A-DIET used existing data, collected new data and developed new software packages. Three human intervention studies were completed: (1) a Discovery study (n=20) whose objectives were to identify biomarkers of specific foods, (2) the Validation study’s (n=61) aim was to validate the putatively identified biomarkers from the Discovery study and (3) the CONFIRM study (n=175) which was used to determine dietary patterns. Analysis of the data from existing data and the new data from these studies has resulted in a number of key publications.
We demonstrated that a urinary biomarker could determine food intake in a number of examples. Here I highlight the example of citrus intake. Calibration curves were constructed with the urinary proline betaine concentration against the known orange juice intake (g/day). Excellent agreement was observed between estimated intakes and actual intakes with the agreement assessed through Bland and Altman analysis. A correlation of 0.92 was reported between actual intake and predicted intake again highlighting the high level of agreement. Further to this, the ability of the biomarker to estimate intake was tested in an independent cross sectional. Using the calibration curves determined in the controlled intervention study the intake (g/day) was estimated from the urinary concentration of proline betaine. There was excellent agreement between the self-report intake and the estimated intake from the biomarker. The significance of this lies with the fact that it clearly demonstrates how biomarkers may be used in a larger cohort/population setting to estimate food intake. Following, this we applied a similar method to other foods.
Dietary patters: we developed a model based on urinary metabolomics data that could classify individuals into four dietary patterns: moderately unhealthy, convenience, moderately healthy, and prudent. The moderately unhealthy and convenience patterns had lower adherence to the alternative healthy eating index (AHEI) and the alternative mediterranean diet score (AMDS) compared to the moderately healthy and prudent patterns (AHEI = 24.5 and 22.9 vs 26.7 and 28.4 p < 0.001). The stability of participants’ dietary pattern membership ranged from 25.0% to 61.5%. This work is important as it demonstrates the potential of classifying individuals into dietary patterns based on urinary metabolites data only.
We developed two software tools (Shiny Apps) and associated R packages. Bio-intake; biomarker guided dietary intake allows users to upload mean daily self-reported citrus intake data (g/day) (estimated from food diaries) and computes calibrated intakes (g/day) based on a biomarker calibration equation (embedded into the package). multiMarker is a web application that infers the relationship between multiple biomarkers and food quantity data from an intervention study and allows prediction of food intake when only biomarker data are available. In addition, the framework allows quantification of the uncertainty in intake predictions.
In addition, to date results from our work have been published in 35 peer reviewed publications and presented at many international conferences and workshops.
