Sign in Subscribe
Mortality

Machine Learning Predicts Survival in Centenarians with Moderate Accuracy

Strategic Insights

— 2 min read

A recent study from the University of Hong Kong used machine learning (ML) to predict mortality in centenarians, a population that has seen significant increases worldwide. The research explored the feasibility of using electronic health records (EHRs) to identify key survival determinants and predict mortality in centenarians. The study analyzed 9,718 centenarians from the population-based EHR database in Hong Kong from 2004 to 2018, utilizing 82 predictors including demographics, diagnoses, prescriptions, and laboratory results.

Key Takeaways:

  • The study used a machine learning approach to predict mortality in centenarians, analyzing 9,718 participants from the population-based EHR database in Hong Kong.
  • The models performed well in predicting 1-year and 2-year mortality, but showed poor calibration for 5-year mortality.
  • The top 3 predictors of mortality included lower albumin levels, more frequent hospitalizations, and higher urea levels.
  • The researchers found that models including these predictors consistently outperformed comorbidity and frailty scores for mortality prediction among oldest-old adults.
  • Further research is needed to determine whether mortality predictors differ across age in the oldest-old population.
  • The University of Hong Kong researchers collaborated with other experts from the Li Ka Shing Faculty of Medicine, University of Hong Kong, and the Department of Pharmacology and Pharmacy, University of Hong Kong.
  • This research was part of a broader effort to understand the challenges and opportunities in aging populations worldwide.

Statistics:

  • The study analyzed 9,718 centenarians from the population-based EHR database in Hong Kong from 2004 to 2018.
  • The researchers used 82 predictors, including demographics, diagnoses, prescriptions, and laboratory results, to train machine learning models.
  • The machine learning models performed well in predicting 1-year and 2-year mortality, with area under the receiver operating characteristic curve (AUROC) values of 0.707 (95% CI = 0.685-0.730) and 0.704 (0.686-0.723), respectively.
  • The researchers identified the top 3 predictors of mortality as lower albumin levels, more frequent hospitalizations, and higher urea levels.

Sources:

  • Machine learning prediction of survival in centenarians after age 100: A retrospective, population-based cohort study. The Journals of Gerontology, Series A, 2025.
  • The Journals of Gerontology, Series A, Oxford Univ Press Inc, Journals Dept, 2001 Evans Rd, Cary, NC 27513, USA.
  • Noel C. Yue, Dept. of Pharmacology and Pharmacy, Li Ka Shing Faculty of Medicine, University of Hong Kong, Hong Kong, People's Republic of China.
  • NewsRx. Reports from University of Hong Kong Advance Knowledge in Machine Learning (Machine learning prediction of survival in centenarians after age 100: A retrospective, population-based cohort study). Journal of Engineering. October 20, 2025; p 2311.

Read more