Letters

Randomised trials useful to find best methods of enhancing clinical practice

EDITORAlthough we are flattered that Keirse devoted his whole editorial to our study on the use of educational visits to enhance the use of systematic reviews, we wish to correct some errors. ^{1 2} Tossing a coin is not inherently a biased method of randomisation, but any open allocation method may lead to selection bias.³ In our cluster trial we randomly assigned 25 hospitals that had already been recruited to either an educational visit or to a control group, and we blinded data collectors to allocation and outreach visitors to the clinical practices designated as markers. Allocation concealment in such a study is superfluous.

Keirse analysed denominators retrospectively to detect differential data completeness. This analysis is not valid as it ignores the clustering. Excluding the use of steroids, data were obtained for 97% and 95% of the target numbers in the intervention and control groups respectively. Further, follow up data were collected blind to the allocation of the unit which prevented bias.

Keirse states that "22 of the 25 units had a rate of use of ventouse extraction at baseline that was either at or outside the 95% confidence interval for the average." We find this a strange criticism as there is no expectation that observations will lie within the confidence interval. The confidence interval expresses a range of uncertainty for the pooled rate and conveys no information about variation among units.

Keirse comments that 30 cases per unit do not provide an adequate snapshot of clinical practice. A larger sample size per unit would have helped to reduce random variation between and within centres but was not feasible. Our interest was the uptake of Cochrane evidence reflected by the four clinical practices designated as markers. We studied 4508 patients, 92 per unit at baseline and another 88 at follow up. Each audit represented two weeks of clinical practice per unit or one unit year of observation at baseline and a further unit year at follow up.

Finally, Keirse states that the reviews whose impact we studied were published years earlier and that practitioners with any interest would have looked up the results. Keirse himself later correctly states that "perhaps it is too simplistic to expect that merely exposing practitioners to evidence will change practice."⁴ Our goal was to promote the interest of lead practitioners in evidence, so we did not distribute review results or dictate which evidence was useful nor how to use it. We agree with Keirse that the best methods for enhancing clinical practice are far from clear, but conclude that rigorous, randomised trials like ours are necessary to explore this issue; one which is important to all healthcare systems.

Jeremy C Wyatt, Senior fellow in health and public policy. 
School of Public Policy, University College London, London WC1E 7HN

Sarah Paterson-Brown, Consultant obstetrician. 
Nicholas M Fisk, Professor of obstetrics and gynaecology. 
Institute for Obstetrics and Gynaecology, Imperial College School of Medicine, Queen Charlotte's and Chelsea Hospital, London W6 0XG

Richard Johanson, Senior lecturer in obstetrics. 
Academic Department of Obstetrics and Gynaecology, City General Hospital, Stoke on Trent ST4 6QG

Douglas G Altman, Head. 
Michael Bradburn, Statistician. 
ICRF Medical Statistics Group, Centre for Statistics in Medicine, Institute of Health Sciences, Headington, Oxford OX3 7LF