Risk Estimation of Severe COVID-19 Based on Initial Biomarker Assessment Across Racial and Ethnic Groups.

Context.—: Disease courses in COVID-19 patients vary widely. Prediction of disease severity on initial diagnosis would aid appropriate therapy, but few studies include data from initial diagnosis.

Objective.—: To develop predictive models of COVID-19 severity based on demographic, clinical, and laboratory data collected at initial patient contact after diagnosis of COVID-19.

Design.—: We studied demographic data and clinical laboratory biomarkers at time of diagnosis, using backward logistic regression modeling to determine severe and mild outcomes. We used deidentified data from 14 147 patients who were diagnosed with COVID-19 by polymerase chain reaction SARS-CoV-2 testing at Montefiore Health System, from March 2020 to September 2021. We generated models predicting severe disease (death or more than 90 hospital days) versus mild disease (alive and fewer than 2 hospital days), starting with 58 variables, by backward stepwise logistic regression.

Results.—: Of the 14 147 patients, including Whites, Blacks, and Hispanics, 2546 (18%) patients had severe outcomes and 3395 (24%) had mild outcomes. The final number of patients per model varied from 445 to 755 because not all patients had all available variables. Four models (inclusive, receiver operating characteristic, specific, and sensitive) were identified as proficient in predicting patient outcomes. The parameters that remained in all models were age, albumin, diastolic blood pressure, ferritin, lactic dehydrogenase, socioeconomic status, procalcitonin, B-type natriuretic peptide, and platelet count.

Conclusions.—: These findings suggest that the biomarkers found within the specific and sensitive models would be most useful to health care providers on their initial severity evaluation of COVID-19.