Purchase this article with an account.
J.S. Ng, M.A. Bearse, Jr., Y. Han, M.E. Schneck, S. Barez, C. Jacobsen, A.J. Adams; Modeling the Development of Non–Proliferative Diabetic Retinopathy Over 2 Years . Invest. Ophthalmol. Vis. Sci. 2006;47(13):988.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To formulate a quantitative model to describe the local development of non–proliferative diabetic retinopathy (NPDR) in patients over a 2 year time period using multifocal electroretinogram (mfERG) implicit times as the only locally predictive variable.
20 eyes of 20 diabetics were examined at baseline (T0) and for 2 consecutive annual follow–ups (T1, T2). At each examination, mfERGs were recorded using VERIS 4 (from the central 45 deg, 10–100Hz filtering), fundus photography was performed, and blood glucose was measured. First–order mfERG implicit times (IT) were derived using a template stretching method (Hood & Li, 1997). The determination of locations with retinopathy was done masked to the mfERG data. Locations of retinopathy were mapped relative to the mfERG stimulus array. 35 zones were constructed from the 103 element stimulus array and each zone was assigned the maximum Z score within it based on 30 age–similar controls. 42 zones showed initial retinopathy and these were excluded from further analysis. The probability that the remaining 658 zones developed retinopathy anytime (T1 or T2) during the 2 year period was modeled based on the mfERG implicit time Z score and other baseline factors: duration of diabetes, blood glucose level, age, diabetic type, and presence of retinopathy. Univariate and multivariate regression models for the development of retinopathy were performed using generalized estimating equations to account for zone correlations within each eye.
New instances of retinopathy developed in 47 zones in 11 eyes during the 2 year period (at T1 or T2). The univariate model using only mfERG IT as a predictor had a sensitivity and specificity of 72% and 69%, respectively and an odds ratio of 5.80 (p<0.001; 95% CI = 2.0–11.23). The area under the ROC curve was 0.75 for this model. Of the baseline factors, diabetic duration and blood glucose level were significant (p<0.05) in the multivariate model. These additional factors improved the model substantially, having a sensitivity of 81% and a specificity of 82% (odds ratio = 19.7; 95% CI 9.2–41.9). The area under the ROC curve for this multivariate model was 0.88.
The multivariate model using abnormal mfERG ITs, diabetic duration, and blood glucose fits the longitudinal data quite well. The univariate and multivariate models produced similar sensitivity, specificity, and AUC values to previous 1–year models (Han, et al. 2004). The 1–year and 2–year multivariate models were similar in that they both identified mfERG IT, diabetic duration, and blood glucose as significant risk factors for the development of NPDR.
This PDF is available to Subscribers Only