Digital Retinal Images as Biomarker to Predict Type 2 Diabetes, Cardiovascular Risk Factors and Microvascular Complications among Caucasians and South Asian Populations

  • Aditya Shivram Nar

Student thesis: Doctoral ThesisDoctor of Philosophy


Background: Retinal screening is an integral part of eye care for people living with diabetes. Due to advancement in computational methods, retinal vascular features (RVFs) can be measured from retinal images. Emerging evidence suggests RVFs can show changes due to cardiovascular risk factors (CRFs), type 2 diabetes (T2D) and microvascular complications. However, studies utilizing population-based diabetic retinal screening images to understand biomarker capability of RVFs with CRFs, T2D and microvascular complications in Caucasians and South Asians are limited.

Aim: To explore RVFs measured from diabetic screening retinal images are capable to predict T2D, CRFs, microvascular complications, specifically diabetic retinopathy (DR), diabetic peripheral neuropathy (DPN) and chronic kidney disease (CKD) among Caucasians and South Asian individuals with type 2 diabetes.

Methods: Participants from three datasets for whom retinal images are measured from semi-automatic VAMPIRE software are selected: Genetics of Diabetes Audit and Research in Tayside Scotland, UK (GoDARTS) (n=6,498), Madras Diabetes Research Foundation, India (MDRF) (n=2,062) and Telemedicine pRoject for screENing Diabetes and its complications in rural Tamil Nadu (TREND) (n=959). Retrospective and prospective data from date of retinal screening was available for GoDARTS, retrospective for MDRF and cross-sectional for TREND study participants. CRFs include blood pressure, HbA1c, lipids and BMI. Microvascular complication for this study includes diabetic retinopathy (DR), diabetic peripheral neuropathy (DPN) and chronic kidney disease (CKD) – stage 3 or higher.

Mean values three years before the date of retinal image was evaluated for CRFs, and multivariable linear regression was performed to evaluate relationship between CRFs and RVFs in both the population i.e. GoDARTS and MDRF. Using crosssectional study design, relationship between RVFs and microvascular complications was explored by using stepwise backward logistic regression among participants from both the sites and to evaluate discrimination accuracy of RVFs to detect DR, DPN or CKD – Stage 3 or higher. Additionally, retrospective follow-up study to assess relationship of RVFs was implemented for GoDARTS participants who were followed-up until first record of DR, DPN and CKD – Stage 3 after date of image and stepwise backward cox proportional hazard model was used.

Furthermore, cross-sectional analysis of participants from TREND study was evaluated for relationship between T2D and RVFs, and examined discrimination accuracy of RVFs to identify people with T2D.

Results: Relationship between CRFs and RVFs: In both the population groups, we found per 10 mm of Hg increase in mean arterial blood pressure is associated with narrower arteriolar caliber (per SD); and 1% increased HbA1c and smoking is associated with wider CRVE (per SD). Additionally, in GoDARTS participants we found relationship between each year increase in duration of T2D and venular tortuosity and arteriolar fractal from main vessels in Zone C and venular and arteriolar fractals from all vessels in Zone C.

RVFs to identify people with T2D (TREND Study): In the TREND study we found that addition of RVFs measurements significantly increases discrimination accuracy to detect diabetes among participants.

Longitudinal and cross-sectional relationship between RVFs and microvascular complications: Utilizing both the study designs we found increased Tort-V is associated with CKD – Stage 3 GoDARTS participants. Utilizing cross-sectional study design we found that increased Tort-A and wider CRVE is associated with any DR and PDR respectively in both the population groups. We also showed that addition of RVFs increases discrimination accuracy to detect any DR and DPN in GoDARTS participants.

Conclusion: From this thesis, we found that RVFs are associated with CRFs and microvascular complication in GoDARTS and MDRF participants; improves discrimination accuracy to identify people with T2D among participants of TREND study; and any DR and DPN among GoDARTS participants when utilizing in addition to established risk factors. This thesis suggests that RVFs extracted from population based retinal screening does have potential to be utilized beyond screening for diabetic retinopathy. However, further multi-ethnic studies among people living with T2D are required to replicate these findings.
Date of Award2022
Original languageEnglish
SupervisorAlexander Doney (Supervisor) & Manuel Trucco (Supervisor)

Cite this