Measurement variability had to be rigorously accounted for, and we dealt with it in several ways. First, we introduced a novel reference plane that allows rim area to be analyzed reproducibly. By this, variability is expected to be significantly less and not appreciably affected by glaucoma or testing involving different operators and visits, compared with other conventional HRT (Heidelberg Engineering, GmbH) reference planes.
25 Variability was thus simpler to account for. Second, we did not presume that variability is uniform across the ONH and so estimated variability in each sector separately. Third, estimation of variability was unique to each eye—necessary because, apart from ONH morphology, individual factors, such as media opacity
32 and ability to maintain fixation,
33 may affect variability differently in each eye. Fourth, only change that was repeatable in two of three tests was attributed to disease progression. This confirmation test strategy had fewer false-positive results than the single-test strategy, and sensitivity in confirming change was not significantly compromised. Hence, confirming change by this criterion resulted in fewer eyes being misidentified as having progressive glaucoma. The two-of-three criterion also had a favorable balance between sensitivity and the false-positive rate compared with other tested criteria. Fifth, variability was estimated using all image data from each series, with limits of variability updateable to factor in data from subsequently acquired images. This is desirable because the estimation of variability can be expected to improve with more δ values. Higher degrees of freedom of δ would result in narrower limits of variability, with the
t-statistic decreasing as degrees of freedom increase
(equation 3) . Sixth, although obtaining three single topographies per visit is advised, more images could be acquired per visit if the number of δ has to be increased quickly. This is possible because the number of δ increases exponentially according to
m C
r (equation 2) . For example, three images per visit yield 3 δ, but six images yield 15 δ (a 5-fold increase) and nine images yield 36 δ (a 12-fold increase). Thus, robust estimates of variability can be obtained rapidly, giving flexibility in clinical situations where indicated. However, fatigue of the patient and its consequences on variability must be considered when many images are acquired at a single sitting.