Prediction of mortality in cancer patients with COVID-19 using machine learning methods

This study aimed to predict mortality in cancer patients diagnosed with COVID-19 using machine learning (ML) algorithms and identify the clinical and laboratory parameters associated with mortality. Demographic, clinical, and laboratory data of cancer patients diagnosed with COVID-19 in the emergency service of Dr Abdurrahman Yurtaslan Ankara Oncology Training and Research Hospital were used. Seven ML algorithms, including decision tree, random forest, k-nearest neighbor, Naïve Bayes, eXtreme Gradient Boosting, Adaptive Boosting (AdaBoost), and support vector machines, were used to calculate the mortality risk of patients. Data balancing was achieved using the synthetic minority oversampling technique. Special libraries in the Python 3.8 programming language (Phyton Sofware Foundation, Fredericksburg) were used to determine descriptive statistics, model creation, and model measurement. Mortality risk was calculated using clinical, demographic, and laboratory data related to COVID-19. Data from 306 patients with cancer and COVID-19 were analyzed. Of these, 246 survived, and 60 died. The average age of the patients was 62.1, and 53.6% were male. A total of 60.1% of patients had comorbid conditions. 81.4% had solid malignancies, and 18.6% had hematological malignancies. The best prediction model, in terms of performance metrics such as accuracy (85.86%), sensitivity (86.37%), specificity (85.92%), and F1-score (85.83%), was the random forest algorithm, which was found to be superior to other algorithms, and feature importance analysis was performed using this algorithm. In this analysis, the most important clinical and laboratory parameters determining mortality were ferritin, D-dimer, lactate dehydrogenase, lymphocyte count, C-reactive protein, neutrophil count, lactate, neutrophil-to-lymphocyte ratio, shortness of breath, fever, and loss of taste and smell, which were shown to contribute significantly to model performance. Based on these findings, reliable classification models can be developed using ML methods for cancer patients with COVID-19, and decision-support modules can be created to guide clinicians and healthcare professionals in prioritizing patients based on their mortality risk.