Background:
As classification entails comparing actual information with representations from previous encounters to enable recognition and differentiation, the ability to correctly diagnose atypical disease presentations is partly grounded in prior exposure to case examples.
This suggests the potential benefit of introducing systematic variation in learning examples and varying demands on discrimination between candidate disease categories, as well as a target for individualized learning. Moreover, it raises the question of potential benefits by optimizing the distribution of these varying examples over time. We hold that carefully designed Virtual Patient Simulations provide means for systematic variation of case attributes - and thus discrimination demand, which can be used to improve conditions for classification learning. With this study, we explore its effects on diagnostic performance and learning. We investigate the effect of feature based manipulation of learning examples to increase or decrease the demand of discriminating between competing disease categories in a given learning example. Moreover we distinguish between different instances of repeated exposure of learners to similar examples: discrimination demand where only the target diagnosis is identical to a previously encountered learning example and instances where additionally, the distractor category is repeated.

Method:
We applied previous formalized conceptions of typicality to Virtual Patients to vary discrimination demand between candidate diagnoses, thus creating two versions (typical/atypical) for each of 12 patient cases.
In a mixed 2×2 (time × typicality) experimental design, we will vary the typicality of patient cases on two measures (first set / second set), where each set consisted of a series of six Virtual Patients, yielding a total of four conditions.
We will measure the accuracy of diagnostic performance in 90 medical students receiving a knowledge test and a series of Virtual Patients with the task of diagnosing.

Planned Analysis:
We exclude participants with scores below 50% on the knowledge test and perform analyses of variance on measures of diagnostic accuracy while adjusting for prior knowledge.

Projected Results:
We expect typical cases to yield higher performance than atypical cases. We also expect a hybrid interaction between case typicality and time.

Discussion:
This study investigates the effects of systematically varying features of learning examples for the deliberate design of discrimination demand as a promising target to foster learning of disease classification.

Keywords: Simulation-based learning, medical education, discrimination learning, diagnostic competence.