Contact Dermatitis, 1999, 40, 14–18 Copyright C Munksgaard 1999 Printed in Denmark . All rights reserved ISSN 0105-1873 The association between size of test chamber and patch test reaction: a statistical reanalysis O G 1 ,A P 1 ,J G 2 ,J B 3 W U 4 1 Department of Medical Statistics, University of Go ¨ ttingen, Germany; 2 Information Network of Departments of Dermatology, University of Go ¨ ttingen, Germany; 3 Department of Dermatology, University of Kiel, Germany; 4 Department of Dermatology, Environmental Medicine and Health Theory, University of Osnabru ¨ck, Germany A recent study by Brasch and co-workers reported on the association between size of test chamber and patch test reaction. The investigators interpreted their data on 495 patients as having conclus- ively shown that standard preparations of fragrance mix, wool wax alcohols, Kathon CG and formaldehyde led to more positive test reactions when large Finn Chambers were used for patch testing. We have scrutinized the statistical aspects of this study and conclude that the authors should have adopted a statistical approach suitable to analyse dependent samples. After explaining the correct methodological way of dealing with quadratic contingency tables formed by 2 depend- ent samples, we reanalyze the data accordingly and compare the results to those of the original paper. Based on this reanalysis, the conclusions are more complex: the reaction pattern for the fragrance mix and wool wax alcohols is significantly different between small and large test chambers; however, this discrepancy arises primarily from changing weak positive reactions with small chambers to strong positive reactions with large chambers. For formaldehyde, no relationship between chamber size and patch test reaction was found in the data, while for Kathon CG, statisti- cal evidence is borderline that more positive test reactions are yielded by large test chambers than by small ones. Key words: dependent two-sample design; patch testing technique; reaction pattern; statistics; test chamber size; fragrance mix; wool wax alcohols; Kathon CG; formaldehyde. C Munksgaard, 1999. Accepted for publication 6 August 1998 Valid conclusions from empirical data require valid statistical procedures for analyzing the data. The appropriate statistical technique has to be selected carefully to reflect precisely the research question to be answered and the study design employed to obtain the data. In a recent paper in this journal, Brasch et al. (1) considered the interesting topic in patch testing as to whether the size of the test chamber affects the test reaction. Based on a study comprising 495 patients who were synchronously tested with 4 standard allergens (fragrance mix, wool wax alcohols, Kathon CG, formaldehyde) with small (8 mm diameter) and large (12 mm di- ameter) test chambers, they compared the test re- sults. They observed that ‘‘for each of the 4 aller- gens, a significantly higher number of stronger re- actions was seen with the large chambers as compared to the small ones’’ (1, p. 118) and thus concluded ‘‘that large test chambers may be useful in the detection of weak sensitization to allergens that are characterized by a low reproducibility of patch tests with small chambers’’ (1, p. 120). We reconsider the same data, paying special attention to the statistical problems raised by the analysis of dependent samples. Particular empha- sis is laid here on a detailed description of the methodological aspects of analysing such patch test data, which also apply to other empirical re- search in patch testing utilizing a similar design. For the data by Brasch et al. (1), reanalysis em- ploying the correct statistical approach for depend- ent samples yields different conclusions with re- spect to the effect of the size of the chambers on the test reactions to those published in the original paper. Materials and Methods Details of study design, selection of patients, prep- aration of allergens for the 2 sizes of test chambers