Prediction of mortality risk and duration of hospitalization of COVID-19 patients with chronic comorbidities based on machine learning algorithms

Background: The severity of coronavirus in patients with chronic comorbidities is much higher than in other patients, which can lead to their death. Machine learning (ML) algorithms as a potential solution for rapid and early clinical evaluation of the severity of the disease can help in allocating and prioritizing resources to reduce mortality. Objective: To predict the mortality risk and length of stay (LoS) of patients with COVID-19 and history of chronic comorbidities using ML algorithms. Methods: This retrospective study reviewed medical records of COVID-19 patients with chronic comorbidities (Afzalipour Hospital, Kerman, Iran; March 2020–January 2021). Outcomes were discharge (alive) or death, and LoS. Filter-based feature scoring and well-known ML algorithms (including ensemble methods) were used to predict mortality risk and LoS. Performance was evaluated using F1, precision, recall, accuracy, and ROC/AUC; reporting followed TRIPOD. Results: 1291 patients (900 alive, 391 dead) were analyzed. Common symptoms were shortness of breath (53.6%), fever (30.1%), and cough (25.3%). Common comorbidities were diabetes mellitus (DM; 31.3%), hypertension (HTN; 27.3%), and ischemic heart disease (IHD; 14.2%). Twenty-six key factors were extracted. Gradient boosting achieved the best mortality prediction accuracy among ensembles (accuracy 84.15%); SVM performed best among base classifiers (accuracy 80.00%, ROC 0.85). For LoS prediction, a multilayer perceptron (MLP) with ReLU activation performed best (MSE 38.96). The most important factors for mortality included hyperlipidemia, diabetes, asthma, and cancer; shortness of breath was highlighted for LoS. Conclusion: ML algorithms can effectively predict mortality risk and LoS for COVID-19 patients with chronic comorbidities using physiological, symptomatic, and demographic data. Gradient boosting and MLP can rapidly identify patients at risk of death or prolonged hospitalization to inform timely clinical interventions.

Parastoo Amiri, Mahdieh Montazeri, Fahimeh Ghasemian, Fatemeh Asadi, Saeed Niksaz, Farhad Sarafzadeh, Reza Khajouei