Flawed meta-analysis creates doubt when answers are knownBMJ 2011; 342 doi: http://dx.doi.org/10.1136/bmj.d3348 (Published 31 May 2011) Cite this as: BMJ 2011;342:d3348
- Paul Walsh, research director1,
- Stephen J Rothenberg, senior investigator2,
- Dale Robbins, physician assistant1,
- John Caldwell, research pharmacist3,
- Stephen Friese, emergency physician4,
- Agustina Garzon, research physician1
- 1Department of Emergency Medicine, Kern Medical Center, Bakersfield, CA 93306, USA
- 2Centre for Research in Population Health, National Institute of Public Health, Cuernavaca, Morelos, Mexico
- 3Pharmacy Department, Kern Medical Center, Bakersfield, CA 93306, USA
- 4Department of Emergency Medicine, Community Memorial Hospital Center, Ventura, CA 93003, USA
We were disappointed to see a Cochrane review recommend adrenaline and steroids in the outpatient management of bronchiolitis.1 The results of the meta-analysis reflect selection criteria excluding randomised controlled trials which do not support the author’s beliefs rather than the available data.
The stated rationale for excluding studies in which infants have had previous wheezing was to minimise including infants who might later develop asthma (undefined). This is illogical. Firstly, many viruses cause bronchiolitis. Infection does not prevent re-infection and a repeated episode of bronchiolitis. Secondly, studies that have included infants with previous episodes have used this variable and controlled their analyses accordingly.2 Thirdly, the correct assignment of the “first episode” depends on accurate parental understanding of wheezing.
Even if excluding infants with recurrent bronchiolitis were valid, individual infants rather than entire studies could have been excluded. Similarly, including only studies with outcomes at days 1 and 7 excludes those with primary outcomes at day 3, and severely limits this meta-analysis.
One excluded study (n=75) shows no significant difference between adrenaline and placebo.3 Another shows a relative risk of 1.18 favouring salbutamol, even after adjustment for recurrent episodes.2 This well-designed trial (n=703), dwarfs the six outpatient studies (combined n=295) that were included. The question has been answered: salbutamol is the better bronchodilator.
We also disagree with the assertion: “These two large (steroid) trials . . . provide a strong signal for further synthesis work.”1 The study of Corneli et al (n=600) shows no benefit from steroids; even their post-hoc subgroup analysis (adjusted for multiple comparisons) was not significant.4 It was important to ask if steroids would help. The answer is No.
Many of us have spent years testing these hypotheses. Now we should follow the data rather than insist that there is some subgroup in whom our initial beliefs are correct.
Cite this as: BMJ 2011;342:d3348
Competing interests: The authors have conducted original randomised controlled trials and related research in the field.