Karakosta and associates highlight the importance of taking into account interocular correlation when both eyes of the same subject are included in statistical calculations. We would like to add that such an adjustment also could be important when dealing with summed data that can contain multiple contributions from individual eyes. When assessing characteristics of lesions of interest (tumors, nevi, retinal breaks, corneal opacities, etc.) within the eye, where one eye may contain multiple such lesions, each lesion should not necessarily be treated as an independent entity, given that it may well be the case that lesion characteristics within the same eye could be correlated. The degree of correlation will differ for different parameters.
Epidemiologic studies of choroidal nevi, for example, have yielded proportions of nevi with various characteristics, without explicitly adjusting for intraocular correlation in all cases. Given the low proportion of eyes with multiple nevi, such a correlation may not be expected to affect the overall comparisons significantly, although the quantification of such correlations may be interesting in understanding pathogenesis. Studies of patterns of retinal detachments similarly often have counted the total number of retinal breaks over a cohort and then quantified the proportion in each quadrant, neglecting the likelihood that multiple breaks within the same eye may not be independent with respect to location, size, and so forth. The dynamics of vitreous detachment in a particular eye may be such that multiple breaks result in the same quadrant; simply to total the number of breaks in a single quadrant in a number of eyes would mean this becomes a confounding factor. A better method, when seeking to answer whether retinal breaks are more likely to be, for example, inferonasal in one population compared with another, may be to quantify the proportion of eyes with presence (or absence) of an inferonasal break in each population (rather than simply comparing the number of inferonasal breaks in total), and then to consider numbers of breaks separately.
Thus, some consideration of within-eye correlation not only may affect the validity of calculated P values and thus the presumed statistical significance of clinical findings, but also may shed light on the pathogenesis of those findings.