Confidence intervals, P values, and statistical significanceBMJ 2015; 350 doi: https://doi.org/10.1136/bmj.h1113 (Published 27 February 2015) Cite this as: BMJ 2015;350:h1113
- Philip Sedgwick, reader in medical statistics and medical education1
- 1Institute for Medical and Biomedical Education, St George’s, University of London, London, UK
The efficacy of nicotine patches in pregnant women who smoked was investigated using a randomised placebo controlled trial. The intervention was the administration of 16 hour nicotine patches until the time of delivery. Participants were pregnant women over 18 years who smoked at least five cigarettes a day and whose babies were between 12 and 20 weeks’ gestation. In total, 402 women were recruited from 23 maternity wards throughout France. Participants were randomised to the intervention (n=203) or placebo patches (n=199).1
The outcome measures included achievement of complete abstinence until delivery and birth weight. Complete abstinence was achieved by 5.5% (n=11) of women in the nicotine patch group and 5.1% (n=10) in the placebo group (odds ratio 1.08, 95% confidence interval 0.45 to 2.60). The mean birth weight was higher in the nicotine patch group (3065 (standard error 44 g) v 3015 g (44 g); difference 50 g, −71.1 to 172.3).
Which of the following statements, if any, are true?
a) The odds ratio for abstinence until delivery was statistically significant at the 5% level because the associated 95% confidence interval did not straddle zero
b) The difference between treatment groups in mean birth weight was not statistically significant at the 5% level because the associated 95% confidence interval straddled zero
c) A 95% confidence interval provides a test of the statistical hypotheses at the 5% level of significance.
Statements b and c are true, whereas a is false.
The odds ratio comparing the intervention group with the placebo group in smoking abstinence until delivery was 1.08 (0.45 to 2.60), whereas mean birth …