A Comparison of Three Pulmonary Artery Oximetry Catheters in Intensive Care Unit Patients (8)

A 95 percent confidence interval was determined for each catheters average bias using two methods. The first method used the SD, a, of all differences over all patients and the standard normal distribution (mean bias ±1.965 a). This method assumes that the measurement error is independent of the patient measured. The second method used the ESD as the variance estimate, thereby factoring out the intersubject variability. A Students t distribution with (10-1) degrees of freedom was used to construct this interval (mean bias ± 2.262 ESD). Trends over time were assessed using a univariate repeated measures of analysis of variance for the difference.
Similarity of catheter performance to the manufacturers published specifications, ± 2 percent (± 1 SD), was tested using the null hypothesis that each catheter (in vivo) differs from COOximetry (in vitro) by more than ±2 percent. Statistical inference procedures for comparing groups detail the use of sample information to infer information about population comparisons within certain probabalistic limits. In a usual t test of group differences, the null hypothesis of “no difference” between groups is tested. Substantial sample evidence of a difference is needed to conclude with high probability that there is a nonzero population difference. In studies intended to examine group similarity or agreement, such as generic drug bioequivalence studies or instrument comparison studies, the null hypothesis states that the groups differ by more than some predetermined meaningful amount.