Accepted Manuscript The Feasibility and Validity of Mobile Phones to Assess Dietary Intake Darren B. Sharp, BMedSc Margaret Allman-Farinelli, PhD, FDAA PII:
S0899-9007(14)00125-7
DOI:
10.1016/j.nut.2014.02.020
Reference:
NUT 9238
To appear in:
Nutrition
Received Date: 4 December 2013 Accepted Date: 17 February 2014
Please cite this article as: Sharp DB, Allman-Farinelli M, The Feasibility and Validity of Mobile Phones to Assess Dietary Intake, Nutrition (2014), doi: 10.1016/j.nut.2014.02.020. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
Title Page Title The Feasibility and Validity of Mobile Phones to Assess Dietary Intake
Authors Darren B Sharp1 Margaret Allman-Farinelli1
RI PT
Running Title Validation of mobile phones for dietary assessment
[email protected] [email protected] SC
Author Affiliations 1 Department of Nutrition & Metabolism, School of Molecular Bioscience, The University of Sydney, Camperdown, Australia
TE D
Qualifications and position title: Darren B Sharp BMedSc Position title: Masters candidate
M AN U
Corresponding Author: Margaret Allman-Farinelli Post: Molecular Bioscience Building G08, The University of Sydney, NSW 2006, Australia Email:
[email protected] Phone: +61 2 90367045 FAX: we do not have one
EP
Margaret Allman-Farinelli PhD FDAA Position title: Associate Professor
AC C
Authorship: D.B.S. and M.A-F. contributed to conception, study design, generation, analysis and interpretation of data, revision and approval of the final manuscript. D.B.S. drafted first version of the manuscript. Word count: 4,850 Tables: 2 Figures: 1
ACCEPTED MANUSCRIPT The Feasibility and Validity of Mobile Phones in Assessing Dietary Intake
Abstract Current limitations of conventional dietary assessment methods restrict the establishment of
RI PT
diet disease relationships and efficacy of dietary interventions. Technology, in particular the use of mobile phones, may help resolve methodological limitations, in turn improving the validity of dietary assessment and research and associated findings. This review aims to
SC
evaluate the validity, feasibility, and acceptability of dietary assessment methods that have been deployed on mobile phone platforms. In August 2013, electronic databases for health
M AN U
sciences were searched for English, peer reviewed, full text articles, published from January 1 2001 onward; and accompanied by a hand search of available relevant publications from universities and government bodies. Studies were not limited by design, length, setting or population group. Twelve of 194 articles met eligibility criteria: mobile phone as the dietary
TE D
recording platform and validation of energy and/or macro-nutrient intake against another dietary or biological reference method. Four dietary recoding methods had been validated on mobile phone platforms: Electronic food diary, food photograph assisted self-administered
EP
24h recall, food photograph analysis by trained dietitians, and automated food photograph
AC C
analysis. All mobile phone dietary assessment methods showed similar, but not superior validity or reliability when compared with conventional methods. Participants’ satisfaction and preferences for mobile phone dietary assessment methods were higher than those for conventional methods, indicating the need for further research. Validity testing in larger and more diverse populations, over longer durations is required to evaluate the efficacy of these methods in dietary research.
Keywords: dietary assessment, Smartphone, food photograph, food diary, reliability,
ACCEPTED MANUSCRIPT
Introduction The complex relationships between diet and disease aetiology are not yet fully understood. Among the many barriers to ascertaining these answers lies the accuracy of dietary
RI PT
assessment methods. Methods of dietary assessment available today are fraught with
inherent and extrinsic methodological problems making accurate measures of nutritional intake extremely difficult [1]. This has the potential to lead to erroneous associations
SC
between diet and disease and inhibits the ability to assess efficacy of dietary interventions
M AN U
[2,3]. The development of accurate assessment methods has been part of ongoing research in the fields of dietetics and epidemiology. Recently research has focused on harnessing technology as a viable solution to current methodological shortcomings.
Technology has the potential to reduce researcher and participant burden, improve adherence
TE D
and communication, automate and standardise coding, and upgrade data quality, [4-6]. In turn these features are likely to reduce costs associated with dietary related research and
EP
healthcare. Of particular interest is the use of mobile phones in aiding dietary assessment.
AC C
Ownership of mobile phones has grown exponentially over the past 10 years with an estimated 3.2 billion active users in 2012 [7]. Mobile phones have a variety of advantageous technological features such as wireless communication, built-in cameras, global positioning systems (GPS), accelerometers, high speed microprocessors, portable designs and connectivity to external devices via blue tooth and infra-red, making them a convenient and suitable platform for dietary research.
These features have successfully been implemented in dietary management and intervention
ACCEPTED MANUSCRIPT studies [8-11]. Less research, however, has investigated the validity of mobile phones for assessing dietary intake. Personal digital assistant (PDA) devices employing food photographs and electronic food diaries have previously been shown to be a valid measure of macro and micro-nutrient intake [12,13]. In the past decade the use of PDA devices has been
RI PT
superseded by mobile phones, in particular Smartphones. The aim of this review is to assess the current literature surrounding the validity, feasibility and acceptability of the methods used to record individual's dietary intake using mobile phones, with the intent of informing
AC C
EP
TE D
M AN U
SC
future research and practice.
ACCEPTED MANUSCRIPT
Methods: A structured literature search was undertaken during August 2013, with a predefined search protocol establishing inclusion and exclusion criteria. Studies utilising mobile phones as the
RI PT
dietary recording platform and published in peer-reviewed English journals from January 1 2001 onward were included. The year 2001 was chosen to reflect the mainstream inception of the term 'Smartphone'. All study designs, lengths, settings and population groups were
SC
considered. Validation of energy and/or macro nutrient intake against another dietary or biological reference method was considered the primary outcome measure. For newer
M AN U
technologies incorporating automated image analysis; accuracy of automated processes (food segmentation, classification, volume/weight estimation and energy calculation) were reported. Search terms were established through preliminary research and reference to the MESH database. They were divided into three hierarchical categories (ranked in order of
TE D
highest to lowest importance) and searched in their complete and truncated form: DEVICE: smart phone*, mobile phone*, mobile telephone*, mobile device*, mobile application*, cellular phone* and cell phone*; DIETARY ASSESSMENT: food diary, food record*, food
EP
intake*, diet record*, dietary record*, dietary assessment*, dietary intake*, dietary
AC C
measurement*, energy intake*, caloric intake*, nutrient intake*, nutritional intake* and nutrition* assessment*; IMAGING/SENSORS: camera*, image*, digital imag*, digital picture*, image analysis, photo*, photograph*, photography, photographing, video*, scanner* or sensor*, bar code*, segmentation* and food volume*. Categories were searched individually and together with one or both other categories (Table 1). Database searches took place in Medline, Web of Science, CINAHL, Scopus and Science Direct. A hand search of universities and government bodies document repositories for available relevant dissertations and publications was conducted using the above search terms. A manual review of main
ACCEPTED MANUSCRIPT authors and selected articles, in the references of manuscripts identified in the search was also
AC C
EP
TE D
M AN U
SC
RI PT
conducted.
ACCEPTED MANUSCRIPT
Results The database search yielded 20,764 publications when search strings were combined. 194 full text articles were extracted into Endnote X6 (Thomson Reuters) for further review. Four
RI PT
additional articles were retrieved following a manual search of authors and screening of selected articles and bibliographies. After applying inclusion and exclusion criteria 15 eligible studies were selected. Hand searching of universities and government bodies yielded one
SC
additional dissertation containing four studies. One study had previously been published and retrieved during the database search and another one was selected after meeting eligibility
M AN U
criteria (figure 1). A total of 16 eligible studies were included and categorised as follows: Self-directed dietary recording (n=2) [14,15] food photograph analysis by trained research dietitians (n=8) [16-22] and automated food photograph analysis (n=6) [23-28]. Study characteristics and results are summarised in Table 2 and further described throughout the
AC C
EP
TE D
review.
ACCEPTED MANUSCRIPT Self-reported dietary recording: Electronic Food diary with image capability Carter and colleagues assessed the validity of a Smartphone electronic food diary application 'My Meal Mate (MMM) with imaging capability [15]. Participants recorded dietary intake
RI PT
by selecting the type and quantities of food and drink items from a food database containing 40,000 generic and branded food items. The ability to capture photographs was also
incorporated to assist with dietary recall when food data was unable to be entered at the time
SC
of consumption. However the relative use of food photographs was not reported. Validation was assessed in a sample of university staff and student volunteers. Means of two day MMM
M AN U
records showed moderate to strong correlations for energy and macro-nutrients when compared with the means of corresponding 24h recalls (r = 0.69 to 0.86, P < 0.001). Correlations however, were lower when 7 day MMM means were analysed (r = 0.64 to 0.75, P < 0.001). Participants’ incorrect portion-size estimations were reported as the methods
TE D
largest source of error. Bland-Altman Plots showed wide limits of agreement indicative of individual variation but these were considerably smaller than previously reported by Beasley using a PDA [13]. Notably daily within person variation was smaller using MMM compared
EP
with the 24h recall. Overall these results compare similarly to previous PDA studies using
AC C
electronic food diaries [13, 29]. In a related study the MMM app showed superior adherence, usage, convenience, social usability, and overall satisfaction when compared with a conventional pen and paper food diary over a 6 month period [30].
Food Photographs used to aid in dietary recall: Arab and colleagues explored the use of passively captured food photographs to later aid participants memory and portion size estimations in a web based self-administered 24h recall [14]. Images were automatically captured using a camera enabled Smartphone worn around
ACCEPTED MANUSCRIPT participants necks. Daily median energy intakes compared very well with criterion measures of doubly labelled water (DLW) (2359kcal versus 2377kcal respectively). Participants considered the use of the device as somewhat cumbersome, with many unsure if they were capturing appropriate food images during meal occasions. Reportedly this may have
RI PT
influenced eating behaviour. The high frequency of image capture also resulted in excessive battery drain, requiring several battery replacements.
SC
Food photograph analysis by trained research dietitians Labelled Food Photographs
M AN U
Wang and colleagues validated the use of a PDA device with camera and phone card known as Wellnavi [16,21,22]. This device had equivalent functionality to today’s Smartphone and therefore was included within this review. Using Wellnavi participants capture a before and after image of their meal. The images along with short descriptions of the ingredients used in
TE D
meals (via directly writing onto the touch screen using a stylus) are then sent to a server via the phone card attachment. Images are retrieved, analysed and coded by research dietitians in near real-time. Inter-rater reliability between dietitians for all 35 nutrients was good to
EP
excellent (r = 0.55 to 0.92, P < 0.05; median = 0.78) [22]. Comparison of a one day Wellnavi
AC C
record with same day weighed food record (WFR) showed moderate to strong correlations for energy and macro-nutrients (r = 0.57 to 0.88, P < 0.01), but the volunteers were young female food and nutrition students [22]. Moderate to strong correlations for the 31 micro nutrients under investigation was also observed (r = 0.38 to 0.93; median = 0.77). Weaker correlations were observed in a subsequent study for macro and micro-nutrients (significant r range: 0.39 to 0.86, P < 0.05) when comparing Wellnavi with two same day weighed food records six months apart [21]. Compared to the WFR, correlations for most nutrients in the study were lower using Wellnavi than with corresponding one day 24h recalls. Participants
ACCEPTED MANUSCRIPT reported Wellnavi as the least burdensome and least time consuming of the three dietary assessment methods. When implemented among the general population comparative validation showed correlations were considerably lower for energy and macro-nutrients (r = 0.557 to 0.692) [16] than in the nutrition students [22]. This trend was also observed for the
RI PT
24 micro nutrients investigated (r= 0.081 to 0.776). Inadequate photograph angles and
insufficient descriptive information provided by participants reportedly hindered accuracy of
SC
the assessment [16,22].
The remote food photography method (RFPM) involves a similar concept to the Wellnavi
M AN U
system [17,18]. Photographs of foods consumed by participants were captured and labelled with short text descriptions before being sent via the cellular network to a server for analysis and coding by research dietitians. Ecological momentary assessment (EMA) prompts were used to remind and encourage participants to capture and send their food photographs along
TE D
with labelled information to research dietitians. Comparison of the RFPM with laboratory weighed foods in a group of healthy adults under laboratory (lunch and dinner) and freeliving (lunch in laboratory, pre-weighed dinner at home) conditions over 3 non-consecutive
EP
days demonstrated excellent correlations with energy intake (r = 0.93 and 0.95 respectively),
AC C
but significantly under-estimated energy intake in both conditions (mean difference: -368kj (88Kcal) or -4.7%, and -406kJ (-97kcal) or -6.6% respectively) [18]. The RFPM showed high reliability over the 3 day period when compared to laboratory weighed foods (ICC laboratory = 0.62; ICC free-living = 0.68, P < 0.0001). Good agreement between the three research dietitians for energy and macro nutrients (r= 0.85 to 0.92) was also observed. Most (78.8%) participants were satisfied with using the RFPM, and 96.6% preferred the RFPM over conventional pen and paper diaries.
ACCEPTED MANUSCRIPT Integration of a semi-automated process for image analysis was later used to refine the method [31-34]. Automated customised EMA prompts were also incorporated following the results from a previous study [17]. The RFPM under-estimated energy intake in a group of mostly overweight adults over a consecutive six day study period under free-living conditions
RI PT
(152 ± 694kcal/day; 3.7%), but did not differ significantly compared with doubly labelled water (DLW) (P = 0.16) [17]. This level of accuracy compares favourably with previous studies validating various self-reported dietary intake methods against DLW across different
SC
population groups [2,35]. Notably a backup recording method (pen and paper food diary) was used on 8.9% of study days, constituting 9.7% of participants' total estimated energy
M AN U
intakes. Compared with two laboratory weighed meals, the RFPM showed no significant differences for energy, macro or micro-nutrient intakes except for vitamin A and cholesterol (P