The Use of Mobile-Based Ecological Momentary Assessment (mEMA) Methodology to Assess Dietary Intake, Food Consumption Behaviours and Context in Young People: A Systematic Review

Mobile-based ecological momentary assessment (mEMA) offers a novel method for dietary assessment and may reduce recall bias and participant burden. This review evaluated mEMA methodology and the feasibility, acceptability and validity as a dietary assessment method in young people. Five databases were searched from January 2008 to September 2021 for studies including healthy young people aged 16–30 years and used mEMA for obtaining dietary intake data, food consumption behaviours and/or contextual factors. Data on the method used to administer mEMA, compliance with recording and validation were extracted. A total of 46 articles from 39 independent studies were included, demonstrating a wide variation in mEMA methods. Signal-contingent prompting (timed notification to record throughout the day) was used in 26 studies, 9 used event-contingent (food consumption triggered recordings), while 4 used both. Monitoring periods varied and most studies reported a compliance rate of 80% or more. Two studies found mEMA to be burdensome and six reported mEMA as easy to use. Most studies (31/39) reported using previously validated questions. mEMA appears to be a feasible and acceptable methodology to assess dietary intake and food consumption in near real time.


Introduction
In the transition from adolescence to young adulthood, young people experience significant personal development, increased independence and freedom of choice [1]. These major life transitions also present health challenges, including increased vulnerability to weight gain [1,2]. Compared to other age groups, young people have the highest mean gain in body mass index, placing them at a higher risk of overweight and obesity [3]. This weight trajectory is concerning as poor dietary behaviours and choices developed at a young age often persist into adulthood, affecting health outcomes later in life [1,2].
Identifying the behaviours and contextual factors that influence patterns of dietary intake in young people has been challenging to capture accurately due to the limitations of traditional dietary assessment methods [4]. For instance, most of the current methods used in practice, including 24 h recalls, food frequency questionaries and diet records, are subject to either recall and/or social desirability biases which reduce their validity [4]. Such selfreport methods are often burdensome and subject to misreporting [5]. To overcome some of these limitations, assessment methods such as digital food diaries and image-based dietary assessments have emerged [4]. Further advances in technologies and their widespread societal adoption have created new opportunities to obtain and consider food consumption behaviours and the contextual factors surrounding eating events as they occur in everyday life [5,6].
Ecological momentary assessment (EMA) is a real-time data capture method originally used for psychological assessments that can monitor human phenomena as they occur in their natural environment [5,7]. EMA has appeared useful in obtaining social, psychological and environmental contexts surrounding dynamic patterns of diet behaviours simultaneously while removing the need for recall memory [6]. Most EMA research in recent times has been delivered over mobile technology due to its ubiquity, particularly with young people [8]. Mobile ecological momentary assessment (mEMA) has the advantage of being incorporated seamlessly into daily living by engaging the individual to provide samples of information in short bursts as it occurs in real time [9]. Sampling approaches to obtain information can be defined as signal-contingent or event-contingent. Signal-contingent sampling is a time-based approach that involves signalling the participant with a prompt to complete the mEMA (i.e., to recall the dietary intake or context that occurred within the recent time interval). Prompts can be sent at fixed times (intervals) or at random times throughout the day [5,10]. The other approach, event-contingent, involves the participant completing the mEMA when a relevant event has occurred (e.g., an eating or drinking event) [5]. For eventcontingent sampling, this can be further differentiated into self-initiated or device-initiated. Self-initiated assessment requires the participant to self-initiate an mEMA recording in response to a specific event or behaviour of interest in which they engaged (e.g., eating). Device-initiated refers to the mobile device auto-initiating the mEMA in response to the detection of an event or behaviour (e.g., GPS tracking or wrist accelerometry) [10].
mEMA has potential to detect nutrition-related problems, allowing for earlier interventions in real-life settings. Previous reviews of EMA studies have focused on psychological and health-related behaviours such as emotions [11], alcohol use [12], craving and substance use [13], physical activity [10], sedentary activity [10] and dietary behaviours [14] across diverse age groups ranging from children and adolescents [15] to older adults [16]. Yet, few reviews have investigated the broad processes of mEMA to capture food consumption and related contextual factors of eating/drinking in young people. Additionally, young people are an understudied population with fast-changing priorities, high technology consumption and increased autonomy around eating and drinking, often leading to poor dietary choices; thus, it was deemed appropriate to target individuals aged 16-30 years [3,8]. Therefore, the current review aimed to close this knowledge gap by focusing on three key objectives: (1) examining the methodology of studies using mEMA to measure various aspects of food consumption, (2) evaluating the administration methods of mEMA and (3) assessing its feasibility, acceptability and validity as a dietary assessment method.

Materials and Methods
This systematic review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) checklist [17]. The review protocol is registered with the Open Science Framework (registration DOI: 10.17605/OSF.IO/WPC7Y; accessed on 24 January 2022).

Search Strategy
The following electronic databases were searched from 1 January 2008 to 8 September 2021: MEDLINE via Ovid, Embase via OvidSP, CINAHL via Ebsco, PsycINFO via Ovid and Scopus. The year 2008 was chosen because it marked the introduction of applications (apps) to the digital marketplace [18]. Key concepts (young people, mEMA and outcomes such as dietary intake, food consumption behaviours and contextual factors) and related terms were searched in all five databases using their appropriate syntax including truncations (*) and wildcards ($). The search strategy was limited to the English language and humans. The MEDLINE search strategy is presented in Table 1 and the full search strategies from the other four databases are available upon request.

Eligibility Criteria
The inclusion and exclusion criteria of the review were developed using a modified PICO framework based on population, EMA measurement method, setting and outcomes. The population was young people aged between 16 and 30 years old. This age range was identified as a time of growing independence and freedom of choice in young people before the typical life changes of marriage and children [1]. There was no restriction to participant characteristics such as gender, sex and race/ethnicity, except participants needed to be healthy without presenting chronic health conditions. The intervention criteria included any form of mEMA delivered on a portable electronic device. The primary outcomes included assessment of nutrition and diet such as dietary intake, food consumption behaviours and context if reported. The search strategy included all peer-reviewed primary research study designs conducted in humans. Qualitative studies and systematic reviews were excluded. All relevant studies from 2008 to present were included. There was no restriction placed on geographical location. The exclusion criteria included studies that were not peer reviewed, not in the English language and studies on young people with chronic health conditions such as eating and psychological disorders, diabetes, chronic dieting and post-bariatric surgery recipients, as these conditions do not represent typical food consumption behaviours [19]. Studies that solely focused on alcohol consumption were also excluded.

Screening and Selection of Studies
The identified studies obtained from all five databases were exported to the Endnote X9 citation management software [20] and then transferred to the Covidence platform [21]. Duplicate studies were screened and removed by automation tools firstly in Endnote and again in Covidence. Titles, abstracts and full texts were screened against the eligibility criteria. Studies that did not meet the inclusion criteria were excluded by following the hierarchy of the exclusion criteria and appropriate reasoning(s) provided. To reach a consensus, disagreements concerning eligibility were resolved through discussions between all review authors (BB, LL, MAF, SJ and SRP). The selection process of included studies was documented following the PRISMA flow diagram, presented in Figure 1.

Screening and Selection of Studies
The identified studies obtained from all five databases were exported to the Endnote X9 citation management software [20] and then transferred to the Covidence platform [21]. Duplicate studies were screened and removed by automation tools firstly in Endnote and again in Covidence. Titles, abstracts and full texts were screened against the eligibility criteria. Studies that did not meet the inclusion criteria were excluded by following the hierarchy of the exclusion criteria and appropriate reasoning(s) provided. To reach a consensus, disagreements concerning eligibility were resolved through discussions between all review authors (BB, LL, MAF, SJ and SRP). The selection process of included studies was documented following the PRISMA flow diagram, presented in Figure 1.

Data Extraction
The data extraction table was designed and modified according to the two standardised forms-the Adapted STROBE Checklist for Reporting EMA studies 'CREMAS' used in previous reviews [5,10]. Two reviewers (BB and LL) independently extracted key information from all included studies. Data included: authors, year and country of publication, title, DOI, study design and overview, study aim(s), sample size, target population, par-

Data Extraction
The data extraction table was designed and modified according to the two standardised forms-the Adapted STROBE Checklist for Reporting EMA studies 'CREMAS' used in previous reviews [5,10]. Two reviewers (BB and LL) independently extracted key information from all included studies. Data included: authors, year and country of publication, title, DOI, study design and overview, study aim(s), sample size, target population, participant characteristics and eligibility criteria. In addition, details of mEMA methodology were also extracted, including purpose/health domain, if mEMA training was provided, delivery mode, sampling approach, prompt frequency, prompt interval, reminders, prompt deactivation, the time required to complete one mEMA, monitoring period, latency, compliance rate, reactivity, missing data and incentives. Furthermore, attrition rate, user experience and participant burden, validity and outcomes of interest (such as dietary intake, food consumption behaviours and contextual factors) were extracted. Any discrepancies between the reviewers were resolved through discussion and consulting third experts (MAF, SP and SJ) when necessary.

Data Synthesis and Analysis
The study, sample characteristics and mEMA methodology of all included studies were summarised. The feasibility, acceptability and validity of each independent study was also summarised in tabular form. The findings were synthesised into a narrative review.

Quality Assessment
The Joanna Briggs Institute (JBI) critical appraisal checklist was used to assess the quality of each study according to its study design [22]. The included papers were either cross-sectional surveys or longitudinal cohort designs. Two authors (BB and LL) independently appraised each paper and together reached an overall agreement. Any discrepancies were discussed and resolved between all review authors (LL, BB, MAF, SP and SJ). The cross-sectional survey checklist had 8 items and the cohort 11 items. Each item was answered with either a 'yes', 'no', 'unclear' or 'not applicable'. To determine the overall quality of evidence presented in the study, the following criteria were adapted from Shi et al. and used: 'good' (only 'yes' or 'not applicable' ratings), 'fair' (1 to 2 'no' or 'unclear' ratings) and 'poor' (3 or more 'no' or 'unclear' ratings) [23].

Study Selection
A total of 6615 records were identified from the five databases ( Figure 1). After the automated and manual removal of duplicates, a total of 4203 records were screened. Then, another set of records (n = 3805) and duplicates (n = 59) were manually excluded, leaving a total of 339 full-text articles. Of these, 71 were abstracts only; thus, 268 full-text articles were assessed for eligibility, resulting in 45 studies meeting the inclusion criteria. An additional record was identified through forward chaining, which resulted in a total of 46 articles being included in this review. Data were extracted from each of the 46 articles and then merged with their respective studies where necessary (seven in total) and reported herein as independent studies (n = 39) [8,9,.
In most studies, mEMA was used as the sole method for collecting dietary information. Eleven studies reported using mEMA in combination with other technologies to collect data on health outcomes. One study used mEMA concurrently with a chewing sensor and digital food scale as part of the development of a 'mobile-or mHealth system' to evaluate the acceptability, usability and limitations of this system to self-monitor dietary habits [42]. Two studies combined mEMA with an accelerometer to assess the interrelations of physical activity and dietary intake [45,60] and one with a smartphone sensor [48] to collect additional data on physical activity and sedentary behaviour passively. Two studies collected saliva samples with each mEMA data entry to assess physiological stress markers: cortisol, alpha-amylase and flow rate [61,62]. One study added the additional function of photographic food records which were self-initiated by the participant separate from the mEMA items [33]. Four of the applications used for mEMA had geographical location (GPS) or information systems (GIS) to collect information on food environments and daily activity [25,33,48,60]. One study directly compared mEMA to handwritten EMA to ascertain differences in compliance [26]. Table 2 presents the data input modalities employed by the studies. A smartphone application was the most common delivery mode used by 26 out of 39 studies [8,9,24,25,30-33, 35,37,41,42,45-49,51-54,56-60,65-67]. Of the remaining studies, six used short text message service (SMS) [26,[38][39][40]50,55], three used a personal digital assistant (PDA) [27][28][29]43,64], one used email [36], one used an iPod Touch [61,62] and another used a palmtop computer [63]. One study gave participants the option to be sent the EMA survey via email or SMS with a link to the survey [44]. The smartphone applications selected by researchers varied among studies with only four using the same application (MovisensXS ® ) [41,45,65,66]. The other applications were unique to each study. The SMS delivery modes utilised automated bulk messaging systems to send a hyperlink to the EMA survey at each prompt [38][39][40]55]. One study did not report the automated bulk text messaging service used but specified using SurveySignal ® for the EMA survey [50]. One survey used SMS to prompt via mProve ® software and required participants to directly send a text reply with their responses [26].

Quality Assessment
Among the 36 independent studies that used cross-sectional designs, 16 were assessed as fair quality [27][28][29][30][31][32][33]38,39,[44][45][46]52,53,[56][57][58][61][62][63], 19 were poor [9,26,36,[40][41][42][43][49][50][51]54,55,59,60,[64][65][66][67] and only 1 study was rated as good quality [35] (Table 4). For the poorquality studies, it was hard to determine the inclusion criteria, exposure/outcome measures, confounding factors and type of statistical analysis used. Of three longitudinal cohort studies, two were of fair quality [24,48] and one was rated as poor quality [37]. Regarding the fair-quality studies, the strategies for dealing with the confounders were unclear or not stated, and follow-up information was not addressed. The poor-quality study did not include eligibility criteria and further details on follow-up were not explained clearly. Table 4. Quality assessment of eligible mobile-based ecological momentary assessment (mEMA) studies (n = 39) in young adults using the Joanna Briggs Institute Critical Appraisal Tools for analytical cross-sectional and longitudinal cohort studies [22]. Adapted based on the quality assessment table presented in Shi, Davies and Allman-Farinelli [23].  Cross-sectional questions (1-8): 1. Were the criteria for inclusion in the sample clearly defined? 2. Were the study subjects and the setting described in detail? 3. Was the exposure measured in a valid and reliable way? 4. Were objective, standard criteria used for measurement of the condition? 5. Were confounding factors identified? 6. Were strategies to deal with confounding factors stated? 7. Were the outcomes measured in a valid and reliable way? 8. Was appropriate statistical analysis used?
Cohort questions (1-11): 1. Were the two groups similar and recruited from the same population? 2. Were the exposures measured similarly to assign people to both exposed and unexposed groups? 3. Was the exposure measured in a valid and reliable way? 4. Were confounding factors identified? 5. Were strategies to deal with confounding factors stated? 6. Were the groups/participants free of the outcome at the start of the study (or at the moment of exposure)? 7. Were the outcomes measured in a valid and reliable way? 8. Was the follow up time reported and sufficient to be long enough for outcomes to occur? 9. Was follow up complete, and if not, were the reasons for loss to follow up described and explored? 10. Were strategies to address incomplete follow up utilized? 11. Was appropriate statistical analysis?

Discussion
This review shows that mEMA can be used to collect data on food and beverage intake, dietary habits, food behaviours and contexts such as the physical environment, emotions and social interactions in young adult populations. Most studies employ smartphone applications to deliver signal-contingent prompts and collect data from participants rather than event-triggered prompts, perhaps because of the need for memory to trigger recording. Compliance was above 80% in half the studies with varying schedules of prompts, reminders and frequency and duration of data collection. Attrition from studies ranged from almost none to one in two participants, but overall attrition and compliance would appear to support the feasibility and acceptability of mEMA. A major limitation of the current evidence base is that the quality of studies is overall poor, with only one rated as good quality, and most assessment of dietary intake has not been validated. Thus, further studies may be needed to clarify findings and determine the most effective protocols to administer EMA in young adults maximising compliance and participation.
Factors that would improve compliance might include training in the system, lower participant burden and shorter study durations. More than half of the studies reported training in advance of the study and Stone and Shiffman have previously made recommendations that encourage reporting on training status when documenting EMA protocols in methods [68]. The study with the lowest compliance of <1% monitored for 16 weeks once per day over a 22-week period. However, other studies ran for longer times with better compliance. It remains unclear if event-contingent (self-initiated) or signal-contingent prompts lead to better recording. The former depends on participants' memory to record in response to food ingestion, which may be problematic but may also mean improved accuracy as it is in real time. Conversely, sending prompts to record may occur more distal to food consumption and hence rely on memory to recall food and beverage intake. With signal-contingent sampling, the frequency of prompts and programming of prompts around reported mealtimes reduces the time period between ingestion and recording, thereby eliminating the need to recall what was eaten over a longer period of time (i.e., 24 h, 3 days) [4]. It should be noted only one study used event-contingent sampling via the use of a wearable device that detected eating. A previous review of children and adolescents reported no advantage of wearables over mobile EMA only [69]. However, as a recent review noted there are few wearable sensors available that could be used for event-contingent mEMA.
Finding the appropriate number and duration of sampling to maintain compliance to the protocol remains a challenge for many researchers. In the current review of young adults, it was found among the eight studies with the highest compliance rates (>90%) [26,37,39,54,57,58,61,62] that five signal-contingent collections of data a day was the most common frequency employed. A previous review and meta-analysis of EMA in children and adolescents reported that the average daily number of prompts was 4.2 for non-clinical participants and 3.6 for clinical participants, and the weighted average compliance rate was 78.3%. The duration of the sampling period did not alter compliance [69]. In a metaanalysis by Williams et al., 68 data sets (41 non-clinical and 27 clinical) in adults were included and it was estimated that overall compliance to mEMA was 81.9% [70]. The median number of prompts per day was found to be five in non-clinical data sets and four in clinical data sets. Interestingly, less frequent prompting of 1-3 prompts per day increased compliance in non-clinical participants compared to 6 or more with 87% and 79.4% compliance rates, respectively. No significance was found in clinical data sets. The meta-analysis by Williams et al. has given insight into compliance relating to prompt frequency; however, the focus of their review was not solely on diet but rather health-related behaviours, which included a small proportion of studies on eating behaviours [70]. Schembre et al. conducted a systematic review on mEMA focusing on diet studies inclusive of both children and adults. They found prompt frequencies of the included signal-contingent studies ranged from 3 to 14 prompts per day with a mean response rate of 79% [4]. Similarly, Maugeri and Barchitta conducted a review in children and adults and found the prompt frequency ranged from 1 to 14 times per day [14].
There were only four studies that performed validation studies for the use of mEMA as a dietary assessment method by comparison to traditional methods such as food records, 24 h recalls and a novel measure of energy expenditure instead of the traditional doubly labelled water. Hence, this limits the ability to recommend mEMA as an effective dietary assessment method. However, there are other advantages of the mEMA method in studying food consumption in that it allows real time evaluation of the social, emotional and environmental context in a way a 24 h recall cannot. Thus, the decision of whether to employ mEMA should depend on the motive for monitoring diet. Clearly mEMA would not appear to be the method of choice for large epidemiological cohort studies of diet disease relationships as outlined in a previous systemic review [4]. However, it may well be useful to monitor changes in nutrition behaviours in different contexts and in response to an intervention or to plan an intervention in an individual or population. Collection of food and beverage consumption to yield macronutrient and micronutrient data is not required in many scenarios in order to improve an individual's dietary behaviour.
Overall, as noted by others [5,71] assessing the mEMA methodology and its feasibility, acceptability and validity remained challenging in this review due to the inconsistencies and absence of key factors in reporting across studies. The major strengths of the current review include an extensive search of multiple databases, resulting in an ample number of mEMA studies for screening and selection. The review focused on mEMA, which is applicable to the digital-savvy 16-to 30-year-olds. A strength of this review is compliance with recognised standards for reporting in systematic reviews and EMA studies and the quality appraisal [5,17]. However, it is acknowledged that this systematic review is not without some limitations. The year 2008 was selected as the year applications were introduced to the digital marketplace and accessible on smartphones, but this may mean some EMA studies with personal digital assistants and text messages were omitted. However, our purpose was to provide an evidence base for those wishing to employ mEMA in the current digital environment. In addition, the target population was young people, so the findings cannot be extrapolated to other age groups such as early adolescents and older people. Only English studies were included, yielding a language bias as per the eligibility criteria.

Conclusions
The current review of 39 independent studies offers unique insights into the uses of mEMA in young people aged 16 to 30 years old. This population is well documented to have poor diet quality and experience the most weight gain among adults. Measuring their food consumption and the context of this consumption is an important step in formulating intervention. mEMA has demonstrated potential to become a feasible and acceptable methodology in assessing food and beverage consumption with the advantage of providing social, emotional, and food environment contextual information. Further research in the technology of wearables and detection of eating as well as validated questionnaires for data collection would advance the field.