Data synthesis
We systematically searched PubMed, Scopus and Web of Science, from 1st January 2020 to 15th May 2020, for observational studies of patients admitted to hospital with COVID-19. Meta-analysis was performed using random-effects modeling. A total of 83 eligible studies with 78,874 hospitalized patients with laboratory-confirmed COVID-19 were included. The pooled prevalence of established diabetes was 14.34% (95% CI 12.62–16.06%). However, the prevalence of diabetes was higher in non-Asian vs. Asian countries (23.34% [95% CI 16.40–30.28] vs. 11.06% [95% CI 9.73–12.39]), and in patients aged ≥60 years vs. those aged <60 years (23.30% [95% CI 19.65–26.94] vs. 8.79% [95% CI 7.56–10.02]). Pre-existing diabetes was associated with an approximate twofold higher risk of having severe/critical COVID-19 illness (n = 22 studies; random-effects odds ratio 2.10, 95% CI 1.71–2.57; I2 = 41.5%) and ~threefold increased risk of in-hospital mortality (n = 15 studies; random-effects odds ratio 2.68, 95% CI 2.09–3.44; I2 = 46.7%). Funnel plots and Egger's tests did not reveal any significant publication bias.