Reader Comments

Post a new comment on this article

Referee comments: Referee 1

Posted by PLOS_ONE_Group on 10 Apr 2008 at 16:20 GMT

Referee 1's review:

N.B. These are the comments made by the referee when reviewing an earlier version of this paper. Prior to publication the manuscript has been revised in light of these comments and to address other editorial requirements.

PLoSONE Review

Paul et. Al., Atorvastatin for RRMS

The authors report the results of a clinical trial examining the effects of atorvastatin for the treatment of RRMS. The trial is important and the results should be published. I have several comments however that should be appropriately addressed prior to publication, so as to make the results transparent.

Major Comments:

I strongly suggest that the manuscript be prepared and structured using CONSORT principles. The PLoS HUB for Clinical Trials (formerly PLoS Clinical Trials) has examples and guidelines for reporting clinical trials.

I suggest having a CONSORT-type tree diagram displaying the flow of participants in the trial.

I strongly suggest providing confidence intervals for effect sizes with all cited p-values.

The ITT population is not properly defined. The ITT population includes all participants that you intended to treat (41) and not patients that were treated (36). You might call the 36 an “as treated” or “per protocol” population. All 41 patients are part of the treatment strategy. Although it “does not seem reasonable” to hold the treatment responsible for disease exacerbation, it is also not reasonable to simply exclude patients that do not do well. Perhaps they are not doing well because the treatment strategy is failing. In accordance with ITT principles, the primary analyses should be conducted on all 41 patients whereas the “as treated” analyses can be secondary. View it this way: if a clinician has a patient entering treatment and we want to describe the expected outcome of the treatment decision, then we do not know if this patient will be more like the 36 or the 5. Thus providing the patient will only information about the 36 is a distortion, as they may more like the 5. You can then perform sensitivity analyses using various assumptions about the 5.

A clear explanation of how the number of CEL are calculated (pre and post) is needed. MRI scans were conducted 3 times/month and summarized over 2 months at baseline (i.e., 6 sequential scans) and 3 months at the end of the treatment period (i.e., 9 sequential scans). It is unclear how a single number is derived for both the pre and post periods.

I was unable to follow the statistical analysis strategy. More detail is needed regarding how the within-patient change was calculated, or the within patient correlation was accounted for, and how the multiple timepoints were used (or a composite was created). The statistical section refers to 3 baseline measurements and 4 treatment timepoints, but I do not understand how this relates to 3 MRI scans/month for a 2 month baseline period and a 3 month post treatment period. Please clarify the dependent variable and independent variables for the MANOVA and the scale of these variables.

Subgroup analyses: suggest clarifying that this was exploratory. Essentially these analyses should only be conducted when interaction effects are detected. I suggest looking for these interaction effects before concentrating on the subgroup analyses.

It is inconsistent to conduct a 2-sided test (and quote a 2-sided p-value) for a global effect, and then conduct 1-sided test for subgroups. I suggest that all tests be 2-sided.

Were MRI data collected in months 0-6? If so, these data should be analyzed.

Results: Treatment adherence and tolerability section, 1st sentence, I do not understand how one can make reliable adherence conclusions based on cholesterol.

Results: Treatment adherence and tolerability section, a need to add information and discussion about the attribution to atorvastatin of the events is needed.

Since this study has no control group, a discussion of the natural history of this disease is needed.

Tables, suggest using medians with Q1, Q3, min, and max as the order statistics are consistent with your nonparametric analysis and are more informative. Please also add effect sizes (confidence intervals).

Table 4: are the zeros really zeros or are these missing data? There’s a big difference!

Minor Comments:

Introduction: there is a reference to an atorvastatin trial in rheumatoid arthritis. Please clarify the relevance of this.

A clear description (or table) of the schedule of events may be helpful.

Please clarify the blinding and centralization of MRI reading. Was the reading done blinded to time (i.e., pre or post)?

Please be careful with the stated conclusions from non-significant tests. Please avoid statements of “no suppression”, “remain unchanged” “did not change”, as non-significant results do not imply no effect or no change. Confidence intervals will be needed to see what effect sizes can be “ruled out”.

Methods: Sample Size: please justify the assumptions (of the mean and SD for pre and post treatment) or cite where the estimates were derived.

Please comment on the fact that the CEL # for baseline was collected over 2 months, whereas the CEL # for post-treatment was calculated based on a 3-month interval. What is the potential relevance of this?

Please clarify how the “annualized relapse rate” is calculated.

Statistical analysis section: when defining the ITT population, replace, “the number of patients” with “the group of participants”.

Results, 2nd paragraph, I did not understand the sentence, “Better MSFC scores …”. Please re-phrase.

Discussion: 2nd paragraph, 1st sentence, if the MRI scans that were used as a baseline were different than the scans used for screening (I believe this was the case as this is why your baseline is a 2-month period rather than 3) then there is no risk for regression to the mean.

Funding, insert “of” after “to all” in the last sentence.