Reply to Bustamante et al. and Sauerland and Korenkov

We appreciate the additional calculations based on our data offered by Bustamante et al. [1], which provide further evidence that the sensitivity of capsule endoscopy was superior to that of push enteroscopy in our study.

Like Sauerland and Korenkov [2], Bustamante et al. criticize the application of Student’s t-test to dichotomous data (”0 = source of bleeding not detected” and ”1 = source of bleeding detected”). The analysis of dichotomous data using the t-test is well justified in statistical theory, in which the t-test and chi-squared test are regarded as variants of the same analysis model. The major alternatives to the t-test are tests for symmetry, mainly the Bowker symmetry test [3] for contingency tables - but not the McNemar test as recommended by Bustamante et al., which is a test for 2 × 2 tables. The Bowker test yields a chi-squared figure of 12.0192 (applying continuity correction) with three degrees of freedom; the corresponding P value is P = 0.0073, which also impressively indicates the greater sensitivity of CE in comparison with PE. The reason why we preferred the t-test and did not use the Bowker test was that, with one exception, all five of the other expected frequencies in Table 4 were below a critical value of 5 - a condition that in a strict sense does not allow the application of a chi-squared test. In recommending the chi-squared contingency test or Fisher’s exact test, Sauerland and Korenkov disregard the fact that these tests require independent observations; however, in our study, the assessments of capsule endoscopy and push enteroscopy were dependent measures derived from the same patients. Independently of that argument, it should be noted that Fisher’s exact test is applicable to 2 × 2 tables, while our data include three categories for each diagnostic measure. The same argument based on dependent capsule endoscopy and push enteroscopy measures from the same patients also applies to the calculation of odds ratios, as recommended by Sauerland and Korenkov.

In our study, capsule endoscopy was administered first, followed by push enteroscopy. The reason for this procedure was to avoid lesions induced by push enteroscopy in the upper small bowel that might have been incorrectly regarded as possible bleeding sources (for example, tiny angiodysplasias are difficult to distinguish from push enteroscopy-induced lesions of this type). It was therefore also not possible to mark the deepest point reached on push enteroscopy and then to carry out an analysis of the diagnostic yield of push enteroscopy and capsule endoscopy in this part of the small bowel. We agree with the authors that this is an interesting point, which cannot be answered by our study.

Randomization of the two diagnostic methods would be one way of controlling systematic bias resulting from the sequence of the two methods. We used blinding of the evaluation as an alternative. Push enteroscopy and capsule endoscopy were carried out by independent physicians who were unaware of the findings with the other method. However, these authors’ recommendations might be considered for further trials as an additional way of controlling possible bias.

Our data can be used for sample size calculation in forthcoming studies, as suggested by Bustamante et al. At the time when we started the trial, no information regarding a valid formal sample size calculation was available except for an animal study. We therefore conducted the trial with the cohort of all patients with chronic gastrointestinal bleeding who presented at our department over a period of 6 months. In the process of scientific research, exploratory studies of this type are justified to govern both the direction of further research and the responsible application of new methods, both in our own patients and in those referred to us by other physicians. We would certainly encourage further (multicenter) studies investigating the sensitivity and specificity of the diagnostic approach using capsule endoscopy - taking different segments of the small bowel into account, for example.

References

• 1 Bustamante M, Galvez C, Higón M D, Siles S. Controlled trial of wireless capsule endoscopy versus push enteroscopy: control not yet perfect.  Endoscopy. 2003;  36 71-72
• 2 Sauerland S, Korenkov M. Capsule endoscopy in chronic gastrointestinal bleeding.  Endoscopy. 2003;  36 71
• 3 Bowker A H. A test for symmetry in contingency tables.  J Am Stat Assoc. 1948;  43 572-574