To the Editor I read with interest the article by Birnbaum et al titled “Correlation Between Clinical Signs and Optical Coherence Tomography With Enhanced Depth Imaging Findings in Patients With Birdshot Chorioretinopathy.”1 The authors examined 14 patients; we are told that 21 eye examinations were done in symptomatic patients, but there was no mention of the number of examinations done on asymptomatic patients. Even though there were 14 patients, the authors stated there were 58 eyes. For example, their Table 1 states that 33 eyes did not have photopsias while 25 did. (Photopsias were graded on a 7-step scale ranging from 0-3 by half-step increments using an undisclosed method; 25 eyes had a score >0.) Similarly, 58 eyes’ worth of optical coherence tomographic data were generated for 4 different features as shown in their Table 2. The implication is not only that there was a potential to use both eyes of a patient but also that the same eye was measured more than once for some patients. Since 58 is not evenly divisible by 14 or 21, there must have been a variable number of examinations per patient and a variable number of eyes used per patient. The statistical tests used in the study (the Pearson product moment correlation and the Spearman rank correlation) have as a basic assumption independence of the observations from each other. These statistical tests would be appropriate if 58 eyes of 58 patients were analyzed; instead, both eyes of an untold number of patients were measured more than once an untold number of times. These steps can lead to biased estimates of the strength of the association because the cases with multiple measurements are no longer random samples from some larger population. The estimates of the significance of the correlation are also incorrect because there is an artificial expansion of the number of cases by measuring eyes more than once, making us more likely to believe the observed correlation would not have happened by chance alone. The analysis of data with this structure requires a more sophisticated approach. Methods with generalized estimating equations have been used for this type of data.2 The authors should provide additional analysis that correctly accounts for the inconsistent use of 2 eyes per patient as well as the measurement of the same eye repeatedly in some patients. In addition, it would be helpful to know how the authors retrospectively graded photopsias into 7 levels of severity.
Spaide RF. Fourteen Patients With Fifty-Eight Eyes. JAMA Ophthalmol. 2015;133(3):357. doi:10.1001/jamaophthalmol.2014.4650