LAUSR

I read the article entitled “Characterizing the learning curve of the MRI–US fusion prostate biopsies” by Halstuch et al. [1] with great interest. The authors used both process and outcome measures as proxies of learning, allowing for an accurate and valid evaluation of learning. To analyse the learning curve, they calculated the fractional change in outcome for each consecutive procedure, and defined the transition point as that where the fractional change was <10%, that was maintained for over 90% of subsequent cases. The most commonly used method to analyse a learning curve is the group-splitting method, which has several disadvantages correctly identified by the authors including arbitrary group sizing and an inability to detect a transition point. The method employed in this study is significantly superior to the group-splitting method, and indeed provides an estimate for the transition period (110–125 cases). However, I note that their learning curve method also arbitrarily uses a 10% fractional change value as the determinant for the transition point. The requirement that 90% of subsequent cases meet this condition is also arbitrary. Employing a segmented linear regression technique may have provided a more accurate and objective description of the learning curve. Segmented linear regression has been previously utilised in learning curve analysis and can test several different learning models to detect the best descriptor of learning [2]. It not only detects a transition period, but describes the precise rate of learning before and after the transition point. The method employed in the current paper assumes a plateau is achieved, whereas a slower constant learning phase may be the true description of learning that follows the transition point. I would like to commend the authors for employing a mathematical method for learning curve analysis that is significantly superior to the more commonly used groupsplitting method, but would recommend employing a segmented linear regression technique with multiple model options for a more objective and detailed description of learning.