Evidence supporting convalescent plasma (CP), one of the first investigational treatments for coronavirus disease 2019 (COVID-19), has been inconclusive, leading to conflicting recommendations. The primary objective was to perform a comparative effectiveness study of CP for all-cause, in-hospital mortality in patients with COVID-19.

The multicenter, electronic health records–based, retrospective study included 44,770 patients hospitalized with COVID-19 in one of 176 HCA Healthcare–affiliated community hospitals. Coarsened exact matching (1:k) was employed, resulting in a sample of 3774 CP and 10,687 comparison patients.

Examination of mortality using a shared frailty model, controlling for concomitant medications, date of admission, and days from admission to transfusion, demonstrated a significant association of CP with lower mortality risk relative to the comparison group (adjusted hazard ratio [aHR] = 0.71; 95% CI, 0.59–0.86; P < 0.001). Examination of patient risk trajectories, represented by 400 clinico-demographic features from our real-time risk model (RTRM), indicated that patients who received CP recovered more quickly. The stratification of days to transfusion revealed that CP within 3 days after admission, but not within 4 to 7 days, was associated with a significantly lower mortality risk (aHR = 0.53; 95% CI, 0.47–0.60; P < 0.001). CP serology level was inversely associated with mortality when controlling for its interaction with days to transfusion (HR = 0.998; 95% CI, 0.997–0.999; P = 0.013), yet it did not reach univariable significance.

This large, diverse, multicenter cohort study demonstrated that CP, compared with matched controls, is significantly associated with reduced risk of in-hospital mortality. These observations highlight the utility of real-world evidence and suggest the need for further evaluation prior to abandoning CP as a viable therapy for COVID-19.

This research was supported in whole by HCA Healthcare and/or an HCA Healthcare–affiliated entity, including Sarah Cannon and Genospace.