Decreasing Uncertainty in Radiology Reports: Clairvoyance (or Artificial Intelligence) Required
Radiologists shouldn't need a crystal ball to reduce uncertainty in their reports.
If your answer to the question "Should radiologists strive to decrease uncertainty in radiology reports?" is yes, then the next question to ask yourself is "Can uncertainty in radiology reports be decreased?" By reviewing the medical record, talking to the patient, and calling providers, you may get accurate, high-quality information to help you decrease uncertainty. However, if every imaging study required that much sleuthing, very little work would get done.
There are times when no input of information is needed to provide reporting certainty. For example, when you see an obvious tension pneumothorax, you immediately pick up the phone to convey this critical finding without first pondering why the patient has the condition. There are times when a small but important amount of information determines imaging study eligibility (such as age and smoking pack-years for lung cancer screening CT) and when there are standardized reporting and management recommendations that reduce uncertainty (like Lung-RADS™).
But too often, the information provided is either "minimalist", e.g. indication is "pain" without additional information like the location of pain and if related to an injury or not, or "maximalist", like this actual indication for a chest radiograph I saw this week:
They lost me somewhere between "DM" and "COUGH." Although the provider was very conscientious about providing clinical context, it took longer to read this than it took me to interpret the chest radiograph. I felt quite certain in my interpretation for this radiograph but at other times had a degree of uncertainty I felt likely could have been reduced with more or better context provided.
Reducing uncertainty in reports has perhaps been most challenging for asymptomatic incidental findings, although a growing collection of published white papers from the ACR Incidental Findings Committee is helping radiologists increase reporting uniformity and management consistency.
How many times have you wished you could use ESP at the time of image interpretation to know the relevant clinical information, the specific concerns of the patient and referring provider, and other important data? If radiologists can't compel anyone to provide this information, but are simultaneously asked to reduce uncertainty, we can either develop clairvoyance or look to innovative solutions. But first we have to better understand the problem.
Fortunately, in a recent JACR article, Bruce Reiner, MD, has done it for us. He lists sources of uncertainty in radiology, including the sixteen steps of the medical imaging cycle, from study ordering to results communication, as well as the people and technologies involved.
When you see all these input variables (many of which are beyond direct radiologist control) and you add to that ten factors impacting reporting variation among radiologists, the size and complexity of the challenge of reducing reporting uncertainty may appear rather daunting.
Dr. Reiner mentions that data-mining technologies may help identify uncertainty language in reporting, but cannot, at this time, contextualize why the uncertainty exists. Perhaps in the future, artificial intelligence will mine the electronic medical record for quantitative data, text (including from past imaging reports), provider ordering habits, patient factors (e.g. body habitus), etc. to calculate a patient- and context-specific pre-test uncertainty before an imaging study is even done. Then, after the imaging occurs, the technology could recalculate the post-test uncertainty, whether higher or lower than the pre-test uncertainty.
Until that happens, I'm going to keep taking clairvoyance lessons with Madame Zelda. Right now, I'm picturing a short journey to a familiar place where everyone instantly recognizes me and, for some reason, asks "What's for dinner?" Yes, it's time to leave work and go home.