Card and colleagues (Gut 2004;53:246–50) reported that the use of antibiotics could increase the risk of Crohn’s disease, particularly those prescribed 2–5 years prior to the diagnosis (odds ratio 1.32 (95% confidence interval 1.05–1.65)). As the use of drugs acting on the central nervous system and of other prescription drugs such as oral contraceptives was also associated with a diagnosis of Crohn’s disease, the authors concluded that this association is non-specific. Nevertheless, some methodological issues could have had an important impact on the results.

One methodological aspect of the study design that could have biased the findings is that calendar time was not accounted for properly. This is highlighted by the fact that the median time available prior to the index date, the period during which exposure to antibiotics was measured, was 6.4 years for the cases compared with 8.2 years for the controls. This difference of almost two years is due to the fact that the index date was correctly taken to be the date of diagnosis of Crohn’s disease for the cases but was taken to be the date of the end of data analysis for the controls. Consequently, a case diagnosed in 1992 could potentially be compared with a control whose date of end of data analysis was in 1998. As a result, the odds ratio based on the comparison of cases and controls for exposure to antibiotics may be biased by comparing exposures to antibiotics that may have changed over calendar time. Thus any trends over time in the patterns of use of antibiotics during the calendar time span of the study, namely the 1990s, would bias the estimate. Such bias cannot be excluded, particularly for an odds ratio as small as 1.32, because the design did not match the index date on calendar time nor did the data analysis adjust for calendar time. In fact, such widely changing patterns have been observed in this population for oral contraceptive usage after a pill scare in the mid-1990s.¹

A second limitation of the study is the unclear accounting of age. On the one hand, there is the issue of matching on age of cases and controls, while on the other there is the issue of adjustment for age in the analysis. Firstly, it is not clear at what time these two ages were measured, and particularly whether age was taken at the index date. We suspect this choice because cases were 2.1 years younger than controls, which is similar to the difference of almost two years in the index dates of cases and controls. Secondly, age matching was based on 20 year age bands, which may be too wide to control fully for confounding, unless the precise age was used in adjustment.