A. Toloff1, S. Bradley1, D. Bouwman1, D. A. Edelman1 1Wayne State University,Department Of Surgery,Detroit, MI, USA
Introduction: Learning curves (LCs) are powerful tools. Selection of an appropriate model allows determination of parameters useful for management of training. LCs are used commonly in educational research but rarely in clinical teaching due to their dependence on high quality assessment at frequent intervals. With the availability of video-assessment of all practice events of the FLS transfer task, we hypothesized that a best model could be selected from a list (linear,exponential, logarithmic,and power) using Akaike information criteria (AIC).
Methods: All practice events occurring in our FLS training curriculum are video logged. We scored performance on all practice events for all novice trainees from July 2015 through June 2016 who completed at least 30 trials of the FLS transfer task (49 trainees). Task completion time [TCT]) was the performance measure (it decreases with acquisition of skill). Each TCT was linked to a chronologically sequenced trial number (TN). The group average TCTs plotted by TN were fitted 4 times to commonly proposed learning curve functions (linear, exponential, logarithmic, and power) using least squares (LS) estimation. The AIC for each model was compared. A lower AIC means more information is preserved and identifies a superior model. Individual trainee curves were also generated for each model with a frequency count of the model providing best fit. Group TCTs at TN1 were calculated for each model and compared. Individual estimated TCTs at TN1 were computed and compared using post-hoc testing after ANOVA.
Results: The grouped data is represented best by the power function which displays the lowest AIC. Fitting individual data, the power function provides the best fit for 30 (61%) individuals. The power function provides the best estimate of the TCT at TN1 for the group data. For individuals, the power function estimate of TCT at TN1 is significantly different from the linear and exponential estimates.
Conclusion: For the group of novice learners, skill acquisition for the FLS transfer task was fit better by the power function than linear, logarithmic, or exponential curves. The skill acquisition data for the majority of individual trainees was also fit best by the power function. The learning curves based on the power linking function are superior to the three other candidate models we assessed and form the basis for analysis of both group and individual data.
