|Title||WHEN ARE MULTIFOCAL ERG MEASURES MOST USEFUL IN PREDICTING DIABETIC RETINOPATHY?|
|Author, Co-Author||Jason Ng, Jr Bearse, Marilyn Schneck, Shirin Barez, Tony Adams|
|Abstract|| PURPOSE: Determine the optimal time interval to use multifocal electroretinogram (mfERG) measures to predict local development of NPDR.
METHODS: 36 eyes of 18 diabetics were examined at baseline and 3 annual follow-ups. At each visit mfERG recording and fundus photography were performed. 35 zones were constructed from the 103-element mfERG stimulus, and each was assigned the maximum P1 implicit time (IT) and minimum P1 amplitude (AMP) Z-scores within it based on 30 controls. Zones with baseline NPDR were excluded from analysis. Three NPDR incidence outcomes were investigated: (1) any (transient + recurring NPDR), (2) transient (present at only 1 follow-up), and (3) recurring (present at 2 or 3 follow-ups). Logistic regression was used to model each outcome’s relationship with mfERG IT and AMP. ROC curves were analyzed to assess each model’s predictive accuracy, yielding an area under the curve (AUC; an overall measure of discrimination).
RESULTS: 77 of 1208 zones developed NPDR lesions, 52 transient and 25 recurring.
Outcome (1): IT models had AUCs of 0.86 (p<0.05), 0.48 (p>0.05), and 0.74 (p<0.05) at 1, 2, and 3-year time points, respectively. AMP models had significant (p<0.05) AUCs of 0.53, 0.59, and 0.57; though as expected these had poor predictive accuracies (sensitivities/specificities ~55%).
Outcome (2): IT models had significant AUCs of 0.88, 0.60, and 0.59; and AMP models had significant AUCs of 0.63, 0.64, and 0.68. However, only the 1-year model using IT performed substantially better than chance.
Outcome (3): IT models had significant AUCs of 0.83, 0.86, and 0.82 and good predictive accuracy (sensitivities/specificities ~80%). AMP models were not significant with AUCs of 0.52, 0.53, and 0.49, respectively.
CONCLUSIONS: Multifocal ERG implicit times most accurately predict NPDR over a 1-year period (transient or recurring). Over longer time intervals they also predict areas of recurring retinopathy with good accuracy. By comparison, multifocal ERG amplitudes are not clinically useful for the prediction of diabetic retinopathy.
ADDITIONAL COMMENTS: NIH/NEI Grants EY02271 & EY07043
|Affiliation of Co-Authors||University of California Berkeley, School of Optometry, University of California Berkeley, School of Optometry, University of California Berkeley, School of Optometry, University of California Berkeley, School of Optometry|