Brno University of Technology Continuous Glycaemia, ECG, Activity and Food Intake Database (BUT CGM)

@article{PhysioNet-but-cgm-1.0.0,
  author = {Šaclová, Lucie and Nemcova, Andrea and Goldschmidt, Tobiáš and Kralik, Martin and Chlíbková, Daniela and Smital, Lukas and Smisek, Radovan and Hubálek, Jaromír},
  title = {{Brno University of Technology Continuous Glycaemia, ECG, Activity and Food Intake Database (BUT CGM)}},
  journal = {{PhysioNet}},
  year = {2026},
  month = nov,
  note = {Version 1.0.0},
  doi = {10.13026/hv8f-3w60},
  url = {https://doi.org/10.13026/hv8f-3w60}
}

@article{pollard_physionet_2026, title = {{PhysioNet} as a global platform for biomedical research}, volume = {1}, issn = {3005-0693}, url = {https://doi.org/10.1038/s44360-026-00096-z}, doi = {10.1038/s44360-026-00096-z}, abstract = {Twenty-five years after its first public repository was established, the PhysioNet resource for biomedical and clinical data has enabled tens of thousands of research papers. Further efforts are ongoing to advance secure data sharing, community engagement and scalable research infrastructure.}, number = {8}, journal = {Nature Health}, author = {Pollard, Tom and Moody, Benjamin E. and Lehman, Li-wei H. and Gow, Brian J. and Fernandes, Chrystinne and Xie, Chen and Johnson, Alistair and Mark, Roger G. and Heldt, Thomas}, month = aug, year = {2026}, pages = {792--795},}

Abstract

The Brno University of Technology Continuous Glycaemia, ECG, Activity and Food Intake Database (BUT CGM) is a multimodal dataset collected in free-living conditions from a healthy predominantly physically active population. The resource was created to address the lack of publicly available datasets that jointly capture interstitial glucose, high-resolution cardiac electrophysiology, physical activity (sports activity, daily steps), sleep, and detailed dietary intake in everyday life. Data from 60 participants were acquired approximately ten consecutive days using a continuous glucose monitor (CGM), single-lead electrocardiogram (ECG), accelerometry (ACC), entry and daily questionnaires, and detailed food logging, and are complemented by laboratory-based cardiopulmonary exercise testing with ventilatory threshold annotation. The dataset enables research on cardio-metabolic interactions, glycaemic dynamics, and lifestyle–performance relationships under real-world conditions.

Background

Continuous Glucose Monitoring (CGM) sensors are now used not only for diabetes management but also among healthy adults, in sports contexts, and for wellness tracking, where rapid glycaemic excursions can be observed in response to meals, physical activity, stress, and sleep in everyday life [1]. In professional and hobby sports, there is growing demand for personalized, real-time monitoring systems that support performance optimization, early detection of physiological risks, and long-term health maintenance. A recent scoping review of CGM-based automatic ingestive-activity detection similarly concludes that most published approaches operate on CGM alone or with minimal contextual information, and that time-aligned diet, activity, and physiological signals are critical for robust, real-time detection suitable for just-in-time adaptive interventions [2]. Moreover, publicly available datasets that support secondary analyses rarely combine high-fidelity cardiac electrophysiology, free-living behavior, and interstitial glucose at research-grade resolution.

The BUT CGM dataset was created within the DistriMuSe (Distributed Multi-Sensor System for Human Safety and Health) project to address this gap by providing a time-synchronized, free-living physiological resource in healthy adults that links electrocardiogram (ECG), accelerometry (ACC), CGM, physical activity (steps, sports activity), and rich behavioral context. The dataset is used to provide detailed insights into the physiological state of the measured subjects. The aim is to provide user-friendly feedback to the sporting population and their coaches, with the ultimate goal of supporting performance enhancement and reducing health risks associated with overtraining, inadequate nutrition, or improper recovery.

Methods

Altogether, 60 participants were recruited for this study from a healthy sporting population with varying levels of physical fitness and performance. 18 (30%) were women, and 42 (70%) were men. Participants' ages ranged from 20 to 46 years, with a mean of 25.7±5.9 years. We conducted 10 days of continuous measurements in free-living conditions.

During the initial laboratory visit, all the sensors (described below) were properly attached to participants. Participants were given instructions on how to care for the sensors, how to change electrodes daily, and when to recharge the sensors to ensure maximum user comfort and data quality. The entry questionnaire was completed by the participants.

Two raw biosignals were measured — single-lead ECG and three-axis ACC — using a Bittium Faros 180 recorder (Bittium Corporation, Finland). The recorder was placed on the chest using Ag/AgCl microporous electrodes (Diagramm Halbach GmbH & Co. KG, Germany). The ECG sampling frequency was set to 1,000 Hz. The ACC was measured at a sampling frequency of 100 Hz with a dynamic range of ±16g.

In ECG signals, the QRS complexes were automatically detected using a previously published robust QRS detection algorithm [7]. It combines three independent QRS detectors based on wavelet, phasor and Stockwell transforms followed by cluster analysis.

Every 5th minute, the interstitial glucose value was recorded using a Dexcom G7 CGM sensor (Dexcom, Inc., USA) placed on the non-dominant upper arm (triceps area), following the manufacturer's guidelines and secured with the provided adhesive patch.

Participants performed Cardiopulmonary Exercise Testing (CPET) using the Graded Exercise Test (GET) in a laboratory during the 10-day measurement. It was done on a bicycle ergometer or treadmill. The GET protocol began with a 5-minute warm-up of low-intensity followed by an increase in load every 2 minutes until exhaustion, and concluded with a 5-minute cool-down. The first and second ventilatory thresholds (VT1, VT2) were identified using standard gas-exchange procedures, including the V-slope method, VE/VO2, VE/VCO2, PetO2, and PetCO2 [3,4]. The determination of VTs was performed independently by two trained annotators. In case of disagreement, a joint reconciliation session was held to resolve discrepancies via consensus. Five participants did not take the test during the 10-day measurement period (105, 152, 153, 154, 155), and for 6 participants (135001, 156001, 157001, 158001, 159001, 160001) the test is missing.

In addition, volunteers recorded their physical activities, steps, and sleep using various sports watches. Based on the CPET results, physical activities were annotated according to intensity zones. For each aerobic activity, the time spent in individual intensity zones is reported.

Step counts were aggregated to produce daily totals. For sources like Samsung Health that provide multiple entries per day (e.g., from different devices or time segments), the data were grouped by date and summed up to ensure a single, accurate step count for each day.

Sleep data were processed to generate one summary record per night. For fragmented sleep records (common in Samsung and Apple Health data), the start and end times were determined by the earliest sleep onset and latest wake time for that date, and total sleep duration was calculated or summed to represent the full night's rest.

Food intake was recorded using the smartphone application “Kalorické tabulky” (calorie tables, Dine4Fit, a.s., Czech Republic) with an emphasis on recording meals immediately after consumption to ensure accurate temporal alignment with other data. Part of the meal data were recorded as individual food items (e.g., bread, cheese, salami). Items consumed within a 10-minute time window were aggregated and treated as a single meal event. Nutritional values (calories, proteins, carbs, sugar, fats) were summed for each window. The data were subsequently aggregated into daily summaries by computing total daily macronutrient intake.

Each day, participants completed project-specific daily questionnaires assessing their activities, health status, perceived daily demands, and overall well-being.

Two participants (IDs 154 and 155) were measured repeatedly (4 and 3 times, respectively), ca. 1× (10 days) per month. The protocol was the same as the one designed above. There was only a single CPET conducted outside the designated measurement period. The dataset from this single test was then replicated across all measurement folders for both ID 154 and ID 155.

Data Description

All the data are structured into folders with 6-digit IDs – the first three numbers represent participant ID and the last three numbers represent measurement ID.

To reduce re-identification risk, all calendar dates were shifted into the years 2100–2199 using a participant-specific random offset of whole weeks. This preserves days of the week, clock times, and all within-participant timing (such as ECG, CGM, and sleep alignment, as well as intervals between measurement periods). Time relationships between different participants are not preserved.

In each folder, the ECG, ACC, and HR data are stored as separate files for each calendar day. Each ECG and ACC day file covers 24 h from 00:00:00 to 23:59:59 local time; periods without a recording (e.g., before the start and after the end of the measurement, during recharging the device) are filled with NaN. Sample k corresponds to the time start_datetime + k / sampling frequency:

• *_date_ECG.h5 are ECG signals, * means ID, date is in format YYYY-MM-DD (e.g. 139001_2106-05-06_ECG.h5)
• *_date_ACC.h5 are ACC signals
• *_date_HR.csv includes automatically detected QRS complexes in samples from the beginning of the signal, beat-by-beat calculated HR in beats per minute (bpm), and length of RR intervals in ms (both based on detected QRS positions).

And the data associated with each subject for the whole session are present:

• *_CGM.csv includes timestamp and glycaemia value
• *_activities.csv includes activity type, date and time, duration, and information on how much time was spent in each training zone (based on the CPET result), and the type of activity (basic, tempo, or interval) is specified in the Note column.
• *_sleep.csv includes the date of waking up, start and end times of the sleep and its duration.
• *_CPET_data.csv includes 19 important selected CPET features – oxygen uptake (VO2), carbon-dioxide output (VCO2), oxygen uptake per kilogram (VO2_kg), oxygen pulse (VO2_HR), heart rate (HR), work rate (WR), ventilatory equivalents for oxygen and carbon dioxide (VE_VO2, VE_VCO2), respiratory exchange ratio (RER), minute ventilation (VE), tidal volume (VT), breathing frequency (BF), overall energy expenditure (EE), substrate-specific oxidation rates for fat, carbohydrate, and protein (EEFAT, EECHO, EEPRO), excess carbon dioxide (ExCO2), end-tidal partial pressures of CO₂ and O₂ (PetCO2, PetO2) and time (t), test phase (WarmUp, Test, CoolDown).
• *_CPET_metadata.csv includes subject, test, and environment info
• *_CPET_summary.csv includes CPET feature values for VT1, VT2, and VO2peak + units for all features.
• *_meals.csv includes the time of each meal, description of the meal in the original language, calories, protein, carbs, sugar, and fats for each meal.
• *_daily_meal_summary.csv includes a summary of daily calories, protein, carbs, sugar, and fats.
• *_daily_steps_summary.csv includes a summary of daily steps.

The file measurement_starts_and_ends.csv includes starts and ends of measurements (based on CGM data where possible, otherwise based on ECG data) for all measurement IDs.

The file daily_questionnaire.csv provides information from the daily questionnaire for each subject. It includes date of measurement, sex, menstruation status, menstrual pain, objective physical load, subjective physical load, mental load, any sport activity that day, objective sport load, subjective sport load, illness symptoms, physical pain, digestive issues, sleep quality, and information about non-sport physical activity that day. File daily_questionnaire_code.csv includes coding information.

The file subject-info.csv provides basic information for each subject from entry questionnaire: age, sex, weight, height, body mass index (BMI), body fat, residence type, info about regular exercise and which sports, sports level, how many years the subject do sports, weekly training load, fulfillment of performance self-expectation, whether they achieving goals, work activity level, daily activity level, info about diet and food allergies, chronic illnesses, past illnesses still affecting life, micronutrient deficiencies, frequency of consuming alcohol and its amount, number of cigarettes smoked per day, number of other nicotine products used per day, medications, systolic and diastolic blood pressure, total, low-density lipoprotein (LDL), high-density lipoprotein (HDL) cholesterol, triglyceride level, and smartwatch brand. The file subject-info_code.csv includes coding information.

In data_completness.csv is indicated which types of data are present for the subject and which are not. This file indicates the completeness of the database. The file data_completness_code.csv includes coding information.

Usage Notes

The multimodal architecture of the BUT CGM database makes it a reusable resource for investigating cardio–metabolic coupling, autonomic regulation, and physiology-aware modeling in free-living healthy adults. The synchronized integration of CGM, high-resolution ECG, accelerometry, sport annotations, sleep summaries, and nutritional intake enable mechanistic analyses of glycemic variability in relation to cardiac electrophysiology and behavioral context. It allows researchers to examine how interacting lifestyle factors shape postprandial glucose dynamics, autonomic balance, and short-term metabolic risk trajectories across multiple days [1, 2, 5, 6].

A distinctive methodological advantage of this dataset is the availability of ventilatory-threshold–calibrated (VT1, VT2) exercise intensity domains derived from laboratory CPET. This enables stratification of glycaemic responses, heart rate variability patterns, and recovery kinetics according to objectively defined internal load rather than self-reported activity categories or external workload alone. Such annotation supports development of individualized predictive models of glucose dynamics under varying metabolic stress, investigation of intensity-dependent substrate utilization effects, and training of explainable machine-learning frameworks that incorporate physiologically meaningful anchors. The dataset is therefore suitable not only for secondary statistical analyses but also for validation of multimodal digital biomarkers and adaptive performance-support algorithms.

Participants 154 and 155 were measured repeatedly (records 154001-154004 and 155001-155003), so the dataset contains 65 records from 60 unique participants. The same CPET test is shared across all records of each of these participants. To avoid data leakage in machine learning applications, data should be split at the participant level (first three digits of the folder ID), keeping all records of one participant within a single subset. In statistical analyses, the repeated measurements should be accounted for, for example by using mixed-effects models or only one record per participant.

There are several limitations in BUT CGM data use. Some data are missing which is indicated in data_completness.csv file. Five participants did not take the CPET test during the 10-day measurement period, for 6 participants, the test is missing. Data from the daily questionnaire and activity data may differ, as a specific activity may not have been recorded by the smartwatch (or vice versa). However, solely the smartwatch logged activities are within activity annotations. It is possible that data about energy intake was missing or may be inaccurate. Data completeness depended on participant compliance and self-reporting diligence protocols. For some subjects, a portion of the CGM data may be missing.

Custom code/application for data visualization is available on GitHub https://github.com/BioSig-BUT/BUT-CGM-supplement.

Release Notes

Version 1.0.0

Initial release

Ethics

Before the start of the study, ethics approval was obtained from the Ethics Committee of the Centre of Sports Activities, Brno University of Technology, on January 10, 2025 (IRB Protocol reference number EK:1/2025). To ensure voluntary participation, all aspects of the pilot — including its purpose, procedures, and potential risks and benefits — were explained in detail to participants, and all questions were answered. Participants were given sufficient time to consider their involvement. Informed written consent was obtained from all participants during the initial laboratory visit.

Acknowledgements

This work was developed under the Horizon KDT JU project DistriMuSe no. 101139769-2/9A24005.

Conflicts of Interest

The authors have no conflict of interest.

References

1. Holzer, R., Bloch, W., & Brinkmann, C. (2022). Continuous Glucose Monitoring in Healthy Adults—Possible Applications in Health Care, Wellness, and Sports. Sensors, 22(5), 1-10. https://doi.org/10.3390/s22052030.
2. Brummer, J., Glasbrenner, C., Hechenbichler Figueroa, S., Koehler, K., & Höchsmann, C. (2024). Continuous glucose monitoring for automatic real-time assessment of eating events and nutrition: a scoping review. Frontiers in Nutrition, 10, 1-21. https://doi.org/10.3389/fnut.2023.1308348.
3. American Thoracic Society & American College of Chest Physicians. (2003). ATS/ACCP Statement on Cardiopulmonary Exercise Testing. American Journal of Respiratory and Critical Care Medicine, 167(2), 211-277. https://doi.org/10.1164/rccm.167.2.211.
4. Beaver, W. L., Wasserman, K., & Whipp, B. J. (1986). A new method for detecting anaerobic threshold by gas exchange. Journal of Applied Physiology, 60(6), 2020-2027. https://doi.org/10.1152/jappl.1986.60.6.2020.
5. Hansen, C. S., Færch, K., Jørgensen, M. E., Malik, M., Witte, D. R., Brunner, E. J., Tabák, A. G., Kivimäki, M., & Vistisen, D. (2019). Heart Rate, Autonomic Function, and Future Changes in Glucose Metabolism in Individuals Without Diabetes: The Whitehall II Cohort Study. Diabetes Care, 42(5), 867-874. https://doi.org/10.2337/dc18-1838.
6. Das, A., Kerr, D., Glantz, N., Bevier, W., Santiago, R., Gutierrez-Osuna, R., & Mortazavi, B. J. (2025). CGMacros: a pilot scientific dataset for personalized nutrition and diet monitoring. Scientific Data, 12(1), 1-9. https://doi.org/10.1038/s41597-025-05851-7.
7. Smital, L., Marsanova, L., Smisek, R., Nemcova, A. & Vitek, M. (2020). Robust QRS Detection Using Combination of Three Independent Methods. Computing in Cardiology Conference (CinC). https://doi.org/10.22489/CinC.2020.100.