Objective: To determine whether a risk prediction tool developed and implemented in March 2020 accurately predicts subsequent hospitalizations.
Design: Retrospective cohort study, enrollment from March 24 to May 26, 2020 with follow-up calls until hospitalization or clinical improvement (final calls until June 19, 2020)
Setting: Single center telemedicine program managing outpatients from a large medical system in Atlanta, Georgia
Participants: 496 patients with laboratory-confirmed COVID-19 in isolation at home. Exclusion criteria included: (1) hospitalization prior to telemedicine program enrollment, (2) immediate discharge with no follow-up calls due to resolution.
Exposure: Acute COVID-19 illness
Main Outcome and Measures: Hospitalization was the outcome. Days to hospitalization was the metric. Survival analysis using Cox regression was used to determine factors associated with hospitalization.
Results: The risk-assessment rubric assigned 496 outpatients to risk tiers as follows: Tier 1, 237 (47.8%); Tier 2, 185 (37.3%); Tier 3, 74 (14.9%). Subsequent hospitalizations numbered 3 (1%), 15 (7%), and 17 (23%) and for Tiers 1-3, respectively. From a Cox regression model with age ≥ 60, gender, and self-reported obesity as covariates, the adjusted hazard ratios using Tier 1 as reference were: Tier 2 HR=3.74 (95% CI, 1.06-13.27; P=0.041); Tier 3 HR=10.87 (95% CI, 3.09-38.27; P<0.001). Tier was the strongest predictor of time to hospitalization.
Conclusions and Relevance: A telemedicine risk assessment tool prospectively applied to an outpatient population with COVID-19 identified both low-risk and high-risk patients with better performance than individual risk factors alone. This approach may be appropriate for optimum allocation of resources.