Trigger Algorithm Streamlines Lung Imaging Follow-Up Review

Researchers led by physicians from the Michael E. DeBakey Veterans Affairs Medical Center and Baylor College of Medicine in Houston are perfecting an electronic health record-based trigger algorithm that can flag abnormal lung imaging results and signal which patients may not be receiving timely follow-up care.


Researchers led by physicians from the Michael E. DeBakey Veterans Affairs Medical Center and Baylor College of Medicine in Houston are perfecting an electronic health record-based trigger algorithm that can flag abnormal lung imaging results and signal which patients may not be receiving timely follow-up care.

While the algorithm they employed demonstrated a productive prediction rate of slightly above 50 percent in a retrospective study, they maintain such numbers are a significant advance over the current industry best practice.

"Most institutions don't have rigorous tracking mechanisms in place to find out how many of their patients are getting lost in follow-up after abnormal imaging," said Hardeep Singh, M.D., associate professor of medicine at Baylor and staff physician at the DeBakey center. "So even if we have a 50-plus percent predictive value, that's actually pretty good, because the computer has already scanned thousands of records to tell you which ones to look at."

Singh and colleagues Daniel Murphy, M.D., Eric Thomas, M.D., and Ashley Meyer, M.D., devised the trigger algorithm and published the results of their research in Radiology. They are continuing their work, testing whether the algorithm can work in real time.

"We are not going to be able to catch all the needles in the haystack," Singh said. "We're trying to do two things – we're trying to make the haystack smaller and we're using a magnet to attract our needles. We're going to miss some needles, and we will miss some people who truly have an abnormality that hasn't been followed up. But that's still better than zero."

Also See: HIE Cuts Redundant Imaging by 25 Percent

The results of the retrospective study illustrate both the promise of EHR-enabled analytics and also how far the industry has to go to achieve optimum performance goals, especially in specialties such as radiology, in which so much critical data resides in unstructured formats.

In terms of promise, the authors employed the computerized trigger on 24,829 patients at a large Veterans Affairs facility who had at least one chest radiograph or CT image obtained during the one-year study period. Of those, 538 had findings that radiologists flagged as being suspicious. The trigger reduced the number of records needed to review to confirm follow-up delays to 131, less than one-quarter of those with suspicious findings. Of those 131, 75 patients, or 57.3 percent, were found to have actually suffered a delay in follow-up. Four of the 75 were diagnosed with primary lung cancer during the following two years.

Murphy, the study's lead author, said results can improve further as natural language processing capabilities improve. However, the study was also aided by an existing feature in the VA's Vista EHR that enhances data review.

That feature is a structured “suspicious for malignancy” code to identify an imaging result that is suggestive of cancer. A similar system for rating mammograms (the Breast Imaging Reporting and Data System, or BI-RADS) is nearly universal throughout the United States, but the use of other structured ratings systems for imaging results is uncommon, according to the authors.

While Singh said the American College of Radiology is developing a tool similar to BI-RADS for lung imaging called Lung-RADS, feedback he has heard from colleagues in the field is that the coding infrastructure is simply too immature at this point to hint at quick progress.

However, both said as the industry moves to value-based payments, the relative simplicity of configuring algorithms like theirs could convince C-suite executives an investment in such technology is worthwhile.

"Once an organization's leadership has had to deal with a malpractice claim or done a root cause analysis, their awareness on these issues goes up quite a bit, and they are looking for solutions," Singh said. "I think right now they think, 'We're just going to have to make the providers more savvy on follow-up and more vigilant' and so on. But what we're saying here is the system needs to step up as well. An organization or healthcare institution will have to have these backup systems in place so when patients fall through the cracks we can catch them in a safety net."

The radiology algorithm is the latest such research Singh and Murphy have done; they previously did similar research on flags for prostate and colorectal cancer and want to expand their work into other organizations to prove their concepts' value.

"I think there is opportunity for translation here," Singh said. "Not just for the integrated systems, but as we get more integrated with ACOs, and people are measured more and more on quality – well, this is quality to us, and if you can't use your EHR to identify issues such as failure to follow up, then that's not progress. We need to make progress, we need to be innovating in this area. We have digital data.  We need to put it to good use."

More for you

Loading data for hdm_tax_topic #better-outcomes...