Understanding how physicians think – a way to reduce diagnostic error

Analysis of fMRI images shows which neural circuits were activated while subjects were diagnosing conditions and prescribing treatment and identified a mechanism that could lead to premature conclusions.

By Karina Toledo  |  Agência FAPESP – Diagnosing a disease and prescribing a treatment on the basis of written information are tasks that activate the same neural circuits in a doctor’s brain as those used by anyone when naming objects or animals.

Researchers at the University of São Paulo’s Medical School (FM-USP) in Brazil reached this conclusion after studying the brain functioning of 31 primary care physicians. The investigators used functional magnetic resonance imaging (fMRI), a procedure that measures brain activity by detecting changes associated with blood flow in response to neural activity. 

The results of the study, which was supported by FAPESP, were published in May in Scientific Reports, an online journal owned by Springer Nature.

“In this study we also identified mechanisms that might lead to a premature diagnostic conclusion. This kind of information can contribute to the development of tools capable of reducing such errors in medical practice,” said Marcio Melo, a researcher at FM-USP’s Medical Informatics Laboratory and first author of the article.

The participants in the study were tested in two different experiments, Melo explained. In the first, each physician was shown a written set of symptoms and asked to identify the disease associated with them (for example, the symptoms fever, productive cough and pulmonary condensation should lead to a diagnosis of pneumonia). For comparison, they were shown interspersed information about animals or objects and asked to name them (for example, the words “meow,” “domestic animal” and “black fur” suggested a cat).

In the second experiment, the physicians were shown the names of diseases, and the task was to say what treatment would be most suitable for each disease.

The participants were positioned inside the MRI scanner, and the written information was displayed to them via a system of mirrors. Brain images were collected as the physicians performed the tasks, and their answers were recorded.

“Our analysis shows a remarkable similarity of cortical activity in all three tasks – diagnosis, prescription and naming of objects or animals – which corroborates our initial hypothesis,” Melo said.

The findings match those of a previous study published by the group in the journal PLOS ONE, Melo noted. The previous study investigated the diagnostic process in the visual field by having radiologists diagnose lesions in chest X-rays and name animals embedded in the X-ray images. Here, too, the brain areas activated during the diagnostic process were very similar to those activated by the naming of animals.

“It’s important to stress that in this latest study 80.7% of the participants chose more than one diagnosis at least once during the experiment,” Melo said. “For example, in response to the symptom ‘despondent,’ one participant chose ‘depression’ and ‘hypothyroidism.’ This shows that a complex process evoking different diagnoses can occur in a matter of seconds.”

Too much certainty can be bad

Both diagnosing diseases and prescribing treatment are decision-making processes, according to the group at FM-USP. Uncertainty in such decisions is rife to begin with, but a confidence threshold is reached as evidence builds and the physician is able to reach a conclusion.

In these experiments, the decision was considered made when the physician verbalized the diagnosis or treatment.

Melo said the MRI images showed that when the physicians had to think about non-specific diagnostic information that could be associated with more than one disease (e.g., fever), activity increased in the brain system known as the frontoparietal attention network (FPAN). 

However, attention monitoring by the FPAN was reduced if the participant was shown information strongly associated with a particular disease, such as positive HIV test results, at the start of the experiment. 

This supports the hypothesis that a decrease in the degree of uncertainty, signaled by the reduction in FPAN activity, is involved in triggering a decision.

“Our analysis suggests a decision may be made prematurely if the physician initially receives information with high diagnostic power,” Melo said. “If blood work shows low thyroxine, for example, the physician may correctly diagnose hypothyroidism. On the other hand, diagnostic certainty may lead to premature interruption of the investigation and prevent detection of associated depression. A premature diagnostic conclusion is a major and frequent cause of medical errors.”

One way to prevent premature closure of the investigation, he added, would be to  show the physician a list of diagnostic options using a computer-aided support system, which could be coupled to electronic patient records, for example. “This would increase the degree of uncertainty and focus attention on the assessment process,” Melo said.

Another important conclusion in the article is that physicians apparently become conscious of a decision when they begin verbalizing it.

According to Melo, the images revealed “an unexpected and remarkable” change in brain activity between the decision-making period and the onset of response verbalization. “As they began speaking, we detected a sharp increase of activity in a broad network of brain structures involved in consciousness and, concurrently, in areas engaged in auditory monitoring.”

This increase in activity, which takes a fraction of a second, could be detected only thanks to a novel methodology developed by the group at FM-USP to enhance the temporal resolution of fMRI data analysis. In the study, an image comprising 43 brain slices with a width of 3 mm was collected every 2.3 seconds.

“The conventional method is to base the analysis on average brain activity for these 43 slices. Our innovation consisted of introducing activity in each of the 43 slices into the mathematical model used to analyze the data. This enabled us to investigate brain activity in periods of 400 milliseconds,” Melo explained.

The results, he added, suggest that the physicians who took part in the study needed to hear their own responses in order to become aware of their diagnostic conclusions.

“We need to hear ourselves speak, out loud or in our imagination, in order to know what we think,” Melo said. “This hypothesis has been mooted before, of course, but we’ve now presented the first experimental evidence to support it. More experiments will have to be performed to corroborate it by addressing the process more specifically. It could have wider implications. For example, it might help us understand more generally how people become aware of what they’re thinking.”

The article “How doctors diagnose diseases and prescribe treatments: an fMRI study of diagnostic salience” can be read at nature.com/articles/s41598-017-01482-0.