29 Nov Validation of the accuracy of PetPace pulse rate measurements whitepaper
Validation of the accuracy of PetPace™ pulse rate measurements
Mickey Scheinowitz1 and B. Duncan X. Lascelles2
1Department of Biomedical Engineering, Tel Aviv University, Israel
2Department of Surgery and Pain Management, North Carolina State College of Veterinary Medicine, USA
Heart rate, or pulse rate if measured peripherally, is an essential part of patient examination, assessment and monitoring. The most commonly used medical devices rely on ECG or pulse-oximetry technologies to capture the pulse signal. They typically require physical attachments to the patients (e.g. clips, stickers), preparations (e.g. shaving) and an immobile patient. Few devices, such as a Holter monitor, allow pulse monitoring while the patient is away from the hospital. However, the Holter’s data is not available in real-time and requires offline extraction and interpretation.
As early as 1991 Cain  and colleagues have developed a system to monitor environmental factors (including heart rate) to reduce stress produced to horses during transportation. Years later, Fletcher et al  developed a new wearable wireless device for heart rate, autonomic activity (heart rate variability – HRV), activity and temperature monitoring device. The sensors were incorporated into a vest, which was tested on monkeys during laboratory and ambulatory conditions. They showed the feasibility of their newly developed sensor, however they did not compare their data to a standard method. Another example was published by Landis-Hanna et al in 2014 , showing the use of an ultra-wide band (UWB) device worn on canine’s neck to measure resting heart rate. The reported accuracy was >80% when comparing the device to a reference ECG. However, the device can be used to measure pulse only once a day and the patient has to be completely immobile. Furthermore, the device in question is no longer available commercially.
PetPace™ had developed a smart collar for dogs and cats that monitors a range of parameters, one of which is pulse. To the best of our knowledge, there is currently no other device that allows real-time, continuous, remote monitoring of pulse in pets. In addition, there is no other device that can monitor pulse without any attachments, preparations or requirement for immobility of the patient. The ability to continuously and remotely capture pulse, among other parameters, such as respiration, temperature, activity and HRV, provides an unprecedented opportunity to assess pets’ health remotely.
The purpose of the present study was to evaluate the accuracy of the pulse determination algorithm used by the PetPace™ collar on pets by comparing the collar data with a standard electrocardiogram and pulse oximetry.
Dogs and cats undergoing anesthesia for surgical or imaging procedures at the Veterinary Specialty and Emergency Center Chavat Daat, Beit Berl, Israel, were included in this study.
All pets that presented to the surgery service during the study time, regardless of age, weight, sex, breed or medical condition, were eligible for inclusion in the study, except pets that had neck lesions that precluded the placement of a collar on their neck. We assumed that pets under anesthesia represent a simple and convenient way of collecting both pulse measurements from the PetPace™ collar and reference data from medical monitoring devices. In addition, pets undergoing anesthetic procedures typically show a wide range of pulse rates, allowing a comprehensive test of the collar’s accuracy.
Body weights were taken and PetPace™ smart collars (https://petpace.com/) were fitted to each pet according to its body weight (size): small, medium or large. Data collection using PetPace™ collars started once pets were sedated and connected to electrocardiographic (ECG) and pulse oximetry monitors. Data were collected continuously throughout the entire procedure and until the animal was disconnected from the medical monitoring devices.
A designated technician was present in the surgery suite and directly observed the procedure and the ECG and pulse-oximetry devices that were both connected to the animal.
Reference data collection
The ECG and pulse oximetry data served as the reference measurements for the PetPace™ collar. The designated technician determined the reference pulse every 1 minute by looking simultaneously at the two devices (ECG and pulse oximetry) and concluded what was the dominant pulse rate value during that one minute period. That value was recorded manually in the PetPace™ system as the reference value to be compared.
If the pulse rate fluctuated significantly during any one minute period and no dominant pulse value could be determined to accurately represent that period, a note would be made in the log. If the ECG and pulse oximetry measurements were not in agreement, the reason for the disagreement was investigated and corrected by the attending anesthesia staff (for example, reapplying the pulse oximetry sensor on the dog’s tongue or re-attaching the ECG clips), and data collection resumed.
PetPace™ pulse collection technology
PetPace™ collars use acoustic sensors to detect pulse waves as they travel through large arteries in the neck. The analog acoustic signal is recorded for two minutes and analyzed, using proprietary algorithms, to determine the number of pulses occurring during those two minutes. It then reports the pulse rate normalized to a one minute unit of measure. For example, if the collar detected 120 pulses during the two minutes recording, it will report a pulse rate of 60 pulses per minute.
PetPace™ collar data filtering
Acoustic sensors, as used by PetPace™, are by nature affected by interfering noises, such as mechanical devices used on the pets during surgery (e.g. surgical drilling or use of hot air blower to maintain animal temperature). Interferences are identified as such by the PetPace™ signal processing and internal quality assurance algorithms.
PetPace™ data with identified noise interferences were confirmed by observation of the surgical procedure and finding direct correlations with specific events, and were consequently excluded from the analysis.
In addition, pets and/or time periods without reliable reference measurements, due to problems with the ECG or pulse oximetry devices, were also excluded from the analysis.
Each PetPace™ pulse measurement comes with an integrated timestamp that is automatically attached to the signal. Each measurement is recorded in the database with its timestamp.
The PetPace™ web app also automatically adds a timestamp to the manually entered reference pulse and it becomes part of the record. Events during the procedures, such as the use of surgical devices or manipulations of the animals, were also logged in the web app along with their timestamps.
The timestamps served as the basis for comparison of the two sets of data (PetPace™ data and reference data). Data points were compared based on having identical timestamps. Other data points, not having a paired data point with the same timestamp, were excluded from the analysis.
All pulse data points, including the PetPace™ readings and the manual reference entries, were downloaded from the PetPace™ web app using the CSV download feature and transferred to an Excel file. The downloaded file also included all manually logged surgical events.
Statistical analysis of the data was performed offline using Excel software. Mean and standard deviation (SD) for PetPace™ pulse measurement and reference measurement were calculated, as well as the differences (% change) between PetPace™ pulse and the matching paired reference pulse. Data was referenced to surgical/procedural time. A Bland Altman graph was produced to compare between PetPace™ and the reference pulse rate measurements.
49 pets in total were recruited to the study. 16 pets (32%) were excluded from the final analysis if at least one data source did not produce readings, such as a malfunction in the ECG monitor, or mechano-electrical disturbances that interfered with the PetPace™ acoustic pulse sensors. Thus, 30 dogs and 3 cats were included in the final analysis, of which 16 animals were males and 17 females with mean age of 5.7±3.7 years (range: 3 months to 14 years) and body weight of 53.1±28.8 lbs. (range: 4 to 114.4 lbs.). 10 pets were fitted with a Small size collar; 11 with a Medium size collar, and 13 with a Large size collar.
For the entire sample of 33 pets, there were a total of 1,551 paired (PetPace™ and reference) pulse rate measurements that were included in the final analysis. The average number of paired pulse rate measurements per pet was 47 (ranging from 4 to 183). Average procedure time was 139±91 min. The correlation between PetPace™ and the reference measurement was 0.680. The overall difference (% change) between PetPace™ and reference pulse rate measurements was 5.7±11.4% ranging between -32% to +26% (figure 1).
Figure 1: Differences for individual pets between PetPace™ collar and reference pulse rate measurements: average difference (% change) between all paired data points (PetPace™ and reference measurements) per animal.
Bland-Altman graph (figure 2) shows that most data points are within +/- 1 and 2 STD with some error measurements above that standard deviation values.
Figure 2: Bland-Altman graph of all pulse rate measurements.
The aim of this study was to measure the accuracy of the PetPace™ collar pulse rate measurement. Collar data from dogs and cats was compared with standard ECG and pulse oximetry methods in a surgery setting. Choosing pets under anesthesia provided several advantages, including a varied pet population, wide range of pulse rates, convenient reference standard for comparison and a relatively stable environment with minimal interferences.
This study demonstrates that the average difference between the PetPace™ pulse measurements and a reference measurements stand on 5.7%. The level of accuracy was not affected by species (dog or cat), age, body weight, sex, breed, or medical condition.
This difference is of no clinical significance. Furthermore, some of the difference in the measurements may have resulted from errors with the reference measurements. However, for the purposes of this study we attributed the entire difference to an error in the PetPace™ device. Therefore, it is reasonable to assume that the true accuracy of the PetPace™ collar is actually somewhat better than the above value.
The study’s limitations include the fact that pets were under anesthesia and having a small sample size for cats. Further work is recommended to assess the accuracy of the PetPace™ pulse rate measurements on non-anesthetized pets under natural conditions and activities.
PetPace™ smart collars are accurate for pulse measurement of pets (dogs and cats) under anesthesia, regardless of characteristics such as size, breed, weight and sex. Further study is required to assess the accuracy on non-anesthetized pets.
|||J. J. S. B. Cain PW, “Portable data acquisition cart for equine transportation stress study. Biomed Sci Instrum.,” Biomedical sciences instrumentation, 1991.|
|||A. K. G. M. G. A. Fletcher RR, “Wearable wireless sensor platform for studying autonomic activity and social behavior in non-human primates,” Conf Proc IEEE Eng Med Biol Soc, 2012.|
|||J. W. M. K. P. T. A. G. A. Landis-Hanna, “Investigation of novel technology to evaluate heart rate and respiratory rate,” J Vet Intern Med, pp. 28:976–1134, 2014 ACVIM Forum (abstract)., 2014.|