Controlled Trial of Wireless Capsule Endoscopy versus Push Enteroscopy: Control Not Yet Perfect

We read with great interest the article by Ell et al. on ”The first prospective controlled trial comparing wireless capsule endoscopy with push enteroscopy in chronic gastrointestinal bleeding” [1], and would like to make some comments on it. Previously, a randomized trial in dogs [2] and a few case reports [3] had suggested that capsule endoscopy is effective, but there had been a lack of substantial evidence for including the wireless capsule in the diagnostic work-up for patients with chronic gastrointestinal bleeding. Publication of a well-designed trial in patients with gastrointestinal bleeding was therefore eagerly awaited.

Although the push enteroscopy technique is far from representing a true gold standard, it is the reference method against which any new device for exploring the small bowel needs to be compared. However, the fact that push enteroscopy only reaches the proximal segments of the small bowel has to be taken into account; obviously, the global sensitivity will be always higher with capsule endoscopy, as it is able to explore the entire small bowel. This was shown in the report by Appleyard et al. [2] using an animal model, in which the overall sensitivity of push enteroscopy was 37 % compared with 64 % for capsule endoscopy; within the range of the push enteroscope, however, the sensitivity rates were 94 % vs. 53 %, respectively. In the report by Ell et al. [1], the global diagnostic yield of capsule endoscopy was significantly higher (66 % vs. 28 %), but the sensitivity in the area within the range of the push enteroscope was not reported. It appears from the paper that only one proximal lesion detected with push enteroscopy was missed on capsule endoscopy, but a better description of the lesions and their location would have been needed to provide an idea of the diagnostic yield of capsule endoscopy within the range of the push enteroscope. The difference in sensitivity in the proximal small bowel might not be very significant in clinical practice, but its actual relevance and magnitude can only be assessed with a good study design.

Another problem is that when the expected differences are so minute, the size of the sample being used to detect them is decisive. For instance, if a 90 % rate of positive diagnoses is expected for capsule endoscopy and a 98 % rate is expected for push enteroscopy within the instrument’s range (as would be expected if only the proximal bowel were to be analyzed), then approximately 134 patients would be needed to have a 95 % of chance of detecting the difference. However, if we use the calculated sensitivity values for the whole bowel (66 % and 28 %), only 23 patients would be needed. In a controlled trial, an a priori calculation of the sample size would therefore have been necessary in order to be relatively sure that capsule endoscopy does not miss any relevant lesions in the proximal bowel.

Finally, we are concerned about the description of the statistical methods. The authors code the diagnosis as ”0” or ”1” in a binary way, so that proportions were being compared. The McNemar test would therefore have been appropriate, instead of Student’s paired t-test, which is suitable for comparison of continuous variables between paired groups [4].

References

• 1 Ell C, Remke S, May A. et al . The first prospective controlled trial comparing wireless capsule endoscopy with push enteroscopy in chronic gastrointestinal bleeding.  Endoscopy. 2002;  34 685-689
• 2 Appleyard M, Fireman Z VI, Glukhovsky A. et al . A randomized trial comparing wireless capsule endoscopy with push enteroscopy for the detection of small-bowel lesions.  Gastroenterology. 2000;  119 1431-1438
• 3 Hahne M, Adamek H E, Schilling D. et al . Wireless capsule endoscopy in a patient with obscure occult bleeding.  Endoscopy. 2002;  34 588-590
• 4 Dawson B, Trapp R G. Basic and clinical biostatistics. 3rd ed.  Norwalk, CT; McGraw-Hill/Appleton and Lange 2000