Clinical Open-Datasetlong-COVID Prediction Using Machine Learning

Abstract

The COVID-19 pandemic, declared by the World Health Organization (WHO, 2020) posed an unprecedented challenge to global health.Now, years after that critical period of time, there is growing concern about long-term covid sequelae in recovered patients not only in Europe and USA, but in other countries.This study focuses on developing machine learning tools to predict the severity of both acute symptoms and post-COVID sequelae, aiming to provide software tools for healthcare specialists, not only in Mexico but all over the world.Using a tailor-made dataset from a number of clinical open datasets, we trained decision tree models on 488 records featuring pre-existing conditions as variables -smoking, diabetes, hypertension.The models predicted symptom and sequelae severity -moderate, severe, critical-with average accuracies as high as 95%.A validation via normalized confusion matrices and ROC curves was also carried out.These results, first, confirm the feasibility of using interpretable AI models to support clinical prognosis and, second, highlight the need for more comprehensive datasets, particularly for long-covid critical cases.

Ähnliche Arbeiten