Abstract:

In the field of medicine, deriving a diagnosis from gathered facts potentially incurs considerable costs when going awry. Diagnosis involves both conscious deliberation as well as rapid and automatic processes. Mentally organizing information about a given medical case relies on experience to facilitate swift and automatic processes. According to both exemplar and prototype theory, classification, as an inherently representational process, entails comparing actual information with representations from previous encounters to enable recognition and differentiation.

We propose introducing systematic variations in example-based learning to facilitate the development of representations that allow for both generalization and specificity, promoting a more robust classification in the face of:
a) contextual variations,
b) distortion of prototypicality or frequent presentation, and
c) cases involving high proximity between competing categories.

Case-based learning in medical education can be effectively realized by employing virtual patients, which offer opportunities for highly standardized and systematically varied examples. We suggest using virtual patients to deliberately manipulate learning examples’ case features in a classification task: context features with no or low predictive value (context generalization), clinically relevant features with high predictive value for a given diagnosis (robustness to distortion), and systematically introducing distractor features to increase proximity to competing categories for diagnosis (discrimination learning). We emphasize the importance of this variation in promoting the development of knowledge structures that enable increasingly robust diagnostic categorization and identify potential for implementation.

Keywords:

Example based learning, virtual patient.