Identification of Blood Biomarkers of a Healthy Dietary Pattern As Facilitated by Cluster Analysis in Patients from the MEDDINI Study: a Pilot Randomised Trial

Dietary biomarkers may help objectively assessing dietary pattern adherence. This study performed K-means clustering analysis on quantitative food diary data from a dietary intervention study. Standardised dietary data (134 food diaries) from 57 participants were K-means clustered stepwise until fully optimised and cross-validated. The primary endpoint was to develop distinct dietary clusters and to evaluate the performanceof 90 plasma metabolites. The secondary endpoint was to analyse the biomarker-food groups relationships from those distinct dietary patterns. The final two cluster models comprised of 6 specific food types. Cluster 1 included participants with higher intake of fruit and vegetables, legumes, fish and whole grain cereals, and lower intake of meat and sweet foods than Cluster 2. Ten plasma metabolites significantly differed between the clusters (p < 0.05; q < 0.05) with reasonable biomarker performance (receiver operating characteristic (ROC): 0.64-0.72). Docosahexaenoic acid (DHA), eicosapentaenoic acid (EPA), α-linolenic acid, citric acid and vitamin C were significantly higher in Cluster 1, whereas adrenic acid, osbond acid, cholesterol, dihomo-γ-linolenic acid (DGLA) and triglycerides were higher in Cluster 2. Five additional metabolites also showed significant differences (p < 0.02; q < 0.11) and were included: palmitic acid, tyrosine, β-carotene, α-carotene and betaine. The DHA-to-Osbond acid ratio was an optimal indicator distinguishing healthy from unhealthy dietary patterns (ROC: 0.78). Combining clustering and metabolite profiling methods effectively identifies biomarkers of particular dietary patterns and highlights several robust food-metabolite correlations.