COPE

COPE predicts in-hospital mortality and 28-day ICU admission risk for emergency department patients with suspected COVID-19 using routinely collected clinical variables.

Key Features:

• Dataset: Retrospective analysis of 5,831 patients admitted to four Dutch hospitals between March and August 2020.
• Predictor selection: Predictors were selected based on prior literature and comprise readily obtainable patient characteristics, vital parameters, and blood tests.
• Predictors: Includes age, respiratory rate, C-reactive protein (CRP), lactic dehydrogenase (LDH), albumin, and urea as model inputs.
• Model type: Logistic regression models were used to generate predicted probabilities for outcomes.
• Shrinkage: Post-hoc uniform shrinkage was applied to model coefficients.
• Outcomes predicted: In-hospital death and ICU admission within 28 days of hospital admission.
• Validation: Temporal validation was performed on a separate cohort from September to December 2020 (second wave).
• Performance evaluation: Model performance was evaluated using Area Under the Receiver Operating Characteristic curve (AUC) and calibration plots, with mortality AUCs of 0.79–0.83 across hospitals and ICU admission AUCs of 0.81–0.84 in a subset with complete data.

Scientific Applications:

• ED risk stratification: Identifies high-risk patients who may require immediate ICU care among those presenting with suspected COVID-19.
• Triage support: Distinguishes intermediate-risk patients appropriate for ward admission and low-risk patients who may be considered for discharge.
• Modeling contribution: Provides a validated predictive model that emphasizes objective, quickly obtainable parameters to reduce bias associated with prior predictive models.

Methodology:

Retrospective analysis of 5,831 patients from four Dutch hospitals (March–August 2020); predictors selected from prior literature; logistic regression with post-hoc uniform shrinkage produced probability estimates for in-hospital death and ICU admission within 28 days; performance assessed by AUC and calibration plots with temporal validation on a cohort from September–December 2020.