Abstract Background Activins are members of the transforming growth factor-β superfamily implicated in the pathogenesis of several immunoinflammatory disorders. Based on our previous studies demonstrating that overexpression of activin-A in murine lung causes pathology sharing key features of coronavirus disease 2019 (COVID-19), we hypothesized that activins and their natural inhibitor follistatin might be particularly relevant to COVID-19 pathophysiology. Methods Activin-A, activin-B, and follistatin were retrospectively analyzed in 574 serum samples from 263 COVID-19 patients hospitalized in 3 independent centers, and compared with demographic, clinical, and laboratory parameters. Optimal scaling with ridge regression was used to screen variables and establish a prediction model. Result The activin/follistatin axis was significantly deregulated during the course of COVID-19, correlated with severity and independently associated with mortality. FACT-CLINYCoD, a scoring system incorporating follistatin, activin-A, activin-B, C-reactive protein, lactate dehydrogenase, intensive care unit admission, neutrophil/lymphocyte ratio, age, comorbidities, and D-dimers, efficiently predicted fatal outcome (area under the curve [AUC], 0.951; 95% confidence interval, .919−.983; P <10−6). Two validation cohorts indicated similar AUC values. Conclusions This study demonstrates a link between activin/follistatin axis and COVID-19 mortality and introduces FACT-CLINYCoD, a novel pathophysiology-based tool that allows dynamic prediction of disease outcome, supporting clinical decision making.