Novel Energy Balance Tracking to Support Personalised AI Health Coaching: A Real-World Evaluation of the ENHANCE Framework

Novel Energy Balance Tracking to Support Personalised AI Health Coaching: A Real-World Evaluation of the ENHANCE Framework | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Technical Report Novel Energy Balance Tracking to Support Personalised AI Health Coaching: A Real-World Evaluation of the ENHANCE Framework Arthur Daw, Maxime Petit, Scott Willis, Lewis James, James King This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6878171/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 9 You are reading this latest preprint version Abstract Background: Energy balance (EB) is the key determinant of fat gain, yet accurate EB tracking is difficult outside laboratory settings. Traditional methods are burdensome (e.g. food-logs) or lack daily resolution (e.g. body weight monitoring), limiting suitability for integration with free-living AI-powered health-coaching. Objective: To introduce ENHANCE—a novel framework prioritising interpretability and temporal accuracy—and demonstrate its use as a low-burden, accurate method for tracking EB using smart devices and minimal self-report, suitable for AI coaching. Methods: This 4-week observational study spanned the Christmas to New Year2024/25 festive period. Participants submitted daily blinded body weight measurements via Wi-Fi scales and EB-related questions via a mobile app, taking <2 minutes. Data were used to generate five weight trends: raw (from scales), smoothed (±3-day average), piecewise (3-segments), predicted (from EB), and corrected. The correction aligned predicted and smoothed trends, using proximity and noise-weighted adjustments, producing enhanced data for AI coaching. An end-of-study questionnaire assessed acceptability and behavioural reactivity. Results: Of 23 participants, 18 were analysed. Five were excluded due to illness (n = 4) or bereavement (n = 1). Participants completed 94% (5.1%) of body weight measurements and 100% of EB-related submissions. Questionnaire results showed low burden (1.8/5) and behavioural reactivity (1.5/5). Group-level predicted trends explained 90.4% of smoothed trend variance (R² = 0.904; mean absolute error [MAE]: 93 g). Corrected trends aligned more closely with piecewise segments than raw trends (MAE: 46 g vs 77 g). Individual-level mean EB corrections were +41 kcal/day—just 2% of reported intake. The corrected trend enhanced interpretability and plausibility while preserving real-world validity. Calculated mean net fat weight change during the monitoring phase was +0.8 kg (0.4 kg); mean net EB was +223 kcal/day (130 kcal/day). Conclusions: This scalable method delivers the accuracy and practicality needed for real-world EB tracking—laying the foundation for continuous personalised AI coaching. Health sciences/Health care/Weight management Health sciences/Health care/Disease prevention/Lifestyle modification Health sciences/Health care/Disease prevention/Preventive medicine Artificial Intelligence Digital health Mobile health Energy Balance Energy intake Obesity Weight management Behaviour change. INTRODUCTION Artificial Intelligence (AI) is transforming healthcare( 1 ). A promising application is personalised AI coaching apps that monitor health using smart device data, including physical activity, sleep, blood pressure, stress, and weight. Energy balance (EB) is often overlooked but is central to cardiometabolic health( 2 ) and foundational to weight control models, embodying principles of energy conservation and regulatory biology( 3 ). Accurate free-living EB tracking is therefore an essential requirement of AI-based coaching platforms( 4 ) yet remains challenging to measure( 2 ). EB can be estimated from body weight changes using predictive equations, providing a low-burden solution( 5 ), but lacks the resolution needed for daily feedback. Conversely, daily EB tracking using estimates of energy intake (EI) (via food-tracking) and energy expenditure (EE) (via smart device data) improves precision but requires ongoing user engagement( 6 ). Although these methods are initially accepted, user engagement drops rapidly, limiting long-term effectiveness( 7 ). Sustained AI-based support requires reliable EB tracking, even during low-engagement high-risk periods( 4 ), such as holidays when 0.5–1 kg fat gain is common( 8 ). To maximise the potential of AI coaching, valid methods are needed to reliably track EB year-round, including during high-risk periods( 9 )( 10 ). This report introduces: ENHANCE Framework : ENHANCE (Estimation and Normalisation for Holistic Alignment, Nuance, and Contextual Enhancements) improves tracking accuracy and interpretability by combining holistic temporal data with physiological and behavioural relationships. Its aim is to enhance user decision-making, engagement, and outcomes in free-living settings—not to generate a ground truth comparable to laboratory measurements. ENHANCE EB Tracking : A low-burden daily EB tracking method combining smart device data with minimal self-report, implemented using the ENHANCE framework. We present the conceptual framework, implementation, and preliminary evaluation—delivered through the Twists.com AI coaching platform ( 11 )—and evaluate its feasibility for continuous AI-based coaching in real-world settings. METHODS Ethical Approval Ethical approval was granted by Loughborough University Ethics Advisory Committee (Ref: 21271) and the study complied with the Declaration of Helsinki(12). All participants gave digital consent. Data Architecture and Anonymisation Participants received a unique identification number and pseudonym during onboarding. Body weight and app data were transmitted over Wi-Fi and stored on a Microsoft database hosted on Amazon Web Services. Access was restricted to authorised researchers via a secure software-as-a-service interface. Questionnaire responses were collected via Google Forms and linked to participants via date of birth only. Study Design and Setting This remote observational study was conducted between 08 December 2024 to 15 January 2025. The Christmas to New Year festive period is a high-risk EB disruptor—due to increased EI, reduced EE, and competing personal demands(8)—making it ideal to evaluate method performance. Study Phases Onboarding (08 December–10 December): Participants connected their Wi-Fi scales, installed the Twists app, and estimated typical EB under habitual conditions. Baseline (08 December–14 December): Participants established a baseline body weight trend while familiarising themselves with the scales and app. Baseline data configured the platform but were excluded from results. Monitoring (15 December–10 January): Participants recorded blind daily body weight measurements and answered EB-related questions. Feedback was withheld to preserve naturalistic behaviour. Feedback (10 January–15 January): Participant acceptability and behavioural reactivity were assessed using an end-of-study multiple-choice questionnaire. Participants and Recruitment Participants were recruited online from Gloucestershire, UK. Inclusion criteria were: men and women, 20–49 years, BMI 18.5–29.9 kg/m², stable body weight, and not travelling overnight >1 night. Twelve is a recommended sample for feasibility (13), but 23 were enrolled to support individual-level prediction error analysis. To preserve naturalistic behaviour, participants were only informed that the study was investigating how the festive period affects population EB. Weight monitoring was not discussed, and readings were blinded. An end-of-study questionnaire assessed behavioural reactivity. Data Collection Protocol Participants received Wi-Fi-enabled scales and proprietary Twists app for daily EB-related submissions—both scales and app were developed by Surpic Limited. Twists provides structured, goal-oriented tracking and AI coaching, but was modified for this study to ensure participants remained blinded to EB and weight feedback. During onboarding, participants watched a demonstration video and received a text-based guide. Each morning, participants recorded blinded weight and self-reported the prior day’s EB-related data. Daily tasks took <2 minutes. Notifications were sent at 6:00 AM (silent)—for measurement reminders, 2:00 PM—if self-reports were missed, and 4:00 PM—if measurements were missed. Weight Measurements via Wi-Fi Scales Each morning, participants weighed themselves after toileting, ideally nude or wearing consistent minimal clothing. Scales were placed on hard, flat surfaces. Displays confirmed measurement and Wi-Fi upload but did not show weight. The scales were manufactured by Surpic Limited (Cheltenham, UK) and assembled and UKCA-certified by Zhongshan Frecom Electronic Limited (Guangdong, China). Raw body weight data were recorded to 0.1 kg resolution. Scale accuracy was independently validated by Société Générale de Surveillance (Geneva, Switzerland), confirming ANSI/ASQ Z1.4 (2003, R2018) standards at AQL Level II(14), and precision ±0.2–0.3 kg(15). Self-Reporting via Mobile App Three daily EB-related metrics were collected via the Twists app: eating (%), steps (count), and exercise (kcal). To minimise time burden and enhance resilience to device issues, participants could choose from multiple simple input methods per metric. Eating was estimated as a percentage deviation from ‘typical’ intake. Participants were weight stable at onboarding, therefore typical EE was used to estimate typical EI. Participants either selected a deviation percentage from five preset options or entered a custom value from −100% to +100%. This method was intentionally approximate and not intended to provide precise estimates of EI, but to generate a plausible trend line with minimal input for further analysis. The five preset options were: Well Above: ~66% more than typical Above: ~33% more than typical Typical: ~0% from typical Below: ~33% less than typical Well Below: ~66% less than typical Steps were estimated using mobile data, wearable devices, four preset levels (2,500–10,000 steps), or manual entry—data were used to calculate non-exercise activity thermogenesis (NEAT)(16). Exercise estimates could be imported from wearable devices, entered manually or calculated using a guided wizard based on Metabolic Equivalent of Task (MET) codes(17)—data were used to calculate exercise activity thermogenesis (EAT). For steps and exercise, participants were encouraged to use device-based inputs to minimise bias. But to improve resilience to device unreliability, they could choose an alternative method if they considered it more accurate on that day. Daily Energy Balance Calculation Daily EB was calculated as follows: EB = EI − EE EI = Typical EI × Self-reported day modifier . Typical intake was estimated during onboarding from typical EE. This approach is novel, and balances analytical accuracy with real-world practicality. EE = REE + TEF + NEAT + EAT . Resting energy expenditure (REE) was calculated using the Mifflin–St Jeor equation(18). Thermic effect of feeding (TEF) was estimated at 10% of EI(3). NEAT was derived from mobile, wearable, or self-reported step counts using MET codes(17). EAT was imported from manufacturer wearable data or estimated using MET codes. Weight Trend Generation and Analysis Five weight change trends were generated to support the analysis, each offering a different perspective on EB: Raw Trend: Unprocessed body weight data collected each morning from the Wi-Fi scales. Although objective, these measurements vary with natural hydration shifts and data anomalies, making them misleading if interpreted directly as fat weight change. Smoothed Trend: Generated by removing outliers from raw data using an Interquartile Range (IQR) method (IQR×1.5) and applying a simple moving average (±3 days). This reduces noise while retaining directional trajectories, providing a cleaner reference for further processing. Piecewise Trend: Segmented the smoothed trend into three linear phases, each representing a consistent rate of change: pre-Christmas (15-23 December), festive peak (24 December–1 January), and post-New Year (2–10 January). These segments provided a benchmark to evaluate alternative trends. Predicted Trend: Calculated from self-reported EB values (EI − EE) to estimate daily fat weight change—estimated at 1 kg per 7,700 kcal EB(19). While unaffected by hydration shifts, it is influenced by reporting error and bias. The prediction serves as an input for further processing, rather than an exact estimate of fat change. Corrected Trend: Generated by aligning predicted and smoothed trends to enhance results. A ±5-day weighted average from the smoothed trend prioritised nearby low-noise values. The result was used to refine the predicted data to produce a corrected trend grounded in user behaviour and biological plausibility, while aligning with observed weight trends. Method Evaluation As a novel method, performance was evaluated through trend comparisons to assess improvements in stability, interpretability, and biological plausibility. Raw vs Piecewise Trend: Evaluates how well raw data aligns with the piecewise trend. Large deviations of >±0.35 kg imply noise from hydration shifts or data anomalies(19)(20). Corrected vs Piecewise Trend: Assesses how closely the corrected trend tracks the piecewise, confirming reduced noise without introducing artefacts. Corrected vs Raw Trend: Confirms reduced noise and recovery of coherent weight signals. Suppressing daily fluctuations >±0.35 kg enhances interpretability while preserving longer-term fat weight trends(19)(20). Signal-to-Noise Ratio (SNR) : Calculated as the ratio of overall variability (SD of weight) to short-term fluctuation (SD of daily changes). Higher post-correction SNR indicates a clearer reflection of fat weight change. Weight Trend Autocorrelation: Calculated as the lag-1 Pearson correlation of daily changes, this reflects directional consistency. Higher autocorrelation post-correction suggests more physiologically coherent progression. Weight Trend Plausibility: Biologicalplausibility was assessed by comparing implied EB to realistic bounds (−100% to +250% of typical EE). The same thresholds were applied post-correction to ensure results remained within plausible limits. RESULTS Participants and Data Completeness Of the 23 participants who enrolled and completed the protocol, 18 were included in the final analysis (Table 1). Five were excluded due to atypical behaviour from illness (n = 4) or bereavement (n = 1). At the individual level, adherence to daily body weight measurements was high, with a mean of 94.0% (5.1%). A small proportion of measurements were excluded due to lateness (3.2% after midday) or outliers (3.9% outside IQR threshold), retaining 87.4% of possible scans for analysis. Compliance with EB-related self-reporting of eating, steps and exercise estimates was 100%. Calculated individual-level mean net fat weight change during the monitoring phase was +0.8 kg (0.4 kg); mean net EB was +223 kcal/day (130 kcal/day). Trend Performance Raw Trend: Group-level signal quality was low (SNR: 2.69), and autocorrelation was negative (−0.25), indicating limited coherence and poor behavioural interpretability. Daily changes in raw weight were highly volatile, with an individual-level mean of 0.62 kg (0.22 kg), often masking meaningful physiological trends (Figure 1). Individual-level false direction reversals—where raw weight changes contradicted reported EB, reducing interpretability—occurred on 38.9% (6.7%) of days. Biological plausibility was low, with only 57% of days within realistic bounds (−100% to +250% of typical EE). Piecewise Trend: Group-level piecewise analysis identified three segments, each with distinct fat weight trends: a modest rise pre-Christmas (+4.3 g/day), a sharp increase during the festive peak (+58.2 g/day), and modest gain post-New Year (+14.9 g/day). The corrected trend aligned more closely with these segments than either raw or predicted. Predicted Trend: Group-level predicted trajectory explained 90.4% of the variance in smoothed trends (R²: 0.904), showing high internal consistency. Errors were low: mean absolute error (MAE) 93 g (63 g), root mean square error (RMSE) 112 g, and mean bias error (MBE) −21 g. This accuracy and stability justified its use by the correction method. Corrected Trend: The correction method refined daily EB estimates by aligning predicted and smoothed trends using a proximity- and noise-weighted smoothing window. This improved stability, interpretability and biological plausibility. Corrected values showed reduced individual-level variability in daily changes: 98 g (48 g); and improved group-level signal quality – SNR: 9.52; and stronger autocorrelation: 0.87—all indicative of a more coherent behavioural signal. Compared to predicted and raw, group-level corrected trends showed the lowest deviation from piecewise (MAE: 102 g, 77 g, and 46 g, respectively), and the highest alignment (r = 0.88)—see Figure 1 for visual comparisons. At the individual level, false direction reversals were eliminated. The final EB corrections were modest: the mean adjustment was +41 kcal/day (189 kcal/day)—just 2% (6.3%) of reported intake. Daily corrections exceeding 30% were rare, occurring on just 7.8 % (8.4 %) of days—supporting the method’s restraint and interpretability. Additional comparisons are shown in Table 2. Questionnaire Findings Post-study questionnaires (n = 18) indicated high user acceptability and low burden (Table 3). Confidence in self-estimates was high across eating, steps, and exercise. Participants reported negligible behavioural reactivity or impact on Christmas enjoyment. DISCUSSION Principal Findings This study evaluates the ENHANCE framework and a novel low-burden method for estimating daily EB using smart devices and minimal self-report. Tested during the demanding festive period, it achieved high adherence and low participant burden. Raw weight data were too volatile and behaviourally implausible for effective use in AI-based coaching. Despite intentionally approximate EB inputs, predicted trends aligned closely with piecewise segments, providing a credible basis for correction. Aligning predicted trends with device measurements produced corrected values that more accurately tracked observed weight changes. This pragmatic design balances scientific rigour with real-world feasibility. By improving stability, interpretability, and biological plausibility, this method offers a superior EB input for AI-driven coaching. Strengths of Study Design The study design prioritised ecological validity and feasibility without compromising data quality. By avoiding intrusive food tracking and blinding participants to feedback, it ensured naturalistic behaviour and sustained adherence during a disruptive and high-risk period for fat gain(21). Limitations and Considerations Firstly, in the absence of laboratory body composition measurements, weight changes were interpreted as shifts in water or fat only, not muscle—reflecting the low likelihood of meaningful muscle gain over short timescales. Second, the sample size was relatively small; larger cohorts will be needed in future studies to assess generalisability. Implications for Health Research and Personalised Healthcare This method represents a foundational step toward a new paradigm in EB tracking. It demonstrates that personalised health insights can be derived from lightweight, behaviourally passive inputs when enhanced by data processing. Crucially, the approach is both automated and capable of real-time operation, making it well suited for integration within AI coaching platforms. Its ability to track energy dynamics during lifestyle disruptors (e.g. holidays) with high acceptability has important implications for intervention design, public health surveillance, and long-term digital coaching(10). Future research should prioritise larger cohorts and integration of machine learning to enable adaptive models that learn individual patterns and variability, to provide superior EB tracking and feedback. CONCLUSION This study introduces the novel ENHANCE framework and evaluates its application for enhanced EB tracking, implemented by combining smart device data with minimal self-reporting. Evaluated during the demanding festive period, it achieved high adherence and low burden while improving accuracy, interpretability, and biological plausibility. By treating self-report as a signal—not ground truth—this method mirrors the challenges faced by health professionals working with imperfect data(22). Removing the need for burdensome food logs makes this method uniquely suited to free-living use. Automated and scalable through Twists and similar platforms, it can deliver a pioneering capability to support personalised AI-based coaching in real-world settings. Future work should include larger cohorts and explore adaptive models that learn from individual variability—unlocking truly personalised AI coaching. Declarations Acknowledgement The research was supported by the National Institute for Health Research (NIHR) Leicester Biomedical Research Centre. The views expressed are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health and Social Care. Authors Contribution AD and JK developed the initial concept and led data collection. AD led data analysis with support from JK. AD, MP and JK generated the first draft of the manuscript. SW and LJ edited the draft. All authors approved the final version of this manuscript and are accountable for all aspects of the work. Data Availability Statement The datasets generated during the current study are not publicly available due to data privacy restrictions. However, anonymised data are available from the corresponding author on reasonable request. References Bajwa J, Munir U, Nori A, Williams B. Artificial intelligence in healthcare: transforming the practice of medicine. Future Healthc J. 2021;8(2):e188–94. Romieu I, Dossus L, Barquera S, Blottière HM, Franks PW, Gunter M, et al. Energy balance and obesity: what are the main drivers? Cancer Causes and Control. 2017;28(3):247–58. Hall KD, Sacks G, Chandramohan D, Chow CC, Wang YC, Gortmaker SL, et al. Quantification of the effect of energy imbalance on bodyweight. Vol. 378, The Lancet. Elsevier B.V.; 2011. p. 826–37. Rutjes H, Willemsen MC, Feijt MA, IJsselsteijn WA. The Influence of Personal Health Data on the Health Coaching Process. Front Big Data. 2022;5. Hall KD. Estimating human energy intake using mathematical models. Vol. 100, American Journal of Clinical Nutrition. American Society for Nutrition; 2014. p. 744–5. Bailey RL. Overview of dietary assessment methods for measuring intakes of foods, beverages, and dietary supplements in research studies. Vol. 70, Current Opinion in Biotechnology. Elsevier Ltd; 2021. p. 91–6. Butryn ML, Godfrey KM, Martinelli MK, Roberts SR, Forman EM, Zhang F. Digital self-monitoring: Does adherence or association with outcomes differ by self-monitoring target? Obes Sci Pract. 2020;6(2):126–33. Díaz-Zavala RG, Castro-Cantú MF, Valencia ME, Álvarez-Hernández G, Haby MM, Esparza-Romero J. Effect of the Holiday Season on Weight Gain: A Narrative Review. Vol. 2017, Journal of Obesity. Hindawi Limited; 2017. Daw AB, Hinchcliffe CN, James LJ, King JA. Lifestyle instability: an overlooked cause of population obesity?: Epidemiology and Population Health. Int J Obes [Internet]. 2025 Apr 9 [cited 2025 Jun 4];1–2. Available from: https://www.nature.com/articles/s41366-025-01787-5 Nahum-Shani I, Smith SN, Spring BJ, Collins LM, Witkiewitz K, Tewari A, et al. Just-in-time adaptive interventions (JITAIs) in mobile health: Key components and design principles for ongoing health behavior support. Annals of Behavioral Medicine. 2018;52(6):446–62. Twists – Every story has twists start sharing yours today [Internet]. [cited 2025 May 26]. Available from: https://www.twists.com/ WMA Declaration of Helsinki – Ethical Principles for Medical Research Involving Human Participants – WMA – The World Medical Association [Internet]. [cited 2025 May 25]. Available from: https://www.wma.net/policies-post/wma-declaration-of-helsinki/ Julious SA. Sample size of 12 per group rule of thumb for a pilot study. Pharm Stat. 2005;4(4):287–91. ANSI/ASQ Z1.4 & Z1.9 Sampling Plan Standards for Quality Control | ASQ [Internet]. [cited 2025 May 13]. Available from: https://asq.org/quality-resources/z14-z19 SGS Internal Report. 2024. Adamakis M. Validity of Wearable Monitors and Smartphone Applications for Measuring Steps in Semi-Structured and Free-Living Settings. Technologies (Basel). 2023;11(1). Ainsworth BE, Haskell WL, Herrmann SD, Meckes N, Bassett DR, Tudor-Locke C, et al. 2011 compendium of physical activities: A second update of codes and MET values. Vol. 43, Medicine and Science in Sports and Exercise. 2011. p. 1575–81. Mifflin MD, St Jeor ST, Hill LA, Scott BJ, Daugherty SA, Koh YO. A new predictive equation for resting energy expenditure in healthy individuals. Am J Clin Nutr [Internet]. 1990 Feb 1 [cited 2025 May 13];51(2):241–7. Available from: https://www.sciencedirect.com/science/article/abs/pii/S0002916523166986 WISHNOFSKY M. Caloric equivalents of gained or lost weight. Am J Clin Nutr [Internet]. 1958 Sep 1 [cited 2025 May 13];6(5):542–6. Available from: https://pubmed.ncbi.nlm.nih.gov/13594881/ Bhutani S, Kahn E, Tasali E, Schoeller DA. Composition of two-week change in body weight under unrestricted free-living conditions. Physiol Rep. 2017;15(13). Madigan CD, Daley AJ, Lewis AL, Aveyard P, Jolly K. Is self-weighing an effective tool for weight loss: A systematic literature review and meta-analysis. Vol. 12, International Journal of Behavioral Nutrition and Physical Activity. BioMed Central Ltd.; 2015. Miotto R, Wang F, Wang S, Jiang X, Dudley JT. Deep learning for healthcare: Review, opportunities and challenges. Brief Bioinform. 2017;19(6):1236–46. Additional Declarations There is NO conflict of interest to disclose Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: revise 26 Nov, 2025 Review # 2 received at journal 12 Nov, 2025 Reviewer # 2 agreed at journal 18 Sep, 2025 Review # 1 received at journal 25 Aug, 2025 Reviewer # 1 agreed at journal 05 Aug, 2025 Reviewers invited by journal 15 Jul, 2025 Submission checks completed at journal 12 Jun, 2025 First submitted to journal 12 Jun, 2025 Editor assigned by journal 12 Jun, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6878171","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Technical Report","associatedPublications":[],"authors":[{"id":485908107,"identity":"7bb94c83-d8cb-4a66-b46b-7ca6bb660331","order_by":0,"name":"Arthur Daw","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAvklEQVRIiWNgGAWjYHACAwaGCosEOFeCOC1nJEjVwthGihb+BuaNn3nnSeTxSx8+wPCjhiFxZgMBLRIH2IqlebdJFEv2pSUw9hxjSJxN0FkHeAykc7dJJG44w2PAwNvAkDiPkA75AzzGv3PnQLQw/iVGi8EBHjPp3AaIFmaQLQQdZniYrcz6zzGgX3rYEg7LHJMwJuh9uePNm2/OqLHJ4+dhPvjwTY2N7IwDhKxhRmIfIC4iR8EoGAWjYBQQBAB4qzjUZwk1dwAAAABJRU5ErkJggg==","orcid":"","institution":"Loughborough University","correspondingAuthor":true,"prefix":"","firstName":"Arthur","middleName":"","lastName":"Daw","suffix":""},{"id":485908108,"identity":"364996cf-d497-4936-951b-6b5835c9f199","order_by":1,"name":"Maxime Petit","email":"","orcid":"","institution":"Loughborough University","correspondingAuthor":false,"prefix":"","firstName":"Maxime","middleName":"","lastName":"Petit","suffix":""},{"id":485908109,"identity":"317e35ed-98cd-4be3-b9c4-495f5cee6bda","order_by":2,"name":"Scott Willis","email":"","orcid":"https://orcid.org/0000-0001-8624-1427","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Scott","middleName":"","lastName":"Willis","suffix":""},{"id":485908110,"identity":"29eceffe-6f93-4d7f-994f-0e00ad8519f5","order_by":3,"name":"Lewis James","email":"","orcid":"","institution":"Loughborough University","correspondingAuthor":false,"prefix":"","firstName":"Lewis","middleName":"","lastName":"James","suffix":""},{"id":485908111,"identity":"00b373ab-f124-46b8-a1ac-860af259b051","order_by":4,"name":"James King","email":"","orcid":"https://orcid.org/0000-0002-8174-9173","institution":"[email protected]","correspondingAuthor":false,"prefix":"","firstName":"James","middleName":"","lastName":"King","suffix":""}],"badges":[],"createdAt":"2025-06-12 08:25:07","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6878171/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6878171/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":87360604,"identity":"810d425c-4742-4980-bd60-b35d1060d06d","added_by":"auto","created_at":"2025-07-23 05:48:05","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":742426,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6878171/v1/2c97e5d7-5d3c-43a3-a032-97ab9e3227e6.pdf"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e conflict of interest to disclose","formattedTitle":"Novel Energy Balance Tracking to Support Personalised AI Health Coaching: A Real-World Evaluation of the ENHANCE Framework","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003eArtificial Intelligence (AI) is transforming healthcare(\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e). A promising application is personalised AI coaching apps that monitor health using smart device data, including physical activity, sleep, blood pressure, stress, and weight. Energy balance (EB) is often overlooked but is central to cardiometabolic health(\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e) and foundational to weight control models, embodying principles of energy conservation and regulatory biology(\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e). Accurate free-living EB tracking is therefore an essential requirement of AI-based coaching platforms(\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e) yet remains challenging to measure(\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eEB can be estimated from body weight changes using predictive equations, providing a low-burden solution(\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e), but lacks the resolution needed for daily feedback. Conversely, daily EB tracking using estimates of energy intake (EI) (via food-tracking) and energy expenditure (EE) (via smart device data) improves precision but requires ongoing user engagement(\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e). Although these methods are initially accepted, user engagement drops rapidly, limiting long-term effectiveness(\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e). Sustained AI-based support requires reliable EB tracking, even during low-engagement high-risk periods(\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e), such as holidays when 0.5\u0026ndash;1 kg fat gain is common(\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e). To maximise the potential of AI coaching, valid methods are needed to reliably track EB year-round, including during high-risk periods(\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e)(\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThis report introduces:\u003c/p\u003e\u003cp\u003e\u003col\u003e\u003cspan\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eENHANCE Framework\u003c/b\u003e: ENHANCE (Estimation and Normalisation for Holistic Alignment, Nuance, and Contextual Enhancements) improves tracking accuracy and interpretability by combining holistic temporal data with physiological and behavioural relationships. Its aim is to enhance user decision-making, engagement, and outcomes in free-living settings\u0026mdash;not to generate a ground truth comparable to laboratory measurements.\u003c/p\u003e\u003c/li\u003e\u003c/span\u003e\u003cspan\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eENHANCE EB Tracking\u003c/b\u003e: A low-burden daily EB tracking method combining smart device data with minimal self-report, implemented using the ENHANCE framework.\u003c/p\u003e\u003c/li\u003e\u003c/span\u003e\u003c/ol\u003e\u003c/p\u003e\u003cp\u003eWe present the conceptual framework, implementation, and preliminary evaluation\u0026mdash;delivered through the Twists.com AI coaching platform (\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e)\u0026mdash;and evaluate its feasibility for continuous AI-based coaching in real-world settings.\u003c/p\u003e"},{"header":"METHODS","content":"\u003cp\u003e\u003cstrong\u003eEthical Approval\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eEthical approval was granted by Loughborough University Ethics Advisory Committee (Ref: 21271) and the study complied with the Declaration of Helsinki(12). All participants gave digital consent.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Architecture and Anonymisation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eParticipants received a unique identification number and pseudonym during onboarding.\u0026nbsp;\u0026nbsp;Body weight and app data were transmitted over Wi-Fi and stored on a Microsoft database hosted on Amazon Web Services. Access was restricted to authorised researchers via a secure software-as-a-service interface. Questionnaire responses were collected via Google Forms and linked to participants via date of birth only.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStudy Design and Setting\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis remote observational study was conducted between 08 December 2024 to 15 January 2025. The Christmas to New Year festive period is a high-risk EB disruptor—due to increased EI, reduced EE, and competing personal demands(8)—making it ideal to evaluate method performance.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStudy Phases\u003c/strong\u003e\u003c/p\u003e\n\u003col start=\"1\" type=\"1\"\u003e\n \u003cli\u003e\u003cstrong\u003eOnboarding\u003c/strong\u003e (08 December–10 December): Participants connected their Wi-Fi scales, installed the Twists app, and estimated typical EB under habitual conditions.\u003c/li\u003e\n\u003c/ol\u003e\n\u003col start=\"2\" type=\"1\"\u003e\n \u003cli\u003e\u003cstrong\u003eBaseline\u003c/strong\u003e (08 December–14 December): Participants established a baseline body weight trend while familiarising themselves with the scales and app. Baseline data configured the platform but were excluded from results.\u003c/li\u003e\n\u003c/ol\u003e\n\u003col start=\"3\" type=\"1\"\u003e\n \u003cli\u003e\u003cstrong\u003eMonitoring\u003c/strong\u003e (15 December–10 January): Participants recorded blind daily body weight measurements and answered EB-related questions. Feedback was withheld to preserve naturalistic behaviour.\u003c/li\u003e\n\u003c/ol\u003e\n\u003col start=\"4\" type=\"1\"\u003e\n \u003cli\u003e\u003cstrong\u003eFeedback\u003c/strong\u003e (10 January–15 January): Participant acceptability and behavioural reactivity were assessed using an end-of-study multiple-choice questionnaire.\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003e\u003cstrong\u003eParticipants and Recruitment\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eParticipants were recruited online from Gloucestershire, UK. Inclusion criteria were: men and women, 20–49 years, BMI 18.5–29.9 kg/m², stable body weight, and not travelling overnight \u0026gt;1 night. Twelve is a recommended sample for feasibility (13), but 23 were enrolled to support individual-level prediction error analysis. To preserve naturalistic behaviour, participants were only informed that the study was investigating how the festive period affects population EB. Weight monitoring was not discussed, and readings were blinded. An end-of-study questionnaire assessed behavioural reactivity.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Collection Protocol\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eParticipants received Wi-Fi-enabled scales and proprietary Twists app for daily EB-related submissions—both scales and app were developed by Surpic Limited. Twists provides structured, goal-oriented tracking and AI coaching, but was modified for this study to ensure participants remained blinded to EB and weight feedback. During onboarding, participants watched a demonstration video and received a text-based guide. Each morning, participants recorded blinded weight and self-reported the prior day’s EB-related data. Daily tasks took \u0026lt;2 minutes. Notifications were sent at 6:00 AM (silent)—for measurement reminders, 2:00 PM—if self-reports were missed, and 4:00 PM—if measurements were missed.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eWeight Measurements via Wi-Fi Scales\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eEach morning, participants weighed themselves after toileting, ideally nude or wearing consistent minimal clothing. Scales were placed on hard, flat surfaces. Displays confirmed measurement and Wi-Fi upload but did not show weight. The scales were manufactured by Surpic Limited (Cheltenham, UK) and assembled and UKCA-certified by Zhongshan Frecom Electronic Limited (Guangdong, China). Raw body weight data were recorded to 0.1\u0026nbsp;kg resolution. Scale accuracy was independently validated by Société Générale de Surveillance (Geneva, Switzerland), confirming ANSI/ASQ Z1.4 (2003, R2018) standards at AQL Level II(14), and precision ±0.2–0.3\u0026nbsp;kg(15).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSelf-Reporting via Mobile App\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThree daily EB-related metrics were collected via the Twists app: eating (%), steps (count), and exercise (kcal). To minimise time burden and enhance resilience to device issues, participants could choose from multiple simple input methods per metric.\u003c/p\u003e\n\u003cp\u003eEating was estimated as a percentage deviation from ‘typical’ intake. Participants were weight stable at onboarding, therefore typical EE was used to estimate typical EI. Participants either selected a deviation percentage from five preset options or entered a custom value from −100% to +100%. This method was intentionally approximate and not intended to provide precise estimates of EI, but to generate a plausible trend line\u0026nbsp;with minimal input for further analysis.\u003c/p\u003e\n\u003cp\u003eThe five preset options were:\u003c/p\u003e\n\u003col\u003e\n \u003cli\u003e\u003cstrong\u003eWell Above:\u003c/strong\u003e ~66% more than typical\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eAbove:\u003c/strong\u003e ~33% more than typical\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eTypical:\u0026nbsp;\u003c/strong\u003e~0% from typical\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eBelow:\u003c/strong\u003e ~33% less than typical\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eWell Below:\u003c/strong\u003e ~66% less than typical\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eSteps were estimated using mobile data, wearable devices, four preset levels (2,500–10,000 steps), or manual entry—data were used to calculate non-exercise activity thermogenesis (NEAT)(16). \u0026nbsp;Exercise estimates could be imported from wearable devices, entered manually or calculated using a guided wizard based on Metabolic Equivalent of Task (MET) codes(17)—data were used to calculate exercise activity thermogenesis (EAT). For steps and exercise, participants were encouraged to use device-based inputs to minimise bias. But to improve resilience to device unreliability, they could choose an alternative method if they considered it more accurate on that day.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDaily Energy Balance Calculation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDaily EB was calculated as follows:\u003c/p\u003e\n\u003cul type=\"disc\"\u003e\n \u003cli\u003e\u003cstrong\u003eEB = EI\u003c/strong\u003e \u003cstrong\u003e− EE\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003cul type=\"disc\"\u003e\n \u003cli\u003e\u003cstrong\u003eEI = Typical EI × Self-reported day modifier\u003c/strong\u003e. Typical intake was estimated during onboarding from typical EE. This approach is novel, and balances analytical accuracy with real-world practicality.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cul type=\"disc\"\u003e\n \u003cli\u003e\u003cstrong\u003eEE = REE + TEF + NEAT + EAT\u003c/strong\u003e. Resting energy expenditure (REE) was calculated using the Mifflin–St Jeor equation(18). Thermic effect of feeding (TEF) was estimated at 10% of EI(3). NEAT was derived from mobile, wearable, or self-reported step counts using MET codes(17).\u0026nbsp;EAT was imported from manufacturer wearable data or estimated using MET codes.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cstrong\u003eWeight Trend Generation and\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003eAnalysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFive weight change trends were generated to support the analysis, each offering a different perspective on EB:\u003c/p\u003e\n\u003col\u003e\n \u003cli\u003e\u003cstrong\u003eRaw Trend:\u003c/strong\u003e Unprocessed body weight data collected each morning from the Wi-Fi scales. Although objective, these measurements vary with natural hydration shifts and data anomalies, making them misleading if interpreted directly as fat weight change.\u003c/li\u003e\n\u003c/ol\u003e\n\u003col start=\"2\"\u003e\n \u003cli\u003e\u003cstrong\u003eSmoothed Trend:\u003c/strong\u003e Generated by removing outliers from raw data using an Interquartile Range (IQR) method (IQR×1.5) and applying a simple moving average (±3 days). This reduces noise while retaining directional trajectories, providing a cleaner reference for further processing.\u003c/li\u003e\n\u003c/ol\u003e\n\u003col start=\"3\"\u003e\n \u003cli\u003e\u003cstrong\u003ePiecewise Trend:\u003c/strong\u003e Segmented the smoothed trend into three linear phases, each representing a consistent rate of change: pre-Christmas (15-23 December), festive peak (24 December–1 January), and post-New Year (2–10 January). These segments provided a benchmark to evaluate alternative trends.\u003c/li\u003e\n\u003c/ol\u003e\n\u003col start=\"4\"\u003e\n \u003cli\u003e\u003cstrong\u003ePredicted Trend:\u003c/strong\u003e Calculated from self-reported EB values (EI − EE) to estimate daily fat weight change—estimated at 1 kg per 7,700 kcal EB(19). While unaffected by hydration shifts, it is influenced by reporting error and bias. The prediction serves as an input for further processing, rather than an exact estimate of fat change.\u003c/li\u003e\n\u003c/ol\u003e\n\u003col start=\"5\"\u003e\n \u003cli\u003e\u003cstrong\u003eCorrected Trend:\u0026nbsp;\u003c/strong\u003eGenerated by aligning predicted and smoothed trends to enhance results. A ±5-day weighted average from the smoothed trend prioritised nearby low-noise values. The result was used to refine the predicted data to produce a corrected trend grounded in user behaviour and biological plausibility, while aligning with observed weight trends.\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003e\u003cstrong\u003eMethod Evaluation\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAs a novel method, performance was evaluated through trend comparisons to assess improvements in stability, interpretability, and biological plausibility.\u003c/p\u003e\n\u003col start=\"1\" type=\"1\"\u003e\n \u003cli\u003e\u003cstrong\u003eRaw vs Piecewise Trend:\u003c/strong\u003e Evaluates how well raw data aligns with the piecewise trend. Large deviations of \u0026gt;±0.35 kg imply noise from hydration shifts or data anomalies(19)(20).\u0026nbsp;\u003c/li\u003e\n\u003c/ol\u003e\n\u003col start=\"2\" type=\"1\"\u003e\n \u003cli\u003e\u003cstrong\u003eCorrected vs Piecewise Trend:\u003c/strong\u003e Assesses how closely the corrected trend tracks the piecewise, confirming reduced noise without introducing artefacts.\u003c/li\u003e\n\u003c/ol\u003e\n\u003col start=\"3\" type=\"1\"\u003e\n \u003cli\u003e\u003cstrong\u003eCorrected vs Raw Trend:\u003c/strong\u003e Confirms reduced noise and recovery of coherent weight signals. Suppressing daily fluctuations \u0026gt;±0.35\u0026nbsp;kg enhances interpretability while preserving longer-term fat weight trends(19)(20).\u0026nbsp;\u003c/li\u003e\n\u003c/ol\u003e\n\u003col start=\"4\" type=\"1\"\u003e\n \u003cli\u003e\u003cstrong\u003eSignal-to-Noise Ratio (SNR)\u003c/strong\u003e\u003cstrong\u003e:\u003c/strong\u003e Calculated as the ratio of overall variability (SD of weight) to short-term fluctuation (SD of daily changes). Higher post-correction SNR indicates a clearer reflection of fat weight change.\u003c/li\u003e\n\u003c/ol\u003e\n\u003col start=\"5\" type=\"1\"\u003e\n \u003cli\u003e\u003cstrong\u003eWeight Trend Autocorrelation:\u003c/strong\u003e\u0026nbsp; Calculated as the lag-1 Pearson correlation of daily changes, this reflects directional consistency. Higher autocorrelation post-correction suggests more physiologically coherent progression.\u003c/li\u003e\n\u003c/ol\u003e\n\u003col start=\"6\" type=\"1\"\u003e\n \u003cli\u003e\u003cstrong\u003eWeight Trend Plausibility:\u003c/strong\u003e Biologicalplausibility was assessed by comparing implied EB to realistic bounds (−100% to +250% of typical EE). The same thresholds were applied post-correction to ensure results remained within plausible limits.\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"RESULTS","content":"\u003cp\u003e\u003cstrong\u003eParticipants and Data Completeness\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eOf the 23 participants who enrolled and completed the protocol, 18 were included in the final analysis (Table 1). Five were excluded due to atypical behaviour from illness (n = 4) or bereavement (n = 1). At the individual level, adherence to daily body weight measurements was high, with a mean of 94.0%\u0026nbsp;(5.1%). A small proportion\u0026nbsp;of measurements were excluded due to lateness (3.2% after midday) or outliers (3.9% outside IQR threshold), retaining 87.4% of possible scans for analysis. Compliance with EB-related self-reporting of eating, steps and exercise estimates was 100%. Calculated individual-level mean net fat weight change during the monitoring phase was +0.8 kg (0.4 kg); mean net EB was +223 kcal/day (130 kcal/day).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTrend Performance\u003c/strong\u003e\u003c/p\u003e\n\u003col\u003e\n \u003cli\u003e\u003cstrong\u003eRaw Trend:\u0026nbsp;\u003c/strong\u003eGroup-level signal quality was low (SNR: 2.69), and autocorrelation was negative (−0.25), indicating limited coherence and poor behavioural interpretability. Daily changes in raw weight were highly volatile, with an individual-level mean of 0.62 kg (0.22 kg), often masking meaningful physiological trends (Figure 1). Individual-level false direction reversals—where raw weight changes contradicted reported EB, reducing interpretability—occurred on 38.9% (6.7%) of days. Biological plausibility was low, with only 57% of days within realistic bounds (−100% to +250% of typical EE).\u003c/li\u003e\n\u003c/ol\u003e\n\u003col start=\"2\"\u003e\n \u003cli\u003e\u003cstrong\u003ePiecewise Trend:\u0026nbsp;\u003c/strong\u003eGroup-level piecewise analysis identified three segments, each with distinct fat weight trends: a modest rise pre-Christmas (+4.3\u0026nbsp;g/day), a sharp increase during the festive peak (+58.2\u0026nbsp;g/day), and modest gain post-New Year (+14.9\u0026nbsp;g/day). The corrected trend aligned more closely with these segments than either raw or predicted.\u003c/li\u003e\n\u003c/ol\u003e\n\u003col start=\"3\"\u003e\n \u003cli\u003e\u003cstrong\u003ePredicted\u003c/strong\u003e \u003cstrong\u003eTrend:\u0026nbsp;\u003c/strong\u003eGroup-level predicted trajectory explained 90.4% of the variance in smoothed trends (R²: 0.904), showing high internal consistency. Errors were low: mean absolute error (MAE) 93\u0026nbsp;g (63\u0026nbsp;g), root mean square error (RMSE) 112\u0026nbsp;g, and mean bias error (MBE) −21\u0026nbsp;g. This accuracy and stability justified its use by the correction method.\u003c/li\u003e\n\u003c/ol\u003e\n\u003col start=\"4\"\u003e\n \u003cli\u003e\u003cstrong\u003eCorrected Trend:\u0026nbsp;\u003c/strong\u003eThe correction method refined daily EB estimates by aligning predicted and smoothed trends using a proximity- and noise-weighted smoothing window. This improved stability, interpretability and biological plausibility. Corrected values showed reduced individual-level variability in daily changes: 98 g (48 g); and improved group-level signal quality – SNR: 9.52; and stronger autocorrelation: 0.87—all indicative of a more coherent behavioural signal. Compared to predicted and raw, group-level corrected trends showed the lowest deviation from piecewise (MAE: 102\u0026nbsp;g, 77\u0026nbsp;g, and 46\u0026nbsp;g, respectively), and the highest alignment (r = 0.88)—see Figure 1 for visual comparisons. At the individual level, false direction reversals were eliminated. The final EB corrections were modest: the mean adjustment was +41\u0026nbsp;kcal/day (189 kcal/day)—just 2% (6.3%) of reported intake. Daily corrections exceeding 30% were rare, occurring on just 7.8 % (8.4 %) of days—supporting the method’s restraint and interpretability. Additional comparisons are shown in Table 2.\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003e\u003cstrong\u003eQuestionnaire Findings\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePost-study questionnaires (n = 18) indicated high user acceptability and low burden (Table 3). Confidence in self-estimates was high across eating, steps, and exercise. Participants reported negligible behavioural reactivity or impact on Christmas enjoyment.\u003c/p\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003e\u003cstrong\u003ePrincipal Findings\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study evaluates the ENHANCE framework and a novel low-burden method for estimating daily EB using smart devices and minimal self-report. Tested during the demanding festive period, it achieved high adherence and low participant burden. Raw weight data were too volatile and behaviourally implausible for effective use in AI-based coaching. Despite intentionally approximate EB inputs, predicted trends aligned closely with piecewise segments, providing a credible basis for correction. Aligning predicted trends with device measurements produced corrected values that more accurately tracked observed weight changes. This pragmatic design balances scientific rigour with real-world feasibility. By improving stability, interpretability, and biological plausibility, this method offers a superior EB input for AI-driven coaching.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStrengths of Study Design\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study design prioritised ecological validity and feasibility without compromising data quality. By avoiding intrusive food tracking and blinding participants to feedback, it ensured naturalistic behaviour and sustained adherence during a disruptive and high-risk period for fat gain(21).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLimitations and Considerations\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFirstly, in the absence of laboratory body composition measurements, weight changes were interpreted as shifts in water or fat only, not muscle—reflecting the low likelihood of meaningful muscle gain over short timescales. Second, the sample size was relatively small; larger cohorts will be needed in future studies to assess generalisability.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eImplications for Health Research and Personalised Healthcare\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis method represents a foundational step toward a new paradigm in EB tracking. It demonstrates that personalised health insights can be derived from lightweight, behaviourally passive inputs when enhanced by data processing. Crucially, the approach is both automated and capable of real-time operation, making it well suited for integration within AI coaching platforms. Its ability to track energy dynamics during lifestyle disruptors (e.g. holidays) with high acceptability has important implications for intervention design, public health surveillance, and long-term digital coaching(10). Future research should prioritise larger cohorts and integration of machine learning to enable adaptive models that learn individual patterns and variability, to provide superior EB tracking and feedback.\u003c/p\u003e"},{"header":"CONCLUSION","content":"\u003cp\u003eThis study introduces the novel ENHANCE framework and evaluates its application for enhanced EB tracking, implemented by combining smart device data with minimal self-reporting. Evaluated during the demanding festive period, it achieved high adherence and low burden while improving accuracy, interpretability, and biological plausibility. By treating self-report as a signal—not ground truth—this method mirrors the challenges faced by health professionals working with imperfect data(22). Removing the need for burdensome food logs makes this method uniquely suited to free-living use. Automated and scalable through Twists and similar platforms, it can deliver a pioneering capability to support personalised AI-based coaching in real-world settings. Future work should include larger cohorts and explore adaptive models that learn from individual variability—unlocking truly personalised AI coaching.\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe research was supported by the National Institute for Health Research (NIHR) Leicester Biomedical Research Centre. The views expressed are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health and Social Care.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors Contribution\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAD and JK developed the initial concept and led data collection. AD led data analysis with support from JK. AD, MP and JK generated the first draft of the manuscript. SW and LJ edited the draft. All authors approved the final version of this manuscript and are accountable for all aspects of the work.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets generated during the current study are not publicly available due to data privacy restrictions. However, anonymised data are available from the corresponding author on reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eBajwa J, Munir U, Nori A, Williams B. Artificial intelligence in healthcare: transforming the practice of medicine. Future Healthc J. 2021;8(2):e188\u0026ndash;94.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRomieu I, Dossus L, Barquera S, Blotti\u0026egrave;re HM, Franks PW, Gunter M, et al. Energy balance and obesity: what are the main drivers? Cancer Causes and Control. 2017;28(3):247\u0026ndash;58.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHall KD, Sacks G, Chandramohan D, Chow CC, Wang YC, Gortmaker SL, et al. Quantification of the effect of energy imbalance on bodyweight. Vol. 378, The Lancet. Elsevier B.V.; 2011. p. 826\u0026ndash;37.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRutjes H, Willemsen MC, Feijt MA, IJsselsteijn WA. The Influence of Personal Health Data on the Health Coaching Process. Front Big Data. 2022;5.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHall KD. Estimating human energy intake using mathematical models. Vol. 100, American Journal of Clinical Nutrition. American Society for Nutrition; 2014. p. 744\u0026ndash;5.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBailey RL. Overview of dietary assessment methods for measuring intakes of foods, beverages, and dietary supplements in research studies. Vol. 70, Current Opinion in Biotechnology. Elsevier Ltd; 2021. p. 91\u0026ndash;6.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eButryn ML, Godfrey KM, Martinelli MK, Roberts SR, Forman EM, Zhang F. Digital self-monitoring: Does adherence or association with outcomes differ by self-monitoring target? Obes Sci Pract. 2020;6(2):126\u0026ndash;33.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eD\u0026iacute;az-Zavala RG, Castro-Cant\u0026uacute; MF, Valencia ME, \u0026Aacute;lvarez-Hern\u0026aacute;ndez G, Haby MM, Esparza-Romero J. Effect of the Holiday Season on Weight Gain: A Narrative Review. Vol. 2017, Journal of Obesity. Hindawi Limited; 2017.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDaw AB, Hinchcliffe CN, James LJ, King JA. Lifestyle instability: an overlooked cause of population obesity?: Epidemiology and Population Health. Int J Obes [Internet]. 2025 Apr 9 [cited 2025 Jun 4];1\u0026ndash;2. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.nature.com/articles/s41366-025-01787-5\u003c/span\u003e\u003cspan address=\"https://www.nature.com/articles/s41366-025-01787-5\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNahum-Shani I, Smith SN, Spring BJ, Collins LM, Witkiewitz K, Tewari A, et al. Just-in-time adaptive interventions (JITAIs) in mobile health: Key components and design principles for ongoing health behavior support. Annals of Behavioral Medicine. 2018;52(6):446\u0026ndash;62.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTwists \u0026ndash; Every story has twists start sharing yours today [Internet]. [cited 2025 May 26]. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.twists.com/\u003c/span\u003e\u003cspan address=\"https://www.twists.com/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWMA Declaration of Helsinki \u0026ndash; Ethical Principles for Medical Research Involving Human Participants \u0026ndash; WMA \u0026ndash; The World Medical Association [Internet]. [cited 2025 May 25]. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.wma.net/policies-post/wma-declaration-of-helsinki/\u003c/span\u003e\u003cspan address=\"https://www.wma.net/policies-post/wma-declaration-of-helsinki/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJulious SA. Sample size of 12 per group rule of thumb for a pilot study. Pharm Stat. 2005;4(4):287\u0026ndash;91.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eANSI/ASQ Z1.4 \u0026amp; Z1.9 Sampling Plan Standards for Quality Control | ASQ [Internet]. [cited 2025 May 13]. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://asq.org/quality-resources/z14-z19\u003c/span\u003e\u003cspan address=\"https://asq.org/quality-resources/z14-z19\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSGS Internal Report. 2024.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAdamakis M. Validity of Wearable Monitors and Smartphone Applications for Measuring Steps in Semi-Structured and Free-Living Settings. Technologies (Basel). 2023;11(1).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAinsworth BE, Haskell WL, Herrmann SD, Meckes N, Bassett DR, Tudor-Locke C, et al. 2011 compendium of physical activities: A second update of codes and MET values. Vol. 43, Medicine and Science in Sports and Exercise. 2011. p. 1575\u0026ndash;81.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMifflin MD, St Jeor ST, Hill LA, Scott BJ, Daugherty SA, Koh YO. A new predictive equation for resting energy expenditure in healthy individuals. Am J Clin Nutr [Internet]. 1990 Feb 1 [cited 2025 May 13];51(2):241\u0026ndash;7. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.sciencedirect.com/science/article/abs/pii/S0002916523166986\u003c/span\u003e\u003cspan address=\"https://www.sciencedirect.com/science/article/abs/pii/S0002916523166986\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWISHNOFSKY M. Caloric equivalents of gained or lost weight. Am J Clin Nutr [Internet]. 1958 Sep 1 [cited 2025 May 13];6(5):542\u0026ndash;6. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://pubmed.ncbi.nlm.nih.gov/13594881/\u003c/span\u003e\u003cspan address=\"https://pubmed.ncbi.nlm.nih.gov/13594881/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBhutani S, Kahn E, Tasali E, Schoeller DA. Composition of two-week change in body weight under unrestricted free-living conditions. Physiol Rep. 2017;15(13).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMadigan CD, Daley AJ, Lewis AL, Aveyard P, Jolly K. Is self-weighing an effective tool for weight loss: A systematic literature review and meta-analysis. Vol. 12, International Journal of Behavioral Nutrition and Physical Activity. BioMed Central Ltd.; 2015.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMiotto R, Wang F, Wang S, Jiang X, Dudley JT. Deep learning for healthcare: Review, opportunities and challenges. Brief Bioinform. 2017;19(6):1236\u0026ndash;46.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"international-journal-of-obesity","isNatureJournal":false,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"ijo","sideBox":"Learn more about [International Journal of Obesity](http://www.nature.com/ijo/)","snPcode":"41366","submissionUrl":"https://mts-ijo.nature.com/cgi-bin/main.plex","title":"International Journal of Obesity","twitterHandle":"@intjobesity","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Artificial Intelligence, Digital health, Mobile health, Energy Balance, Energy intake, Obesity, Weight management, Behaviour change.","lastPublishedDoi":"10.21203/rs.3.rs-6878171/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6878171/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground:\u003c/strong\u003e Energy balance (EB) is the key determinant of fat gain, yet accurate EB tracking is difficult outside laboratory settings. Traditional methods are burdensome (e.g. food-logs) or lack daily resolution (e.g. body weight monitoring), limiting suitability for integration with free-living AI-powered health-coaching.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eObjective:\u003c/strong\u003e To introduce ENHANCE—a novel framework prioritising interpretability and temporal accuracy—and demonstrate its use as a low-burden, accurate method for tracking EB using smart devices and minimal self-report, suitable for AI coaching.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods:\u003c/strong\u003e This 4-week observational study spanned the Christmas to New Year2024/25 festive period. Participants submitted daily blinded body weight measurements via Wi-Fi scales and EB-related questions via a mobile app, taking \u0026lt;2 minutes. Data were used to generate five weight trends: raw (from scales), smoothed (±3-day average), piecewise (3-segments), predicted (from EB), and corrected. The correction aligned predicted and smoothed trends, using proximity and noise-weighted adjustments, producing enhanced data for AI coaching. An end-of-study questionnaire assessed acceptability and behavioural reactivity.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults:\u003c/strong\u003e Of 23 participants, 18 were analysed. Five were excluded due to illness (n = 4) or bereavement (n = 1). Participants completed 94% (5.1%) of body weight measurements and 100% of EB-related submissions. Questionnaire results showed low burden (1.8/5) and behavioural reactivity (1.5/5). Group-level predicted trends explained 90.4% of smoothed trend variance (R² = 0.904; mean absolute error [MAE]: 93 g). Corrected trends aligned more closely with piecewise segments than raw trends (MAE: 46 g vs 77 g). Individual-level mean EB corrections were +41 kcal/day—just 2% of reported intake. The corrected trend enhanced interpretability and plausibility while preserving real-world validity. Calculated mean net fat weight change during the monitoring phase was +0.8 kg (0.4 kg); mean net EB was +223 kcal/day (130 kcal/day).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusions:\u003c/strong\u003e \u0026nbsp;This scalable method delivers the accuracy and practicality needed for real-world EB tracking—laying the foundation for continuous personalised AI coaching.\u003c/p\u003e","manuscriptTitle":"Novel Energy Balance Tracking to Support Personalised AI Health Coaching: A Real-World Evaluation of the ENHANCE Framework","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-23 05:39:55","doi":"10.21203/rs.3.rs-6878171/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"revise","date":"2025-11-26T11:47:07+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"This content is not available.","date":"2025-11-12T07:05:04+00:00","index":2,"fulltext":"This content is not available."},{"type":"reviewerAgreed","content":"This content is not available.","date":"2025-09-19T00:04:45+00:00","index":2,"fulltext":"This content is not available."},{"type":"editorInvitedReview","content":"This content is not available.","date":"2025-08-25T21:15:24+00:00","index":1,"fulltext":"This content is not available."},{"type":"reviewerAgreed","content":"This content is not available.","date":"2025-08-05T21:57:31+00:00","index":1,"fulltext":"This content is not available."},{"type":"reviewersInvited","content":"","date":"2025-07-15T17:10:33+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-06-12T11:24:50+00:00","index":"","fulltext":""},{"type":"submitted","content":"International Journal of Obesity","date":"2025-06-12T08:20:05+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-06-12T08:20:05+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"international-journal-of-obesity","isNatureJournal":false,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"ijo","sideBox":"Learn more about [International Journal of Obesity](http://www.nature.com/ijo/)","snPcode":"41366","submissionUrl":"https://mts-ijo.nature.com/cgi-bin/main.plex","title":"International Journal of Obesity","twitterHandle":"@intjobesity","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"2c5012f1-f5ef-4bd6-98b6-bf0315dc51e1","owner":[],"postedDate":"July 23rd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":51584827,"name":"Health sciences/Health care/Weight management"},{"id":51584828,"name":"Health sciences/Health care/Disease prevention/Lifestyle modification"},{"id":51584829,"name":"Health sciences/Health care/Disease prevention/Preventive medicine"}],"tags":[],"updatedAt":"2026-04-22T07:22:23+00:00","versionOfRecord":[],"versionCreatedAt":"2025-07-23 05:39:55","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6878171","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6878171","identity":"rs-6878171","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}