The blog series is expanding! No doubt soley inspired by now running the magnificent @ADC_JC, @davidking83 has taken up the challenge of exploring a critical appraisal nugget/thorn in response to an appraisal session.
You too could be part of our team – tweet @ADC_BMJ or find us on Facebook to get in touch – but for now, let’s settle back, imaging the warm smell of pastries and coffee, switch off our pagers and enter the Journal Club…
We were recently discussing a paper in my local journal club at Sheffield Children’s hospital,which was a double-blind randomised controlled trial (RCT) examining the effect of pulse oximetry on the admission rate of infants with mild or moderate bronchiolitis. This is a fascinating paper and well worth a read if you have not already. Essentially, it showed that in infants with mild to moderate bronchiolitis, clinicians are less likely to admit children if their oxygen saturations are artificially raised by 3%.
However, one of the points raised from the journal club was the possible effect of selection bias on the results of the trial. The authors state that the parents of a large number of eligible infants refused to take part and this may have introduced selection bias.
The question was raised as to if this made the study less valid.
Part of the problem here, is exactly what is meant by selection bias. It’s mainly used to describe the selection of patients to receive a therapy – so in the context of an RCT, preferentially allocating patients with similar prognostic factors to the same arm of the clinical trial. This would result in a treatment appearing more or less effective than it actually is. In the trial mentioned above, an example of this type of selection bias would be deliberately allocating more babies with mild rather than moderate respiratory distress to the group with falsely elevated oxygen saturations.
True randomisation, followed by good allocation concealment, by definition, should not allow this to occur. A properly double-blind RCT will have and this type of selection bias should therefore be eliminated in a double-blind RCT. However, there is a subtype of selection bias, known as ‘sampling bias’ which could create problems. To stick with this paper, hypothetically it could be that the study was only performed in infants with oxygen saturations of >95%, mild respiratory distress and in families with non-smokers as these were the people most likely to agree to take part in the research study*. This would make the results less generalisable to your own population but otherwise it does not affect the validity of the results.
In summary, providing randomisation and blinding has occurred properly in a double-blind RCT most types of selection bias should not be a problem. However, sampling bias and therefore whether the results are generalisable to your own population, is something of which you need to be wary.
* the study population was actually much more representative of patients with mild to moderate bronchiolitis than in this example