Letter in the Lancet: Misleading data Analyses in Salmeterol (Serevent) Study

October 8, 2005

The Lancet, Volume 366, Issue 9493, 8 October 2005-14 October 2005, Pages 1261-1262, Peter Lurie and Sidney M. Wolfe

In 1996, due to reports of paradoxical bronchospasm associated with the use of GlaxoSmithKline's (GSK's) newly approved, long-acting inhaled β agonist salmeterol, and previous epidemics of asthma-related deaths in patients taking other long-acting β agonists,[1] the company initiated a randomised trial comparing salmeterol (Serevent, known as Advair when combined with the steroid fluticasone) to placebo. The results have never been published.

We only know the detailed results of this crucial study because the drug came before the US Food and Drug Administration's (FDA's) Pulmonary-Allergy Drugs Advisory Committee on July 13, 2005, and because the FDA now provides the Advisory Committee briefing materials, including reviews by its own medical officers, on its website before the meeting.[2] These documents provide insight into the manner in which the company manipulated the data it submitted to the FDA in an apparent attempt to convince the agency that the drug's risks were smaller.

The Salmeterol Multicenter Asthma Research Trial (SMART) was designed to randomise 60,000 asthma patients to either salmeterol 42 μg twice daily or placebo. The study duration was 28 weeks, with patients examined about every 4 weeks. Investigators were also asked to report (but not actively seek out) any serious adverse event that occurred within 6 months after the patient completed the trial. The primary outcome was combined respiratory-related deaths and life-threatening experiences; secondary outcomes included asthma-related deaths, asthma-related deaths or life-threatening experiences, and all-cause deaths.

In September, 2002, the SMART Data Safety Monitoring Board (DSMB) met to review the data on the 25,858 patients who had been enrolled to date. Based on nearly significant treatment-related increases in both the primary outcome variable and asthma-related deaths, the DSMB offered two alternatives: either increase the target sample size by at least 10,000 patients or terminate the study “with dissemination of findings as quickly as possible.” GSK selected the latter option.

GSK presented the interim analysis of the SMART data at the 2003 meeting of the American College of Chest Physicians, claiming that “The interim analysis was inconclusive.”[3] Based on the interim data, on Aug. 11, 2003, the FDA modified the labelling for both Serevent and Advair to include a black box warning of a “small but significant increase in asthma-related deaths.” With the interim results in hand, the company finalised a statistical methods appendix, which it submitted to the agency for the first time on Aug 29, 2003, along with the final study data. In a significant deviation from the original protocol, the appendix permitted including in the study dataset adverse events that were reported in the 6 months after the termination of the trial.

The table below presents the results of the SMART study according to whether or not the post-trial results are included. For the primary outcome variable, as well as for each of the three major secondary outcome variables, the inclusion of the post-trial data produces an apparent diminution of the dangers associated with salmeterol.

Table: Relative risks of primary outcome and major secondary outcomes in SMART study with and without inclusion of post-trial data


Relative risk (95% CI)

28-week study

28-week study plus 6 months

Respiratory-related deaths and life-threatening experiences

1·39 (0·91–2·13)

1·16 (0·78–1·72)

Asthma-related deaths

4·33 (1·24–15·21)

2·50 (0·97–6·44)

Asthma-related deaths or life-threatening experiences

1·68 (0·99–2·85)

1·52 (0·92–2·52)

All-cause deaths

1·31 (0·83–2·08)

1·04 (0·70–1·55)

GSK did not specify to the FDA which dataset it had submitted. Only on April 26, 2004, after the FDA specifically enquired as to which dataset had been provided, did the agency learn that the company had provided only the dataset that included the 6-month data. The medical officers state that “The Division presumed the data represented [only] the 28-week period as the 28-week period is clinically the period of interest”. The FDA recommended that any further labelling changes be based on the protocol-specified 28-week study, which it forced the company to provide.

At the July 13, 2005, Advisory Committee meeting, the committee recommended strengthening the labels for both Serevent and Advair to encourage concomitant inhaled corticosteroid use.[4] The agency has not made a final decision on the label.

It is now approaching 3 years since the SMART study was terminated. The results have still not found their way into print (GSK told the Advisory Committee that they have been accepted for publication in Chest) and the drugs' labels have still not been finalised. Instead, the company, seemingly under little pressure from the FDA, has succeeded in drawing out the process and initially misleading the agency, physicians, and patients with not-per-protocol analyses that diminished the drug's apparent risks. In the absence of the transparency associated with Advisory Committee meetings, these deceptions would never have come to public attention. In 2001, however, only 21% of new drug approvals were preceded by Advisory Committee meetings, allowing most drugs to avoid similar public scrutiny.[5]

We declare that we have no conflict of interest.

[1] Stolley PD. Why the United States was spared an epidemic of deaths due to asthma. Am Rev Respir Dis 1972; 105: 883–90.

[2] US Food and Drug Administration. Medical officer review. Provided for: Meeting of Pulmonary-Allergy Drugs Advisory Committee, July 13, 2005. http://www.fda.gov/ohrms/dockets/ac/05/briefing/2005-4148B1_03_02-FDA-Sm...

[3] Knobil K, Yancey S, Kral K, Rickard K. Salmeterol Multi-center Asthma Research Trial (SMART): results from an interim analysis. Meeting of American College of Chest Physicians, Orlando, FL, USA; Oct 27, 2003.

[4] Anon. Serevent, Foradil use with steroids should be strongly urged in labeling. The Pink Sheet July 18, 2005: 12–13.

[5] Office of the Inspector General. FDA’s review process for new drug applications: a management review. Washington, DC: Department of Health and Human Services, 2003.