Integration of the circulating miRNome and clinical information to predict 90-day mortality in elderly COVID-19 subphenotypes

Elderly patients with COVID-19 exhibit high mortality rates. We assessed whether microRNA (miRNA) profiling provides prognostic information in this population. This multicenter study included hospitalized COVID-19 patients aged ≥65 years (n=763). Clinical subphenotypes were identified using clinical data through k-prototypes algorithm. Plasma miRNome was profiled using qPCR. Clinical and miRNA-based models predicting 90-day mortality were developed using VSURF. Median age was 79 years, 44.0% were female and 90-day mortality was 26.7%. Although 13 candidate miRNAs were identified during screening (n=39), none were associated with mortality in the full derivation cohort (n=340), leading to a subphenotype-stratified analysis. Three subphenotypes (eCOVID-1, -2 and -3) with distinct clinical features and mortality risks were identified. In eCOVID-2, miR-106b-3p was the strongest predictor of 90-day mortality and improved model discrimination beyond clinical variables alone (AUC 0.75 vs. 0.66). Plasma miRNAs may provide complementary prognostic information when integrated with clinical variables in elderly COVID-19 subphenotypes.