Development and External Validation of a Prognostic Prediction Model for Hospitalization in SARS-CoV-2-Infected Ambulatory Patients.

Background: Accurately predicting hospitalization risk in severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) positive outpatients is critical for triage and resource planning. We developed and externally validated a clinical prediction model for 30-day COVID-19 hospitalization using data from symptomatic adults who tested positive in the outpatient setting. The derivation cohort included 22 859 patients from 185 outpatient clinics in a large Utah health care network between December 2021 and March 2023, during Omicron variant predominance. Among these patients, 281 (1.2%) were hospitalized for COVID-19 within 30 days. Methods: We fit random forest and multivariable logistic regression models incorporating clinical variables (vital signs, comorbidities, and vaccination status), social determinants of health, seasonality, and air quality indices. We externally validated our model using data from 10 670 patients in a Pennsylvania healthcare network who tested positive between October 2021 and November 2022, among whom 166 (1.6%) were hospitalized. Results: In cross-validation, a random forest model including only clinical predictors performed similarly to an expanded model that also included social, environmental, and seasonal predictors (area under the receiver operator characteristic curve [AUC]: 0.83, 95% CI: 0.77-0.88 for both). A parsimonious logistic regression model with only 3 clinical predictors (respiratory rate, age, and pulse oximetry) achieved an AUC of 0.79 (95% CI: .72-.86) on internal validation and an AUC of 0.88 (95% CI: .85-.90) on external validation. Calibration was robust, and decision curve analysis demonstrated clinical utility at low-risk thresholds. Conclusions: We conclude that a parsimonious 3-predictor model can effectively stratify hospitalization risk in SARS-CoV-2 positive outpatients, offering a practical tool to support clinical decision-making and optimize resource allocation during current and future COVID-19 surges.