Purchase this article with an account.
Shaban Demirel, Chris A. Johnson, Stuart K. Gardiner; Does Glaucoma have a Short Memory?. Invest. Ophthalmol. Vis. Sci. 2012;53(14):215.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To determine whether prediction of functional change in early glaucoma is improved by adding older to recent data.
99 participants (190 eyes) from a longitudinal study (Portland Progression Project; P3) contributed data. All eyes had 12 visits, initially annually, but then biannually. Visits included Humphrey visual fields, Heidelberg Retina Tomography (HRT), IOP and CCT. Different data types were usually collected on the same day. Data were split into 2 sequences; visits 1-6 and 7-12. Four generalized estimating equation models were constructed to predict the rate of functional change (MD linearly regressed over time; MDR) in sequence 2 using stepwise backwards elimination that minimized the Akaike Information Criterion. This algorithm occasionally retains variables with p>0.05 if they are informative. For the 4 models various sets of predictors were used; A) visit 6 only, B) the mean of visits 1-6, C) rates of change during visits 1-6 and D) all of the above (with highly correlated predictors excluded).
Mean±SD age at visit 6 was 61.7±9.6 years and MD was -0.3±2.7 dB. During sequence 2 the MDR was -0.06 (range -3.9 to +0.74) dB/Yr. Model A explained 10% of the variance in MDR and contained 6 predictors but only 4 were significant (GHT not Within Normal Limits (WNL) [p<0.01], Cup Volume [p<0.01], Max Cup Depth [p=0.01], Age [p=0.04]). Model B explained 5% of the variance and contained 3 predictors with 2 significant (mean Cup Area during visits 1-6 [p<0.01], Age [p=0.02]). Model C explained 1% of the variance and contained 1 non-significant predictor (Difference in mean IOP between 2 sequences [p=0.06]). Model D explained 11% of the variance and contained 9 predictors with 5 significant (GHT not WNL [p=0.03], Cup Volume [p=0.01] & Max Cup Depth [p=0.02] at visit 6, rate of MD [p=0.05] and rate of Cup Volume [p=0.02] change during visits 1-6).
Predicting the future rate of functional change in glaucoma is problematic. The best statistical model generated here explained only 11% of the variance. Using historical structural (HRT) and functional (SAP) data in addition to recent HRT and SAP data added little to the predictability of the rate of future functional change. Using historical means or rates alone resulted in even poorer performance. Useful information about the future rate of functional change in glaucoma may come predominantly from recent data, suggesting that glaucoma may have a short memory.
This PDF is available to Subscribers Only