Abstract
Abstract: :
Purpose: To characterise the noise in Heidelberg Retinal Tomograph (HRT) measurement of neuroretinal rim area (RA) in order to make recommendations on the optimal frequency of imaging in follow–up for detecting disease progression in glaucoma. Methods: Patients with primary open angle glaucoma or ocular hypertension (OH) were recruited to two studies: one of test–retest HRT imaging (74 patients) and the other of longitudinal imaging over 6 years (238 patients). Global RA measurement variability was characterised by fitting theoretical distributions to the observed data. The rate of loss in RA was estimated from the longitudinal data using linear regression. Computer simulations of disease progression were performed by specifying noise–free progression rates estimated from the clinical data, and then adding noise from the distribution derived from the clinical data. Rates of sensitivity and specificity in detecting disease progression (defined as a negative regression slope with P<0.05) were investigated for stable eyes and progressing eyes. Fixed follow–up periods and different rates of imaging were investigated. Results: Measurement variability was not Normally distributed and was best characterised by the Hyperbolic distribution, giving good fit to averages whilst allowing for extreme values. Rates of loss of RA were small (mean loss per year in OH patients who converted to glaucoma according to visual fields = 0.015 mm2, standard deviation = 0.018 mm2). Poor rates of detection (sensitivity = 24%, specificity = 95%) were observed by imaging 2 times per year over a follow–up period of 3 years. Detection rates improved by imaging more frequently (for example, 4 times a year for 5 years gave sensitivity = 76%, specificity = 95%). Conclusions: The noise in HRT measurement is well characterised by the Hyperbolic distribution, allowing for experiments on the frequency of testing to detect various rates of disease progression. Sensitivity of detection improves with more frequent testing. However, in the clinical situation follow–up is not fixed but instead continues sequentially until progression is detected, resulting in deteriorating specificity with repeated testing, thus corrective statistical methods are required to maintain specificity. This work is based on average rates of loss, however it is possible to develop similar methods which will allow detection to be tailored for individual patients.
Keywords: imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound) • clinical (human) or epidemiologic studies: biostatistics/epidemiology methodology