BACKGROUND: Use of statistical modelling techniques to identify models that both describe glaucomatous sensitivity decay and allow predictions of future field status. METHOD: Twelve initially normal fellow eyes of untreated patients with confirmed normal tension glaucoma were studied. All had in excess of 15 Humphrey fields (mean follow-up 5.7 years). From this cohort individual field locations were selected for analysis if they demonstrated unequivocal deterioration at the final two fields. Forty-seven locations from five eyes satisfied this criterion and were analysed using curve-fitting software which automatically applies 221 different models to sensitivity (y) against time of follow up (x). Curve-fitting was then repeated on the first five fields, followed by projection to the date of the final field to generate a predicted threshold which was compared to the actual threshold. Competing models were therefore assessed on their performance at adequately fitting the data (R2) and their potential to predict future field status. RESULTS: Models that provide the best fit to the data were all complex polynomial expressions (median R2 0.93). Other simple expressions fitted fewer locations and exhibited lower R2 values. However, accuracy in predicting future deterioration was superior with these less complex models. In this group a linear expression demonstrated an adequate fit to the majority of the data and generated the most accurate predictions of future field status. CONCLUSIONS: A linear model of the pointwise sensitivity values against time of follow-up can provide a framework for detecting and forecasting glaucomatous field progression. Linear modelling allows the clinically important rate of sensitivity loss to be estimated.
BACKGROUND: Use of statistical modelling techniques to identify models that both describe glaucomatous sensitivity decay and allow predictions of future field status. METHOD: Twelve initially normal fellow eyes of untreated patients with confirmed normal tension glaucoma were studied. All had in excess of 15 Humphrey fields (mean follow-up 5.7 years). From this cohort individual field locations were selected for analysis if they demonstrated unequivocal deterioration at the final two fields. Forty-seven locations from five eyes satisfied this criterion and were analysed using curve-fitting software which automatically applies 221 different models to sensitivity (y) against time of follow up (x). Curve-fitting was then repeated on the first five fields, followed by projection to the date of the final field to generate a predicted threshold which was compared to the actual threshold. Competing models were therefore assessed on their performance at adequately fitting the data (R2) and their potential to predict future field status. RESULTS: Models that provide the best fit to the data were all complex polynomial expressions (median R2 0.93). Other simple expressions fitted fewer locations and exhibited lower R2 values. However, accuracy in predicting future deterioration was superior with these less complex models. In this group a linear expression demonstrated an adequate fit to the majority of the data and generated the most accurate predictions of future field status. CONCLUSIONS: A linear model of the pointwise sensitivity values against time of follow-up can provide a framework for detecting and forecasting glaucomatous field progression. Linear modelling allows the clinically important rate of sensitivity loss to be estimated.
Authors: Reuben R Shamir; Yael G Friedman; Leo Joskowicz; Michael Mimouni; Eytan Z Blumenthal Journal: Graefes Arch Clin Exp Ophthalmol Date: 2016-01-07 Impact factor: 3.117
Authors: Felipe A Medeiros; Linda M Zangwill; Christopher Bowd; Kaweh Mansouri; Robert N Weinreb Journal: Invest Ophthalmol Vis Sci Date: 2012-10-05 Impact factor: 4.799