Author(s): GARCIA-AGUIRRE, M.1,2,3, BALTASAR-FERNANDEZ, I.1,2,3,4, ALCAZAR, J.1,2,3, GUADALUPE-GRAU, A.1,2,3, GALEANO, J.5, ALFARO-ACHA, A.2,3,6, ARA, I.1,2,3, RODRIGUEZ-MAÑAS, L.2,7, ALEGRE, L.M.1,2,3, GARCIA-GARCIA, F.J.2,3,6, Institution: UNIVERSIDAD DE CASTILLA LA MANCHA, Country: SPAIN, Abstract-ID: 1261

INTRODUCTION:
Aging constitutes a multifaceted and inevitable process characterized by a gradual decline in physiological functions and an escalation in health risks. Despite a significant increase in life expectancy, the achievement of healthy aging has not kept pace accordingly. This emphasizes the importance of identifying biomarkers associated with healthy aging. Thus, this study aimed 1) to cluster the study population into healthy and unhealthy aging by using machine learning methodologies, 2) to identify straightforward predictive variables for these clusters, and 3) to analyze the relationship between clusters indicative of unhealthy aging and adverse health events and conditions.
METHODS:
A prospective cohort study that included 1852 older adults (>65 years) from the Toledo Study for Healthy Aging. A total of 366 variables (including physical, cognitive, and biochemical outcomes among others) were assessed using machine learning k-means clustering. Gradient boosting, area under the curve (AUC) values, and classification accuracy (CA) were used to evaluate the predictive capacity of healthy and unhealthy aging clusters. Logistic regression analyses adjusted for age, educational level, and comorbidities were used to assess the relationship between the healthy and unhealthy aging clusters and the risk of experiencing cognitive impairment, frailty, hospitalizations, and all-cause mortality over the subsequent 6 years.
RESULTS:
Two clusters of healthy (C1) and unhealthy aging (C2) were identified in both men (C1= 243; C2= 565) and women (C1= 342; C2= 702). Among the 366 variables tested, relative sit-to-stand power, loss of memory in the last two years, and habitual gait speed collectively exhibited excellent predictive capacity for healthy and unhealthy aging in both older men and women according to the AUC values (0.850 in men and 0.875 in women) and CA (0.819 in men and 0.792 in women). After 6 years of follow-up, older adults in the unhealthy aging cluster exhibited a higher risk of experiencing cognitive impairment [OR (95%CI) = 4.7 (1.8, 12.3) in men and 2.9 (1.3, 6.3) in women], frailty [OR (95%CI) = 2.8 (1.2, 6.8) in men and 7.1 (3.3, 15.4) in women], hospitalizations [OR (95%CI) = 1.8 (1.2, 2.6) in men and 2.0 (1.4 , 2.8) in women] and all-cause mortality [OR (95%CI) = 2.0 (1.4, 2.9) in men and 2.2 (1.5, 3.1) in women] compared to those older individuals in the healthy aging cluster.
CONCLUSION:
The machine-learning method identified that relative sit-to-stand power, memory loss, and habitual gait speed were strong predictors of healthy and unhealthy aging in older adults. Those older people within the unhealthy aging cluster exhibited a 2 to 7 times higher likelihood of experiencing an adverse event or condition compared to those in the healthy aging cluster after 6 years of follow-up, emphasizing the importance of assessing and monitoring these variables to ensure healthy aging.