Stanford University Home

Stanford News Archive

Stanford Report, December 10, 2003

Radiology research tackles squishiness of wordiness When radiologists share their findings with clinicians, the meaning may be fuzzy

By MITZI BAKER

While communication difficulties may keep psychologists in business, it can be a big problem when the communication in question is a critical piece of specialized medical information that needs to be conveyed to a patient’s physician.

The field of radiology in particular is heavily dependent on the ability of the highly trained radiologists to relay their interpretations of tests such as X-rays or CAT scans to the clinicians requesting the tests for their patients.

One Stanford radiologist, Anthony Yun, MD, wondered whether the transfer of information from radiologist to clinician might be a source of confusion. "A lot of research and teaching is focused on acquiring and interpreting medical images, yet scant academic attention is paid to communicating the interpretation," said Yun. So he set out to see if he could actually quantitatively measure the differences in values that doctors ascribe to words used in radiology reports.

Yun collaborated with Patrick Lee, MD, a radiology resident; Dan Chao, a Stanford medical student; and Pamela Schraedley-Desmond, a software developer in the Department of Psychology.

"The question we asked is whether commonly used radiologic terms have ambiguous meanings whose values vary among doctors," said Yun, who presented his findings last week at the Radiological Society of North America meeting in Chicago. "Most radiologists have a vague intuition that sometimes what the clinician heard was not what we said."

In scouring the literature from linguistics, computational science and even game theory, Yun came up with a plan to answer his questions. First, by combing through randomly selected radiology reports generated by the radiology department inside of a month, the team determined the 11 most commonly used descriptive words or phrases that signify a degree of probability.

They found that the most frequent terms used were probable, consistent with, consider, likely, suggestive, no definite evidence, suspicious, cannot exclude, not likely, maybe and possible. Using these terms, they created a straightforward survey in which they presented 20 radiologists and 20 internists with a scenario, such as a chest X-ray with the diagnosis in question being lung cancer.

Based on the phony report, the doctors were to supply numerical probabilities for how they interpreted the wording. For example, did probable lung cancer mean that 60 percent of patients would actually have lung cancer or 90 percent? As Yun pointed out, the interpretation of probability may be critical to the physician’s decision on whether to send the patient for a biopsy.

Using the survey and a complex statistical analysis of the responses, Yun’s team found that there was statistically significant variability in how the words were interpreted, not only between the radiologists and the clinicians but also within each of the two groups.

The doctors also had discrepancies in the probability they assigned to a word depending on the diagnosis.

Probability values ascribed to words by doctors were often influenced by the diagnosis in question. For example, radiologists tended to use more aggressive words of probability when the diagnosis in question was cancer as opposed to congestive heart failure.

There were also some words or phrases that turned out to be quite variable, such as consider, and others that were interpreted pretty much the same way, such as no definite evidence.

Identifying the problem is only half the battle, noted Yun. He said he doesn’t foresee the field converting to a numerical probability system any time in the near future for many reasons, not the least of which is the difficulty of obtaining pathologic follow-up.

Until there is an efficient feedback loop back to the radiologist from a pathologist either confirming or disproving the radiologists’ assessment of an image, they don’t find out what percentage of the time their diagnosis was correct.

Now that Yun has provided the first quantitative study examining communication difficulties surrounding the radiology report, he plans to study it in more depth. "The next step to pursue is what factors contribute to that variability, how does it evolve over time and what does it tell us about the nature of language in a medical setting," he said.



Lucas Center digs deep to meet research goals (12/3/03)