Food & You: the utility of a digital cohort for personalized nutrition studies

By Neha MathurReviewed by Sophia CoveneyAug 22 2023

In a recent article posted to the medRxiv* server, researchers used a digital (completely remote) cohort to implement a personalized nutrition study.

*Important notice: medRxiv publishes preliminary scientific reports that are not peer-reviewed and, therefore, should not be regarded as conclusive, guide clinical practice/health-related behavior, or treated as established information.

They gathered high-resolution data on demography, dietary intake, gut microbiota, physical activity, and blood glucose levels of over 1000 participants for two to four weeks between October 2018 and March 2023 in Switzerland and the Netherlands. 

It allowed study participants to collect in situ data daily using a mobile application and wearable sensors. However, to rule out all concerns related to data quality, the researchers focused on assessing data quality and reliability.

Background

Most nutritional epidemiology studies studied the effects of nutrition and a healthy diet at the population level, which informed dietary recommendations and guidelines. More recently, some scientific studies have found the link between microbiota composition and the health benefits of a nutritious diet, thus, raising the potential of personalized nutrition. 

So far, all studies demonstrating the effectiveness of personalized dietary recommendations have examined personal features, for example, gut microbiota compositions, in isolation.

However, there is a need for a more holistic approach to nutritional epidemiology that covers all the factors influencing the ability of the human body to derive maximum benefit from diet, for example, lifestyle factors. 

For example, in patients with diabetes, sedentary behavior is an established risk factor similar to a carbohydrate-rich diet. In addition, the gut microbiome of these patients might be modulating their blood glucose response to food intake.

Thus, nutritional studies examining the association of blood glucose response, specifically postprandial glucose response (PPGR) and insulin resistance (IR) that characterize type 2 diabetes, need data on all relevant factors, preferably in situ and continuously.

Thus, digital cohorts mark the beginning of a major development in the context of nutritional epidemiological and clinical studies. Since digital cohorts are a recent development, there is an urgent need to address questions regarding selection bias, study adherence, and data quality of digitally implemented studies. 

Another challenge to address is the time burden due to digital cohorts. They fatigue some participants, translating to lower adherence to study protocol or poor data quality.

About the study

The researchers attempted to address such questions and challenges by describing the protocol of the “Food and You” digital study and reporting study engagement data from enrollment to completion. They examined the study response and its completion rate. Furthermore, they assessed data quality by comparing nutritional and microbiota data of “Food and You” with data from traditional (on-site) studies.

This study had enrolment, preparatory, tracking, and follow-up phases. Following screening via a short questionnaire in the enrolment phase, a research team member enrolled an eligible participant into this study and instructed them to fill out a series of questionnaires on the “Food & You” website.

Next, they asked them to download MyFoodRepo mobile app, an artificial intelligence (AI)-assisted nutrition tracking app specifically designed for this study, to monitor their food intake for at least three days (trial period).

All participants who completed the trial entered the study. The researchers provided them with the study material, which included a continuous glucose monitoring (CGM) sensor and material for stool collection.

During the tracking phase, all participants wore the CGM sensor and logged their food/beverage intake via the MyFoodRepo app.

Participants were divided into two sub-cohorts digitally: sub-cohort “Basic” cohort B) and sub-cohort “Cycle” (cohort C). These cohorts consisted of non-diabetic participants and non-diabetic women of reproductive age not on hormonal medication or contraceptives, respectively.

Study cohorts B and C answered two daily surveys for 14 and 28 days, respectively.

Per the study protocol, all participants consumed standardized breakfasts from days 2 to day 7 during the first week and refrained from physical activity for the next two hours.

Cohort B participants performed an oral glucose tolerance test on days 6 and 7, while Cohort C participants performed it on days 6 and 7 and days 21 and 22. Participants provided one stool sample during the tracking days. Eventually, participants uploaded their CGM data and physical activity levels on the study website. During the follow-up, all participants filled out a feedback questionnaire to log their overall experience.

Results and conclusion

This study had a high completion rate, with over 60% of enrolled participants completing it. In both cohorts, the completion rates of subjects with dietary restrictions were above 80%. Compared to other digital health studies, the retention rate for 14 and 28 days was also high in this study.

In both cohorts, except for physical activity and sleep, data availability was high for most indicators, e.g., diet, implying good adherence to the study. Besides response fatigue, technical issues with sensor devices or Apps might have impacted adherence to the study.

A study annotator reviewed every submitted data point on nutrition, implying data quality was good. The researchers found that all participants appropriately and timely logged their food intake, and missing inputs were low. The authors noted expected patterns concerning the time of food intake, glucose curves, etc., on weekdays and weekends.

Encouragingly, the MyFoodRepo App received overall positive feedback. It fetched dietary data with a high resolution of 315,126 food dishes constituting more than 46 million kcal, giving a reliable representation of the dietary patterns of over 1000 participants for at least two weeks. The researchers also had high-resolution data from 49,110 survey responses, 23,335 participant days, and 1,470,030 blood glucose measurements for analysis. 

Furthermore, the authors collected 1,024 stool samples for gut microbiota analysis. They attributed the observed variations in gut microbiota across two study cohorts, originating from Switzerland and the Netherlands to geographical differences. They used 16S ribosomal ribonucleic acid (rRNA) sequencing to analyze self-collected stool sample data. Further analyses of the microbiome and its link to other data are ongoing.

To summarize, it is likely that digital nutrition cohorts might become the preferred study design for large-scale personalized nutritional studies as they have the potential to help collect a large amount of high-quality data with temporal resolution.

*Important notice: medRxiv publishes preliminary scientific reports that are not peer-reviewed and, therefore, should not be regarded as conclusive, guide clinical practice/health-related behavior, or treated as established information.