Comparative performance of two methods that assess the quality of international normalized ratio measures

Abstract

Two methods, Petersen's error grid analysis and Shermock's method to detect clinically important differences, were recently developed to advance the assessment of analytic performance of point-of-care INR devices. Both methods predict when alternate INR measures lead to different clinical decisions. Our goal was to compare their performance characteristics. Performance characteristics were assessed by comparing the models' predictions to clinical decisions that were directly measured in a previous experiment. Shermock's method (82% of predictions correct) demonstrated superior predictive performance compared with the error grid analysis (75% of predictions correct, p=0.008). Shermock's method was particularly superior at identifying the clinical decisions that actually disagreed (79% for Shermock's method vs. 47% for error grid). Consequently, Shermock's method was superior at identifying a POC device with poor performance (79% accuracy vs. 70%, p=0.006). Shermock's method had superior performance characteristics and should be integrated into analytic strategies to assess POC INR devices.