Individual patient data meta-analysis in intensive care medicine and contextual effects

Excerpt

The review by Reade et al. [1] stated that “In intensive care, the context in which an intervention is delivered may be as influential as the intervention itself.” I wish to further emphasize the need for cautious screening for possible contextual effects as part of any meta-analysis, but particularly a meta-analysis using individual patient data and also of interventions undertaken in intensive care units. I use as an example the meta-analysis of antibiotic prophylaxis to prevent respiratory infections [2] cited by these authors [1]. Figure 1 displays the control and intervention group ventilator-associated pneumonia-incidence proportion (VAP-IP) data for the studies as abstracted in this meta-analysis [2] in relation to a funnel plot derived using the VAP-IP data of 44 non-intervention studies [3] that demonstrate three contextual issues relevant to the interpretation of the intervention under study [2]. Firstly, the VAP-IPs among the control groups of this meta-analysis [2] are highly variable, much more so than for the VAP-IPs of the intervention groups. Secondly, half of these control groups have a VAP-IP in excess of 45% in contrast to the VAP-IP of 184 control and intervention groups of 96 non-antibiotic VAP prevention studies for which a VAP-IP >45% is rare [3]. Thirdly, this variability is not confined to the smaller groups and hence is not explained by publication bias. These contextual effects had not been explained in this nor subsequent meta-analyses of this topic and would be further obscured in a meta-analysis using individual patient data. [2].

…