unchanged when antibiotic-sensitive GNB are detected. The ‘sensitivity’ of surveillance cultures employed in this manner has little clinical value although it may demonstrate intestinal translocation or exogenous infection. Thirdly, all screening tests are assessed by the harm: benefit ratio. Performing complete identification on multiple GNB will significantly increase ‘harm’ by diverting resources with only a modest effect on ‘benefit’. Hence, we believe the absence of identification is not a weakness. “Public Health Laboratory and TDepavtment of Child Health, St George’s Hospital, London S W17 O&T, UK
P. Y. C. Lee* R. E. Holliman” E. G. Daviesj-
We believe that the statistical analysis methodology described by Palmer et al.’ that they used to compare the two regimes in patients undergoing colorectal surgery has been carried out incorrectly. The error may have come from a misunderstanding about how a two-sided P-value should be calculated for this test. The two-sided P-value reported in the paper seems to be based on twice the one-sided P-value. While this would be the right thing to do for many tests, it is not correct for the Fisher’s exact probability test. On correct recalculation the P-value becomes 0.045. Hence, there is a statistically significant difference at the 5% significance level in postoperative infection rates in this type of surgery in favour of the cefuroxime plus metronidazole arm. It is essential to point out this mistake as this study provides evidence that, for patients undergoing colorectal surgery, co-amoxiclav gives a statistically inferior cover to cefuroxime plus metronidazole. Failure to appreciate this may put patients at risk. A. Cheesbrough
Statistics and Data Management, Glaxo Laboratories Ltd, Uxbridge, Middlesex UBll IBT, UK
References 1. Palmer BV, Mannur KR, Ross WB. An observer blind trial cefuroxime plus metronidazole in the prevention of postoperative general surgery. J Hosp Infect 1994; 26: 287-292.