An artificial intelligence (AI)-based software tool can make radiology clinical decision-support (CDS) software more user-friendly to ordering clinicians, according to research published October 1 in the Journal of the American College of Radiology.
Researchers from the University of Virginia Health System in Charlottesville shared their experience after incorporating a commercially available AI application for their radiology CDS software. This tool enables clinicians to enter the reason for their advanced imaging order via free text and then choose from a list of AI-predicted structured indications.
The AI-based CDS software was not only used more often than the traditional process for ordering radiology exams, but it also significantly improved the rated appropriateness of their orders.
"Our experience with an AI tool found it to successfully predict structured indications for ordering providers based upon their free-text entry, providing potential means to decrease the burden on ordering clinicians while ensuring [Protecting Access to Medicare Act] compliance," wrote the authors led by first author Dr. David Gish.
Their institution had adopted the CareSelect Imaging CDS software (Change Healthcare) in 2014, and then added the firm's AI tool in February 2020. After the tool was added, providers were also -- even after initially using the free-text method -- still able to use the traditional approach of searching directly for structured indications, as well as bypass the CDS process entirely.
To test their hypothesis that the free-text AI approach would be used more frequently and that the method would successfully facilitate selection of structured indications, the researchers reviewed 40,053 advanced imaging orders for the first three months after it was implemented.
In the first three months following adoption of the AI tool, more than half of the orders were completed after using the new approach. What's more, the appropriateness of the entered orders also was significantly higher.
| Improved CDS results from using free-text entry for imaging order indications | ||
| Traditional searching method for indications | AI-based free-text approach | |
| Percentage of overall orders after adoption of AI tool | 58.9% | 41.1% |
| Orders considered "usually appropriate" | 70.2% | 71.7% |
| Orders considered "may be appropriate" | 18.9% | 20.9% |
| Orders considered "usually not appropriate" | 10.9% | 7.4% |
"Providers significantly more often elected the new free-text-AI approach to order entry for CDS, suggesting provider preference over the traditional approach," the authors wrote. "The AI tool commonly predicted indications acceptable to ordering providers."
In other findings, the AI tool yielded alerts with predicted indications in 91.7% of orders that used the free-text method. Also, providers selected one of the AI-predicted indications 57.7% of the time, according to the authors.
When structured indications were predicted and presented to the ordering providers by the free-text-AI tool, providers went on to select a matching structured indication from an additional direct search in only 2.5% of cases.
"Thus, although structured indication sets for CDS were not comprehensive enough to encompass all clinical scenarios, the AI approach provided sufficient access to the available choices," the authors wrote.
