Advancing nutrition risk assessment in middle-aged and older individuals with diverse food cultures: A data-driven personalized approach to predict incident hypertension, diabetes and mortality

Background: Personalized nutrition risk assessment is crucial in addressing the association between healthy dietary habits across the life course and the development of disease, functional capacity, and healthy aging, as specific dietary pattern recommendations may not be suitable for diverse food cultures. Objective: To develop a data-driven, personalized nutrition risk assessment algorithm linked to incident hypertension, diabetes, and all-cause mortality utilizing the food frequency questionnaire among middle-aged and older individuals. Methods: A retrospective, population-based cohort study conducted between 1999 and 2015 utilized the nationally representative Taiwan Longitudinal Study on Aging (TLSA) survey to examine personalized dietary risk clusters and their associations with health outcomes. Latent class analysis was performed to derive the dietary diversity clusters among community-dwelling middle-aged and older individuals. Outcomes were defined as new-onset hypertension, diabetes mellitus and all-cause mortality at 4-, 8-, 12- and 16-year follow-ups. Results: Data from 1,811 participants (58.14% males, 43.90% aged 50−64 years) showed that around one-third of participants reported being illiterate, 21.98% widowed, and 51.46% engaging in regular physical exercise. Four dietary diversity clusters were identified: “least diverse”, “fish and meat”, “dairy, fruit, and vegetable”, and “most diverse”. The “most diverse” cluster was characterized by a high consumption of protein-rich foods, while the “dairy, fruit, and vegetable” cluster had the highest consumption of dairy products and beans/legumes. The “least diverse” cluster had the lowest intake of protein-rich foods, and dark-colored vegetables and fruits. The “most diverse” cluster had a significantly lower risk of hypertension development at the 4-year (aOR 0.58; p < 0.02) and 8-year (aOR 0.57; p < 0.01) follow-up and diabetes at the 4-year (aOR 0.44; p < 0.03) follow-up. Participants in the “most diverse” clusters exhibited lower risks of 8-year, 12-year, and 16-year mortality than those in the “least diverse” cluster (aOR 0.67, p < 0.05; 0.67, p < 0.03; and 0.50, p < 0.01, respectively). Conclusion: The personalized nutrition risk assessment algorithm from the food frequency questionnaire can effectively stratify personal health risks among diverse middle-aged and older individuals, making it a valuable tool in lifestyle modification and intervention studies.