Sirs, The article by Thalmaier et al.1 on +1059G/C polymorphism in the C-reactive protein (CRP) gene and Crohn's disease, which was recently published in Alimentary Pharmacology & Therapeutics, came to the conclusion that this CRP gene polymorphism is associated with Crohn's disease. We argue that this conclusion was based on inappropriately applied statistical analyses.
The comparisons of data shown in Table 3 assumed independence between the observations; in fact there is not. An individual, whether case or control, can be GG, GC or CC. Thus, the appropriate chi-square value should be derived using the entire 2 × 3 table and not for each genotype individually. Thus, for the 2 × 3 table the chi-square value is 4.2, which is not significant (2 d.f.).
Likewise, within any given ‘disease localization’ (Table 2), an individual can be GG, GC or CC. Thus, the percentages should have been obtained for all subjects with a given disease localization (n = 147 people for L1 and L3) and not the other way around, i.e. for those with CC what type of disease localization they have (n = 5). Done in that manner the authors found that within five subjects with CC, two (40%) have terminal ileum (L1), two have ileocolon (L3, 40%) and one had colon involvement. Indeed, if the table is set up properly, it would have shown that the majority of patients with ileal disease are not CC (n = 4) but rather GG (128 of 147, or 87%); when properly analysed (chi-square in a 2 × 3 table with continuity correction for small number in some cells), the P-value does achieve significance but the interpretation is totally different from the one presented by the authors, i.e. the majority of patients with ileal involvement are GG. Moreover, even if one accepts the analyses depicted in Table 2, it is dangerous to draw conclusions based on five subjects who are CC.