Dealing With Deaths in Clinical Trials and Meta-Analyses
========================================================

* Euan J McCaughey
* Peter Humburg
* Angus J McLachlan

Gutiérrez-Arias et al1 undertook an interesting and highly relevant review into the potential efficacy of neuromuscular electrical stimulation to reduce mechanical ventilation duration for patients who are critically ill. The authors should be commended for the amount of work that obviously went in to this review. However, research in the ICU is fraught with difficulty, with high death rates and study participants not reaching primary end points, which typically results in missing data. In taking a study that looks at the use of neuromuscular electrical stimulation to reduce mechanical ventilation duration as an example, in which the intervention period has a fixed duration, the participants who do not reach an end point (in this case, liberation from mechanical ventilation) by the end of the study would normally be treated as censored, that is, they did not wean from mechanical ventilation within the intervention period. Dealing with deaths in a study is more difficult. One option is to also censor these participants when estimating the risk of being liberated from ventilation (and, conversely, to estimate the risk of death). This would result in a cause-specific hazard analysis. Although this approach can be suitable, it is important to note that it assumes that the risks are independent. The fact that a participant has reached one end point is treated as uninformative for the risk of an alternative end point.

Alternatively, the approach of subdistribution hazards by Fine and Gray2 can be used to handle competing risks. Compared with the censoring approach, this method does not treat the competing events as censoring but instead includes these observations in the “at-risk” group. This can be understood as considering the (hypothetical) risk of a person being liberated from mechanical ventilation if the first event (death) had not occurred. The approach by Fine and Gray2 is problematic in the context of a causal analysis but tends to give better predictions than censoring.3 Which type of analysis is more appropriate depends on the context and aims. Both methods can produce misleading results if the model is misspecified, that is, if not all relevant risk factors are included in the model, especially if they affect multiple outcomes.

The scale of the problem of death in ICU studies is highlighted in the recently published study by Gutiérrez-Arias et al.1 In one of their included studies, that by Routsi et al,4 50 of the 142 participants (35.2%) randomized to the study died. Neither Gutiérrez-Arias et al1 nor the authors of the studies reported in the meta-analysis adopted the approach of censoring or subdistribution hazards to account for the proportion of the participants who were either not liberated from mechanical ventilation or died during these studies. Instead, all of these participants were excluded from the analysis. As such, the reporting of the studies in the meta-analysis by Gutiérrez-Arias et al1 is inherently biased, with results only reported for those who were liberated from mechanical ventilation. We suggest that the use of a shared parameter model for the meta-analysis would help to reduce the bias due to competing risks.5

It should also be noted that there are a number of reporting errors in the article of Gutiérrez-Arias et al.1 When considering the mechanical ventilation duration in the intervention and control groups for the study conducted by Dall’Acqua et al,6 the mean ventilation duration for the experimental group is incorrectly reported as 10 days, as opposed to the 7 days reported in the original article. By following the reporting method used by Gutiérrez-Arias et al,1 this correction would further strengthen the significance of the effect of neuromuscular electrical stimulation to reduce mechanical ventilation duration, which the authors currently describe as very weak. It should also be noted that the *P* value for the study by McCaughey et al7 is incorrectly reported at .40, rather than .04. We hope that this letter provides some insight into how to deal with competing risks in clinical trials, and thank Gutiérrez-Arias et al1 for their interesting article.

## Footnotes

*   Correspondence: Euan J McCaughey MEng PhD CEng, Neuroscience Research Australia, 139 Barker St, Randwick, New South Wales 2031, Australia. E-mail: e.mccaughey{at}neura.edu.au
*   Dr McLachlan is employed by Liberate Medical, a medical device company that is developing an abdominal muscle stimulator. Dr McCaughey discloses a relationship with Liberate Medical. Dr Humburg has no conflicts to disclose.

*   Copyright © 2021 by Daedalus Enterprises

## References

1.  1.Gutiérrez-Arias RE, Zapata-Quiroz CC, Prenafeta-Pedemonte BO, Nasar-Lillo NA, Gallardo-Zamorano DI. Effect of neuromuscular electrical stimulation on the duration of mechanical ventilation. Respir Care 2021;66(4):679-685.
    
    [Abstract/FREE Full Text](http://rc.rcjournal.com/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6ODoicmVzcGNhcmUiO3M6NToicmVzaWQiO3M6ODoiNjYvNC82NzkiO3M6NDoiYXRvbSI7czoyNDoiL3Jlc3BjYXJlLzY2LzkvMTUwMy5hdG9tIjt9czo4OiJmcmFnbWVudCI7czowOiIiO30=) 

2.  2.Fine JP, Gray RJ. A proportional hazards model for the subdistribution of a competing risk. J Am Stat Assoc 1999;94(446):496-509.
    
    [CrossRef](http://rc.rcjournal.com/lookup/external-ref?access_num=10.2307/2670170&link_type=DOI) 
    
    [Web of Science](http://rc.rcjournal.com/lookup/external-ref?access_num=000081058500019&link_type=ISI) 

3.  3.Lau B, Cole SR, Gange SJ. Competing risk regression models for epidemiologic data. Am J Epidemiol 2009;170(2):244-256.
    
    [CrossRef](http://rc.rcjournal.com/lookup/external-ref?access_num=10.1093/aje/kwp107&link_type=DOI) 
    
    [PubMed](http://rc.rcjournal.com/lookup/external-ref?access_num=19494242&link_type=MED&atom=%2Frespcare%2F66%2F9%2F1503.atom) 
    
    [Web of Science](http://rc.rcjournal.com/lookup/external-ref?access_num=000267887900014&link_type=ISI) 

4.  4.Routsi C, Gerovasili V, Vasileiadis I, Karatzanos E, Pitsolis T, Tripodaki E, et al. Electrical muscle stimulation prevents critical illness polyneuromyopathy: a randomized parallel intervention trial. Crit Care 2010;14(2):R74.
    
    [CrossRef](http://rc.rcjournal.com/lookup/external-ref?access_num=10.1186/cc8987&link_type=DOI) 
    
    [PubMed](http://rc.rcjournal.com/lookup/external-ref?access_num=20426834&link_type=MED&atom=%2Frespcare%2F66%2F9%2F1503.atom) 

5.  5.Thom H, López-López JA, Welton NJ. Shared parameter model for competing risks and different data summaries in meta-analysis: implications for common and rare outcomes. Res Synth Methods 2020;11(1):91-104.
    
    

6.  6.Dall' Acqua AM, Sachetti A, Santos LJ, Lemos FA, Bianchi T, Naue WS, et al. Use of neuromuscular electrical stimulation to preserve the thickness of the abdominal and chest muscles of critically ill patients: a randomized clinical trial. J Rehabil Med 2017;49(1):40-48.
    
    [PubMed](http://rc.rcjournal.com/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Frespcare%2F66%2F9%2F1503.atom) 

7.  7.McCaughey EJ, Jonkman AH, Boswell-Ruys CL, McBain RA, Bye EA, Hudson AL, et al. Abdominal functional electrical stimulation to assist ventilator weaning in critical illness: a double blinded, randomised, sham-controlled pilot study. Crit Care 2019;23(1):261.