A new logistic regression derived combined index for early prediction of in-hospital mortality in COVID-19 patients

BACKGROUND: While the type and the number of treatments for Coronavirus Disease 2019 (COVID-19) have sub-stantially evolved since the start of the pandemic a significant number of hospitalized patients continue to succumb. This requires ongoing research in the development and improvement of early risk stratification tools. METHODS: We developed a prognostic score using epidemiological, clinical, laboratory, and treatment variables col-lected on admission in 130 adult COVID-19 patients followed until in-hospital death (N.=38) or discharge (N.=92). Potential variables were selected via multivariable logistic regression modelling conducted using a logistic regression univariate analysis to create a combined index. RESULTS: Age, Charlson Comorbidity Index, P/F ratio, prothrombin time, C-reactive protein and troponin were the se-lected variables. AUROC indicated that the model had an excellent AUC value (0.971, 95% CI 0.926 to 0.993) with 100% sensitivity and 83% specificity for in-hospital mortality. The Hosmer-Lemeshow calibration test yielded non-significant P values (x2=1.79, P=0.99) indicates good calibration. CONCLUSIONS: This newly developed combined index could be useful to predict mortality of hospitalized COVID-19 patients on admission.