Positive single-center randomized trials and subsequent multicenter randomized trials in critically ill patients: a systematic review

Background: It is unclear how often survival benefits observed in single-center randomized controlled trials (sRCTs) involving critically ill patients are confirmed by subsequent multicenter randomized controlled trials (mRCTs). We aimed to perform a systemic literature review of sRCTs with a statistically significant mortality reduction and to evaluate whether subsequent mRCTs confirmed such reduction. Methods: We searched PubMed for sRCTs published in the New England Journal of Medicine, JAMA, or Lancet, from inception until December 31, 2016. We selected studies reporting a statistically significant mortality decrease using any intervention (drug, technique, or strategy) in adult critically ill patients. We then searched for subsequent mRCTs addressing the same research question tested by the sRCT. We compared the concordance of results between sRCTs and mRCTs when any mRCT was available. We registered this systematic review in the PROSPERO International Prospective Register of Systematic Reviews (CRD42023455362). Results: We identified 19 sRCTs reporting a significant mortality reduction in adult critically ill patients. For 16 sRCTs, we identified at least one subsequent mRCT (24 trials in total), while the interventions from three sRCTs have not yet been addressed in a subsequent mRCT. Only one out of 16 sRCTs (6%) was followed by a mRCT replicating a significant mortality reduction; 14 (88%) were followed by mRCTs with no mortality difference. The positive finding of one sRCT (6%) on intensive glycemic control was contradicted by a subsequent mRCT showing a significant mortality increase. Of the 14 sRCTs referenced at least once in international guidelines, six (43%) have since been either removed or suggested against in the most recent versions of relevant guidelines. Conclusion: Mortality reduction shown by sRCTs is typically not replicated by mRCTs. The findings of sRCTs should be considered hypothesis-generating and should not contribute to guidelines.