Miguel A. Hernán, MD; Sonia Hernández-Díaz, MD; James M. Robins, MD
This article was published online first at www.annals.org on 10 September 2013.
Grant Support: In part by grants R01 HL080644 and R01 AI102634 from the National Institutes of Health.
Potential Conflicts of Interest: Disclosures can be viewed at www.acponline.org/authors/icmje/ConflictOfInterestForms.do?msNum=M13-1455.
Requests for Single Reprints: Miguel A. Hernán, MD, Department of Epidemiology, Harvard School of Public Health, 677 Huntington Avenue, Boston, MA 02115; e-mail, firstname.lastname@example.org.
Current Author Addresses: Drs. Hernán, Hernández-Diaz, and Robins: Harvard School of Public Health, Department of Epidemiology, 677 Huntington Avenue, Boston, MA 02115.
Author Contributions: Conception and design: M.A. Hernán, S. Hernández-Díaz, J.M. Robins.
Drafting of the article: M.A. Hernán, S. Hernández-Díaz, J.M. Robins.
Critical revision for important intellectual content: M.A. Hernán, S. Hernández-Diaz, J.M. Robins.
Final approval of the article: M.A. Hernán, S. Hernández-Diaz, J.M. Robins.
Statistical expertise: M.A. Hernán, S. Hernández-Díaz, J.M. Robins.
Obtaining of funding: M.A. Hernán.
Administrative, technical, or logistic support: M.A. Hernán.
Hernán M., Hernández-Díaz S., Robins J.; Randomized Trials Analyzed as Observational Studies. Ann Intern Med. 2013;159: 560-562. doi: 10.7326/0003-4819-159-8-201310150-00709
Download citation file:
Published: Ann Intern Med. 2013;159(8): 560-562.
Despite what you may have heard, randomized trials are not always free of confounding and selection bias. Randomized trials are expected to be free only from baseline confounding but not from postrandomization confounding and selection bias (1). In this commentary, we describe the settings in which postrandomization confounding and selection bias emerge in randomized trials, discuss the shortcomings of intention-to-treat analyses to handle these biases, and direct readers to more appropriate methods.
The neglect of postrandomization confounding and selection bias in randomized trials is the historical consequence of the fact that many early trials were short, small, double-blinded, tightly controlled experiments in highly selected patients. Most premarket trials still fit this description. In these experiments, randomization makes baseline confounding unlikely, whereas double-blinding, tight control, and short duration minimize postrandomization confounding (due to deviations from protocol or differential use of concomitant therapies) and selection bias (due to differential loss to follow-up). Such trials may be optimal to detect small treatment benefits but not to guide clinical decision making: follow-up too short for clinically relevant outcomes, patients unrepresentative, interventions unrealistic, sample size too small to identify adverse events.
to gain full access to the content and tools.
Learn more about subscription options.
Register Now for a free account.
Research and Reporting Methods.
Results provided by:
Copyright © 2016 American College of Physicians. All Rights Reserved.
Print ISSN: 0003-4819 | Online ISSN: 1539-3704
Conditions of Use
This PDF is available to Subscribers Only