Choroidal thickness measurements in subjects on 3 spectral domain optical coherence tomography machines: common mistake in reproducibility analysis

Excerpt

We were interested to read the paper by Koay C. L. and colleagues published in June 2017 issue of the Int Ophthalmol [1]. The purpose of the authors was to investigate the reproducibility of choroidal thickness measurements in normal subjects on 3 spectral domain optical coherence tomography instruments, namely Zeiss Cirrus HD-OCT (Carl Zeiss Meditec Inc., Dublin, CA), Heidelberg Spectralis (Heidelberg Engineering, Heidelberg, Germany), and Optovue RTVue (Optovue Inc., Fremont, CA). Forty-seven eyes of 47 healthy volunteers without ocular pathology were imaged on the fovea using Cirrus HD 1-line raster, Spectralis enhanced depth imaging, and RTVue retina-cross. Paired t test and Pearson’s correlation were used to compare the results [1]. Although they correctly used modified Bland–Altman plot and limit of agreement; however, applying Pearson’s correlation and paired t test in Tables 2 and 5 are among common mistakes in their reliability analysis [2]. Briefly, for quantitative variable intraclass correlation coefficient (ICCC) and for qualitative variables weighted kappa should be used [2‐6]. Regarding reliability, it is crucial to know that an individual (subject)-based approach instead of global average (mean ± SD) should be considered [2‐6]. In other words, possibility of getting exactly the same average of a variable between two methods with no reliability at all is high [2‐6]. Moreover, they concluded that inter-observer correlation was strong for RTVue and moderate in both Cirrus and Spectralis. It is crucial to know that correlation does not mean reproducibility [2]. Therefore, their conclusion may be a misleading message due to applying inappropriate tests to assess reproducibility. As a take-home message, misdiagnosis and mismanagement of the patients cannot be avoided by using inappropriate tests to assess reproducibility. …