DX-anes/PeriWaveSync-ABG

---

Title: A perioperative electrocardiography and photoplethysmography dataset synchronized with arterial blood gas measurements.

We developed PeriWaveSync-ABG, a prospective four-center perioperative database comprising ECG and PPG waveforms synchronized with ABG measurements. The dataset was created to support collaborative research and methodological development in perioperative medicine. It contains continuous intraoperative ECG and PPG signals acquired using wearable hardware, paired with repeated arterial blood gas measurements obtained during the same operative episode. Because waveform segments were aligned to the timing of blood sampling, the dataset enables investigation of how continuously observed non-invasive signals relate to intermittently measured invasive biochemical markers. In contrast to perioperative databases built primarily around bedside monitoring systems, this resource emphasizes wearable signal acquisition and the joint analysis of waveform data with blood gas and biochemical variables, while retaining the heterogeneity and clinical realism of intraoperative care. The dataset is suitable for a range of downstream applications, including signal processing, multimodal integration, digital biomarker discovery, non-invasive estimation of laboratory variables, and perioperative risk modeling. We expect it to complement existing perioperative data resources and to support further work on precision monitoring in both intraoperative and postoperative settings.

Repository Structure

File/Folder	Description
A.zip / B.zip / C.zip / D.zip	ECG/PPG waveform segments from Research Centers A, B, C, and D, respectively
Demographic.csv	Patient demographics, anthropometrics, and comorbidities
Surgery-related.csv	Surgery- and anesthesia-related variables
ABG.csv	Event-level arterial blood gas records, vital signs
Medicines.csv	Intraoperative medication administration records
Parameters.txt	Full data dictionary (variable names, coding, units, missing-value convention)

All tables, as well as the waveform files, can be linked across the dataset via the de-identified patient identifier (subject_id / Medical Record Number code).

Waveform Files

Each .zip archive (one per research center) contains individual .txt files, one per waveform segment. File names follow the pattern:

{Research_Center}-{subject_id}-{random_integer}.txt

Example: A-0253-368.txt

Here, random_integer is an arbitrary, non-sequential identifier assigned solely to guarantee file name uniqueness within a given patient folder; it carries no temporal or clinical meaning and should not be used for matching.

File Content Format

Each .txt file is a plain-text file with exactly four lines:

Signal start time: YYYY-MM-DD HH:MM:SS.ffffff
ECG1 data: [v1, v2, v3, ...]
PPG1 data: [v1, v2, v3, ...]
PPG2 data: [v1, v2, v3, ...]

Line	Field	Description
1	Signal start time	Timestamp (de-identified via patient-level random time-shift, see below) of the first sample in this segment
2	ECG1 data	ECG signal (modified Lead I), unit: mV
3	PPG1 data	PPG signal, red channel (~660 nm), unit: mV
4	PPG2 data	PPG signal, infrared channel (~940 nm), unit: mV

All three channels are synchronously sampled at 500 Hz. No per-sample timestamps are stored; the timestamp of the n-th sample (0-indexed) within a given array can be computed as:

sample_time(n) = Signal start time + n / 500 seconds

No digital filtering was applied beyond the analog front-end amplification of the recording device; raw physiological artifacts (e.g., motion artifact, lead-off events, sensor saturation) may be present and have been intentionally retained for research purposes (see Usage Notes).

Linking Waveforms to ABG Events

Waveform segments are linked to arterial blood gas (ABG) sampling events in ABG.csv through patient identity and timestamp matching, rather than through an explicit file-name reference column. To identify the waveform segment corresponding to a given ABG event:

1. From ABG.csv, compute the absolute collection timestamp of the ABG event by combining surgery_date and collection_time (Hour/Minute/Second).
2. Within the waveform folder corresponding to the same subject_id (and research center), locate the .txt file whose Signal start time places the ABG collection timestamp approximately at the midpoint of the recorded segment — i.e., each segment spans approximately 5 minutes before and 5 minutes after the actual arterial blood draw (≈10 minutes total).
3. Note that both the ABG collection time and the waveform Signal start time have undergone the same patient-level random time-shift during de-identification (see Figure 3 / De-identification Process), so their relative temporal relationship — and therefore the matching logic above — remains valid, even though absolute calendar dates are not real.
4. If no waveform segment was successfully recorded or matched for a given ABG event, no corresponding file exists; this is not separately flagged in ABG.csv.

Variable Dictionary

See Parameters.txt for the complete list of column names, coding schemes, units, and the comorbidity items comprising the age-adjusted Charlson Comorbidity Index.

Missing Value Convention

Missing values are represented as empty cells across all released tables.

Usage Notes

This dataset is released under the CC-BY 4.0 license. We ask users not to attempt to re-identify any individual in this dataset.

License

CC-BY 4.0