Predict my codes

Deep learning algorithm from Stanford University may help veterinarians code patient records
Published on
information-circle This article is more than 3 years old

Veterinarian using a hand-held deviceRecent work out of Stanford University that uses a deep learning algorithm to predict diagnostic codes from veterinary appointment notes may be the key to creating usable data for an industry that lacks standardized charting.

The system, DeepTag, still requires more development and may have a hard time infiltrating the veterinary community, but those familiar with veterinary medical informatics and the project researchers are hopeful about its possible applications.

"In veterinary medicine, there is a lot of unstructured information. There is no software to turn that into structured information. DeepTag fills in that gap," said James Zou, PhD, DeepTag supervisor and Stanford University assistant professor of biomedical data science. DeepTag turns information from notes into diagnostic codes to allow for research and disease monitoring.

History lesson

The veterinary industry has worked to standardize medical terminology and charting by adapting the Systematized Nomenclature of Medicine Clinical Terms, which was developed for human medicine, and applying it to veterinary medicine in the form of the Veterinary Extension to SNOMED CT, maintained by the Veterinary Terminology Services Laboratory at Virginia-Maryland College of Veterinary Medicine.

"Veterinary involvement in SNOMED has a 50-plus–year history," said Dr. Jeff Wilcke, a professor of veterinary medical informatics at Virginia Tech. "The AVMA endorsed SNOMED for veterinary records systems over 20 years ago."

However, despite the desire for standardization by industry leaders, SNOMED is not universally used.

"SNOMED CT is widely used in human medicine, but not in veterinary medicine. It is the recommended standard by the AVMA—and adoption has expanded some over the past few years—but is not widely used in veterinary practices," said Wayde Shipman, president of the association that administers the Veterinary Medical Databases and a veterinary medical information specialist at Virginia Tech.

Finding a way in

Electronic medical records and standardized charting can improve the quality of records and allow for the analysis of veterinary patient records, but the process can be time-consuming.

"Electronic records and doing data analysis is not going to make you more money; it is more about being a better veterinarian," Dr. Shipman said. "If practices could collect their data in a more standard format, then we could analyze information easier and detect diseases in a more widespread way."

A system such as DeepTag would be able to automatically code notes for veterinarians in a simple way, but it requires data to learn from, and the veterinary industry has very few coded records available.

"There is very little training data," said Dr. Ashley Zehnder, a veterinarian and DeepTag researcher. "Human medical records have an army of medical coders to at least put codes on them, but we don't have that infrastructure in veterinary medicine."

DeepTag was initially trained on a large data set from the Colorado State University College of Veterinary Medicine & Biomedical Sciences.

However, one of the challenges with using an academic caseload to train the system is that the notes are specialized, and private practices don't necessarily have the same coding, Dr. Zehnder said.

DeepTag researchers will try to close the gap between the two so DeepTag can be applied to private practice. However, the gap is not well solved even in human medicine.

That gap was the most surprising discovery for Allen Nie, DeepTag researcher. He asked: "How do we take something that was groomed from the academic side and apply it to the actual industry? There is a gap between the CSU data and private practice data. You can't take an algorithm and train it on a dataset and expect it to work really well everywhere else."

Mr. Nie and Dr. Zehnder
Allen Nie, DeepTag researcher, and Dr. Ashley Zehnder, DeepTag researcher, companion exotics veterinarian, and co-founder and CEO at FaunaBio. (Courtesy of Dr. Zehnder)

The Banfield way

Banfield Pet Hospital uses its own centralized software for pet medical records, PetWare, which its more than 1,000 hospitals use to collect and record data for each pet.

Banfield began developing its proprietary practice management and electronic veterinary health record system to meet its needs in the 1980s. The company reports that it has the largest electronic veterinary medical database.

"Banfield's Veterinary Science team gathers and analyzes medical data year-round to monitor trends in pet health and disease. After analyzing the data, we publish our findings in professional journals and reports to help educate pet owners and advance the veterinary profession," said a Banfield spokesperson.

The company sees the data gathering as an opportunity to evolve the care it delivers and to advance knowledge in the veterinary profession, according to the company.

Looking ahead

Whether private practices begin to use standard terminology for medical records or something like DeepTag to code patient notes is still up in the air.

In the next five years, unless there is some sort of push, the situation will probably be about the same, Dr. Shipman said. There has to be some organized pressure to improve the quality of records through a grassroots movement or a push from a professional organization, he said.

There are examples of how pressure from a third-party organization can effectively influence the greater industry, Dr. Zehnder said. In the United Kingdom, the University of London Royal Veterinary College has developed standards that allow veterinarians to report patient outcomes and uses a system called VetCompass to then analyze the data.

The AVMA is also working toward standardization goals in some areas. The AVMA's Veterinary Economic Strategy Committee, in partnership with the American Animal Hospital Association and Veterinary Management Groups, offers a free tool to standardize finances across the veterinary profession called Chart of Accounts.

The AVMA, VetPartners, and the Veterinary Hospital Managers Association have endorsed the tool as the industry standard for classifying revenue, expense, and balance sheet accounts.

Related JAVMA content:

Veterinary Medical Database adopts SNOMED (April 15, 2002)