Artificial Intelligence Mines EHR Data to Improve Diagnoses

An artificial intelligence algorithm is able to mine EHR data and suggest the best diagnostic approaches, leading to enhanced diagnoses and treatments, according to a study published in the Journal of Biomedical Informatics.

Although AI performs very well when trained on years of human data in specific areas, the technology hasn’t been able to manage the huge number of diagnostic tests and disorders of modern clinical practice, researchers noted.

A team from the USC Viterbi School of Engineering worked to develop an AI algorithm that can learn and think like a doctor, but with essentially infinite experience. The new form of AI mines EHRs in databases to recommend optimal treatment strategies. The algorithm works just like a doctor, the team said.

"The algorithm thinks about what to do next at each stage of the medical work-up," said Gerald Loeb, a professor of biomedical engineering, pharmacy and neurology at USC Viterbi School of Engineering and a trained physician. "The difference is that it has the benefit of all the experiences in the collective healthcare records."

Researchers at the USC Viterbi School of Engineering have spent years applying AI to haptics and building robots to sense and identify objects. In previous research, the team developed a model that was able to 117 objects with 95 percent accuracy – a significant achievement, as the state of AI for haptics was to identify about ten objects with about 80 percent accuracy.

When the team extended the model to 500 objects and 15 different possible tests, the algorithm got even faster and more accurate. Researchers then began thinking about adapting the model for clinical practice.

Conventional AI algorithms typically use a specific algorithm called Bayesian Inference to suggest to physicians the most likely diagnoses given a set of observations. This algorithm uses whatever information is currently available to suggest which diagnoses are the most likely.

The algorithm developed by USC researchers reverses this process. Instead, the new AI model seeks those tests that would most likely identify the correct illness or condition, no matter how obscure. The algorithm can also take into account the costs and delays associated with a range of diagnostic tests.

The team pointed out that the new algorithm could help providers make better diagnostic and testing decisions by suggesting several good options, including some that clinicians may not have considered otherwise. Additionally, the algorithm would automatically update and improve as physicians input more data into EHRs.

The algorithm could also help providers generate and complete accurate medical records more easily. Instead of hunting for codes or working their way through drop-down menus, clinicians could simply select a particular illness or diagnostic procedure suggested by the algorithm, which would automatically input the correct information into the EHR.

Researchers emphasized that providers do have the option to override the AI and go with their own judgment, as the technology is meant to serve as clinical decision support.

“The algorithm isn't meant to make decisions for doctors or replace them," Loeb said. "It's meant to complement and support them."

While researchers believe that the algorithm has the potential to transform medical and testing diagnostics, the group also acknowledges that there are cost and technological challenges associated with applying AI to EHRs. The industry would need the expertise and financial resources to develop and deploy the massive database and user interfaces for widespread adoption of the algorithm.

"If the promise of success is great enough, then people are going to be motivated to do it," Loeb said. "And that's what we think this algorithm provides: the possibility, the promise of offering a solution to a huge problem that wastes a lot of resources, trillions of dollars' worth."