Statistics from Altmetric.com
We appreciate the concerns raised by Sandberg et al in their recent letter1 pertaining to the design and statistical methods of our analysis described in our recent article,2 and we welcome the opportunity to respond to each point raised.
We agree that ideally the best ‘gold standard’ study design would involve the use of matched controls, matching on eye function at baseline, as we plan to do in our upcoming clinical trial. For preliminary pilot studies, however, other study designs, including methodologies that do not have controls, can also be meaningful, especially when the objectives are to assess treatment potential and establish the equipoise necessary for further investigation in a randomised control study. An example that is relevant to our work is presented by an article from one of the authors of the letter (Rosner et al; Section 4.1),3 in which the investigators assessed the potential benefit of treatment and concluded that their findings were suggestive of a lack of effect due to treatment. These researchers did not have a control group, yet they felt comfortable with the conclusions based on this design. Our analytical design and findings are consistent with this example. Our focus was an exploration of potential therapeutic value, and our detailed description of visual function in 14 eyes establishes a possible treatment benefit that merits a randomised control trial.
We believe that the exploration of treatment potential in a sample size of seven is best addressed with detailed descriptions rather than formal tests of statistical significance. Given that significance levels were reported, however, we agree with Sandberg et al that the unit of analysis should have taken into account the correlation structure of the data. We thank Dr Rosner for pointing out the modification of the Wilcoxon signed-rank test for use with …