Abstract
Age is a well-known risk factor to develop severe viral respiratory infections, including severe COVID-19. This study aimed to identify the biological alterations linked to severe disease in elderly patients with COVID-19. For this purpose, we employed a derivation cohort with 450 SARS-CoV-2 infected and unvaccinated patients admitted to hospital wards and a validation cohort with 244 SARS-CoV-2 infected and unvaccinated patients admitted to hospital Intensive Care Unit (ICU). Twenty-one biomarkers were measured in plasma samples from patients upon admission, including SARS-CoV-2 RNA, IgG antibodies, and protein biomarkers. Patient cohorts were divided into two groups based on age: adult (≤ 70 years old) and elderly (> 70 years old) patients. In the derivation cohort, 90-day mortality rate observed in the adult group was 6.0% whereas in the elderly group it rises to 31.6%, same trend was noticed regarding the validation cohort, with 11.2% versus 40.3% 90-day mortality rates for adult and elderly groups, respectively. The machine-learning framework XGBoost-SHAP, fed with the plasma biomarkers information, was used to profile an age-related host response to SARS-CoV-2 infection. Based on SHAP plot, elderly patients had a strong thrombo-inflammatory response profile (significantly elevated plasma levels of: lipocalin-2, endothelin-1, D-dimer) combined with deficient adaptive and cytotoxic antiviral responses. Model performance evaluated with the validation cohort confirmed the robustness and generalizability of the model developed (AUC = 0.710). In conclusion, the machine learning approach we built allowed us to identify the presence of a deranged host response in elderly patients with COVID-19 linked to poor viral control and increased mortality. Graphical Abstract