While radiologists often make recommendations for further imaging, clinical tests, or clinical assessment for abnormal findings on an imaging study, a significant proportion of these patients do not get appropriate follow-up, according to presenter Dr. Martin Gunn.
The group used an automated natural language processing technique to detect conditional statements (i.e., hedges) in recommendations for follow-up in all radiology reports, except for mammography, at their multihospital radiology department between 2010 and 2014.
The use of nonconditional follow-up recommendations strongly and significantly correlated with the probability of follow-up, Gunn and colleagues found. There was 78% follow-up compliance when nonconditional recommendation sentences were used, compared with only 44% when conditional recommendation sentences were used.
Gunn cautioned that correlation is not the same as causation, however.
"Although the use of nonconditional language by radiologists may have compelled some referrers to order follow-up imaging, we believe that this is only likely to have a small impact," he told AuntMinnie.com. "It is most likely that radiologists use conditional clauses in their follow-up recommendation sentences when they believe it is less clinically important that follow-up occurs. This helps us fine-tune our algorithm; we can weight nonconditional follow-up recommendation sentences more strongly than conditional ones."
You can follow-up on this topic by attending this Tuesday presentation.
A little over an hour later, University of Washington and Philips researchers will also present a related poster (IN237-SD-TUB5) that shows significant variation in radiologist recommendation rates between patient settings, subspecialties, and imaging modalities. Find out what else they learned by visiting Lakeside Learning Center, Station 5, from 12:45 p.m. to 1:15 p.m.
