Inadequate statistical power of published comparative cohort studies on ventilator-associated pneumonia to detect mortality differences between the compared groups
Abstract number: O387
Falagas M., Kouranos V., Michalopoulos A., Rodopoulou S., Athanasoulia A., Karageorgopoulos D.
Objective: Comparative cohort studies are often conducted to identify novel therapeutic strategies or prognostic factors for ventilator-associated pneumonia (VAP). We aimed to evaluate the statistical power of such studies to provide statistically and clinically significant conclusions.
Methods: We searched in PubMed and Scopus for comparative cohort studies evaluating the mortality of patients with VAP. We calculated for each of the included studies the statistical power to detect the observed difference in mortality between the compared groups (observed power), as well as 3 expected, clinically relevant, effect sizes (expected power).
Results: We identified 39 (20 prospective) comparative cohort studies on VAP as eligible for inclusion in this analysis. The median observed power of these studies was 17.9% [interquartile range (IQR), 9.852.4%]. The median expected power was 10.0% (IQR, 7.213.6%) for a risk ratio for mortality of 0.85 between the compared groups; 14.7% (IQR, 10.621.8%) for a risk ratio of 0.80; and 7.9% (IQR, 6.310.2%) for a reduction in mortality from 30% to 25%. All expected power measures were significantly lower than the observed power.
Conclusion: The statistical power of most cohort studies to detect the observed difference in mortality between compared groups of patients with VAP is low. The power is even lower when expected, clinically relevant, differences in mortality are considered. For a wiser utilisation of resources allocated to research, we favour the conduction of cohort studies with larger sample size so that potential differences between the compared groups are more likely to be shown.
|Session name:||19th European Congress of Clinical Microbiology and Infectious Diseases|
|Location:||Helsinki, Finland, 16 - 19 May 2009|
|Back to top|