Sir,

We read with interest the article of Tsang CSL et al1 in which they compared the visual field and optic nerve head changes in normotensive patients with obstructive sleep apnoea (OSA) to age- and sex-matched controls without OSA. Unfortunately, we feel that some aspects of their analysis could have led the authors to invalid conclusions.

Comparing eyes in the two groups, the authors found statistically significant differences (P<0.01) in both visual field (VF) indices using unpaired t-tests. However, as most patients have contributed both eyes to the analysis, eye-specific outcomes cannot be considered independent. Failure to account for this dependence results in standard errors that are too small and P-values that are too significant.2 For this reason alone, the author's conclusions must be considered questionable. Further, the authors despite having identified deviations from normality in the distribution of both VF indices in the OSA patients have used a test that assumes normality. One approach that addresses both the non-normality and the dependence in the data would be to construct bootstrap confidence intervals (CIs)3, 4 for the difference in means allowing for clustering at the subject level. The authors have repeatedly used the term CI but none have been reported. From the figures shown, there does not appear to be a large difference in either VF index between the two groups and no estimates of this difference have been provided. Reporting the differences in means with CIs would have demonstrated whether any observed differences are clinically relevant.

There are also inaccuracies in the description of the study design methodology. The term ‘incidence’ of suspicious disc change is referred to in the title and in the text, whereas in fact the study investigates prevalence. This distinction in terminology is important and should not be confused. Furthermore, the term case–control study should be reserved for studies where subjects are selected on the basis of outcome and not exposure (as in this case, ie, presence/absence of OSA).

In summary, this paper demonstrates that collaborations between statisticians and other researchers are vital in order to avoid preventable mistakes that can render the results of otherwise well-conducted studies uninterpretable.