Joschka Wohlgemuth, head of data science and signal processing, and the entire R&D team at Piavita worked hard gathering, testing, and analyzing data to ultimately publish five validation studies around the Piavet System’s ECG, body temperature, and activity measurements, as well as a measurement protocol experiment for respiration rate.
Doing validation studies is obviously important, but gathering enough good data is challenging. These studies were an initial step toward further improving the Piavet System’s vital sign and related health parameter measurements. However, it’s important to note that these studies were done with relatively small datasets. So, while the findings were very promising, they are not fully representative.
We are proud to be working closely with our research partners and customers to continue gathering vital sign data (especially that which is harder to acquire, such as hypo- and hyperthermia data), which will allow us to further validate and continuously improve the Piavet System.
INTERVIEW WITH JOSCHKA WOHLGEMUTH
WHAT WAS THE CATALYST FOR DOING THESE VALIDATION STUDIES?
“A lot of our customers work in larger clinics or research facilities. In most places like this, it is standard procedure that the validity of a used medical measurement device needs to be ensured. Positive personal experiences with the system are in many cases not enough to convince veterinarians to purchase the system for their team.
The second motivation to do the studies was internal. We’ve done a lot of work on the system and made a lot of progress over the last months. It was time for us to take a step back and assess where we stand with regards to the quality of our vital sign measurements and predictions. We wanted to know what already works well and identify the areas that need additional attention.”
HOW WAS THE OVERALL PROCESS DOING VALIDATION STUDIES FOR THE PIAVET SYSTEM?
WHAT WENT WELL?
“Performing validation studies like this in a short time frame and with a limited budget requires an extraordinary team. The Piavita team has proven to be driven, agile and focused. The collaboration between staff veterinarians, external veterinarians, biomedical experts, data scientists, signal processing engineers and content editors went very well. Within eight weeks we completed the data collection, data analysis and documentation of five validation studies.”
WHAT CHALLENGES DID YOU FACE?
“Everyone in the biomedical field knows that the collection of useful clinical data is the most challenging, costly and time consuming—yet the most important—part of the job. We owe the fact that we were able to do it within our ambitious time frame to our wonderful external partners who were willing to share their time and resources with us in the service of science.”
WHAT WERE THE RESULTS? (I.E. HOW DOES PIAVET COMPARE?)
ECG
“The precision of the ECG R wave detection was quite good (99.82%*). The P-QRS-T visibility and the RR interval calculation was on par with a well-known competitor. However, we identified a few minor problems that we will work on in the next weeks.”
TEMPERATURE
“The new machine learning model predicted rectal temperature with a mean absolute error of 0.21 °C* for normothermia , 0.51 °C* for hyperthermia and 0.79 °C* for hypothermia. To further improve the hyper- and hypothermia prediction we will seek out opportunities to gather more data to improve the model. Horses that show an abnormal body temperature are usually immediately medicated which minimises the window of time in which we can obtain the valuable measurement data.”
ACTIVITY
“Our new activity feature, lying down laterally, performed with a precision of 98%*. This is a good start. We are currently working on extending the model to predict lying down sternally as well as rolling. All studies relied on relatively low numbers of test horses which makes more studies necessary to achieve decent statistical significance. The trends however are very promising.”
RESPIRATION RATE
“The mean absolute difference between manually counted respiratory rates and the Piavet System’s respiratory rates was 1.3* breaths per minute. This seems pretty good given that the mean absolute difference between three veterinarians who counted respiration rates manually was 1.6* breaths per minute.”
WHAT LESSONS DID YOU LEARN ALONG THE WAY?
“The frequent direct contact to horses, horse owners and veterinarians is essential for us to successfully develop a product that adds value for our customers. This can’t be done solely from the desk. Every once in while the nerds need to put their rubber boots on and get in touch with our four-legged clientele.”
HOW DO YOU THINK THESE VALIDATION STUDIES WILL IMPACT PIAVITA’S FUTURE?
“I believe that our customers and those who might be interested will be thankful for the information that we generated and encourage us to continue down this path. Continuous improvement needs to be continuously quantified. The frequent internal validation will become a growing part of our development processes. In addition, we hope to find external partners who are willing to put Piavet to the test and publicise the results.”
*99.82% => based on 10 min. window analyses with consistent electrical attachment extracted from 10 ECG recordings (9 healthy horses) ; 0.21 °C => 2 measurements (1 horse) ; 0.51˚C => 13 measurements (13 horses) ; 0.79˚C => 1 measurement ; 98% => Based on 3 independent measurements (3 horses) ; 1.3 => based on 1 measurement of 5 min on one horse
At Piavita, we like happy vets and healthy horses. It’s that simple. Our mission it to transform veterinary care by providing remote health monitoring technology to the veterinary industry. With a non-invasive, sensor-enabled hardware device and sophisticated software platform, the Piavet Solution automates and digitizes repetitive, manual tasks to help vets save time and improve patient outcomes. We are a diverse group of engineers, developers, researchers, and horse people with a passion for delivering meaningful solutions to veterinarians. We operate out of offices in Zürich, Berlin, and North Carolina. Have a question or suggestion? We’d love to hear from you.
