128Study data within systematic reviews of intensive care interventions analysed as a cluster randomized trial

Abstract

\n \n \n Cluster randomized trials (CRT) can address infectious disease research questions at the whole-of-population level, such as herd protection and herd peril effects, that cannot be answered within randomized controlled trials. For adequate power, ideally, a CRT will have at least 40 clusters.\n \n \n \n Studies of decontamination (using either chlorhexidine or topical antibiotics) to prevent ICU-acquired infections among adult patients requiring prolonged mechanical ventilation (MV) have been summarized in\u2009>\u200915 systematic reviews. Of 83 studies of topical antibiotics, 69 had concurrent control (CC) versus non-concurrent control (NCC; n\u2009=\u200914) design (Figure below; solid symbols are intervention group patients). The mean ICU pneumonia incidence for topical antibiotic study CC control groups (34.5%; 95% CI, 29.1-40.4) is higher versus a literature benchmark (26.1%; 23.3%-29.1%), and versus NCC control groups (29.9%; 21.4-40.4) and versus chlorhexidine study CC control groups (25.5%; 19.6 – 32.5; n\u2009=\u200917).\n \n \n \n Benchmarking the event rates among control groups versus the event rate among non-intervention studies for MV patients provide a measure of the contextual effect of the intervention. Differing intervention effects for studies with CC versus NCC design implicates herd effects.\n \n \n \n Systematic reviews of interventions that include studies with CC versus NCC design provide a natural experiment of contextual effects where their study otherwise would be difficult or, for adverse herd effects, unethical.\n