Romanian Society of Pharmaceutical Sciences
 

« Back to Farmacia Journal 1/2026

RISK STRATIFICATION OF COVID-19 SEVERITY IN CANCER PATIENTS USING MACHINE LEARNING ALGORITHMS

ELENA-VICTORIA MANEA (CARNELUTI) 1,#, STEFAN-ALEXANDRU ARTENE 1,#, ANDREEA DENISA HODOROG 2, MIRCEA-SEBASTIAN SERBANESCU 3 , IRINA ANNA-MARIA STOIAN 4 , ILONA GEORGESCU 1,5, CRISTINA PANĂ 6, STEFANA OANA POPESCU 1*, ANICA DRICU 4

1 Department of Biochemistry, Faculty of Medicine, University of Medicine and Pharmacy of Craiova, Str. Petru Rareș nr. 2-4, 710204 Craiova, Romania
2 Department of obstetrics and gynecology, Clinical Hospital Mioveni, Bd. Dacia, nr. 131A, 115400 Mioveni, Romania
3 Department of Medical Informatics and Biostatistics, Faculty of Medicine, University of Medicine and Pharmacy of Craiova, Str. Petru Rares nr. 2-4, 710204 Craiova, Romania
4 Department of Biochemistry, Faculty of Medicine, “Carol Davila” University of Medicine and Pharmacy, Bd Eroii Sanitari 8, 050474 Bucharest, Romania
5 Department of Infectious Diseases, “Victor Babeș” Hospital of Infectious Diseases and Pneumophtisiology, Str. Calea București, nr. 126, 200525 Craiova, Romania
6 Department of Mechatronics and Robotics, Faculty of Automatics, Computers and Electronics, University of Craiova, Bd. Decebal, nr.107, 200776 Craiova, Romania

Download Full Article PDF

This study aimed to explore the use of machine learning models for classifying COVID-19 severity in patients with cancer, based on clinical, radiological, and demographic data collected at hospital admission. A cohort of 235 oncological patients with confirmed SARS-CoV-2 infection was analysed using a four-category severity framework (mild, moderate, severe, critical). Several machine learning approaches were evaluated, including linear, non-linear, and ensemble-based models within an error- correcting output codes (ECOC) framework. Performance was assessed using accuracy and one-vs-rest ROC analysis with area under the curve (AUC). Internal validation was performed through cross-validation. Model performance varied across approaches and severity classes. Linear methods showed limited separation between neighbouring categories, while non-linear and ensemble models provided more consistent identification of advanced disease. Ensemble bagging achieved the most stable performance across classes, whereas RUSBoost showed less consistent results for intermediate stages. Machine learning models may assist in severity stratification in oncological COVID-19 patients; however, results require cautious interpretation and external validation before clinical implementation.