HFOV was introduced into each center between 1998 and 1999. Data were extracted from the records of all patients who were treated at each center, from the time of introduction until January 2002. The data were reviewed by individuals not involved with data extraction, were entered into a database, and were screened for aberrant entries. Statistical analysis was performed in two prospectively defined steps. First, summary data were tabulated with means, SDs, and percentages, as appropriate, and the trends of ventilator settings and physiologic variables were evaluated with repeated-measures analysis of variance. A t test was used to evaluate the changes from baseline compared with specific time points. In the second step, differences in baseline risk and HFOV response associated with survival were explored. Logistic regression was used to identify significant baseline characteristics associated with mortality.
A repeated-measures analysis of covariance was used to identify which HFOV response variables were associated with mortality. For those response variables found to be significantly associated with mortality, stepwise logistic regression was used to determine at which specific time that variable was most significantly associated with mortality. Finally, these significant response variables were stepped into the baseline logistic regression equation to determine whether the response was significant, while controlling for baseline status. One exploratory post hoc analysis also was conducted to identify a dichotomous indication of mortality associated with oxygenation index (OI). We used the t test and x2 test, as appropriate. The analysis of variance used a maximum likelihood method, so that patients would not be excluded if data were not available for all time periods. Statistical analysis was performed using two statistical software packages (SPSS, versions 10 and 11; SPSS; Chicago, IL; and GB-STAT, version 4.5; Dynamic Microsystems, Inc; Silver Spring, MD). For all analyses, p < 0.05 was considered to be significant.