Abstract 13327: Development and Validation of a Clinical Risk Score to Predict the Critical Illness in Hospitalized Patients With COVID-19

Abstract

The Covid-19 pandemic has obligated healthcare systems to triage patients efficiently in order to utilize resources in the proper manner. Robust, validated clinical prediction tools are lacking that identify patients with coronavirus disease 2019 who are at the highest risk of mortality. A cohort of 661 positive COVID patients admitted to the Loyola hospital from 3/11/2020 to 10/12/2020 was used to validate previously published results with the composite outcome of ‘critically ill’, defined as invasive ventilation, ICU admission, or in-hospital death. Epidemiological, clinical, laboratory, and imaging variables ascertained at hospital admission were screened using Least Absolute Shrinkage and Selection Operator (LASSO) and logistic regression to construct a predictive risk score. Accuracy of the score was measured by the area under the receiver operating characteristic curve (AUC). Fifty-two (52) percent of our cohort had critical illness compared to 8% in the Liang paper (the first predictor model which was developed in China). Our population was older, had significantly higher abnormal chest x-ray, dyspnea, and number of comorbidities compared to their population. In addition, our population had higher values for NLR and LDH and lower direct bilirubin. Using the model coefficients presenting in Liang et al, the AUC was only 0.68 (95% CI: 0.64, 0.72), considerably lower than .88 presented in their publication. LASSO variable selection identified 8 variables (HR, SBP, fever, SpO2<90, LDH, ferritin, D dimer, CRP on admission) for a multivariable model. In the multivariable model: SpO2 <90, fever at admission, and increasing LDH were found to be statistically associated with our outcome. We finally added baseline predictors of age, sex, race, and number of comorbidities to the LASSO predictors which resulted in an AUC of 0.75 (95% CI: 0.71, 0.78). In our study, we found that three predictors (SpO2 <90, fever, and increasing LDH on admission) were selected by the LASSO analysis to construct a predictive nomogram. The application of our model would help clinicians make a prompt decision to optimize patient stratification management. However, this quantitative tool needs to be validated by further large-scale prospective studies.