Development and validation of age-specific predictive models on the risk of post-acute mortality within 1 year of COVID-19 infection

Journal article

Background: The existing risk prediction models for COVID-19 associated mortality have not considered the differences in risk factors across different age groups of patients. Aim: To develop age-specific prediction models to forecast the risk of all-cause mortality in patients recovering from COVID-19 infection. Design: Population-based, retrospective cohort study. Methods: Patients with COVID-19 between 1 April 2020 and 31 July 2022 survived beyond the acute phase of infection were stratified into separate age cohorts (<45, 45–64, ≥65) and followed-up for 1 year. Backward stepwise logistic regression and four statistical and machine learning algorithms were employed to develop age-specific models on the risk of post-acute mortality following COVID-19 infection, based on a comprehensive set of clinical parameters including demographics, COVID-19 vaccination status, pre-existing comorbidities and laboratory-test findings. Results: Of the 891 246 patients with COVID-19 identified, 13 578 (1.05%) died within 1 year of the index date. Age, COVID-19 vaccination status and history of acute respiratory syndrome prior infection were identified as predictors in the models for separate age groups. The model for patients aged ≥65 exhibited excellent prediction performance with an AUROC of 0.87 (95% CI: 0.87, 0.88), followed by the model for patients aged 45–64 [AUROC = 0.83 (95% CI: 0.81, 0.85)] and those aged <45 [AUROC = 0.79 (95% CI: 0.72, 0.86)]. Conclusion: The age-specific models accurately predicted the risk of post-acute mortality in their corresponding age groups of patients, providing valuable asset in optimizing clinical strategies and resource allocation in the management of the global burden of Long COVID.

Authors: Lam, ICH; Zhou, J; Liu, W; Man, KKC; Zhang, Q

• Publication status: Published
• Peer review status: Peer reviewed
• Publisher: Oxford University Press
• Journal: QJM: An International Journal of Medicine
• Volume: 119
• Issue: 2
• Pages: 117-128
• Publication date: 2025-09-22
• DOI: 10.1093/qjmed/hcaf218
• EISSN: 1460-2393
• ISSN: 1460-2725
• Language: English
• Keywords: Risk assessment; Predictive modelling; Asset (computer security); Risk management; Risk factor; Resource allocation; Disease