising 20% of the data in a random manner. The present study employed the caret package in the R programming language to develop machine learning models aimed at predicting the length of stay (short or long) in a given context. The performance metrics of these models were subsequently documented. Results: The k-nearest neighbor model produced the best results among the various models. As per the model, the evaluation outcomes for the estimation of hospitalizations lasting for 5 days or less and those exceeding 5 days are as follows: The accuracy rate was 0.92 (95% CI, 0.73-0.99), the no-information rate was 0.67, the Kappa rate was 0.82, and the F1 score was 0.89 (p=0.0048). Conclusion: By applying machine learning into Covid-19, length of stay estimates can be made with more accuracy, allowing for more effective patient management.