Skip to main content
eScholarship
Open Access Publications from the University of California

Dermatology Online Journal

Dermatology Online Journal bannerUC Davis

A neglected issue in interpretation of results of randomized controlled trials: Informative censoring

  • Author(s): Gorouhi, Farzam;
  • Khatami, Alireza;
  • Davari, Parastoo
  • et al.
Main Content

A neglected issue in interpretation of results of randomized controlled trials: Informative censoring
Farzam Gorouhi MD, Alireza Khatami MD MSPH, Parastoo Davari MD
Dermatology Online Journal 15 (1): 13

Center for Research and Training in Skin Diseases and Leprosy, Tehran University of Medical Sciences, Tehran, Iran. gorouhif@gmail.com

Abstract

The presence of informative censoring is a commonly neglected issue that can lead to inaccurate results in randomized controlled trials. It is mandatory to perform an intention to treat analysis to minimize this source of error.


Comment:

In general, well-conducted randomized controlled trials (RCTs) are considered to be the gold standard study designs for determining the efficacy of therapeutic interventions [1]. These studies provide the highest level of resource for systematic reviews [2]. The number of published randomized controlled trials has increased dramatically over the past two decades and considerable effort has been made to improve the methods, interpretation, and reports of RCTs [3, 4]. In RCTs, several strategies have been used to minimize the errors that can occur in other study designs, random and systematic errors [5]. Random errors can be minimized by appropriate calculation of the sample size of the study. Incorrect conclusions can be drawn when there are differences in the mean or proportion of occurrence of an event in one of the study groups and when there are errors in the precision of the assessment of the group differences [6]. Type I error (known as α error and false positive) occurs when the null hypothesis is rejected but it is actually true and type II error (known as β error and false negative) occurs when the null hypothesis is accepted but it is actually false.

Major determinants of the validity of an RCT are: 1) random allocation of the participants, which itself depends on appropriate randomization sequence generation, 2) allocation concealment, 3) blinding, and 4) intention to treat analysis (ITT) [7]. Random allocation is performed to eliminate selection bias; the major purpose of blinding is to reduce information bias in a study. Intention to treat analysis not only provides prognostic balance in study groups but also minimizes the effects of withdrawals, non–compliant participants, and lost-to-follow-up patients [7].

However, an important confounding factor that is often not addressed is informative censoring. This problem occurs when dropout (censored) subjects are either more or less likely to experience the specific event than remaining individuals in the future [8]. Consequently, longer trial periods are even more subject to bias due to informative censoring.

Imagine a trial that shows similar results for two interventions (A and B) in the treatment of basal cell carcinoma, with a primary endpoint of 5-year disease-free survival. By the end of five years, about 25 percent of both treatment groups withdraw. If for intervention A, withdrawal is mainly due to patient non-compliance or treatment failure/side effects, whereas for intervention B, dropout occurs because the patients are completely cured and prefer not to be followed-up further. In this case, the problem is clear. In a standard analysis, disease-free survival rates would be overestimated for intervention A and underestimated for intervention B because they would both be based on the patients who stayed in the study.

Hence, it is impossible to determine whether informative censoring has affected the results without more information about the duration of disease-free survival for censored patients. Informative censoring may bias all outcome measures including measures that are dichotomous or continuous in both directions, depending on which group the censoring favors. Ideally, all patients, including withdrawn patients, must be followed to determine their final situation. Another effective way to minimize this problem is to perform a sensitivity analysis of the main conclusions due to various assumptions when data on dropouts is scanty. Considering "dropout" as an endpoint could be an approach to better quantify this problem in the analysis.

The problem of informative censoring necessitates the application of ITT in any randomized trial. Additionally, it should make trial analyzers more thorough regarding the exact method of ITT used in a specific trial to provide results that are as accurate as possible.

References

1. Machin D. General issues. In: Machin, Day S, Green S, editors. Textbook of Clinical Trials. London 2004 John Wiley & Sons, Ltd. pp. 11-44.

2. Sackett et al. Evidence-Based Medicine. Churchill Livingstone; Edinburgh; 2000:173-7.

3. Devereaux PJ, Yusuf S. The evolution of the randomized controlled trial and its role in evidence-based decision making. J Intern Med 2003 Aug;254(2):105-13. [PubMed]

4. CONSORT. Available at: http://www.consort-statement.org. Last accessed: July 7, 2007.

5. Rothman KJ, Greenberg S, editors. Modern Epidemiology: an introduction. Philadelphia: Lippincott Williams & Wilkins Publishers; 1998.

6. Williams HC. How to critically appraise a study reporting effectiveness of an intervention. Evidence-based dermatology. In: Williams HC, Bigby M, Diepgen T, et al, editors. Evidence-based dermatology. Oxford: BMJ Publishing Group; 2003: 56-63.

7. Juni P, Altman DG, Egger M. Systematic reviews in health care: Assessing the quality of controlled clinical trials. BMJ. 2001 Jul 7;323(7303):42-6. [PubMed]

8. Collett, D, editor. Modeling Survival Data in Medical Research. London: Chapman & Hall; 1994.

© 2009 Dermatology Online Journal