Purchase this article with an account.
Alexander Hua, Walter Johnson, Namrata Nandakumar, Gianna C Teague, Megan E Baldwin, Kameran Lashkari; Linear Discriminate Analysis for Design of High-Value Biomarkers in Proliferative Diabetic Retinopathy. Invest. Ophthalmol. Vis. Sci. 2017;58(8):2511.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Efforts have been made to identify biomarkers to quantify characteristics of various disease processes and to establish therapeutic endpoints. In this study, we aimed to develop and validate a method of combining multiple factors into a composite biomarker using linear discriminant functions. The purpose of a composite biomarker is to provide a useful tool to accurately classify disease-state proliferative diabetic retinopathy (PDR) from disease-free subjects without complicated data interpretation.
Vitreous samples collected from control subjects (n=28) and those with PDR (n=35) were subjected to ELISA or multiplexing to detect levels of 4 distinct factors that have been classically associated with PDR: VEGF-A, PLGF, PAI-1 and IL-8. All four factors were analyzed individually and in pairs in a linear discriminant function to produce a final composite biomarker. Comparison of success rates for correct diagnosis of disease condition, PDR vs. control was performed with training and test sets using the randomized k-fold stratified validation method.
In a discriminant function analysis using one proangiogenic factor (VEGF-A or PLGF) or one pro-inflammatory factor (PAI-1 or IL-8), the results for successful classification of PDR was approximately 82.52% in a 7-fold stratified test data set. In a two-factor linear discriminant function analysis, the pair combination of one proangiogenic and one pro-inflammatory factor into a multi-cytokine biomarker demonstrated approximately 90% success in distinguishing between PDR and control. The derived linear discrimination function of the combination of VEGF-A and IL-8 was tested against independent published data demonstrated a 90% success classification rate in identifying PDR from control.
Our study provides a useful method for identification of disease process in unknown subjects by applying linear discriminant analysis of 4 distinct biomarkers that have been widely accepted to be associated with PDR. The application of multiple factors to construct a composite biomarker provides stronger evidence in identifying disease processes than a single biomarker alone. This methodology may be extended to plasma/serum biomarkers, which ultimately could allow a simple blood test to be used for diagnosis of a specific eye disease.
This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.
Correct Classification Success Rate of PDR using Pro-angiogenic and Pro-Inflammatory Factors
This PDF is available to Subscribers Only