Reviewer 3's Review

-----

1. Would it not have been possible to use a double placebo design to have some chance of true double masking? Without this, it is hard to separate the effects of the treatments from the effects of biases resulting from the lack of masking.

2. What were the results of the seven patients who were improperly randomized? A red flag could be raised if these seven differed markedly from the rest (as in, all responded, or none responded), but this red flag can be ruled out by presenting the responses of this group, and noting its lack of extremity, or difference from the others.

3. The primary variable seems to be the scale with five outcome levels, cure, re-infestation, O failure only, L failure only, double failure. This variable should be tabulated. In addition, it would appear that this is a partially ordered endpoint, as cure>re-infestation>single failure>double failure, but can one objective compare the two single failures? Without this, the endpoint is partially ordered, and special methods of analysis are required. These methods are a generalization of the U-test methodology you already used. See:

Berger, VW, Zhou, YY, Ivanova, A, Tremmel, L (2004). "Adjusting for Ordinal Covariates by Inducing a Partial Ordering", Biometrical Journal 46, 1, 48-55.

This paper treats a different problem, and so is not directly applicable. But it does demonstrate a method for deriving an analysis that is appropriate.

-----

N.B. These are the general comments made by the reviewer when reviewing this paper in light of which the manuscript was revised. Specific points addressed during revision of the paper are not shown.