Interpretation of randomised trials is indeed open to debate

Editor—McCormack and Greenhalgh once argued that the interpretation of randomised controlled trials is open to debate.¹ A recent BMJ news item featuring an overly favourable comment on the diabetes subtrial of the heart protection study illustrates this perfectly.^2-4 We question the study's relevance to general practice on four points.

Generalisibility—The diabetes subtrial is part of the main study published in 2002. The researchers in the original trial excluded more than two thirds of the original 63 603 patients. Most of them opted out or were deemed not to be reliably compliant for the trial.

Bias—Analysis of the study's design raises the possibility of a non-match between treatment and control group⁵; there is a potentially important difference in the dropout rate.

Merging the boundaries—Both studies look at patients with and without preexisting cardiovascular disease. This is misleading as patients with established disease have a higher chance of benefit from lipid lowering treatment. We disagree with the impression in the BMJ news item that the case for using statins in diabetic patients without cardiovascular disease who have low density lipoprotein cholesterol, < 3.0 mmol/l, is made; the 3.1% absolute difference between treatment and placebo group is only marginally significant (P = 0.05%).

Presentation of results—Relative risk reduction and composite end points as seen in both studies exaggerate the potential benefit of a proposed treatment and should not be used in serious medical journals.