Testing the Stable Unit Treatment Variance Assumption (SUTVA) Within Cochrane Reviews of Antimicrobial-Based...

Background/Objectives: Causal inference requires validating the stable unit treatment variance assumption (SUTVA). Whilst antimicrobial-based interventions, being topical chlorhexidine and topical antibiotics prophylaxis (TAP), appear effective in preventing ventilator-associated pneumonia (VAP) among ICU patients receiving mechanical ventilation (MV) within randomized concurrent controlled trials (RCCT), SUTVA has never been tested for this inference. Methods: Caterpillar plots of the VAP incidence proportions for control and intervention groups within RCCTs obtained from recent Cochrane reviews of antimicrobial-based VAP prevention interventions were derived using random effects methods to enable comparison versus the expert VAP incidence reference range (5 to 40%). Results: The summary VAP prevention effect size derived from three categories of 57 RCCTs of antimicrobial-based interventions was recapitulated. The VAP incidences of 24 control groups and 6 intervention group incidences were above, whereas only 1 and 6, respectively, were below the expert VAP incidence reference range (5 to 40%) (p < 0.001; chi-square = 17.42; df = 2). The results after excluding 18 low-quality studies were similar. Paradoxically, the 90% prediction limits in association with the summary control group incidences are each approximately 20 percentage points wider than for those associated with the intervention group summaries. Conclusions: Control group VAP incidences above and more dispersed versus the expert opinion VAP range are common within the Cochrane reviews of antimicrobial-based VAP prevention interventions. Recognition requires an arms-based analysis using caterpillar plots. The SUTVAs that underlie the inference of prevention from the effect size estimates are not valid.