In a significant advancement in the field of gerontology, researchers have employed machine learning techniques to distinguish the factors contributing to healthy aging from those associated with chronic disease risks. This study, detailed in the journal Nature Aging, represents a pivotal step in understanding the complex interplay between aging and age-related diseases.
The research team, analyzing health data from a vast cohort of 4.57 million individuals aged 30 to 85 years, hailing from Israel, the United Kingdom, and the United States, has successfully identified key markers that could serve as early indicators of healthy aging. Â
The study’s foundation lies in the “geroscience hypothesis,” which posits that targeting universal aging processes could not only promote healthy aging but also reduce the prevalence of age-related diseases such as type 2 diabetes mellitus, cardiovascular disease, chronic kidney disease, liver disease, and chronic obstructive pulmonary disease.
The challenge, however, has been the co-occurrence and correlation of these diseases with aging, making it difficult to model causality. This is where the study’s use of machine learning comes into play, offering an unbiased approach to dissecting the relationship between healthy aging and age-related diseases. Â
The researchers utilized electronic health records (EHRs) to capture the health trajectories of patients over time. Despite the limitations of existing data, which spans up to 20 years, the study marks a significant leap in our understanding of the relationship between aging, disease, and disease risk. Prior studies in this domain lacked the use of a longitudinal model, a gap that this research aimed to fill.
By developing a machine learning-based model, the team was able to identify predictive clinical markers for disease-free healthy aging and revisited the heritability and genetic associations of phenotypes linked to longevity. The methodology involved developing a machine learning model using a three-year history of patients aged above 80 years and analyzing laboratory tests correlating with longevity.
The model assessed longevity potential across ages by inferring longitudinal trajectories using partial patient histories. This approach allowed for the determination of a longevity potential score for each age, predicting five-year mortality or a change in longevity potential. To understand how lifelong disease predisposition potentially affected the longevity score, the researchers implemented an extended disease risk Markov model using disease-onset data.
They investigated the physiological processes underlying longevity potential in very healthy individuals using clinical markers over a more than 10-year follow-up. The model was then tested on the UK Biobank and National Health and Nutrition Examination Survey population databases. The results were revealing. The three-year history model could discern a detailed spectrum of risk levels, highlighting significant prognostic differences even within the top 4% of healthy patients.
Laboratory tests identified markers like red blood cell distribution width, C-reactive protein, and albumin as continually associated with prognosis. The model provided a generalizable metric for health, classifying patients as healthy or unhealthy and encouraging the use of models that quantitatively track changes in health potential. Â
Clinical markers contributing to the longevity score varied across ages. For instance, alkaline phosphatase impacted younger adults, while glucose and cholesterol seemed to affect mid-adulthood. In older ages, albumin and red blood cell distribution width were more influential. Key features like overweight, blood sugar and cholesterol played a significant role in predicting lifelong disease risk.
Markers of chronic disease risk were consistently low in very healthy individuals. A high longevity score was indicated by low levels of neutrophils, alkaline phosphatase, and the ratio of microcytic and hypochromatic red blood cells, as well as medium levels of body mass index, creatinine, and liver enzymes. Â
The study found that the model’s predictive power increased with age, particularly in identifying high-risk individuals for diseases like type 2 diabetes at ages 50–60 due to improved sensitivity from routine tracking. The estimated lifelong disease predispositions were strongly associated with each other and correlated with the longevity score. However, a subset of individuals exhibited variation in longevity potential despite low disease risk. Â
The longevity scores were robust across Israeli, US, and UK populations, demonstrating significant predictive power for longevity in individuals without known predisposition to diseases. Furthermore, the degree of disease predisposition varied between populations at age 50. Parents of highest longevity scoring-individuals were found to have a one-year increase in lifespan. The study suggests that genetic variation may also contribute to longevity. Â
In conclusion, this study marks a significant stride in our understanding of the interplay between aging and major chronic diseases. It paves the way for comprehensive, longitudinal models to replace static representations of healthy aging and common diseases. Further research is required to quantify a “healthy state” and investigate the physiological processes underlying the disease-related findings highlighted in the study. This research not only enhances our understanding of aging but also opens new avenues for promoting health and longevity.Â
Journal ReferenceÂ
Cohen, N. M., Lifshitz, A., Jaschek, R., Rinott, E., Balicer, R., Shlush, L. I., … Tanay, A. (2023). Longitudinal machine learning uncouples healthy aging factors from chronic disease risks. Retrieved from https://www.nature.com/articles/s43587-023-00536-5
