Impact of dietary component clusters identified by K-means++ on renal function decline in a Taiwanese cohort

Dietary intake influences renal health, but the impact of overall dietary patterns on renal outcomes remains unclear. The K-means++ algorithm, which improves the stability of traditional K-means clustering, may provide a robust approach for identifying real-world dietary behaviors. We analyzed 24,820 adults from the MJ Health Research Database in Taiwan. Dietary intake was assessed using a validated food frequency questionnaire, and patterns were derived with the K-means++ algorithm. Three clusters - non-healthy, normal, and healthy - were identified at baseline and follow-up. Renal outcomes were evaluated by estimated glomerular filtration rate (eGFR), with worsening defined as an annual decline >1 mL/min/1.73 m². At baseline, three distinct dietary clusters were identified. The non-healthy cluster was characterized by high intake of sugar-sweetened beverages, processed foods, and condiments, whereas the healthy cluster was characterized by greater consumption of nutrient-rich foods such as dairy, eggs, beans, and vegetables. Over follow-up, participants transitioning from a non-healthy to a healthy cluster exhibited a modestly lower risk of renal function worsening (adjusted odds ratio = 0.81, 95% CI: 0.67-0.99, p = 0.042). The K-means++ algorithm effectively identified meaningful dietary patterns and revealed clinically relevant associations with renal outcomes. These findings suggest that dietary modification may contribute to renal health preservation and demonstrate that the principal component analysis (PCA)-based clustering approach with K-means++ initialization provides a stable and appropriate framework for identifying dietary patterns in nutritional epidemiology.