Bayesian Network Multimorbidity Models in COVID-19 Mortality

Abstract

Clusters of multimorbidities indicate common molecular and physiological causal mechanisms, which can help to identify promising targets for novel drugs and treatments. We analyzed the comorbidities for COVID-19 death cases in 2020 in Hungary (9537 cases). The daily, public reports of anonymous cases contained age, sex and free-text description of comorbidities for each officially COVID-19 deceased. We cleaned, clinically encoded and aggregated the reported morbidities resulting in 22 medical categories, which are also used in a reference national database with normal Hungarian population. Using Bayesian networks, we analyzed the dependency models of reported comorbidities in different population cohorts, such as in the first wave (2020 March-August) versus in the second wave (September- December) or above versus below the expected life expectancy (76 years). To explore latent morbidities, not yet manifested or not reported, but potentially already influencing COVID-19 mortality risk, we predicted excess multimorbidity for deceased. Our results suggest significant excess multimorbidity, which also calls for an improved and more transparent data dissemination policy. Multimorbidity models, especially models also based on multimorbidity data from normal population could provide vital risk prediction for an improved health care and vaccination programs.