The Pharmaceutical Journal
Vol 271 No 7258 p85
19 July 2003

This page
Reprint
Photocopy

PDF* 110K

  Unlicensed medicines
  Branding
  Patient packs
  Workforce
  Veterinary pharmacy
  Modernisation
  The Society
  CPD
  The Journal

Letters to the Editor

The Journal

Absence of evidence is not evidence of absence

From Dr P. H. Rowe

A recent news item in The Journal (12 July, p39) provides a nice example of the way in which non-significant statistical results can be misunderstood. The item concerns a study of oral contraceptive use and stroke. The study result was reported as a 95 per cent confidence interval for the odds ratio. The null hypothesis (exactly equal risk of a stroke in users and non-users of oral contraceptives) would be represented by an odds ratio of 1.0. The upper and lower limits were 0.86 to 3.61. The magic figure of 1.0 is therefore included within this range and we are forced to conclude that the null hypothesis cannot be discounted.

However, the title of the article claims "Study dispels fears over contraceptive pill use and stroke". This implies that safety has been positively demonstrated — an extension that cannot be justified. If we wanted to go that extra step and "dispel fears", the correct approach would be to agree upon what odds ratio would be small enough not to be a source of fear and then inspect the upper limit of the confidence interval. If, for example, we had agreed that anything less then a four-fold increase in the odds was acceptable, then the upper limit of "only" 3.61 would be tolerable and our fears would truly be dispelled.

The general point is that any non-significant result only means that we have not shown an effect to be present, it does not mean that we have shown it to be absent. Or, as is frequently said in bars up and down the country, "absence of evidence is not evidence of absence".

Phil Rowe
Reader in Pharmaceutical Computing
John Moores University

Send your letter to The Editor

Previous Topic (CPD)