Kestrel Labs (United States)

To the Editor The recent article by Park et al1 includes the study conclusion that the Radical-7 “…can be useful as a trend monitor...” This conclusion, however, appears to be unsupported by the method of analysis and the lack of a priori criteria for acceptable trend performance. Figure 1, adapted from Figure 1B in the article, plots the magnitude of the change in total hemoglobin as measured by the Radical-7 (ΔSpHb) against the magnitude of the actual change in total hemoglobin (tHb) measured on any given study subject over some unspecified time interval. Data points inside the bolded square at the center of the plot are excluded because they represent changes too small to “reflect trending ability.” To support their conclusion of useful trend performance, the authors noted that 93% (53/57) of the data pairs outside the exclusion zone demonstrated “correct directionality,” meaning that they fell in the upper right or lower left quadrants.Figure 1: Adapted from Figure 1B in the associated article.1 This figure clearly shows the trend data point pairs that fall outside the suggested ±1 g/dL limits. These points represent pairs in which the inaccuracy of the test instrument exceeded 1 g/dL.Clinical decision making depends not only on correct directionality of the change in tHb but also on the accuracy of the estimated magnitude of that change. By the article’s methodology, the point from Figure 1 at a ΔtHb of −2.4 g/dL and a ΔSpHb of −0.1 g/dL would be considered acceptable because it falls into the first or third quadrants, even though the change in SpHb is in error by greater than 2 g/dL. A more appropriate analysis technique might be to draw a pair of parallel lines on both sides of the line of identity at a cutoff level selected a priori, perhaps at ±1 g/dL as shown in Figure 1, and then calculate the percentage of points that fall outside this band. This percentage can then be compared against an a priori error threshold (based in part on the sample size). Performing an analysis graphically from Figure 1 yields the error rates for various cutoff levels listed in Table 1. It is questionable whether the error rates shown in Table 1, given the associated magnitude of these errors, are indicative of a clinically useful trend monitor.Table 1: Error Rates for 3 Different Cutoff Levels (Predetermined Distances from the Line of Identity) for the Data in Figure 1