Statistical analysis
Univariate comparisons of proportions, means, and medians were respectively done using Fisher exact (or chi-square), Student t, and Wilcoxon rank-sum tests. Logistic regression models for mortality included covariates with P < 0.1 on univariate analysis. Year of ICU admission was included as a categorical variable to evaluate any association of mortality across time.
To understand how PP and SV changes contributed to the determination of fluid challenge results, we collected more detailed data for the fluid challenges done using PBFM for patients in 2012, only if the arterial line was connected to the Flotrac-Vigileo device. These fluid challenges could be done at any time, including beyond the first 24 h of ICU stay. We assessed the sensitivity, specificity, percent agreement, and kappa of positive PP1-2 (versus positive SV1-2 as a reference standard), positive PP2-3 versus positive SV2-3, and positive PP1-2 or PP2-3 versus positive SV1-2 or SV2-3 (23). Finally, using overall protocol-defined volume response as the reference, we recomputed the test characteristics of consecutive 250-mL fluid challenge boluses and 500-mL fluid challenges for both PP and SV end points.
Statistical significance was taken as P < 0.05. Parts of this study were submitted as abstracts to the Singapore Health and Biomedical Congress 2014 and the Asian Pacific Society of Respirology Congress 2014.