Lancet Respiratory Medicine has issued an expression of concern for a meta-analysis on tracheostomy in the intensive care unit that they published earlier this year.
The paper, “Effect of early versus late or no tracheostomy on mortality of critically ill patients receiving mechanical ventilation: a systematic review and meta-analysis“, came from a group at Harvard, Weill Cornell and the University of Athens. The authors purported to find that:
The synthesised evidence suggests that early tracheostomy is associated with lower mortality in the intensive-care unit than late or no tracheostomy; a finding that might question the present practice of delaying tracheostomy beyond the first week after translaryngeal intubation in mechanically ventilated patients. However, the scarcity of a beneficial effect on long-term mortality and the potential complications associated with tracheostomy need careful consideration; thus, further studies focusing on long-term outcomes are warranted.
However, according to the notice, the researchers appear to have misinterpreted the work:
Following the publication online on June 27, 2014, of the Article “Effect of early versus late or no tracheostomy on mortality of critically ill patients receiving mechanical ventilation: a systematic review and meta-analysis” by Siempos and colleagues, The Lancet Respiratory Medicine received a letter from Dr Gusmao-Flores and Dr Barreto (Hospital Universitário Professor Edgard Santos, UFBA, Salvador, Bahia, Brazil) that highlighted some possible data discrepancies related to intensive-care unit (ICU) mortality for one trial (Zheng and colleagues) included in the paper. These concerns were put to the authors and after investigation by the authors and discussion between the editors and Dr Siempos, the authors have indicated “While extracting data on ICU mortality from two trials (Zheng et al and Terragni et al), we made an incorrect assumption; we assumed that patients who were not discharged from the ICU, died in the ICU. These extracted data appeared in figure 2, lines 6 and 8 of our paper. We have made every effort subsequently to acquire the correct data on ICU mortality for both trials, but have been unable to obtain data for ICU mortality. The original paper reported that all-cause mortality in the ICU was significantly lower in patients assigned to the early versus the late or no tracheostomy group (OR 0·72, 95% CI 0·53–0·98; p=0·04). If one repeats the analysis by using the earliest timepoint for mortality provided in the trials by Zheng and colleagues (10 days) and Terragni and colleagues (28 days) as an approximation for ICU mortality, one calculates a pooled odds ratio of 0·76 (95% CI 0·55–1·05) using a random-effects model, and 0·83 (95% CI 0·69—0·99) using a fixed effect model. In our original paper, we used a random effects model. We (the authors) cannot be sure that our finding regarding ICU mortality is not misleading.” The editors therefore wish to alert our readers that the ICU mortality findings are incorrect. A panel of experts has been convened to discuss the findings, and we will inform readers as soon as we have the thoughts of this group.
This is unfortunate, but it doesn’t seem as bad as most of the retractions featured on this site.
The assumption that patients who were not discharged from the ICU died in the ICU seems reasonable, though I’d be interested to hear from someone who has more expertise than I about whether/why this was an especially bad assumption.
In any case, a retraction should be made if the conclusions cannot be verified or an erratum published if the results change upon re-analysis.
Also, I wonder how much fault lies with the previous studies for publishing data that led to a faulty assumption and that cannot be readily verified once the faulty assumption was noticed.
The topic seems important and useful to ICU doctors, so I hope that however it shakes out, good conclusions can be reached.
I haven’t looked at the original papers, but most patients aren’t discharged from they ICU – they’re transferred to a non-intensive care ward and discharged home from that ward. Maybe when they say “discharged” they mean transferred, but that could be part of the issue.