Enhancing clinical reasoning using probabilistically generated VP cases

By Supriya Krishnan and Sian Claire Owen

During a presentation at the 2nd International Conference on Virtual Patients and MedBiquitous Annual Meeting, Dr Jeroen Donkers, from the University of Maastricht, The Netherlands described his research looking at using artificial intelligence principles and probability to generate virtual patients.

Here, he described the link between clinical reasoning and uncertainty. As Jeroen explained to eViP: “The problem with doctors is that when they do diagnostic reasoning they have to deal with a lot of uncertainties. They see the symptom, but what is the probability that the diagnosis belongs to that symptom? It’s an important ability that students have to develop.”

“For instance if you see a lot of patients who have a headache caused by drinking too much, the next time you see a patient with a headache you’ll think that he drinks too much. On the other hand, if you only see patients with headaches because of brain tumours, the next time you see one [patient with headache] you immediately think of brain tumour.”

Often, the doctor will be required to make quick decisions and use subjective probabilities that are based on the following:

  • Unconscious reasoning;
  • Reasoning based on experience;
  • Reasoning that is biased by many factors, including exposure to a wide range of patients with a given disease state; and
  • Reasoning only partly supported by scientific and epidemiological data.

Errors in subjective probabilities may lead to wrong decisions in medicine. “Problem based learning and evidence-based medicine are important for improving clinical reasoning,” says Donkers. Furthermore, he adds: “There is evidence to suggest that extensive case descriptions can bias subjective probabilities.”

Donkers and colleagues encoded information about diabetes mellitus, demographical and epidemiological data into a probabilistic network – the Bayesian Belief Network (BBN) – to generate a set of around 100 virtual patients. These short scenarios were presented to students in order to help them estimate key probabilities and achieve a better range of subjective probabilities.

“Subjective probabilities play a major role in clinical reasoning, but they can be easily biased,” he said. “VP populations created using the Bayesian Belief Networks could lead to improved subjective probabilities.”

You can listen to Jeroen talk about his research here:

Comments are closed.