Infectious Disease
Mediation Analysis for Calcium, Vitamin-D, and COVID-19 Severity Ayse Ulgen* Ayse Ulgen Sirin Cetin Wentian Li
Background: Hypocalcemia and vitamin-D deficiency are known to be associated with a worst outcome of COVID-19 patients, but their causal direction is unclear. Methods: In this paper, blood samples of more than 1300 COVID-19 patients from Tokat, Turkey are analyzed. A new strategy is adopted; instead of inferring a simple causal model, the task to estimating causal path proportions with a given assumed causal model is assumed. For a given causal model, if a causal path is not statistically significant and/or if a causal path has a very low proportion, a particular causal model is unlikely to be true. Therefore, choices among all possible causal models may be narrowed. With three variables: vitamin-D and calcium level, and indicator variable on if a patient is in critical state or not, standard mediation analysis is used to investigate the possible relationship among them. Since age contributes to almost every risk factors as well as to COVID-19 prognosis, age (and gender) are used as co-variates in the mediation analysis. Results: 1) Association between being critically-ill by COVID-19 and calcium (but not vitamin-D) is independent from osmolality: Logistic regressions show that osmolality, calcium, and vitamin-D are all significantly associated with critically ill status conditional on age and gender, with p-values 2.6E-7 (osmolality), 5.3E-10 (calcium), 2.7E-4 (vitamin-D) and 2E-6 (log-vitamin-D. 2) Reduction of vitamin-D level at different age groups: It can be seen from Fig.1 that the reduction of vitamin-D scale is less in the critical group for older age group (e.g. 80 or above), compared to a relatively younger group (e.g., in the age of 60s). Conclusions: Single regression analyses is not adequate for inferring causality. Assuming that cause is the independent variable and effect is the dependent variable in a regression analysis is only an assumption and may not reflect reality.
