QoL_Stress: A Multimodal Dataset of Physiological and Self-Reported Stress Responses

@article{PhysioNet-qol-stress-1.0.0,
  author = {Calvo, Amaia and Cejudo, Ander and {Martín Andonegui}, Cristina},
  title = {{QoL\_Stress: A Multimodal Dataset of Physiological and Self-Reported Stress Responses}},
  journal = {{PhysioNet}},
  year = {2026},
  month = oct,
  note = {Version 1.0.0},
  doi = {10.13026/vqtz-pm70},
  url = {https://doi.org/10.13026/vqtz-pm70}
}

@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

This dataset contains physiological signals and self-reported measures collected during a controlled stress protocol. Sixty-six healthy adults participated in standardized experimental sessions consisting of a relaxation phase followed by stress induction using cognitive and interactive tasks (Stroop Color Test and Geometry Dash). Physiological recordings include electrocardiogram (ECG, 250 Hz), electrodermal activity (EDA), and heart rate (HR, 1 Hz) obtained from wearable devices (Fitbit Charge 6). Participants also completed validated self-reported questionnaires, including the Perceived Stress Scale (PSS-14) and the State-Trait Anxiety Inventory (STAI-S). The recordings are linked by participant identifiers, phase-annotated (RELAX/STRESS), de-identified, and provided in structured collections. The dataset supports research on stress detection, physiological signal analysis, multimodal modeling of stress responses, and machine learning applications for emotion and stress recognition.

Background

Stress is a major public health concern, with profound physiological and psychological effects. Acute stress responses involve complex interactions between the autonomic nervous system and endocrine mechanisms, reflected in physiological signals such as heart rate, ECG, and EDA [1,2]. These signals are widely used to study stress and develop automated detection systems.

Recent advances in wearable sensing technologies enable continuous, non-invasive monitoring of physiological signals in both laboratory and real-world contexts [3,4]. Devices capturing ECG, EDA, and HR allow analysis of stress-related physiological patterns with high temporal resolution. Complementary psychological instruments such as PSS-14 and STAI-S provide validated assessments of perceived stress and state anxiety [5,6].

Despite growing interest in wearable-based stress detection, publicly available datasets combining multimodal physiological recordings with validated self-reported measures remain limited. Existing datasets, such as WESAD, provide benchmarks but differ in sensing modalities, protocols, and annotation strategies [7,8].

QoL_Stress addresses this gap by providing synchronized physiological and self-reported data collected during controlled relaxation and stress-induction protocols, enabling research in stress detection, psychophysiology, and affective computing.

Methods

Measurement Device

The Fitbit Charge 6 is a wrist-worn wearable device designed for continuous real-time physiological monitoring, suitable for research applications [3,9]. It includes an optical photoplethysmography (PPG) sensor providing continuous heart-rate measurements. It also incorporates multipurpose electrical sensors for single-lead ECG and EDA for assessing sympathetic arousal.

Physiological and Self-Reported Measures

Physiological signals were recorded as follows:

• Electrocardiogram (ECG): single-lead ECG recordings were acquired at the end of each experimental phase. Each recording lasted 30 seconds, according to the Fitbit acquisition procedure. These measurements were used to characterize the physiological state following each experimental condition and to derive HR and HRV indices [9,10].

• Electrodermal Activity (EDA): EDA measurements were recorded to characterize skin conductance and sympathetic activation. EDA recordings lasted up to 2 minutes, which was the maximum duration selected for the study [10].

• Heart Rate (HR): recorded continuously throughout the session at per-second and per-minute resolution [3].

Self-reported measures included:

• Perceived Stress Scale (PSS-14): 14-item questionnaire evaluating perceived stress over the past month [5].

• State Anxiety (STAI-S): 20-item scale assessing state anxiety; threshold of 36 distinguishes low/high state anxiety [6].

All measures were linked through participant identifiers and, where available, temporal information and experimental phase (RELAX/STRESS), enabling multimodal analyses.

Population

Sixty-six healthy adults (39 males, 27 females; mean age 29.7 ± 6.9 years) participated. Inclusion criteria included age ≥ 18 years and ability to provide informed consent. Exclusion criteria included cardiovascular or dermatological conditions, psychoactive medication use, or inability to complete the protocol. Written informed consent was obtained. The study was approved by the Ethics Committee for Research of the Gipuzkoa Health Area in accordance with applicable ethical standards and data protection laws.

Experimental Protocols

Sessions lasted ~30 minutes and followed a structured two-phase protocol:

• Relaxation Phase: participants first completed the PSS-14 and watched a guided relaxation video. At the end of the relaxation phase, ECG (30 seconds) and EDA (up to 2 minutes) were recorded to characterize the physiological state following relaxation, followed by the STAI-S assessment.

• Stress Induction Phase: participants completed the Stroop Color Test and Geometry Dash gameplay to induce stress. At the end of the stress phase, ECG (30 seconds) and EDA (up to 2 minutes) were recorded again, followed by the second STAI-S assessment. HR was continuously monitored throughout the session.

Data Collection and Quality Control

Signals were checked for recording availability and quality. No temporal interpolation or resampling was applied to the EDA observations. The original time_sec values and repeated timestamps present in the device export were retained. Eight EDA recordings were not successfully obtained because of acquisition errors and were therefore not included in eda.csv.

Data Preparation for Analysis

Data were de-identified, temporally organized where applicable, and labeled by experimental phase and participant ID. All data are exported in CSV format, organized by collection (e.g., ecg.csv, eda.csv, fitbit_heart_rate_per_second.csv), and structured to enable linking physiological signals with self-reported measures [3,4].

To reduce the risk of re-identification, absolute calendar dates were removed from the publicly released tabular data. Time-of-day information was retained where necessary to support temporal alignment between physiological recordings and self-reported measures. Thus, date information was removed from the Fitbit heart-rate files, ECG recordings, and questionnaire files, while the corresponding time information was retained where applicable. The EDA data already use relative time in seconds (time_sec) and do not contain calendar dates. Participant demographic data also contain no date information.

Data Description

The QoL_Stress dataset is organized into collections exported as CSV files. Each file contains data for all participants; phase labels (RELAX/STRESS) and participant identifiers allow linking physiological signals with self-reported measures, enabling multimodal analyses.

File Naming

Files are named after the collection, e.g., ecg.csv, eda.csv, fitbit_heart_rate_per_second.csv, and participants.csv. Each file contains data for the participants included in that collection. Phase labels (RELAX/STRESS) and participant identifiers allow linking physiological signals with self-reported measures.

Data Format

All files are provided in CSV format and can be analyzed using standard tools such as Python (Pandas, NumPy), R, or MATLAB. Temporal information is provided where appropriate to support alignment between physiological and self-reported measures.

In ecg.csv, waveform_samples contains the complete sequence of ECG samples for each recording within a single CSV field. In eda.csv, time_sec is a relative temporal marker from the device export. The interpretation and handling of this field are described in the EDA time representation section below.

Participant Identifiers

The username field is a pseudonymous SHA-256 identifier used to link records belonging to the same participant across data collections. It does not contain the participant's original username or directly identifying information.

Data Dictionary

The following dictionary describes the variables required to interpret the released CSV files. Data types refer to the values stored in the CSV files. The Role column indicates whether a field is a primary analysis variable, an analysis-related variable, a quality-control variable, a linking/temporal variable, or metadata/informational information.

ecg.csv

eda.csv

Other CSV files

EDA time representation

The time_sec field in eda.csv is the relative temporal marker provided by the Fitbit EDA export. It is not a continuously sampled timestamp and does not represent a one-second sampling interval.

Several consecutive EDA observations may share the same time_sec value. The temporal marker typically advances by approximately 5 seconds, with some increments of 6 seconds. These values reflect the temporal representation of the device export and were retained without interpolation or resampling.

Users should therefore preserve the original time_sec values when performing temporal analyses. Consecutive rows should not be assumed to be equally spaced one-second samples, and users should not reconstruct a one-second time axis from this field unless applying an explicitly documented processing method.

Usage Notes

The QoL_Stress dataset can be used for stress detection, physiological signal analysis, multimodal modeling, and machine learning, deep learning, and LLM-based approaches.

The dataset has several limitations. It includes 66 healthy adults and was collected in a controlled experimental setting using specific stress-induction tasks (Stroop Color Test and Geometry Dash). Therefore, the results obtained from this dataset may not generalize directly to other populations or real-world stress conditions. In addition, ECG was recorded at the end of each experimental phase rather than continuously during the stress-induction tasks. This protocol was designed to characterize the physiological state following each condition and enable comparison between the relaxation and stress phases. Consequently, transient cardiovascular responses occurring during the tasks may not be captured.

The dataset has also been used in an ongoing study investigating stress prediction using machine learning, deep learning, and large language models. The corresponding manuscript is currently under preparation and has not yet been published.

For comparative studies, users may consider related publicly available datasets such as WESAD, which also provides multimodal physiological data for stress and affect detection.

Release Notes

Version: 1.0.0

Initial release of the data

Ethics

The study was approved by the Ethics Committee for Research of the Gipuzkoa Health Area, in accordance with Spanish Law 14/2007 on Biomedical Research, the ethical principles of the Declaration of Helsinki, and all other applicable ethical standards. The study was also conducted following the General Data Protection Regulation (GDPR; EU 2016/679) and the Spanish Organic Law 3/2018 on the Protection of Personal Data and Guarantee of Digital Rights (LOPDGDD).

All participants provided written informed consent prior to participation. The consent form included detailed information about the study objectives, experimental procedures, potential risks and benefits, and the voluntary nature of participation, allowing participants to withdraw at any time without justification.

The study did not involve any invasive or medical procedures. All publicly released data were de-identified, and participants consented to the sharing of demographic and physiological data for research purposes.

Acknowledgements

We thank all participants and the Vicomtech team for their contributions.

Conflicts of Interest

The authors declare no competing interests.

References

1. Abd‑Alrazaq, A., Alajlani, M., Ahmad, R., AlSaad, R., Aziz, S., Ahmed, A., … Sheikh, J. (2024). The performance of wearable AI in detecting stress among students: Systematic review and meta‑analysis. Journal of Medical Internet Research, 26, e52622. https://doi.org/10.2196/52622
2. Al Abdi, R., AlKaabi, S., Elsifi, S., & Yousaf, J. (2026). Mental stress detection using physiological sensors and artificial intelligence: A review. Sensors, 26(5), 1616. https://doi.org/10.3390/s26051616
3. Almadhor, A., Sampedro, G. A., Abisado, M., Abbas, S., Kim, Y.‑J., Khan, M. A., … Baili, J. (2023). Wrist-based electrodermal activity monitoring for stress detection using federated learning. Sensors (Basel), 23(8), 3984. https://doi.org/10.3390/s23083984
4. Zhu, L., Spachos, P., Ng, P. C., Yu, Y., Wang, Y., Plataniotis, K., & Hatzinakos, D. (2023). Stress detection through wrist-based electrodermal activity monitoring and machine learning. IEEE Journal of Biomedical and Health Informatics, 27(5), 2155–2165. https://doi.org/10.1109/JBHI.2023.3239305
5. Cohen, S., Kamarck, T., & Mermelstein, R. (1983). A global measure of perceived stress. Journal of Health and Social Behavior, 24(4), 385–396. https://doi.org/10.2307/2136404
6. Spielberger, C. D. (1983). Manual for the State-Trait Anxiety Inventory (STAI). Consulting Psychologists Press.
7. Schmidt, P., Reiss, A., Duerichen, R., Marberger, C., & Van Laerhoven, K. (2018). Introducing WESAD, a multimodal dataset for wearable stress and affect detection. In Proceedings of the 20th ACM International Conference on Multimodal Interaction (pp. 400–408). https://doi.org/10.1145/3242969.3242985
8. Pinge, A., Gad, V., Jaisighani, D., Ghosh, S., & Sen, S. (2024). Detection and monitoring of stress using wearables: A systematic review. Frontiers in Computer Science, 6, 1478851. https://doi.org/10.3389/fcomp.2024.1478851
9. Crosswell, A. D., & Lockwood, K. G. (2020). Best practices for stress measurement: How to measure psychological stress in health research. Health Psychology Open, 7(2), 2055102920933072. https://doi.org/10.1177/2055102920933072
10. Iqbal, T., Elahi, A., Redon, P., Vazquez, P., Wijns, W., & Shahzad, A. (2021). A review of biophysiological and biochemical indicators of stress for connected and preventive healthcare. Diagnostics, 11(3), 556. https://doi.org/10.3390/diagnostics11030556