AI-Enhanced Prediction of Multi Organ Failure in COVID-19 Patients

The occurrence of multi-organ failure (MOF) in COVID-19 patients constitutes a critical complication, markedly elevating the mortality risk compared to patients without MOF. Consequently, early identification and timely intervention for these patients are crucial. In this research, we utilized a substantial dataset derived from the multicenter observational study "Coagulopathy associated with COVID-19 (CA-COVID-19)," covering 26 UK NHS Trusts and involving 8,032 COVID-19 patients aged 18 years and older. Previously, numerous analyses have been conducted to assess clinical outcomes and their predictive factors, utilizing data from the CA-COVID-19 study through standard statistical methods. However, Artificial Intelligence (AI) models have not been used on this data for predicting clinical outcomes. This paper introduces an AI driven approach to predict the onset of multi-organ failure (MOF) in patients diagnosed with COVID-19. We implemented six AI models including (i) Artificial Neural Network with Backpropagation, (ii) XGBoost, (iii) Support Vector Classifier, (iv) Stochastic Gradient Descent Classifier, (v) Random Forest, and (vi) Logistic Regression. The models underwent evaluation through a 5-fold cross-validation technique, employing various metrics for assessment. The findings revealed that the Support Vector Classifier surpassed all other models in terms of overall performance, consistently achieving a score of 0.98 across accuracy, precision, F1 score, and recall metrics. Additionally, this model attained the lowest loss score at 0.082 and the highest AUC score of 0.951, outperforming all competing models. Leveraging a distinctive feature selection method, we identified that certain factors such as major bleeding, thrombosis, prior malignancy, lung disease history, smoking status, Asian ethnicity, and elevated levels of platelets, D-dimer, LDH, and Troponin I, significantly contribute to the development of multi-organ failure in COVID-19 patients. The insights garnered from this study could enable clinicians to promptly identify patients at heightened risk of developing multi-organ failure, facilitating timely interventions that may enhance clinical outcomes.