[Helicobacter eradication: an expensive Sisyphus task]

The benefits of eradicating H. pylori infection vary about the antimicrobial regimen. In contrast, comparing many clinical studies the overall outcomes seem independent of the kind of treatment. Therapy depends on accurate diagnosis. Therefore, the observed differences and correspondences can be caused by not obviously but systematic influences using noninvasive diagnostic tests.

Simulations concerning noninvasive tests with variable specificity and sensitivity were performed. The results were compared to the hypothetical estimate that 1/3 of the population of Germany were H. pylori infected. This group should be traced and checked again after a treatment with 90% efficacy.

In comparison to the a priori conditions most decisions according to noninvasive tests were not valid. Both sensitivity and specificity differ significantly from 100%, therefore the accuracy was low. The differences between true positives and true negatives were unpredictably accelerated if the inclusion criteria depend on the poor sensitivity of the first test. For the second test which is commonly used regarding its specificity, there is a change in the statistical behavior. As a result, the homogeneity that is true and false negatives and positives of the non-infected or infected groups differed considerably. One of the examples was a procedure using a test with 79% specificity and 85% sensitivity. If the prevalence of the infection was 33% the simulations demonstrated a therapeutic effect of about 76.5% instead of a priori efficacy of 90%. Under these conditions 14% of the population become false positive. Furthermore, the simulations revealed a second effect. Using the same test with low accuracy twice the investigator got the impression of superior therapeutic considerations. The misleading conclusion was caused by diminished sensitivity if the specificity remained constant. Thus, neither serological tests nor the 13C-breathtests can be assumed as validated methods for both to screen infected people and to check the treatment. Furthermore, the noninvasive tests are not useful for the determination of the prevalence of H. pylori infection. The influence of variable sensitivity and specificity of a diagnostic test on the outcome of a clinical trial can become a bias or can be used to manipulate the outcome.

The effect of a therapeutic regimen depends on the prevalence of the H. pylori infection. Because it is difficult to separate true positives, in clinical trials the main effects and interactions can be improved under the conditions used in "experimental design". Some earlier results must be reevaluated because of the bias resulting from inadequate diagnostic tests.