A Veterinary Vital Sign Time-Series Database Collected During General Anesthesia

@article{PhysioNet-veterinary-vitals-anesthesia-1.0.0,
  author = {Cha, Jeesoo and Lee, Hyung-Chul},
  title = {{A Veterinary Vital Sign Time-Series Database Collected During General Anesthesia}},
  journal = {{PhysioNet}},
  year = {2026},
  month = oct,
  note = {Version 1.0.0},
  doi = {10.13026/wvfp-p831},
  url = {https://doi.org/10.13026/wvfp-p831}
}

@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 resource contains physiological time-series recordings from dogs and cats undergoing inhalational general anesthesia at VIP Animal Medical Center, Cheongdam, Seoul, Republic of Korea. The dataset was assembled from routine clinical recordings obtained in 2026 and includes 226 anesthesia cases (191 canine and 35 feline). Some cases were stored as multiple sequential recording files, and a small number of animals contributed more than one case; animal-level identifiers are not included in this release. Data were acquired using Vital Recorder from Dräger Infinity M540 patient monitors and, where connected to the recording system, Dräger Atlan A350 anesthesia workstations. Available signals differ between recordings: some contain monitor-derived numerical measurements only, whereas others additionally contain physiological waveforms and/or anesthesia-machine measurements and settings. Accompanying clinical metadata describe species, breed, age, sex, neuter status, body weight, physical status, and procedures performed. The release includes 241 de-identified recording files, totaling approximately 369 hours of recording, together with file-level signal inventories, a data dictionary, and an example for reading the recordings. The resource supports exploratory veterinary perioperative physiology research, signal-processing method development, and feasibility assessment for future larger studies. It represents a single-center, availability-based cohort rather than a complete census of anesthetized patients; heterogeneous signal availability and limited subgroup sizes should be considered when interpreting analyses.

Background

Physiological monitoring during veterinary general anesthesia produces repeated measurements of cardiovascular, respiratory, and other physiological variables. The clinical context of those measurements—including animal species, body size, and the procedure being performed—is necessary for meaningful interpretation. Sharing time-series recordings together with structured clinical metadata can enable reproducible exploratory analyses of perioperative physiological changes.

Vital Recorder provides a means of collecting physiological data from multiple anesthesia-related devices [1]. This project applies that recording approach to routine canine and feline clinical care. Its purpose is to make a documented collection of veterinary anesthesia recordings available for reuse, including investigation of data completeness, waveform characteristics, and numerical physiological trends. The dataset is not derived from the human VitalDB cohort.

Recordings were available only when a clinical workstation was connected to the recording system and a usable file was retained. Accordingly, the collection is described as an availability-based convenience sample. It is not claimed to be representative of all animals treated at the hospital or of the wider veterinary population.

Methods

Setting, cohort, and selection

The dataset was assembled at VIP Animal Medical Center, Cheongdam, Seoul, Republic of Korea. The investigator team includes a veterinarian working in anesthesiology and pain management. The recordings were not collected for research purposes; they were obtained during routine clinical anesthesia monitoring and were retrospectively assembled for this resource. Clinical care and data collection occurred between January 2026 and September 2026.

The initial screening source was the hospital's 2026 anesthesia log. Log entries with an explicitly documented Vital Recorder filename were identified and matched against the recording files. Inclusion required a canine or feline patient, documented inhalational general anesthesia, a verifiable correspondence between the clinical case and its recording file(s), and a recording file that could be opened and contained at least two physiological tracks. The final cohort comprised 226 anesthesia cases (191 canine and 35 feline) represented by 241 recording files. A small number of animals underwent more than one anesthesia case during the study period; to reduce re-identification risk, identifiers linking cases from the same animal are not released. When a single case was stored in multiple sequential files, the files share the case identifier and are distinguished by a numeric suffix.

Of the 228 anesthesia-log entries with a documented Vital Recorder filename, 1 was excluded because injectable rather than inhalational anesthesia was recorded and 1 because clinical information was missing, leaving 226 candidate cases. Of these, none were excluded because no matching file was found, because of unresolved duplicate or conflicting file linkage, because a file could not be read or contained no data, or because the minimum file criterion was not met. Exclusion reasons were applied in the order listed, and each case was counted only under the first applicable reason. The final cohort comprised 226 anesthesia cases (191 canine and 35 feline) represented by 241 recording files. No inference about recording availability was made from a patient name or from the mere presence of an entry in the anesthesia log.

Clinical context

Animals underwent clinically indicated surgical and diagnostic procedures, including abdominal soft-tissue surgery, orthopedic surgery, excision of cutaneous and subcutaneous masses, spinal surgery, thoracic and cardiovascular procedures, and diagnostic imaging under anesthesia. The English rendering of each case's procedure is provided in procedure_glossary.csv, where procedure counts are also summarized. Because some cases involved more than one procedure type, category counts sum to more than the number of cases: abdominal soft-tissue 62, orthopedic 60, mass excision 46, diagnostic imaging 33 (imaging only 18), reproductive 26, spinal 12, thoracic/cardiovascular 10, and head/oral/ophthalmic 6. All included cases were managed or supervised by a veterinary anesthesiologist, and no research-specific interventions were performed. Sevoflurane was recorded as the maintenance agent in 204 cases and isoflurane in 20 cases; the maintenance agent was not clearly recorded in the remaining 2 cases, and ambiguous drug-column marks were not converted into confirmed exposure.

Clinical metadata were abstracted from the anesthesia log. Age in months was derived as 12 times the recorded year component plus the additional month component. Entries with a month component of 12 were flagged for source review. The original ASA physical-status entry was retained, and a numerical class was assigned only where the entry was unambiguous. Absence of an emergency marker was not interpreted as proof of elective status. Clinical-log anesthesia and procedure durations are distinct from the duration of the physiological recording. Anesthesia duration was defined as the interval from induction of anesthesia to extubation. Procedure duration was defined as the interval from first incision to completion of wound closure for surgical cases, and from the start of the first acquisition to the end of the last acquisition for diagnostic imaging (CT or MRI) cases.

Recording equipment and acquisition

The clinical equipment consisted of Dräger Infinity M540 patient monitors and Dräger Atlan A350 anesthesia workstations (Dräger, Lübeck, Germany). Numerical parameters were obtained from the M540 monitor, and high-resolution arterial-pressure and electrocardiogram waveforms were digitized through an analog-to-digital converter module (SNUADCM). This module is SNUADC, an open-source 8-channel 10-bit analog-to-digital converter developed by the Vital Recorder development group; its hardware design, part list, and firmware are publicly available [4]. Vital Recorder was used for acquisition [1]. Monitor and anesthesia-workstation software versions and installed modules were not retained in the recordings. The version of Vital Recorder used to create each recording is stored in the header of every .vital file (prog_ver field), as specified in the Vital file-format documentation [2], and was versions 1.15.9 (16 files), 1.15.11 (166 files), 1.18.41 (37 files), and 1.18.43 (22 files).

Recorded tracks are grouped under the device labels Infinity (M540 numerics), MedibusX (Dräger anesthesia-machine data, present only when the workstation was connected), and SNUADCM (digitized analog waveforms). In the released files, the two waveform tracks—invasive arterial pressure (ART, mmHg) and lead-II electrocardiogram (ECG_II, mV)—were stored at 500 Hz. Numerical tracks, including heart rate, pulse oximetry (SpO2 and pulse rate), esophageal temperature, invasive arterial pressure (systolic/diastolic/mean), and noninvasive blood pressure (systolic/diastolic/mean), were stored at approximately 1 Hz. Channel availability varied between files, and not every track was present in every recording.

Sensor and measurement details, where used, were as follows: the electrocardiogram was recorded in lead II; invasive arterial pressure was measured from a dorsal pedal arterial catheter with the transducer referenced to heart level; pulse oximetry was obtained from a lingual or gingival mucosal site; and temperature was measured as esophageal temperature. Pulse oximetry was measured using the Dräger pulse-oximetry function integrated in the M540 monitor, and noninvasive blood pressure was recorded in a subset of cases. Transducer calibration values and any input scaling or filtering applied within the acquisition pathway were not retained with the recordings.

Not all clinical workstations were connected to Vital Recorder. In addition, the connected devices and recorded channels varied by case. Therefore, missing anesthesia-machine data do not imply that no anesthesia workstation was used, and a missing waveform does not imply that the underlying physiological quantity was not monitored clinically. No assumption is made that every file contains the same channels. Acquisition frequency and numerical update intervals are reported from the released records, rather than inferred from display zoom settings or manufacturers' maximum specifications.

Processing, temporal alignment, and quality control

No processing was applied to the contents of the .vital files. The released recordings retain the original sampling rates, timestamps, and any missing or invalid intervals as recorded; no trimming, track removal, unit conversion, resampling, artifact removal, or imputation was performed. File selection and the renaming of files for de-identification were the only operations applied at the file level. A file-level track inventory (track_inventory.csv) listing each file's tracks, record types, units, sampling rates, and record counts is provided for reference. No waveform-quality screening, artifact removal, or exclusion of physiological extremes was performed, so users should assess signal quality themselves for their intended analysis.

Device timestamps are those assigned by each acquisition source (SNUADCM and the Infinity monitor) at recording time. Clock synchronization between devices was not separately verified, and any inter-device timing offset is unknown. Although tracks from different devices are displayed on a common timeline in Vital Recorder, sample-level synchronization across devices should not be assumed. For a case stored in multiple files, the original segment order and elapsed gaps between segments are preserved.

De-identification and ethics

Public tables omit owner names, animal names, hospital chart numbers, staff names, exact clinical dates, identifiers linking repeat cases from the same animal, and the private linkage table. Original filenames, which contained recording dates and times, were replaced with case identifiers. Timestamps, events, string tracks, device names, port fields, and other metadata embedded in the .vital files were inspected and de-identified, since renaming files alone does not remove embedded times. Relative timing is preserved within each case and across its file segments. Because animal-level identifiers are not released, intervals between cases from the same animal are not available. Internal timestamps in every released .vital file were shifted to a common synthetic reference (January 1, 2000), removing all real dates and times while preserving relative timing within each recording; for cases stored in multiple files, the original intervals between segments were preserved.

The institution determined that this project—the secondary use of clinical data generated during routine veterinary care, with no additional animal treatment or intervention—was not subject to formal review by an animal ethics committee, and VIP Animal Medical Center, Cheongdam, approved the use of the data for research and its public sharing on PhysioNet on the condition that the data be de-identified (institutional approval dated September 9, 2026; approving authority, the institutional representative). Because the project used only previously recorded clinical data and involved no additional contact with owners or animals, no separate owner consent was required for this secondary use, as specified in the institutional authorization.

Cohort-size rationale

This is a data-resource release rather than a study designed to test a prespecified treatment effect. Cohort size reflects the eligible recordings available from the connected clinical workstations during the defined period. No formal sample-size or power calculation was performed. The expanded cohort is intended to support exploratory analyses and feasibility work. Its adequacy for a particular analysis depends on the number of cases, usable signal duration, signal availability, and subgroup or event counts; the collection is not presented as sufficient for unrestricted predictive-model development or external clinical validation.

Data Description

The release contains 226 anesthesia cases (191 canine and 35 feline) and 241 recording files; all counts are reported at the case level. Clinical-log characteristics were as follows: median age 105 months (IQR 64–137); median body weight 4.5 kg (IQR 3.3–6.8); male 115 (50.9%); neutered 194 (85.8%); ASA class 1/2/3/4/5 = 48/73/62/36/7; median clinical-log anesthesia duration 113 min (IQR 72–147). Across the 241 recording files, total recording span was approximately 369 hours, with a median per-file span of 88.5 minutes (IQR 55–123, range 5–307). Signal availability, computed at the case level, was: lead-II ECG, heart rate, pulse oximetry, and pulse rate in all 226 cases; esophageal temperature in 225 (>99%); the invasive arterial-pressure waveform in 192 (85%); invasive arterial-pressure numerics in 121 (54%); noninvasive blood pressure in 55 (24%); and anesthesia-machine (MedibusX) data in 52 (23%).

The primary physiological records are stored in .vital format, a documented compressed binary format supporting waveform, numerical, and string records [2]. Clinical metadata and inventories are supplied as UTF-8 CSV files.

The release is organized as follows:

• clinical_metadata.csv: one row per anesthesia case, linked to recording files by case_id.
• record_manifest.csv: one row per recording file, with case_id, filename, segment order, recording span, and file-level signal-availability flags.
• track_inventory.csv: one row per file and track, with device label, track name, record type, unit, waveform sampling frequency where applicable, number of observations, observed duration, and missingness or quality information.
• data_dictionary.csv: definitions, units, allowed values, and missing-data conventions for clinical variables.
• breed_summary.csv and procedure_glossary.csv: breed counts and procedure terminology, including the original procedure text and its English rendering.
• vital/: de-identified physiological recording files.
• read_vital_example.py: example showing how to list and read tracks.
• README.md: package overview and instructions.

Each recording file is named by its case identifier (e.g., 1-31.vital). When a single anesthesia case was stored in multiple sequential files, a numeric suffix indicates segment order (e.g., 1-32-1.vital and 1-32-2.vital belong to case 1-32). Case identifiers that had been duplicated in the source log (2-20, 6-12, 6-15) were reconciled so that every case has a unique identifier.

Recorded tracks (55 distinct signals across the cohort) are grouped by device label. Two high-resolution waveforms were recorded through SNUADCM at 500 Hz: the invasive arterial-pressure waveform (ART, mmHg) and the lead-II electrocardiogram (ECG_II, mV); a plethysmography waveform (PLETH) was additionally present in a subset of recordings. Numerical parameters from the Infinity (M540) monitor, stored at approximately 1 Hz, included heart rate, pulse oximetry (SpO2 and pulse rate), esophageal temperature, invasive and noninvasive blood pressure (systolic/diastolic/mean), and ST-segment and ectopy indices where available. When the anesthesia workstation was connected, MedibusX provided anesthesia-machine data, including capnography and airway-pressure waveforms and numerical ventilation, fresh-gas, and inhalant-agent parameters (for example, end-tidal CO2, inspired/expired sevoflurane or isoflurane, minimum alveolar concentration, tidal volume, and respiratory rate). The complete list of tracks, with record types, units, sampling rates, and per-file availability, is provided in track_inventory.csv.

Missing clinical values are represented by empty CSV fields. The value unknown is used for an explicitly unresolved categorical status, not for a negative observation. The track inventory distinguishes tracks that are absent, empty, or present but unusable.

Usage Notes

The recordings can be opened with Vital Recorder or accessed programmatically through the VitalDB Python library [3]. The accompanying example first lists track names and then reads one explicitly selected track at a user-specified output interval; a resampled array should not be interpreted as the track's native acquisition frequency. The Vital file-format specification provides information for independent parsers [2].

Potential applications include descriptive veterinary perioperative physiology, examination of canine/feline or body-size-related differences, evaluation of signal completeness, and development of exploratory signal-processing methods. Only recordings with the required usable signals should enter a given analysis. Analyses of exposure or intervention effects additionally require detailed treatment timing and confounder information, which should not be assumed to exist in this release.

The dataset has several limitations. It is a single-center, availability-based sample with heterogeneous recording configurations. Some cases are stored in multiple files, and a small number of animals contributed more than one case. Because animal-level identifiers are not provided, cases cannot be assumed to be fully independent; users should consider this when designing statistical analyses or data splits. Device artifacts, disconnections, missing intervals, and clinical case-mix differences may affect results. Clinical-log duration is not interchangeable with recording span, and file-level signal availability does not guarantee complete coverage throughout a case.

No WFDB conversion is included in this release; the physiological recordings are provided in their original .vital format.

Release Notes

Version 1.0.0

The first public release of this resource. It contains 226 anesthesia cases and 241 recording files. Compared with an earlier unpublished submission, this version provides an expanded cohort, detailed clinical metadata, a description of data acquisition, file and track inventories, and usage documentation.

Ethics

This project involves the secondary use of physiologic vital sign data generated during routine veterinary clinical care at VIP Animal Medical Center Cheongdam. No additional animal treatment, testing, or intervention was performed for research purposes, and no additional contact with the owner or animal occurred. The institution reviewed the project plan, scope of data use, and de-identification measures, and determined that the project is not subject to formal review by an animal ethics committee, given that it involves only secondary use of existing clinical records. The institution approved the use of the data for research purposes and its submission and public sharing on PhysioNet, provided that the data are de-identified. All direct identifiers — including owner names, contact details, animal names, medical record numbers, staff names, original filenames, and exact encounter dates/times — were removed prior to external sharing, and random identifiers were assigned to each animal and anesthetic record. Institutional approval was granted on September 9, 2026 (data-use period: November 10, 2025 – December 31, 2026).

Acknowledgements

We thank the veterinarians and veterinary technicians at VIP Animal Medical Center, Cheongdam, who contributed to clinical care, physiological recording, and data organization.

This work received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Conflicts of Interest

The authors have no conflicts of interest to declare.

References

1. Lee, H.-C., & Jung, C.-W. (2018). Vital Recorder—a free research tool for automatic recording of high-resolution time-synchronised physiological data from multiple anaesthesia devices. Scientific Reports, 8, Article 1527. https://doi.org/10.1038/s41598-018-20062-4
2. VitalDB. (n.d.). Vital file format specification. Retrieved September 17, 2026, from https://vitaldb.net/docs/?documentId=VitalRecorder/Vital_File_Format.md
3. VitalDB. (n.d.). Python library. Retrieved September 17, 2026, from https://vitaldb.net/docs/?documentId=API/Python_Library.md
4. VitalDB. (n.d.). SNUADC: An open source 8-channel 10-bit analog-digital converter [Computer hardware]. GitHub. Retrieved September 26, 2026, from https://github.com/vitaldb/snuadc