2021 FDA Science Forum
Categorization of COVID-19 Severity to Determine Mortality Risk
- Authors:
- Center:
-
Contributing OfficeCenter for Drug Evaluation and Research
Abstract
Background
Respiratory support (supplemental oxygen or non-invasive ventilation, O2/NIV, and invasive mechanical ventilation, IMV) in hospitalized patients with COVID-19 appears to be a critical indicator of disease severity when determining inpatient treatment effectiveness. Algorithms for COVID-19 severity using inpatient real-world data (RWD) are needed.
Purpose
Develop an algorithm to determine use of O2/NIV and IMV in inpatient RWD and estimate the risk and incidence rate (IR) of death among subgroups to confirm that patients with greater disease severity at admission are at higher risk for severe outcomes.
Methodology
Using HealthVerity healthcare claims data (April 2020-January 2021), we selected patients hospitalized with a COVID-19 diagnosis or positive SARS-CoV-2 laboratory results and developed an algorithm to categorize mutually exclusive COVID-19 severity levels at admission: no O2 O2/NIV, and IMV. The algorithm included procedure and diagnosis codes indicative of need for respiratory support or procedure-related, and revenue codes indicating O2 use. Patients were followed from admission until death, discharge, or 28-days to report risks, IR, and corresponding 95% confidence intervals overall and for each severity level. Trends for heterogeneity in risk/IR of death across levels were evaluated.
Results
Among 88,967 patients, the COVID-19 severity algorithm categorized 33,579 (37.7%%) as no O2, 47,691 (53.6%) as O2/NIV, and 7,697 (8.7%) as IMV; among 11,010 patients who died, 1,294 had no O2, 6,060 had O2/NIV, and 3,656 had IMV at admission. The risk of death was 12.4% (12.2-12.6%) with an IR per 1000 person-days of 15.75 (15.46-16.05) over a median (IQR) of 5 (3-10) days. The risk among patients with no O2, O2/NIV, and IMV increased with increasing severity level [3.9% (3.7-4.1%); 12.7% (12.4-13.0%); 47.5% (46.4-48.6%); p<0.001]. A similar trend was found for IR per 1000 person-days [6.44 (6.09-6.79); 15.67 (15.28-16.07); 32.85 (31.79-33.92); p<0.001], despite an increase in median (IQR) follow-up days [4 (2-7); 6 (4-10); 14 (6-23)].
Conclusion
Although performance remains to be validated, we observed a positive association between algorithm-defined severity level and 28-day mortality risk and IR, which provides assurance that these severity levels can be used for confounding control in treatment effectiveness studies.
