How Valuable Are Clinical Decision Support Mechanisms?

The clinical decision support (CDS) requirement of the Protecting Access to Medicare Act (PAMA) is currently in an Educational and Operations Testing period, requiring physicians to consult appropriate use criteria (AUC) using CDS tools when ordering advanced imaging for Medicare patients. However, there has been no prior rigorous statistical validation that using a CDS improves imaging appropriateness.

In 2021, a multidisciplinary study on CDS in the Journal of Digital Imaging showed that using a CDS mechanism does, in fact, lead to improved appropriateness scores.

“This study introduced a new scientific approach that enabled the team to get beyond anecdotal and small sample size data regarding CDS,” said study senior author Frank J. Rybicki, MD, PhD, vice chair of operations and quality and professor of radiology and biomedical engineering at the University of Cincinnati.

The study evaluated the CT, MRI, ultrasound, and nuclear medicine requisitions from 244,158 providers at 288 U.S. institutions; each of the first 200 requisitions was placed in chronological order and in relation to the provider’s total exposure to the CDS mechanism.

With new software engineered to assess 7,345,437 advanced imaging requisitions, the team could evaluate the last 10 requisitions in comparison to the first 10, on a per provider basis. The number of requisitions classified as “usually appropriate” increased by 3%, while the number of “may be appropriate” and “usually not appropriate” requisitions decreased by 0.8% and 3%, respectively.         

“Using these methods, we can state with very high statistical confidence that greater exposure to a CDS mechanism really does improve appropriateness scores for high-cost medical imaging,” Dr. Rybicki said.

AI Application Matches Free-Text to CDS

As much as CDS mechanisms can help improve appropriateness of advanced imaging, they can also be frustrating to use.

This is largely because providers must choose from a list of pre-set reasons for the exam, noted Cree Gaskin, MD, professor and vice chair of radiology and associate chief medical information officer at UVA Health in Charlottesville, VA.

Dr. Gaskin and colleagues published a study in the Journal of the American College of Radiology about a possible solution to this issue: the use of AI. 

“The provider writes a free-text reason for the exam, and AI is used to help find a match that will work with the decision support system,” Dr. Gaskin explained. “This allows the ordering provider to share his/her own words with the radiologist and allows everyone to receive credit for using decision support.” 

The AI tool was able to yield alerts with the predicted indications in 91.7% of free-text orders. Providers ultimately selected the indications predicted by AI in 57.7% of cases in which the tool offered them.

As described in another CDS-related study, also published by Dr. Gaskin in the Journal of the American College of Radiology, his institution expanded the use of CDS to voluntarily include patients with private insurance, rather than just Medicare patients.

One private insurance company (Aetna) agreed to support expedited prior authorization—even allowing for exams to be scheduled on the same day—when the CDS tool helped document that the level of study appropriateness was good, he explained.

Once the CDS tool was integrated into the electronic health record, the relative frequency of indicated studies went from 64.5% to 82%, Dr. Gaskin said. 

A Role for RI-RADS

According to Ömer Kasalak, MD, PhD, a radiologist at the University Medical Center Groningen in Groningen, Netherlands, another tool that could improve the quality of radiologic imaging requests is the Reason for Exam Imaging Reporting Data System (RI-RADS).

Dr. Kasalak and colleagues published a study in the European Journal of Radiology on RI-RADS that found that, in a random sample of 673 radiologic examinations performed at a tertiary care center, more than 75% were graded as inadequate according to RI-RADS.    

“It can be argued that referring physicians are busy and have a lack of time to adequately fill in request forms,” Dr. Kasalak said. “However, this either means that the interpretation of the radiologic examination will be suboptimal, or that the radiologist has to spend a lot of time delving into the patient files or contacting the referring physician to get the relevant information before interpreting a scan.” Referring physicians should provide more feedback to improve the quality of imaging requests, asserted Dr. Kasalak.

“An AI system that automatically checks the quality of the imaging request may be a solution, but this also requires willingness of referring physicians to embrace such a system,” he said.

For More Information 

Read previous RSNA News stories on clinical decision support:

• Radiology: Are You Ready for Clinical Decision Support Law in 2020?
• Radiologists Can Lead Shift to Value-Based Care

Access the Journal of Digital Imaging study at springer.com/journal/10278.

Access the Journal of the American College of Radiology studies at jacr.org.

Access the European Journal of Radiology study at ejradiology.com.