(C) PLOS One
This story was originally published by PLOS One and is unaltered.
. . . . . . . . . .



Food & You: A digital cohort on personalized nutrition [1]

['Harris Héritier', 'Digital Epidemiology Lab', 'School Of Life Sciences', 'School Of Computer', 'Communication Sciences', 'Epfl', 'Lausanne', 'Chloé Allémann', 'Oleksandr Balakiriev', 'Victor Boulanger']

Date: 2024-02

Nutrition is a key contributor to health. Recently, several studies have identified associations between factors such as microbiota composition and health-related responses to dietary intake, raising the potential of personalized nutritional recommendations. To further our understanding of personalized nutrition, detailed individual data must be collected from participants in their day-to-day lives. However, this is challenging in conventional studies that require clinical measurements and site visits. So-called digital or remote cohorts allow in situ data collection on a daily basis through mobile applications, online services, and wearable sensors, but they raise questions about study retention and data quality. “Food & You” is a personalized nutrition study implemented as a digital cohort in which participants track food intake, physical activity, gut microbiota, glycemia, and other data for two to four weeks. Here, we describe the study protocol, report on study completion rates, and describe the collected data, focusing on assessing their quality and reliability. Overall, the study collected data from over 1000 participants, including high-resolution data of nutritional intake of more than 46 million kcal collected from 315,126 dishes over 23,335 participant days, 1,470,030 blood glucose measurements, 49,110 survey responses, and 1,024 stool samples for gut microbiota analysis. Retention was high, with over 60% of the enrolled participants completing the study. Various data quality assessment efforts suggest the captured high-resolution nutritional data accurately reflect individual diet patterns, paving the way for digital cohorts as a typical study design for personalized nutrition.

To understand personalized nutrition, detailed individual data collected in the real world are needed. Traditional studies often face challenges in collecting data from participants’ day-to-day lives. To address this issue, digital or remote cohorts have emerged as an alternative, allowing data collection through mobile apps, online services, and wearable sensors. In the study "Food & You," a personalized nutrition study, we implemented a digital cohort where participants tracked their food intake, physical activity, gut microbiota, glycemia, and other data for a period of two to four weeks. Over 1000 participants completed the study, resulting in a very large dataset. The study achieved a high retention rate, with over 60% of enrolled participants completing the study. Furthermore, efforts to assess the data quality suggest that the captured nutritional data accurately reflect individual diet patterns. These findings support the potential of digital cohorts as a typical study design for personalized nutrition research. Digital cohorts offer a promising approach to understanding the complex relationship between nutrition and health by enabling researchers to collect real-world data from participants in their daily lives.

Funding: This work was supported by grants to MS of the Kristian Gerhard Jebsen Foundation, the Seerave Foundation, and the Fondation Leenaards. The funders had no role in the design or execution of this study and will have no role in the analyses, interpretation of the data, or decision to submit results.

The present paper details the study protocol, and reports study engagement data by looking at the individual characteristics of participants on their journey from enrollment to completion. Further, we provide an overview of the data collected in the “Food & You” cohort from October 2018 to March 2023, and describe our efforts to assess data quality, including the comparison of nutritional and microbiota data collected in “Food & You” with data collected in traditional (on-site) studies. We also discuss the challenges of running a complex digital cohort, and how we addressed them. Overall, retention rates were relatively high, with more than 60% of enrolled participants completing the study. Despite certain fatigue over time, adherence was very high, especially for glucose response data, nutritional data, microbiota data, and data from daily surveys. While the study population shows some demographic differences compared to the overall population, the nutrition patterns are in very good agreement with data obtained in another study from a representative sample of the general Swiss population [ 18 ].

Here, we address some of these questions by evaluating the completion rates, adherence, and data quality of the “Food and You” digital cohort on personalized nutrition. The “Food & You” study started in late 2018 and consisted of two distinct digital sub-cohorts; the sub-cohort “Basic” (cohort B) restricted to non-diabetic participants, and the sub-cohort “Cycle” (cohort C) restricted to non-diabetic women of reproductive age who did not use hormonal contraceptive or medication (see Methods for inclusion / exclusion criteria). The study duration for cohort B was 14 days, whereas cohort C participants were enrolled for 28 days, matching the length of a typical menstrual cycle. The study was performed in Switzerland, and all participants were required to have a postal address in Switzerland. Throughout the study, participants were requested to report i) food consumption using an AI-assisted food tracking app (MyFoodRepo), ii) continuous blood sugar levels using a continuous glucose monitor, and iii) physical activity and sleep using either activity trackers or daily surveys. They were furthermore asked to follow a protocol that included a one-time stool sample collection for gut microbiota analysis, and the consumption of standardized breakfasts.

In digital health studies—also called remote or siteless studies—all interactions with participants, as well as data collection, are digital or digitally coordinated. These studies leverage an array of digital devices, wearable sensors, and online services. Digital cohorts and trials have been heralded as a new major development for epidemiological and clinical studies. However, since digital cohorts are a relatively new study approach, open questions regarding selection bias, retention, and data quality remain. Indeed, access to devices connected to the internet and digital literacy may lead to selection bias which in turn may lead to a lack of representativity of the study population compared to the general population. Furthermore, the time burden generated by following the study protocol and collecting the data might create response fatigue, which could in turn translate into lower study adherence, or data quality. Finally, novel data collection methods may not have been thoroughly validated.

Blood glucose response is a particularly interesting outcome measure for nutritional studies due to its association with insulin resistance, metabolic syndrome, and diseases such as stroke, type 2 diabetes, and heart disease [ 11 ]. Reducing blood glucose levels is thus a recommendation of public health authorities around the globe [ 12 ]. Identified risk factors for elevated blood glucose levels include a carbohydrate-rich diet [ 13 ], lack of physical activity [ 14 ], and poor sleep [ 15 ], among others. In addition, studies have begun to investigate the role of the gut microbiome in modulating the blood glucose response to food intake [ 7 , 16 , 17 ]. Nutritional studies trying to understand postprandial glucose response (PPGR) are thus faced with the challenge of obtaining data on relevant factors all at once, ideally continuously and in situ, that is, in the regular environment in which participants’ lives unfold.

Nutrition plays a significant role in moderating the risk and/or severity of several diseases, such as type 2 diabetes [ 1 ], cardiovascular diseases [ 2 , 3 ], or cancer [ 4 ]. Findings from nutritional epidemiology studies have led to dietary guidelines and public health campaigns designed to support healthy diets. However, while these recommendations are generally based on results aggregated at the population level, a more individualized approach to health [ 5 ] has led to the concept of personalized nutrition. For example, a randomized study showed that personalized recommendations improved diet quality as measured by the Healthy Eating Index (HEI) compared to a control group [ 6 ]. Zeevi and colleagues showed how personalized nutrition algorithms could be used to design diets that lower postprandial glucose responses [ 7 ]. Further studies showed the importance of personal features such as gut microbiota compositions on glycemic responses [ 8 , 9 ]. Another intervention study showed significant improvement in the food categories consumed when receiving personalized diet advice, compared to generic or no advice [ 10 ]. These findings highlight the need for a more holistic approach to nutritional epidemiology, encompassing diet, gut microbiota, physical activity, lifestyle, and other factors, with the goal of tailoring dietary guidelines to each person’s unique circumstances.

Methods

Study design and setting The Swiss “Food & You” study is a digital cohort study collecting data on glycemia, nutrition, gut microbiota, lifestyle, and physical activity as well as demographic data (Fig 1). The cohort was open to anyone fulfilling the inclusion criteria listed in S1 Table. The study consisted of four sequential phases: enrollment phase, preparatory phase, tracking days phase, and follow-up phase (Fig 2). In the enrollment phase, interested participants were first required to perform a self-check of their eligibility, and fill out a consent form and a short screening questionnaire. Following this, and upon acceptance by a member of the study team (based on the available capacity to accommodate new participants), the participants were considered enrolled. During the subsequent preparatory phase, enrolled participants were given instructions on the “Food & You” website and asked to fill out a series of questionnaires. Next, they were instructed to download the AI-assisted nutrition tracking app MyFoodRepo (MFR– https://www.myfoodrepo.org) to track their food intake for a trial period of at least three days. After the successful completion of the trial period, participants would order the study material which included, among other things, a continuous glucose monitoring (CGM) sensor for each period of 14 days and stool-sample self-collection material. Upon receipt, they were asked to choose the starting date of their tracking days phase. In the tracking days phase, participants were required to log all their food and drink intake via the MyFoodRepo app, wear the CGM sensor, and answer two daily surveys for 14 (cohort B) and 28 (cohort C) days. In addition, participants were instructed to take standardized breakfasts in accordance with dietary restrictions from the 2nd to the 7th day during the first week. They were asked to avoid altering standardized meals, as well as to refrain from eating or engaging in physical activity for the subsequent two hours. Cohort C participants repeated these instructions on their third tracking week (S2 Table). On days 6 and 7 (additionally on days 21 and 22 for cohort C) they were asked to perform an oral glucose tolerance test by drinking 50g of glucose which they received by mail. In addition, study participants were asked to provide one stool sample collected anytime during the tracking days phase. At the end of the tracking days phase, participants were asked to upload their physical activity and CGM data on the “Food & You” website. In the follow-up phase, they were requested to fill out a feedback questionnaire regarding their experience. Participants were also provided with interactive visualizations of their data (S1 Fig). Cohort C participants were followed for two additional menstrual cycles during which they continued to track their menstrual cycle and fertility-related body signs such as morning temperature and cervical mucus characteristics. PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 1. Data collection. a) Schematic illustrating the data collection process. (Left) Participants track the study variables in situ (from home, work, etc.). (Center) Data or samples are collected via web platforms and apps, or shipped to the lab by mail. (Right) Data is processed in the Food & You database. b and c) Example of data collected by one participant over 5 days. Top panel shows blood glucose levels (orange line), physical activity (turquoise spikes), and sleep (translucent turquoise rectangles). Bottom panel shows time and micronutrient composition (colors) of reported food intake. Like in the top panels, translucent turquoise rectangles show when the participant is asleep. https://doi.org/10.1371/journal.pdig.0000389.g001 PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 2. Study phases with participants per phase and exit numbers. https://doi.org/10.1371/journal.pdig.0000389.g002 The Geneva ethics commission has reviewed and authorized the project (Ethical Approval Number: 2017–02124). The study is registered on the website of the Federal Office of Public Health (SNCTP000002833) and the platform clinicaltrials.gov (NCT03848299).

Data collection Questionnaires: During the enrollment phase, interested subjects were requested to fill out a screening questionnaire with items regarding age, gender, height, weight, type of mobile phone and dietary restrictions. During the preparatory phase, enrolled participants had to fill out a lifestyle and health-related questionnaire. Participants were asked about their general health (smoking, diet, food supplement intake, general hunger levels, health state, past diagnoses, antibiotic intake, and menstrual health), physical activity (exercise frequency, duration, and intensity), sleep (bedtime, wake-up time), sociodemographic variables (nationality, socioeconomic status, job status, and household description), and requested to provide self-measured anthropometric measurements (waist and hip circumference, height, and weight). In the tracking days phase, participants had to fill out a short form each evening to validate their adherence to protocol and document medication intake. Cohort C participants had to answer additional questions on menstrual blood, cervical mucus and provide self-reported temperature measurements on a daily basis. This data will in the future enable us to study menstrual cycle effects on the glycemic response which are still understudied, despite some anecdotal evidence that some diabetic menstruating individuals need to modulate their medication with their menstrual cycle [19], and that menstruating women alter their diet throughout their cycle [20]. Dietary Intake: Participants of the “Food & You” study were asked to log any dietary intake in real time on the MyFoodRepo app (MFR) using one the following options: taking pictures of the food/drink, scanning the product’s barcode (if available), or describing the food item with text. A logged entry is defined as a “dish” and can contain multiple food items. For example, tuna, steamed potatoes, and green beans are all single food items, and together compose a dish. The pictures were automatically segmented and classified by an image recognition algorithm [21]. Portions, segmentations, and food classes were subsequently verified or edited by a team of trained annotators. The MyFoodRepo app also allows annotators to communicate with the participant for clarifications about the food, and participants were able to leave comments through the app. This process ensured that every single dish in the nutritional data was reviewed by a member of the study team. Each food item was linked to a nutritional value database containing 2’129 items built on the Swiss Food Composition Database [22], MenuCH data [23], and Ciqual [24]. When food intake was logged through barcode scanning, nutritional values of the food items were fetched from the Open FoodRepo database API [25]. Manual entries were matched to food items by the annotators. As the aforementioned nutritional values data sources did not provide standard portion sizes, these were manually extracted from the portion list of the WHO MONICA study [26], and the Mean Single Unit Weights of Fruit and Vegetables report [27] by the German Federal Office of Consumer Protection and Food Safety. When a standard portion was not available for a particular food item, we assigned the standard portion of a similar food item. Each dish logged through the MyFoodRepo app carries a timestamp, which enables dietary analysis at a high temporal resolution. Food items were classified into categories based on the menuCH study [23]. Barcoded food items from the Open FoodRepo database were categorized based on the food product description. When such a description was not available, we assigned the category extracted from the Open Food Facts database (https://world.openfoodfacts.org/). While “Food & You” is the first large cohort to use MyFoodRepo, the app annotation quality had previously been validated [28]. Glycemia: Glycemic data was collected by using the Flash Glucose Monitor Freestyle Libre (Abbott Diabetes Care). The system, which has been validated in numerous studies [29–31], consists of a disposable sensor applied to the back of a participants’ upper arm, and a reader device or a smartphone app allowing to collect data from the sensor via NFC technology. It measures glycemia every 15 minutes via a subcutaneous filament carrying enzyme glucose sensors [32]. To encourage high adherence to protocol, we chose a non-blinded glucose monitoring system to allow the participants to see their glycemia in real time. Participants self-applied the sensor at home following explanations provided in writing and video. Cohort B participants wore a single sensor for 14 days, whereas cohort C participants wore two sensors consecutively for a period of 28 days to cover the length of a typical menstrual cycle. Notably, when participants scan the sensor, the data from the previous eight hours is collected. Thus, unless participants scan the sensor at least every 8 hours, some data may remain unretrievable. Gut Microbiota: Participants were requested to collect a stool sample following detailed written and video instructions. They could collect and ship their sample anytime during the tracking days phase. Samples were collected with stool nucleic acid collection and preservation tubes from Norgen Biotek, stored at room temperature and shipped in batches of 100 to 192 samples to Microsynth AG (Balgach, Switzerland) for sequencing and bioinformatics analysis. V4 region of the bacterial 16S rRNA gene was sequenced via creation of two-step Nextera PCR libraries using the primer pair 515F (NNNNNGTGYCAGCMGCCGCGGTAA) and 806R (NNNNNGGACTACNVGGGTWTCTAAT). The primers use 5 bases at their 5´ end to increase diversity of the bases during the first five sequencing cycles. Subsequently, the Illumina MiSeq platform and a v2 500 cycles kit were used to sequence the PCR libraries. The produced paired-end reads which passed Illumina’s chastity filter were subject to de-multiplexing and trimming of Illumina adaptor residuals using Illumina’s real time analysis software included in the MiSeq reporter software v2.6 (no further refinement or selection). The quality of the reads was checked with the software FastQC version 0.11.8. The locus specific V4 primers were trimmed from the sequencing reads with the software cutadapt v2.8. Paired-end reads were discarded if the primer could not be trimmed. Trimmed forward and reverse reads of each paired-end read were merged to in-silico reform the sequenced molecule considering a minimum overlap of 15 bases using the software USEARCH version 11.0.667. Merged sequences were then quality filtered allowing a maximum of one expected error per merged read. Reads that contained ambiguous bases or were considered outliers regarding the amplicon size distribution were also discarded. Samples that resulted in less than 5000 merged reads were discarded, to not distort the statistical analysis. The remaining reads were denoised using the UNOISE algorithm [33] implemented in USEARCH to form amplicon sequence variants (ASVs) discarding singletons and chimeras in the process. The resulting ASV abundance table was then filtered for possible bleed-in contaminations using the UNCROSS [34] algorithm, and abundances were adjusted for 16S copy numbers using the UNBIAS [35] algorithm. ASVs were compared against the reference sequences of the RDP 16S database, and taxonomies were predicted considering a minimum confidence threshold of 0.5 using the SINTAX algorithm [36] implemented in USEARCH. Physical Activity and Sleep: Study participant’s physical activity (PA) and sleep data were collected using one of two methods: objectively via Apple Health, Google Fit, or smart-watches, or subjectively, i.e., self-reported on the study website via the morning and/or evening questionnaire. The different formats of the objective PA and sleep data were harmonized and stored in a single database, and comprised daily step count, daily calories burned, bedtime, and wake-up time. In addition, for PA measured with smart devices, the type of physical activity, the start and end times, the amount of burned calories, the average heart rate, and the maximum heart rate were collected. Participants self-reporting their sleep and PA on the website had to report the times at which they fell asleep and woke up, if they did any physical activity, and if so, when they started and finished, as well as the perceived intensity of their effort.

[END]
---
[1] Url: https://journals.plos.org/digitalhealth/article?id=10.1371/journal.pdig.0000389

Published and (C) by PLOS One
Content appears here under this condition or license: Creative Commons - Attribution BY 4.0.

via Magical.Fish Gopher News Feeds:
gopher://magical.fish/1/feeds/news/plosone/

You are viewing proxied material from magical.fish. The copyright of proxied material belongs to its original authors. Any comments or complaints in relation to proxied material should be directed to the original authors of the content concerned. Please see the disclaimer for more details.