Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2025 Mar 1.
Published in final edited form as: Appetite. 2023 Dec 27;194:107176. doi: 10.1016/j.appet.2023.107176

An explanation for the accuracy of sensor-based measures of energy intake: Amount of food consumed matters more than dietary composition

Tommy Chou a, Adam W Hoover b, Stephanie P Goldstein a, Dante Greco-Henderson a, Corby K Martin c, Hollie A Raynor d, Eric R Muth e, J Graham Thomas a
PMCID: PMC10895650  NIHMSID: NIHMS1958827  PMID: 38154576

Abstract

Understanding and intervening on eating behavior often necessitates measurement of energy intake (EI); however, commonly utilized and widely accepted methods vary in accuracy and place significant burden on users (e.g., food diaries), or are costly to implement (e.g., doubly labeled water). Thus, researchers have sought to leverage inexpensive and low-burden technologies such as wearable sensors for EI estimation. Paradoxically, one such methodology that estimates EI via smartwatch-based bite counting has demonstrated high accuracy in laboratory and free-living studies, despite only measuring the amount, not the composition, of food consumed. This secondary analysis sought to further explore this phenomenon by evaluating the degree to which EI can be explained by a sensor-based estimate of the amount consumed versus the energy density (ED) of the food consumed. Data were collected from 82 adults in free-living conditions (51.2% female, 31.7% racial and/or ethnic minority; Mage = 33.5, SD = 14.7) who wore a bite counter device on their wrist and a used smartphone app to implement the Remote Food Photography Method (RFPM) to assess EI and ED for two weeks. Bite-based estimates of EI were generated via a previously validated algorithm. At a per-meal level, linear mixed effect models indicated that bite-based EI estimates accounted for 23.4% of the variance in RFPM-measured EI, while ED and presence of a beverage accounted for only 0.2% and .1% of the variance, respectively. For full days of intake, bite-based EI estimates and ED accounted for 41.5% and 0.2% of the variance, respectively. These results help to explain the viability of sensor-based EI estimation even in the absence of information about dietary composition.

Keywords: Ecological momentary assessment, eating, dietary monitoring, behavior, technology, energy intake estimation, dietary assessment, free-living, energy intake, ecological momentary assessment

Introduction

Energy intake (EI) is an important factor in the etiology and maintenance of many prevalent health conditions, particularly obesity (Archundia Herrera & Chan, 2018; Wadden et al., 2020). It is therefore a frequent target of research and intervention. Many methods have been developed to measure EI under free-living conditions (i.e., outside of laboratory settings, as an individual goes about their daily life), including assisted and independent dietary recall, doubly labeled water, and the Remote Food Photography Method (RFPM) (Archundia Herrera & Chan, 2018). Assisted and independent dietary recall incurs a high burden for researchers and patients/participants, and is costly to implement. Further, research has shown participants underestimate EI by 10%−50% (Champagne et al., 2002; Westerterp & Goris, 2002). Conversely, though doubly labeled water is less burdensome and more accurate, it remains costly and requires materials and technical expertise that are not widely available (Archundia Herrera & Chan, 2018; Westerterp, 2010). It also provides no information on when foods are eaten, and it provides only mean energy over the period of observation; hence, fluctuations in energy intake and energy intake at the day- or meal-levels are not measured. More recent efforts utilizing RFPM – a methodology involving taking pictures of meals before and after eating, which are then analyzed for caloric content by a trained dietician – provides accurate calorie counts and eliminates reliance on participant estimation and recall (Martin et al., 2012); however, RFPM requires participants to receive specialized training, depends upon their adherence to taking photographs at each meal, and again relies on the availability of personnel with technical expertise. Food diaries are commonly used in obesity treatment to assist in caloric restriction; however, due to well established problems with underreporting and a common pattern of rapid disengagement (Burke et al., 2008; Ravelli & Schoeller, 2020), food diaries are now infrequently used for research purposes.

Altogether, these challenges limit the utility of existing EI estimation methodologies, which in turn impacts the effectiveness of treatments that include EI monitoring as a key component, as well as the strength of conclusions drawn from scientific inquiry (Ravelli & Schoeller, 2020). In contrast, wearable technologies have grown in ubiquity, accessibility, and portability, spurring rising interest in wearable passive sensors as a means to infer aspects of eating behavior, including energy balance, with minimal or no input from the individual. Recent research at the intersection of behavioral medicine and engineering has demonstrated the potential for accurate EI estimation via wearable sensor-based approaches (e.g., Dong et al., 2012; Fontana et al., 2014, 2015). Several such approaches have now been developed and tested, including a neck-worn device with microphones attached to the jaw muscles and neck that listen for sounds of chewing and swallowing (Fontana et al., 2015), and an eyeglasses-mounted optical or strain sensor that detects chewing via contraction of the temporalis muscle (Fontana et al., 2014).

Likewise, research shows that wrist-worn gyroscopic sensors, like those found in most smartwatches and research-grade activity monitors, can accurately count the number of bites taken during a meal by tracking the unique wrist roll motion that occurs when bringing food to the mouth (Shen et al., 2017). Prior literature has established that a consistent relationship exists between bites measured via smartwatch and EI (Dong et al., 2012; Salley et al., 2016, 2019). This relationship is shown to vary by individuals’ characteristics such as biological sex, which allows EI to be estimated from bite counts (i.e., by multiplying number of bites by an average kcals per bite informed by individual characteristics such as biological sex). In fact, a landmark study found that the accuracy of bite-based EI estimates was significantly greater than that of participants’ self-estimates, even when participants were given kcal information about the food that they consumed (p < .001)(Salley et al., 2016). Specifically, when given kcal information about the foods consumed during an ad libitum meal, participants’ self-reported EI estimates had a mean estimation error of −257.36±790.22 kcal. By comparison, bite-based estimates of EI in the same study yielded more accurate estimates, with a mean error of 71.21±562.14 kcal. The accuracy of bite-based EI assessments has been replicated across both laboratory (Dong et al., 2012; Salley et al., 2016; Shen et al., 2017) and free-living studies (Scisco et al., 2014).

The apparent accuracy of bite-based EI estimates, which do not account for dietary composition, is surprising given conventional wisdom (Centers for Disease Control and Prevention, 2022) and prior research (Ledikwe et al., 2006) showing that dietary composition plays a critical role in determining EI. One would expect that to achieve adequate accuracy, a measure of EI would need to account for the energy density (ED) of foods consumed (i.e., kcal/gram), regardless of how much food is consumed. For example, a 275 g meal consisting of lean protein, whole grains, and vegetables might have an ED of 1.1 kcal/g for a total of 310 kcals, while a fast-food cheeseburger, French fries, and desert weighing the same amount might have an ED 5.3 kcal/g for a total of 1,455 kcal.

The current research aims to explain how bite-based estimates sufficiently approximate EI without additional ED information. Specifically, we conducted secondary analysis of data collected from adults recruited for a study to validate the accuracy of bite-based EI estimates in a free-living sample. For a two-week period, participants wore a wrist-based bite counter, and took photographs of their food before and after meals using RFPM, which was used to calculate dietary ED and EI for individual meals and full days of intake. Analyses examined the degree to which bite count accounted for variance in RFPM-measured EI (i.e., meal-to-meal and day-today differences in EI), and the degree to which ED accounted for variance in RFPM-measured EI above and beyond that which is explained by bite count. The presence of a beverage during the eating episode was also evaluated for its impact on EI variance (alone and in combination with ED and bite count). The error in bite-based EI estimates (i.e., the difference between bite- and RFPM-measured EI) was compared within ED quartiles, with a hypothesis that bite-based EI estimates would be less accurate for meals with particularly high or low ED. The findings provide important context for previous studies that have validated bite-based EI estimates, but have not explored how ED could be related to accuracy or variability in estimates.

Methods

2.1. Participants

We conducted secondary analyses on data drawn from a parent study which targeted enrollment of 82 individuals to achieve a sample size of 76 for inclusion in analysis, assuming 8% attrition. All participants provided adequate data and were therefore included in analysis. Participants were recruited via fliers, emails, and word of mouth from the faculty, staff, and student population of Clemson University and the surrounding geographical area. These advertisements directed interested individuals to complete an online screener for eligibility. Purposive sampling was used to selectively recruit for variance in age, sex, and ethnicity representative of the local population, and study criteria were limited to maximize generalizability. Adults age 18 years or older with no history of an anorexia nervosa or bulimia nervosa diagnosis were eligible to participate. Study procedures were reviewed and approved by the Clemson University Institutional Review Board (IRB# 2015–042).

2.2. Procedure

Study staff invited potential participants identified via online screening procedures for an initial training session, during which participants completed additional screening questionnaires and – if eligible – informed consent procedures and baseline surveys. Research staff measured participant height and weight, and provided a Bite Counter wristwatch device and smartphone with the SmartIntake RFPM application. The Bite Counter device is an independent wrist-worn device that allows a participant to denote the start and end of eating via button press, and uses wrist-roll motion to infer the number of bites taken during this known eating episode (Salley et al., 2016). A one-hour training involved orientation to the wrist-worn Bite Counter and the SmartIntake app, which participants used to capture images of food and drink per the RFPM protocol described in Measures. Research staff instructed participants to photograph their food and start Bite Counter tracking via button press at the start of each “eating activity”, defined as any snack, meal, or drink with a discrete start and end time and excluding “grazing” (i.e., otherwise eating and drinking on the go) throughout the day. Participants could eat and drink normally during each eating activity, then photograph the remaining food and end Bite Counter tracking via button press. Using the SmartIntake app, participants could add annotations to their before and after meal images. Upon completing a meal record, the data were automatically transmitted to the researchers. Additionally, participants were asked to upload their Bite Counter data to the researchers via a PC at the end of each day. Research staff reinforced the importance of compliance over the two-week period, indicated that participants could be removed from the study should they demonstrate an inability to adhere to study procedures, and confirmed willingness to move forward with study participation. If after the training participants did not wish to continue with the experiment, they were compensated $5 for their time. Ultimately, no participants were removed due to inadequate adherence.

Participants were asked to complete fourteen days of data collection, starting the day following their training session. To ensure rigorous collection of high-quality data, and to accomplish the parent study’s primary aim of validating bite-based EI estimates, the study team monitored compliance with data collection procedures by checking the transmitted SmartIntake and Bite Counter data. If participants’ photographs were not interpretable based on standards described in their training, research staff problem-solved by phone. Participants were allowed to extend the duration of the data collection period to promote data adequacy in the event that their participation was interrupted due to illness, family emergency, etc. All available data were included in analysis. Participants attended a debriefing session during which they returned their equipment and were compensated $350 for completing 14 days of data collection.

2.3. Measures

Remote Food Photography Method

The Remote Food Photography Method (RFPM) was used to estimate EI (Martin et al., 2012). Prior work has established the validity of RFPM as a measure of EI in both laboratory and free-living conditions (Martin et al., 2009, 2012, 2014). In particular, investigators have shown that estimates of EI based on RFPM do not significantly differ from those obtained using doubly labeled water in free-living conditions. In this study, RFPM was applied to images captured using the SmartIntake app and following procedures described briefly below, and in more detail elsewhere (i.e., Martin et al., 2014). Participants followed app instructions to photograph all food and drink before and after each eating activity. During the training session, participants were oriented to the use of a reference card (for scale), distribution of food on plates and in bowls to ensure clarity for evaluation, use of a 45-degree angle to ensure the photograph provides an optimal three-dimensional representation of food items, and inclusion of drinks in the image. Participants were asked to use SmartIntake’s text description functionality to provide information on specific ingredients, methods of preparation, location of eating, and other details not readily apparent in the photograph. All de-identified photographs and descriptors were uploaded to servers operated by the study team. Registered dieticians trained as RFPM raters accessed participants’ digital images asynchronously, and used the Food Photography Application© (Martin et al., 2014) to view each participant’s food selection at the start of each meal, leftovers at the end of each meal, and images of a standard portion size available through the USDA Food and Nutrition Database for Dietary Studies. Raters approximated the percent of the standard portion represented by foods at the start of the meal and the percent of the standard portion left over at the end of the meal. The Food Photography Application© then used values from the USDA Food and Nutrition Database for Dietary Studies to calculate ED of food selection and leftovers, and estimated EI as the difference between leftovers and original meal selection. Estimates were made for each meal as a whole, and not for individual foods within the meal. For example, if an eating occasion included chicken, rice, and beans, EI and ED were calculated for the whole meal as opposed to individually for each portion within the meal. Additionally, the program calculated ED and EI for food only, though the presence or absence of a beverage was noted.

Sensor-based Bite Count and EI Estimation

Bite count was estimated using an independent wristwatch device, called the Bite Counter, developed by the research team in numerous prior studies. The Bite Counter utilizes a gyroscopic sensor paired with an algorithm described in Dong and colleagues (2012) to infer the number of bites by monitoring the unique wrist rolling motion inherent to bringing food to one’s mouth. Participants were instructed to wear the device on their dominant wrist (primary hand they eat with), and press a button on the Bite Counter to indicate the start of an eating episode. They then ate their meal as usual, and pressed the button again at the end of their eating episode. The Bite Counter deactivated automatically after 60 minutes in the absence of a button press indicating the end of a meal. During monitoring, the Bite Counter screen displayed “ON” to indicate active data collection, but did not display bite count. Consistent with prior work, bite-based EI estimates were generated by multiplying a previously-validated estimate of kcals per bite by the total counted bites per meal (i.e., kcals/bite x total number of bites = bite-based EI estimate). Values of 11 kcals/bite for women and 19 kcals/bite for men were found to be the most accurate in our prior research (Scisco et al., 2014), and so these values were used to estimate bite-based EI based on gender in the current analysis. Previous work has shown that the Bite Counter accurately infers approximately 86% of bites taken in controlled environments where food items and utensil use is largely unrestricted.

2.4. Statistical Analysis

Data were analyzed using IBM SPSS Statistics for Windows, Version 28.01.0. Demographics and eating occasions were characterized using means and standard deviations, or counts with percentages, as appropriate. Linear mixed effect models (LMMs) were used to calculate variance accounted for in RFPM-measured EI by bite-based EI estimates, ED, and the presence of a beverage, using a pseudo R2 statistic calculated using the method described by Nakagawa & Schielzeth (2013). Pseudo R2 functions similarly to a traditional R2 statistic in that it ranges from 0–1 and the value can be interpreted as the proportion of variance in the outcome accounted for by the predictor variables. The marginal R2 statistic is reported given that it only accounts for fixed effects (bite counts, ED, and the presence of a beverage) and not the random effect of subject, which would not typically be measured or known in practical applications of bite counts for EI estimation. Because it only accounts for fixed effects, the marginal R2 is also the more conservative statistic. Two sets of LMMs were generated, one with intake from individual eating occasions as the dependent variable (eating occasion represented at level 1 in the analysis), and one with total EI for the day as the dependent variable (day represented at level 1 in the analysis). The latter could not accommodate a variable for the presence of a beverage because this varied inconsistently across eating occasions within a given day. In all models, the participant was treated as a random effect and an unstructured covariance matrix was used. This approach accounted for non-independence of observations within participant. Model fit was compared using the likelihood ratio test. To test whether the discrepancy between bite-based and RFPM EI estimates was greater for eating events with more extreme values of ED, the error in bite-based EI estimates (i.e., the difference between bite-based EI and RFPM-measured EI) was compared via LMM across eating events in the highest quartile of ED, those in the middle two quartiles of ED, and those in the lowest quartile of ED.

All participants meeting a minimum threshold of ≥10 eating occasions were included in the analysis. This threshold was selected to ensure adequacy of the data and avoid potential bias resulting from participants with few, and therefore potentially non-representative, eating occasions. Critical alpha was set at 0.05 for statistical tests; a correction for multiple comparisons was not implemented due to the exploratory nature of the analysis.

RESULTS

3.1. Participants

Eighty-three participants consented to participate. One opted not to continue after the training session; and eight-two participants engaged in data collection, recorded ≥10 eating occasions, and were therefore included in the analysis. The mean (SD) age of these participants was 33.5 (14.7) years and the average body mass index (BMI) was 28.1 (7.7) kg/m2. About half (n = 42, 51.2%) identified as women and n=50 (61.0%) had a BMI ≥25 kg/m2, putting their BMI in the overweight or obese range. Twenty-six (31.7%) identified as a member of a racial and/or ethnic minority population.

3.2. Eating Occasions

There were 3,094 eating events captured by the Bite Counter, 2,956 eating events captured by RFPM, and 2,884 eating events captured by both and included in analysis. Bite count events for eating events wherein the Bite Counter timed out at 60 minutes (i.e., the participant started tracking via button press at the start of the meal, but did not press the Bite Counter button at the end of the meal) were excluded from analyses. The mean (SD) number of days of data collection per participant was 13.8 (1.0), and the mean number of eating occasions per participant was 35.2 (8.8). A beverage was recorded by the expert raters in 28.2% of the eating occasions. Unconditional LMMs (i.e., with no predictors) indicated that mean (SE) RFPM-measured kcal was 479.5 (14.7) per eating occasion and 1210.4 (44.0) per day. There were 30.2 (1.1) bites recorded per eating occasion and 77.2 (3.4) bites per day. Corresponding bite-based EI estimates were 452.2 (22.6) kcals per eating occasion and 1146.1 (61.6) per day. The average ED was 2.2 (0.05) per eating occasion, and 2.0 (0.04) for all food consumed within a given day.

3.3. Variance accounted for in RFPM-measured EI

As reported in Table 1, bite-based EI estimates independently accounted for 23% of the variance in RFPM-measured EI (R2= 0.234) for individual eating occasions. Energy density independently accounted for 0.2% of the variance in RFPM-measured EI (R2= 0.002). When combined, these two variables accounted for 25.2% of the variance in RFPM-measured EI (R2= 0.252). Adding ED to bite-based EI estimates significantly improved model fit (χ2 = 61.62, p <.001). Adding a beverage indicator to bite-based EI estimates accounted for a total of 23.4% of the variance in RFPM-measured EI (R2= 0.234) and also significantly improved model fit (χ2 = 9.80, p =.002). The three variables combined accounted for 25.3% of the variance in RFPM-measured EI (R2= 0.253) and produced the best model fit compared to all other models (ps < .001). Per Table 1, the same pattern was observed for full days of intake; ED accounted for relatively little additional variance in EI, above and beyond that captured by bite-based EI estimates. As would be expected, variance accounted for by bite count was substantially higher for full days of intake than for individual eating occasions, reflecting greater accuracy when estimating intake over a longer period (R2=0.415 for bite count alone, R2= 0.429 for bite count plus energy density).

Table 1.

Variance (pseudo R2) in RFPM-measured energy intake accounted for by bite-based energy intake estimates, energy density, and presence of a beverage.

Predictor(s) in the model Eating Occasion
 Full Day of Intake
R2 R2
Bite count-based EI (BC) 0.234 0.415
Energy Density (ED) 0.002 0.002
Beverage (Bev) 0.001 --
BC + ED 0.252 0.427
BC + Bev 0.234 --
BC + ED + Bev 0.253 --
Open in a new tab

3.4. Error in bite-based EI estimation by RFPM-measured ED

There was a statistically significant difference in the mean error of bite-based EI estimates by ED quartile (p < 0.001). For meals in the lowest ED quartile, mean error (SE) was 121.82 kcal (20.49), which was significantly greater than the mean error of both the middle two quartiles (p < .001) and the highest ED quartile (p < .001). For meals in the middle two ED quartiles, mean error was −55.02 kcal (18.28), which was significantly higher than that of the highest ED quartile (p < .05). The mean error for meals in the highest ED quartile was −82.60 kcal (20.00).

DISCUSSION

Previous research has shown that a wrist-worn sensor can be used to generate sufficiently accurate bite-based estimates of EI. This study sought to explain how that is possible given that such estimates do not account for dietary composition. To achieve this goal, LMMs were used to test the degree to which bite-based EI, ED from RFPM, and presence of a beverage accounted for meal- and day-level variability in RFPM-measured EI among individuals tracking their food intake over a period of 14 days. While these analyses were conducted secondarily on data not collected for this purpose, the spread in ED aligns with (and exceeds) work conducted in a nationally representative sample drawn from the Continuing Survey of Food Intakes by Individuals dataset including over 9,000 subjects, which estimated an average ED of 1.85 (0.01) for all food consumed within a 24-hour period (Ledikwe et al., 2005). In our dataset, the ED was 1.3 at the 25th percentile, 2.0 at the 50th percentile, 2.8 at the 75th percentile, and 3.8 at the 90th percentile. By comparison, Ledikwe and colleagues (2005) data analyses showed an ED of 1.56 at the 25th percentile, 1.88 at the 50th percentile, 2.23 at the 75th percentile, and 2.54 at the 90th percentile among men. Among women, they found an ED of 1.43 at the 25th percentile, 1.75 at the 50th percentile, 2.11 at the 75th percentile, and 2.47 at the 90th percentile. Thus, these data were well suited for this purpose.

Results showed that bite count-based estimates of EI accounted for the 23.4% of the variance in RFPM-measured EI. Adding ED and beverage intake to bite-based EI estimates significantly improved model fit, but explained only about 2% additional variance in RFPM-measured EI. These results suggest that bite-based estimates of EI are able to achieve a high degree of accuracy because most of the meal-to-meal and day-to-day variability in EI can be captured by measuring how much food an individual consumes. Despite evidence that adequate EI estimates can be generated using only a bite-based measure of the amount of food consumed, dietary composition is not irrelevant.

This study’s results show that for eating events with particularly low ED, EI calculated from bite count significantly overestimated calories consumed. Likewise, bite-based estimates significantly underestimated the number of calories consumed for meals with especially high ED. This pattern aligns with expectations; for less energy dense meals, an EI estimate based on amount of food consumed should yield a value that is too high, and vice versa. While the average magnitude of the error is relatively modest, these results suggest that incorporating information about dietary composition in a bite-based EI estimation approach might improve accuracy.

Notably, even the best-performing models of EI in this study only accounted for approximately 40% of the variance in EI. As described above, we computed the marginal R2 statistic, which is appropriate for this study but produces lower overall values than the conditional R2 statistic, and is therefore more conservative. Aside from measurement error, which is to be expected, there are no additional data available from the primary study to account for the unexplained variance. As such, a critical area for future research is continuing to explore additional factors contributing to differences in EI within and across persons, including but not limited to: social context, performance of secondary activities during eating (e.g., watching TV vs. walking vs. driving), and other characteristics of eating (e.g., duration of an eating occasion) (Braude & Stevenson, 2014; Goldstein et al., 2022; Lock et al., 2016; Ogden et al., 2017; Ruddock et al., 2021).

This secondary analysis had several strengths. The use of previously validated, real-time assessment of bite counts, ED, and EI in free-living distinguishes this study from many others to date, which have typically utilized food diaries and self-report. These innovative dietary assessment approaches allowed the current study to be conducted entirely outside of the lab (e.g., in free-living), which substantially improves the ecological validity of the results.

The results of this study should also be interpreted in the context of important limitations. First, while the RFPM is a rigorous approach to measuring free-living EI and ED, it is not without error. Thus, both bite-based EI and the measure of EI it was compared, and the measure of ED, involve some degree of measurement error. Per above, this likely contributes to only about 40% of the variance in RFPM-measured EI being accounted for in this study. Importantly, the goal of this study was not to validate bite-based EI estimation; that has been the focus of previous research. Rather, this study explored how bite-based EI estimates achieve the accuracy described in prior research without accounting for dietary composition.

Additional limitations of this study include the requirement for participants to press a button on the wearable bite counter device to indicate the start and end of their eating, in addition to taking pictures of their food before and after eating for RFPM. The burden of completing both of these tasks for every eating occasion over 14 days may have contributed to reactivity and/or non-compliance. The average RFPM-measured EI of 1146 kcals/day, despite not including energy from beverages, is not likely enough to maintain participants’ weight. While it is not expected that missing data would bias the estimates of variance accounted for in EI by the independent variables selected for the present research, future replication studies should consider using methods that are less reliant on participant adherence (e.g., devices that automatically detect when the user is eating). Next, while results highlighted the potential importance of incorporating beverage intake to estimate EI, there were limited data on beverage intake available from RFPM. Also, this study utilized a widely-supported and gold-standard measure of ED (i.e., kcal/gram); however, there are several possible ways to calculate ED and so the results described herein could differ depending on the measure of ED being applied (Ledikwe et al., 2005; Stubbs et al., 2000). Lastly, emerging research on passive sensing devices indicates that the accuracy of algorithms for inferring and characterizing eating can be sensitive to context and differ across devices (Bell et al., 2020). Results from this study, which rely heavily on a wearable device and associated algorithms for characterizing eating, should therefore be interpreted with caution and replicated across new settings, diverse groups of individuals, using different devices for eating monitoring, and with different ground truth measures of EI.

Conclusion

The current study demonstrated that bite count accounts for much of the meal-to-meal and day-to-day variability in EI. While a measure of dietary composition (i.e., ED) did not account for much of the variance in EI above and beyond that already explained by bite count, error in bite-based EI estimates was greater for eating events with particularly high or low ED. Indirect estimation of intake remains a major challenge regardless of the method used, but this secondary analysis builds on previous research that validated bite-based methods of estimating EI by exploring how dietary composition is related to bite-based EI estimates, and the conditions under which ED may influence their accuracy.

Acknowledgements

We thank the study participants for their contributions to the research.

The authors report grants from NIDDK (K23 DK136978-01), NIGMS, and NIMH (5T32MH019927) during the conduct of the study and preparation of this manuscript, including NORC Center Grant P30 DK072476 entitled “Nutrition and Metabolic Health Through the Lifespan” sponsored by NIDDK, and grant U54 GM104940 from the National Institute of General Medical Sciences, which funds the Louisiana Clinical and Translational Science Center. Dr. Thomas also reports personal fees from Lumme Health, Inc. and Medifast, Inc.

Funding.

This work was supported by the National Heart, Lung, and Blood Institute (R01HL118181); the National Institute of Diabetes and Digestive and Kidney Diseases (R01DK135679); and the National Science Foundation (2242812).

Footnotes

Publisher's Disclaimer: 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.

Declaration of Interest:

The other authors have no conflicts of interest to declare. The intellectual property surrounding the Remote Food Photography Method© (RFPM) and SmartIntake® app are owned by Pennington Biomedical Research Center/Louisiana State University; CKM is an inventor of the technology.

Ethical Statement

This study was performed in accordance to the Declaration of Helsinki and approved by the Clemson University Institutional Review Board (IRB# 2015–042).

References

  1. Archundia Herrera MC, & Chan CB (2018). Narrative review of new methods for assessing food and energy intake. In Nutrients (Vol. 10, Issue 8). MDPI AG. 10.3390/nu10081064 [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Bell BM, Alam R, Alshurafa N, Thomaz E, Mondol AS, de la Haye K, Stankovic JA, Lach J, & Spruijt-Metz D (2020). Automatic, wearable-based, in-field eating detection approaches for public health research: a scoping review. NPJ Digital Medicine, 3, 38. 10.1038/s41746-020-0246-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Braude L, & Stevenson RJ (2014). Watching television while eating increases energy intake. Examining the mechanisms in female participants. Appetite, 76, 9–16. 10.1016/j.appet.2014.01.005 [DOI] [PubMed] [Google Scholar]
  4. Burke LE, Sereika SM, Music E, Warziski M, Styn MA, & Stone A (2008). Using instrumented paper diaries to document self-monitoring patterns in weight-loss. Contemp Clin Trials, 29(2), 182–193. 10.1016/j.cct.2007.07.004 [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Centers for Disease Control and Prevention. (2022, February 13). Healthy Weight, Nutrition, and Physical Activity https://www.cdc.gov/healthyweight/calories/index.html
  6. Champagne CM, Bray GA, Kurtz AA, Resende Monteiro JB, Tucker E, Volaufova J, & Delany JP (2002). Energy intake and energy expenditure: A controlled study comparing dietitians and non-dietitians. Journal of the American Dietetic Association, 102(10), 1428–1432. 10.1016/S0002-8223(02)90316-0 [DOI] [PubMed] [Google Scholar]
  7. Dong Y, Hoover A, Scisco J, & Muth E (2012). A new method for measuring meal intake in humans via automated wrist motion tracking. Applied Psychophysiology Biofeedback, 37(3), 205–215. 10.1007/s10484-012-9194-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Fontana JM, Farooq M, & Sazonov E (2014). Automatic ingestion monitor: A novel wearable device for monitoring of ingestive behavior. IEEE Transactions on Biomedical Engineering, 61(6), 1772–1779. 10.1109/TBME.2014.2306773 [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Fontana JM, Higgins JA, Schuckers SC, Bellisle F, Pan Z, Melanson EL, Neuman MR, & Sazonov E (2015). Energy intake estimation from counts of chews and swallows. Appetite, 85, 14–21. 10.1016/j.appet.2014.11.003 [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Goldstein SP, Hoover A, & Thomas JG (2022). Combining passive eating monitoring and ecological momentary assessment to characterize dietary lapses from a lifestyle modification intervention. Appetite, 175, 106090. 10.1016/j.appet.2022.106090 [DOI] [PubMed] [Google Scholar]
  11. Ledikwe JH, Blanck HM, Kettel Khan L, Serdula MK, Seymour JD, Tohill BC, & Rolls BJ (2006). Dietary energy density is associated with energy intake and weight status in US adults. The American Journal of Clinical Nutrition, 83(6), 1362–1368. 10.1093/ajcn/83.6.1362 [DOI] [PubMed] [Google Scholar]
  12. Ledikwe JH, Blanck HM, Khan LK, Serdula MK, Seymour JD, Tohill BC, & Rolls BJ (2005). Dietary energy density determined by eight calculation methods in a nationally representative United States population. The Journal of Nutrition, 135(2), 273–278. 10.1093/jn/135.2.273 [DOI] [PubMed] [Google Scholar]
  13. Lock C, Brindal E, Hendrie GA, & Cox DN (2016). Contextual and environmental influences on reported dietary energy intake at evening eating occasions. Eating Behaviors, 21, 155–160. 10.1016/j.eatbeh.2016.03.012 [DOI] [PubMed] [Google Scholar]
  14. Martin CK, Correa JB, Han H, Allen HR, Rood JC, Champagne CM, Gunturk BK, & Bray GA (2012). Validity of the remote food photography method (RFPM) for estimating energy and nutrient intake in near real-time. Obesity, 20, 891–899. 10.1038/oby.2011.344 [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Martin CK, Han H, Coulon SM, Allen HR, Champagne CM, & Anton SD (2009). A novel method to remotely measure food intake of free-living individuals in real time: The remote food photography method. British Journal of Nutrition, 101(3), 446–456. 10.1017/S0007114508027438 [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Martin CK, Nicklas T, Gunturk B, Correa JB, Allen HR, & Champagne C (2014). Measuring food intake with digital photography. Journal of Human Nutrition and Dietetics, 27(SUPPL.1), 72–81. 10.1111/jhn.12014 [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Nakagawa S, & Schielzeth H (2013). A general and simple method for obtaining R2 from generalized linear mixed-effects models. Methods in Ecology and Evolution, 4(2), 133–142. 10.1111/j.2041-210x.2012.00261.x [DOI] [Google Scholar]
  18. Ogden J, Oikonomou E, & Alemany G (2017). Distraction, restrained eating and disinhibition: An experimental study of food intake and the impact of “eating on the go”. Journal of Health Psychology, 22(1), 39–50. 10.1177/1359105315595119 [DOI] [PubMed] [Google Scholar]
  19. Ravelli MN, & Schoeller DA (2020). Traditional Self-Reported Dietary Instruments Are Prone to Inaccuracies and New Approaches Are Needed. In Frontiers in Nutrition (Vol. 7). Frontiers Media S.A. 10.3389/fnut.2020.00090 [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Ruddock HK, Brunstrom JM, & Higgs S (2021). The social facilitation of eating: why does the mere presence of others cause an increase in energy intake? Physiology & Behavior, 240, 113539. 10.1016/j.physbeh.2021.113539 [DOI] [PubMed] [Google Scholar]
  21. Salley JN, Hoover AW, & Muth ER (2019). Between- and Within-Subjects Predictors of the Kilocalorie Content of Bites of Food. Journal of the Academy of Nutrition and Dietetics, 119(7), 1109–1117. 10.1016/j.jand.2018.12.009 [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Salley JN, Hoover AW, Wilson ML, & Muth ER (2016). Comparison between Human and Bite-Based Methods of Estimating Caloric Intake. Journal of the Academy of Nutrition and Dietetics, 116(10), 1568–1577. 10.1016/j.jand.2016.03.007 [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Scisco JL, Muth ER, & Hoover AW (2014). Examining the Utility of a Bite-Count–Based Measure of Eating Activity in Free-Living Human Beings. Journal of the Academy of Nutrition and Dietetics, 114(3), 464–469. 10.1016/j.jand.2013.09.017 [DOI] [PubMed] [Google Scholar]
  24. Shen Y, Salley J, Muth E, & Hoover A (2017). Assessing the Accuracy of a Wrist Motion Tracking Method for Counting Bites Across Demographic and Food Variables. IEEE Journal of Biomedical and Health Informatics, 21(3), 599–606. 10.1109/JBHI.2016.2612580 [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Stubbs J, Ferres S, & Horgan G (2000). Energy density of foods: effects on energy intake. Critical Reviews in Food Science and Nutrition, 40(6), 481–515. 10.1080/10408690091189248 [DOI] [PubMed] [Google Scholar]
  26. Wadden TA, Tronieri JS, & Butryn ML (2020). Lifestyle modification approaches for the treatment of obesity in adults. In The American psychologist (Vol. 75, Issue 2, pp. 235–251). NLM (Medline). 10.1037/amp0000517 [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Westerterp KR (2010). Physical activity, food intake, and body weight regulation: Insights from doubly labeled water studies. Nutrition Reviews, 68(3), 148–154. 10.1111/j.1753-4887.2010.00270.x [DOI] [PubMed] [Google Scholar]
  28. Westerterp KR, & Goris AHC (2002). Validity of the assessment of dietary intake: problems of misreporting. Current Opinions in Clinical Nutrition and Metabolic Care, 5, 489–493. 10.1097/00075197-200209000-00006 [DOI] [PubMed] [Google Scholar]

RESOURCES