COVID-19 Pandemic
Latent Class and K-modes Analyses of Comorbidities of COVID-19 Patients According to CO-RADS Classification Ayse Ulgen* Ayse Ulgen Ulgen Ulgen (1) Department of Mathematics and Physics, School of Science and Technology, Nottingham Trent University, Nottingham N11 8NS, UK; (2) Department of Biostatistics, Faculty of Medicine, Girne American University, Karmi 99428, Cyprus
Background: This pioneering study compares Latent Class and K-modes analyses for stratifying COVID-19 patients by risk, aiming to identify high-risk individuals based on COVID-19 Reporting and Data System (CO-RADS) categories and comorbidities. CO-RADS is a standardized chest CT scoring system that classifies the likelihood of COVID-19 lung involvement from 1 (very low) to 6 (confirmed). Methods: This retrospective study included 600 patients diagnosed with COVID-19, and the analysis utilized CO-RADS levels along with comorbidity variables such as hypertension, diabetes, COPD, etc. for each patient. The objective of the study is to compare the output from Latent class analysis and K-modes clustering. Model performance was measured using AIC/BIC values and the kappa statistic. Results: Both latent class and K-modes analyses consistently identified cardiometabolic disorders and chronic conditions, particularly diabetes mellitus, hypertension, hyperlipidemia, chronic heart and kidney diseases, and respiratory comorbidities such as COPD and asthma, as key characteristics distinguishing high-mortality risk groups among COVID-19 patients (p <0.001). Mortality analyses revealed comparable class distinctions across methods, with Latent Class Analysis indicating a 39.4% mortality rate in the low-risk group and 23.4% in the high-risk group, while K-modes clustering identified similar patterns with mortality rates of 40.2% and 21.3%, respectively. Latent Class Analysis formed homogeneous subgroups using parametric distributions, while K-modes handled heterogeneous data by optimizing intra- and inter-cluster variability. Their 62% concordance underscores complementary strengths and combined utility in clinical datasets. Conclusion: This study integrates advanced statistical techniques to develop a scalable model for understanding and managing complex diseases like COVID-19, offering a robust framework for personalized risk stratification across diverse conditions.
