Panel Paper: Clinical Decision Support for Radiology Imaging

Thursday, November 7, 2019
I.M Pei Tower: Majestic Level, Savoy (Sheraton Denver Downtown)

*Names in bold indicate Presenter

Laura Feeney1,2, Joseph Doyle1, Sarah Reimer3, Sarah Abraham1 and Amy Finkelstein1, (1)Massachusetts Institute of Technology, (2)J-PAL North America, (3)Aurora Health Care

There is widespread concern over the health risks and costs from inappropriate use of medical imaging; prior research suggests that up to 30 percent of diagnostic imaging is unnecessary. This statistic is particularly concerning given the high cost of some imaging orders, such as MRIs and CT scans—a substantial share of these orders may be adding to health care costs while providing minimum value to patients.

In response, clinical decision support (CDS) systems have been designed so that if a health care provider orders a diagnostic scan that is inconsistent with current professional guidelines, they will receive a pop-up notification asking whether they would like to proceed or change the order. Reflecting concerns about inappropriate scanning, beginning January 1, 2020, Medicare will no longer reimburse providers for high-cost scans unless they are ordered using a qualifying clinical decision support system.

We conducted a randomized evaluation to study the impact of a clinical decision support system on the number of certain high-cost medical imaging orders. Working with Aurora Health Care, a large, private, not-for-profit health care system in eastern Wisconsin and northern Illinois, approximately 3,500 health care providers with imaging order permissions were selected to participate in the study. We randomly assigned half of these providers to receive the CDS intervention, with the other half serving as a control group.

The intervention ran for 12 months, from December 2016 to December 2017. We measured scan ordering through administrative data recorded by Aurora’s electronic medical records system and by the National Decision Support Company, the provider of the clinical decision support system.

We found that clinical decision support reduced the number of high-cost scans targeted by the software but did not change the total number of high-cost scans ordered. Results suggest that Medicare’s mandate requiring health care systems to adopt clinical decision support could modestly improve the appropriateness of high-cost imaging orders. Yet contrary to prior observational studies that found large reductions in targeted images, the results of this large-scale randomized evaluation suggest the vast majority of high-cost scans were not eliminated.

The intervention was designed to study whether a software intervention could improve the appropriateness of high-cost-image ordering—an intervention that is easily scalable. These findings underscore the importance of rigorously testing new interventions. An important next step is to evaluate which design features of clinical decision support, such as the information provided at the time of ordering and complementary actions including supervisor reviews, can increase its effectiveness.