To evaluate the efficacy of corticosteroids on coronavirus disease 2019 (COVID-19) patients with different levels of disease severity. In our multicenter study, 543 patients with confirmed COVID-19 were classified as non-severe group and severe group, and then were compared respectively for all-cause mortality and length of hospital stay between those who received corticosteroids and not. By searching in PubMed, Web of Science, Embase, and CNKI, we identified 13 retrospective studies and 6 random control trials eligible for criteria of inclusion, and conducted comprehensive meta-analyses assessing the impacts of corticosteroids on mortality, length of stay, duration of RNA clearance and duration of fever. Our multicenter study demonstrated that low-dose corticosteroids can reduce mortality in the multivariable Cox regression analysis for severe patients (p = .03), while presented no influence in univariable analysis for non-severe patients (p = .14). From multivariable analyses, patients with corticosteroids in non-severe group had longer duration of hospitalization (p = .003), but did not in severe group (p = .18). Moreover, for severe patients, corticosteroids can evidently shorten duration of fever. The same results were summarized in the meta-analyses supplemented with the result that corticosteroids delayed viral clearing in non-severe patients. Corticosteroids should be considered based on patient's condition. For patients with non-severe COVID-19, corticosteroid was not recommended as a routine therapeutic initiative as that presented prolonged duration of hospitalization and delayed viral clearing, as well as no positive impact on prognosis. While low-dose corticosteroids may benefit patients with severe COVID-19 for it can manifestly lower risk of death and improve the clinical status to some extent.