Thus, it is more appropriate to test a null hypothesis of difference, with an alternative hypothesis of similarity. When this more direct null hypothesis is rejected, we conclude that the measures are similar with high probability. The second approach presented herein uses a null hypothesis that each catheter (in vivo) mean differs from CO-oximetry (in vitro) by more than 2 percent of the in vitro mean. The corresponding alternative hypothesis states that the measures are similar within these limits. These tests for equality require that the level of agreement to be considered as “equality” be established before statistical testing. No data in the literature establish an acceptable level of agreement between PA oximetry catheters and bench oximeters. Consequently, we are unable to test whether the performance of catheters in question differs significantly from bench oximetry clinically. However, the performance of these devices does not match, with high probability, the performance specifications published by their respective manufacturers.
Second, statistical analysis for most tests requires independent observations. Limitations of clinical comparison studies such as these require multiple observations in individual patients with subsequent pooling of data, violating the rule of independent observation. All of the comparison studies noted to date also suffer from this flaw. A more appropriate analysis should account for the covariance among repeated measures within patients. The tests for agreement (95 percent confidence limits based on intrasubject variability) used in this study accounted for the repeated measures by using the within-subject standard error of the differences.