It has been suggested that a plot of the difference between methods against their means may be more informative and eliminates the statistical artifact caused by plotting the difference against either value separately. While this concept has been discussed in statistical publications, it is not widely recognized by medical users of statistics.
The Oximetrix 3 and HEM0PR02 systems tended to underestimate SvOa compared with CO-oximetry with average biases (Eq 3) of —1.98 and —2.28, respectively. The SAT-2 system, on the other hand, tended to overestimate Sv02 compared with COoximetry, with an average bias of +1.86. The precision (Eq 4) of the three instruments based on the pooled data across patients was ±3.07, ±3.49, and ±5.24 for the Oximetrix 3, SAT-2, and HEM0PR02, respectively. The ESD (Eq 7) for the Oximetrix 3, SAT-2, and HEM0PR02 systems was ±2.03, ±2.50, and ±2.90, respectively. The 95 percent confidence limits for the three systems are shown in Figure 1. The confidence limits based on total variability and on intrasubject variability are shown and are plotted relative to the overall bias exhibited by each instrument. It should be noted that the 95 percent confidence limits are not affected by the manner in which the data are plotted (ie, [in vivo —in vitro] against in vitm, as in Figure 1a, or [in vivo — in vitro] against [in vivo + in vitro]/2t as in Figure 1b).