Purchase this article with an account.
W. W. Harrison, M. A. Bearse, Jr., J. S. Ng, S. Barez, N. P. Jewell, M. E. Schneck, A. J. Adams; Multifocal Electroretinograms Are Predictive of the Onset of Diabetic Retinopathy in Adult Diabetics. Invest. Ophthalmol. Vis. Sci. 2010;51(13):4673.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Our group previously derived models to predict local formation of new diabetic retinopathy (DR) lesions in adults with diabetes (DM). Those studies included a mix of patients with and without preexisting DR. In this study a multivariate model was derived for the local prediction of DR onset in patients who all had no DR at baseline.
78 eyes from 41 patients with DM, but no DR, were followed and tested annually until either DR developed (n=20) or the study ended (n=21). The presence or absence of DR in the last set of fundus photos was the outcome measure and data from the prior year was used as the baseline for prediction. Logistic regression was used to assess the relationship between DR development and 7 baseline risk factors: Multifocal ERG (mfERG) implicit time (IT) Z-score, gender, diabetes duration, blood glucose, HbA1c, age, and type of diabetes. 35 retinal zones were constructed from the 103 mfERG stimulus elements. Each zone in the fundus photos was graded for DR. Also, based on data from 50 control subjects, the maximum IT Z-score for each zone was calculated. An ROC curve averaged from a 5-fold cross validation was used to determine the model’s accuracy and validity.
The DR that developed was mild, consisting of microaneurysms or dot hemorrhages. DR developed in 80 of the 2,730 retinal zones (3%), in 29 of 78 eyes. Multivariate analysis showed mfERG IT to be the most predictive factor for DR development in a zone, with diabetes type being a significant confounder of this measure. The two factors together produce a model, which after 5-fold cross validation, has an 80 ± 4% sensitivity and 74 ± 4 % specificity for prediction of DR onset. All other candidate factors did not add to the strength of the model or improve its accuracy.
MfERG implicit time is a good predictor of DR onset one year later in patients with DM but no DR at baseline, when controlling for diabetes type. This makes the mfERG a possible tool for assessing risk for DR development and an outcome measure to evaluate novel therapeutics directed at the earliest stages of DR.
This PDF is available to Subscribers Only