Diagnosis of Dry Eye Disease Using Principal Component Analysis: A Study in Animal Models of the Disease

Abstract

Purpose: To evaluate whether principal component analysis (PCA) can assess various diagnostic tests of dry eye disease (DED), providing a simplified, more informative measure of disease status than individual clinical test parameters (ICTP). Materials and Methods: ICTP were analyzed using PCA in two groups of normal rabbits (Groups 1 and 2). Group 3, not truly normal, was also assessed. DED was induced in Group 1 by complete dacryoadenectomy; in Groups 2 and 3 by injection of concanavalin A. Tear break up time, tear osmolarity, Schirmer’s tear test and rose bengal staining were the ICTP measured in all groups. Statistical analysis including descriptive statistics, t test, correlation coefficients and PCA was done. PCA using ICTP data from Group 1 generated axes; Group 2 and 3 were plotted over these axes. Results: All groups had induction of DED. Correlations for all ICTP were in the correct direction and were strongest for Group 1 and weakest in Group 3. PCA clearly separated DED and normal eyes. Principal component (PC) 1, made up of nearly equal contributions from the four clinical tests, explained 73% of the variation and provided a means to separate normal from DED. PC 1 values under 0.52 can be mathematically defined as DED. Of all pairwise comparisons, PC 1 vs PC 2 and PC 1 vs PC 3 were the most informative providing excellent spatial separation and additional information regarding DED status. Conclusions: PCA proved useful for evaluating DED providing a simpler, more comprehensive assessment than ICTP. PC 1 is a valuable, clinically relevant, and informative metric for DED status and severity having superior diagnostic value and statistical strength compared to ICTP. Spatial information on biplots of PC 1 vs PC 3 is also informative. PCA, and specifically PC 1, has the potential to serve as a biomarker for DED.