Using Learning Curve mathematical models to describe the learning of radiograph interpretation
May 19, 2015 8:00 AM – 9:30 AM
Kathy Boutis1, Phd, Martin Pecaric2, MD, Martin Pusic3, PhD, Oleksandr Savenkov4, PhD
1Institute for Innovations in Medical Education, 2University of Toronto, 3Contrail Consulting Services, 4NYU School of Medicine
Background: Deliberate practice reliably increases performance in a wide variety of domains. In explicitly modeling the longitudinal growth nature of learning during well-designed deliberate practice, we hope to demonstrate the feasibility of predicting individual learning trajectories to the bene?t of individuals and groups of learners. In this work, we describe an approach to the statistical modeling of a representative medical classi?cation task.
Methods: We contrast the learning of 20 medical students against that of 6 residents and 12 pediatric emergency medicine fellows. Each participant was presented with 234 randomly assigned ankle radiographs with immediate feedback using a web-based application. Di?erent parametric function were applied: linear, power, and the Thurstone learning equation.
Results: We were able to successfully model both group level and individual level learning. In particular, we found educationally and statistically meaningful differences between the two groups in terms of their y-intercepts, rate of learning and asymptotes (i.e. maximum potential learning).
Discussion: Based on modeling of learning curves, we can predict the number of cases a learners needs to complete in order to achieve a pre-speci?ed competency level.