We thank the authors of the letter to the editor for their comments on our observational target trial of fludrocortisone and hydrocortisone as compared with hydrocortisone during early septic shock.1 As noted in our limitations, we agree that there is risk for immortal time bias (a form of selection bias) due to the “calendar day” granularity of the included data set. Specifically, because day 0 of the target trial emulation was defined as the day of hydrocortisone initiation and because patients assigned to the hydrocortisone–fludrocortisone group were required to be initiated on fludrocortisone also on day 0, it is possible that there was up to a 24-hour interval (if hydrocortisone was started just after 12:00 am) between hydrocortisone initiation and fludrocortisone initiation in patients who received both (ie, treatment assignment to hydrocortisone–fludrocortisone). This time interval between receipt of hydrocortisone and fludrocortisone represents potential immortal time.
During the study design phase,1 we considered several additional analyses to address the potential for immortal time bias, including the proposed landmark analysis that excludes patients who died or were discharged on day 0. However, we decided against the landmark analysis because this approach can introduce 2 additional biases: (1) selection bias by conditioning on a postexposure collider variable (ie, death/discharge on day 0 regardless of whether associated with corticosteroid choice)2 and (2) bias from conditioning on a potential mediator variable (death/discharge on day 0 if associated with corticosteroid choice).3 Because the landmark analysis could introduce additional biases (either toward the null for conditioning on a mediator, or in either direction if conditioning on a collider, without knowledge as to which bias is present), we believed it would not clarify the question of immortal time bias on day 0.
Instead, we chose to conduct alternative analyses to estimate the potential risk for immortal time on day 0 by (1) quantifying the time from hydrocortisone to fludrocortisone in a separate granular data set (average of 2 hours) and (2) use of a negative control outcome analysis (no association between corticosteroid choice and the negative control outcome).1 Both analyses suggested little influence of immortal time on day 0.
Conflict of Interest Disclosures:
Dr Bosch reported grants from the US National Institutes of Health–National Center for Advancing Translational Sciences (NIH/NCATS) during the conduct of the study; and grants from the US Department of Defense, Gilead Sciences, and NIH/NHLBI outside the submitted work. Dr Law reported grants from NIH/NHLBI during the conduct of the study. No other disclosures were reported.
References
- 1.Bosch NA, Teja B, Law AC, Pang B, Jafarzadeh SR, Walkey AJ. Comparative effectiveness of fludrocortisone and hydrocortisone vs hydrocortisone alone among patients with septic shock. JAMA Intern Med. 2023;183(5):451–459. doi: 10.1001/jamainternmed.2023.0258 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Hernán MA, Hernández-Díaz S, Robins JM. A structural approach to selection bias. Epidemiology. 2004;15(5):615–625. doi: 10.1097/01.ede.0000135174.63482.43 [DOI] [PubMed] [Google Scholar]
- 3.Elwert F, Winship C. Endogenous selection bias: the problem of conditioning on a collider variable. Annu Rev Sociol. 2014;40:31–53. doi: 10.1146/annurev-soc-071913-043455 [DOI] [PMC free article] [PubMed] [Google Scholar]