Home Medical field NUMBERS: Don’t you see AI in emergency medicine? Here it is: Emergency Medicine News

NUMBERS: Don’t you see AI in emergency medicine? Here it is: Emergency Medicine News


artificial intelligence, synthetic data


The use cases for synthetic data are not just theoretical; real people are making huge strides in the field. I was lucky enough to be able to speak with two such people, each of whom uses different types of solutions to tackle different problems.

Synthea is a computerized engine operated by MITER, a non-profit corporation that operates federally funded research and development centers to generate realistic, industry-standard artificial health records. Jay Walonoski, one of the founders of Synthea, came up with the idea for Synthea when he brought his daughter in for medical attention and the team caring for her could not easily access her medical information.

“Getting health care data is impossible,” he said. “You have to go through all this bureaucracy, and there are all sorts of protections in place for a good reason. But if a patient comes to your emergency room, how do you get their medical records? They may not come to your hospital regularly. … You don’t know what allergies they have or if they take a certain medication. Healthcare interoperability is a huge issue. We are developing standards and software to solve this problem.

Synthea uses a comprehensive set of modules that describe the disease and recovery cycle and simulate disease processes to generate completely new patient records so that these synthetic records can be used to solve interoperability issues. Recordings made with Synthea look real but can still be used without fear of compromise. Developing a process for sharing sensitive information with outside sources, other hospitals, for example, is difficult, but Synthea enables IT teams to create and test these processes in a safe and convenient way.

“There are other methods of generating synthetic data where you bring your own data to the table and they will do a statistical clone of your data; that’s not what Synthea does,” Walonoski said.

Synthea has the advantage of being open source, which means that the engine is free and transparent, in addition to avoiding any possibility of re-identification of the patient. The modules are also relatively easy to understand and can be modified by the general community.

“Any time someone says we’re going to improve your COVID model or make your cancer model work better or we’re going to add a new disease, we can collaborate and improve Synthea that way,” he said. .

Knowing that such a resource exists can help emergency physicians advocate for interoperability. Mr. Walonoski also suggested exploring the open-source community or trying to develop new emergency conditions using the module builder at https://bit.ly/ModuleBuilder.

Maintaining confidentiality

Gretel.ai approaches the problem from a different angle; it’s an engine that takes real data, uses machine learning technology to create a general picture of the data, and then uses that picture to produce more data that’s completely new but indistinguishable from the original.

“What we focus on is high-quality synthetic data,” said Ali Golshan, CEO and Founder of Gretel.ai. “And that means synthetic data is so statistically similar to your original data that you can make production-grade AI predictions on top of that. You can get the same business outcome as if you were using the original raw data. However, synthetic data allows us to be able to put this data almost in a safe harbor, so that these teams can open it, collaborate and share it with other environments and institutions around them.

Golshan said the company also solves the problem of an underrepresented dataset when institutions simply don’t have enough data. “The University of California at Irvine was doing a large study on a rare heart disease, and they had an overwhelming amount of information and data on male patients, but not on female patients. So we used a feature we call auto-complete for data, where you can actually say, “It’s the data that matters to me, but I don’t have enough of it.” So we learn about that, and then we strengthen that underrepresented dataset.

Gretel.ai uses a concept called “differential privacy,” a constraint on patterns that prevents unique cases from being re-identified to ensure the safety of unique patients in medical datasets.

“We built advanced privacy filters that use differential privacy to inject noise into AI model training,” he said. “We use a deep neural network language model, similar to OpenAI’s GPT-3. As the model learns, with a single button, we allow anyone to inject differential privacy and to create things like outlier filters. [There is a] balance between total confidentiality and 100% data accuracy. This is actually why synthetic data cannot and should not be more than 94-95% similar to your original data. Otherwise, you could essentially derive all the original data from the synthetics.

Golshan said he created a simple one-button model to determine whether the researcher wants more privacy or more quality. “More quality may be needed for your internal use cases where you want much higher efficiency,” he said. “More privacy may be needed to collaborate and share your data openly.”

One of the strengths of the Gretel.ai engine is that it can be used for free on any website. The engine can also be run in a container which allows to take advantage of synthetic data generation without downloading sensitive data to Gretel’s own computers. Try it by going to https://gretel.ai/ and clicking on “Start for free”.

Use cases for synthetic data continue to explode. I encourage emergency physicians to explore these products and think about how they can help our patients by improving data sharing and learning.

Dr. Bélanger has created a web application that uses procedurally generated pediatric patients with infectious respiratory symptoms to see if it would be possible to use synthetic medical data to generate real insights. Read this article, “The patient in the computer”, onhttps://bit.ly/3DzxZIc.

Dr. Belangeris secretary of the Locum Tenens chapter of the American College of Emergency Physicians and an emergency physician in McKinney, TX. Read his past articles onhttp://bit.ly/EMN-numbERs.