We used data-driven approaches to identify independent diet exposures among 45 candidate variables, for which we then probed cross-sectional associations with cardiometabolic risk (CMR). We derived average daily caloric intake and macronutrient composition, daily meal frequencies, and irregularity of energy and macronutrient intake from 7-day food diaries in the Airwave Health Monitoring Study participants (N = 8090). We used K-means and hierarchical clustering to identify non-redundant diet exposures with representative exposures for each cluster chosen by silhouette value. We then used multi-variable adjusted logistic regression to estimate prevalence ratios (PR) and 95% confidence intervals (95%CI) for CMR (≥3 criteria: dyslipidemia, hypertension, central adiposity, inflammation and impaired glucose control) across diet exposure quartiles. We identified four clusters: i) fat intake, ii) carbohydrate intake, iii) protein intake and intake regularity, and iv) meal frequencies and energy intake. Of these clusters, higher carbohydrate intake was associated with lower likelihood of CMR (PR = 0.89, 95%CI = 0.81-0.98; ptrend = 0.02), as was higher fiber intake (PR = 0.76, 95%CI = 0.68-0.85; ptrend < 0.001). Higher meal frequency was also associated with lower likelihood of CMR (PR = 0.76, 95%CI = 0.68-0.85; ptrend < 0.001). Our results highlight a novel, data-driven approach to select non-redundant, minimally collinear, primary exposures across a host of potentially relevant exposures (including diet composition, temporal distribution, and regularity), as often encountered in nutritional epidemiology.
Keywords: Airwave Health Monitoring Study; body mass index; cluster; diet patterns; food intake regularity; meal frequency; nutrient intake.