Abstract
Purpose :
The combination of mechanism-driven modeling and Artificial Intelligence (AI) holds promise to explain statistical correlations in terms of physiological principles, thereby enabling more effective, personalized patient care. We consider the Singapore Epidemiology of Eye Diseases (SEED), where an association between low systolic blood pressure (BP) and primary open angle glaucoma (POAG) was found to be more pronounced in the presence of high intraocular pressure (IOP). Here, we use mechanism-driven modeling to enhance the dataset and understand the physiological implications of this association.
Methods :
SEED involved 9877 participants (19587 eyes), including 213 POAG (293 eyes), who underwent ocular and systemic examinations (Collected Data, Fig1). A validated model of retinal circulation (Guidoboni et al 2014) was used to simulate individualized hemodynamic outputs based on measured BP and IOP (Simulated Data, Fig1). The Enhanced Dataset, comprising Collected and Simulated Data, was analyzed via multivariable linear regression models, adjusted for established risk factors, to examine the association between POAG and two outcomes: vascular pressure (vP) and vascular resistance (vR). Generalized estimating equation (GEE) with exchangeable correlation structures and Gaussian link were used to account for the correlation between pairs of eyes for each individual (P value for significance was set at <0.05).
Results :
Enhanced Dataset analysis shows that POAG is associated with (i) increased vR in the retinal venules (P<0.039) and in the intraocular portion of the central retinal vein (P<0.028). The use of anti-hypertensive medications is associated with decreased vP (P<0.001) and increased vR in the venules (P<0.001), but not in the central retinal vein. These results indicate that the association between low BP and POAG, especially at high IOP, is related to hemodynamic alterations in the venous side of the circulation, where increased vP and vR are indicative of higher susceptibility to collapse.
Conclusions :
The mechanism-driven algorithm for dataset enhancement in AI proved capable of explaining statistical correlations in terms of physiological principles and suggests that future glaucoma studies should focus on the venous side of the circulation.
This is a 2020 ARVO Annual Meeting abstract.