To the Editor:
I read with great interests on the study by Pandya et al1 in CHEST (February 2021), in which they compared the difference between late vs early intubation of patients with coronavirus disease 2019. They found that late intubation was associated with longer length of stay in ICU and duration of mechanical ventilation than the early intubation group. Although it is plausible that the late intubation group may experience prolonged periods of hypoxia that result in pathophysiologic derangements such as hypoxemia and multiorgan dysfunction, the finding may also be attributable to the immortal time bias.2 Immortal time bias refers to a distortion that modifies an association between an exposure and an outcome, caused when a cohort study is designed so that follow up includes a period of time in which participants in the exposed group cannot experience the outcome and are essentially “immortal.”
In the present study, the time from admission to intubation is the immortal time, in which the outcome of mortality cannot occur. When the length of stay in ICU was calculated from admission, this immortal time is attributed inappropriately to the effect of intubation. As a result, the length of stay is prolonged in the late intubation group. A potential solution to the immortal time bias is to reset the time zero of follow up to the time of intubation.3 Because the indication for tracheal intubation should be carried out uniformly in an institution, it is reasonable to consider the time of intubation as the time when the pathophysiologic condition is similar across patients. In contrast, the time of admission may not represent the same stage of coronavirus disease 2019. In other words, some patients may arrive at the hospital at an early stage, but others may arrive at a late stage.
Another possible solution to the immortal time bias is the use of Cox regression model with time-varying covariates.4 In this model, the survival outcome is considered as the time-to-event variable. Intubation is a covariate that can happen at any time during hospitalization. This will allow adjustment for other time-varying confounders.
Furthermore, if we want to consider different probabilities of receiving tracheal intubation during the time course of hospitalization, the time-dependent propensity score matching can be used.5 Because the authors have stated that the intubation is determined by the treating physician without explicit criteria, the propensity of receiving intubation varied across patients during the hospital stay.
Footnotes
FINANCIAL/NONFINANCIAL DISCLOSURES: None declared.
References
- 1.Pandya A., Kaur N.A., Sacher D., et al. Ventilatory mechanics in early vs late intubation in a cohort of COVID-19 patients with acute respiratory distress syndrome: a single center’s experience. Chest. 2021;159(2):653–656. doi: 10.1016/j.chest.2020.08.2084. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Dall’Olio F.G., Di Nunno V., Massari F. Immortal time bias question in the association between toxicity and outcome of immune checkpoint inhibitors. J Clin Oncol. 2020;38(1):105–106. doi: 10.1200/JCO.19.01728. [DOI] [PubMed] [Google Scholar]
- 3.Lévesque L.E., Hanley J.A., Kezouh A., Suissa S. Problem of immortal time bias in cohort studies: example using statins for preventing progression of diabetes. BMJ. 2010;340 doi: 10.1136/bmj.b5087. b5087. [DOI] [PubMed] [Google Scholar]
- 4.Zhang Z., Reinikainen J., Adeleke K.A., Pieterse M.E., Groothuis-Oudshoorn C.G.M. Time-varying covariates and coefficients in Cox regression models. Ann Transl Med. 2018;6(7) doi: 10.21037/atm.2018.02.12. 121-121. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Zhang Z., Li X., Wu X., Qiu H., Shi H. on behalf of AME Big-Data Clinical Trial Collaborative Group. Propensity score analysis for time-dependent exposure. Ann Transl Med. 2020;8(5) doi: 10.21037/atm.2020.01.33. 246-246. [DOI] [PMC free article] [PubMed] [Google Scholar]