Data-driven integration of emerging technologies to generate a Standardised and Objective Dietary Intake Assessment Tool

Nutrition surveys aim to understand how usual diets impact health. The problem with this is that we don't have an accurate tool to assess diet. We rely on people telling us what they have eaten in the last day or month. However, it is difficult to remember what and how much we have eaten. The surveys used also struggle to capture the range of diets in the UK. Often, people we want to talk to about their diets find these methods unsuitable.

There are lots of emerging ways in which we can assess diet. We can use urine and finger-prick blood samples to test for 'markers' of food and drinks. The benefit of these is that they give us objective data. We can also use wearable cameras to assess foods and diets. Artificial intelligence software is used to determine the type and amount of food eaten. Additionally, new online tools are making it easier for us to self-report. However, no single tool can accurately measure all aspects of the diet.

The aim of this project is to develop a combined tool to accurately assess diet. To do so, we will determine the optimal combination(s) of these emerging methods. The final tool will be easy-to-use and low cost. The combined tool will capture all aspects of the diet.

This project involves four expert teams. Our expertise spans each of the emerging tools. This includes nutrition studies, bio-sampling, chemical analysis, wearable camera technology and web-based diet assessment. First, we will assess the performance of each tool in a small group of people (~ 30). Volunteers will attend a clinical unit 2 times for 4 days to eat standard meals. The meals will represent foods eaten in the UK. Blood, urine, self-reported diet, and food images will be captured. This trial will guide the running of a longer remote trial. The larger trial will involve around 120 people and test habitual diet. We will validate the use of the tools in a home setting. The data will then be analysed to identify which food components are measured accurately by each different tool. This will enable us to determine the optimal combination of measurement techniques to enable comprehesive coverage of the UK diet. We will test use of the combined tool in a final remote trial. This study will involve people often excluded from research. For example, we will engage people from disadvantaged backgrounds and minority ethnic groups. This will help us ensure the tool is not only accurate but suitable for the wider UK population.

Throughout the project, we will talk to members of the public to ensure the trials are easy to follow. To ensure uptake of the combined tool, we will hold workshops with key stakeholders. This will include representatives of the nutrition research community and government departments.

The reliable and objective assessment of habitual diet is a major challenge in nutrition research. New technologies are emerging, but unlikely to solve this problem independently. Metabolites derived from food, found in either urine or blood offer objective 'biomarkers' of dietary intake. Similarly, the analysis of images derived from wearable cameras can discern much about eating behaviour. Our overall aim is to evaluate the optimal combination(s) of measurements provided by these emerging technologies to monitor accurately habitual dietary intake in real-life settings.

In a controlled food intervention, we will first calibrate the ability of each technology to monitor exposure to a comprehensive range of foods commonly consumed in the UK. First morning void urine samples collected using vacuum tube technology and capillary blood samples obtained using the OneDraw Blood Collection device will be analysed using already standardised quantitative LC-MS/MS methods to measure the concentration of more than 80 dietary intake and nutritional status biomarkers. Images from unobtrusive cameras will be analysed using deep learning semantic segmentation and AI methods to predict dietary intake.

A second study with 'rolling' recruitment will validate how each diet assessment tool can be deployed unsupervised to assess habitual diet remotely in a larger, free-living population. The subsequent data will be modelled using machine learning to determine the optimal combination of diet assessment tools to assess habitual eating behaviour.

The acceptability and performance of a minimal, low burden novel integrated tool will then be tested both in free-living individuals generally representative of the UK population and in 'hard to engage/low compliance' demographic groups. Throughout the programme we will consult with PPI groups and stakeholders to ensure that the new technology is perceived as highly acceptable, easy to use and cost effective.