Abstract
We compared the efficacy of diclofenac potassium in unpublished clinical study reports (CSRs) and published reports to examine publication bias, industry bias, and comprehensiveness. Novartis provided CSRs of randomised double-blind trials of diclofenac potassium involving postoperative patients following third molar extraction (3 trials, n=519), gynaecological surgery (3 trials, n=679), and dysmenorrhoea (2 trials, n=711) conducted in 1988-1990. Searches identified published reports of 6 trials. Information from 599/1909 patients was not published; trials with 846/1909 patients were published in a defunct journal. Greater methodological information in CSRs contributed to lesser risk of bias than published trials. Numbers needed to treat (NNT) from CSRs for all six postoperative trials for at least 50% of maximum pain relief over 6 h were 2.2 (95% confidence interval, 1.9-2.6) and 2.1 (1.8-2.4) for 50 and 100 mg diclofenac potassium, respectively. A Cochrane review of published trial data reported NNTs of 2.1 and 1.9, and one comprehensive analysis reported NNTs of 2.2 and 2.1, respectively. All analyses had similar results for patients remedicating within 8 h. No data from dysmenorrhoea CSRs appeared in a Cochrane review. CSRs provide useful information and increase confidence. Stable efficacy estimates with standard study designs reduce the need for updating reviews.