Stopping trials early for benefit is dangerous to the public health

Abstract text
Background:Simulations suggest that meta-analyses including RCTs stopped early for benefit (truncated RCTs) will yield unbiased estimates of treatment effect. If, however, truncated RCTs occur early in the sequence of trials, they may result in premature judgments of benefit.

Objectives:To examine the impact on clinical practice guidelines of truncated trials that proved misleading.

Methods:We identified three instances of misleading results from truncated trials and examined the evolution of clinical practice guidelines after the publication of the truncated and subsequent trials.

Results:: In 1999 a truncated RCT of bisoprolol in patients undergoing non-cardiac surgery reported a 91% relative risk reduction (RRR) in death and myocardial infarction. In 2002 a clinical practice guideline recommended beta-blockers for such patients. In 2008 a meta-analysis documented a two-fold increase in non-fatal strokes and a possible increase in all-cause mortality with beta blockers. Guidelines published in 2009 and 2012 continue to recommend beta-blockers.

In 2001, a truncated RCT of intensive insulin regimen in critically ill patients with elevated serum glucose reported a 42% RRR in mortality. Recommendations published as early as 2003 suggested an upper limit of glucose of 150 mg/dl or less. A meta-analysis published in 2008 showed no reduction in mortality and an increased risk of hypoglycemia. The most recent recommendations of a number of guideline groups continue to advocate limits of 150 mg/dl or less.

In 2001, a truncated RCT of recombinant human Activated Protein C (rhAPC) in critically ill patients suffering from sepsis suggested a mortality reduction. In 2004, guidelines first recommended rhAPC and continued the recommendation up to 2009 despite a negative trial published in 2005, The drug was withdrawn from the market after another negative trial in 2011.

Conclusions:Early truncated trials may lead to persistent recommendations for interventions that result in net harm.
Authors
Guyatt G1, Briel M2, Glasziou P3, Bassler D4, Montori V5
1 McMaster University, Canada
2 Institute for Clinical Epidemiology and Biostatistics, Switzerland
3 Bond University, Australia
4 University Children's Hospital, Tuebingen, Germany
5 Mayo Clinic, USA
Presenting author and contact person
Presenting author: 
Gordon Guyatt
Contact person Affiliation Country
Gordon Guyatt (Contact this person) McMaster University, Faculty of Health Sciences Canada
Date and Location
Session: 
Oral session B10O3
Date: 
Tuesday 2 October 2012 - 11:40 - 12:00
Location: