Decreased retinal vascular complexity is an early biomarker for myocardial infarction during the next 5 years after ophthalmic examination

Abstract

Purpose : There is increasing evidence that retinal vascular complexity (D_f) might offer earlier insights into myocardial infarction (MI) progression. Our recent study indicates that D_f predictive power is as important as systolic blood pressure and body-mass index in predicting MI risk. We present here a study analysing D_f utility period on predicting MI over the past 8 years succeeding the ophthalmic examination.

Methods : We used a subset of 516 incident MI cases and 516 random and age-matched controls from UK Biobank (UKB) that had good quality images from both eyes. We obtained D_f from this imaging subset with VAMPIRE 3.1 software and calculated a polygenic risk score for coronary artery disease (PRS_CAD) for each participant using the summary statistics from the CArdioGRAM consortium and PRSice2 software. We developed a MI predictive model combining sex, body-mass index, SBP, smoking, both eyes D_f, PRS_CAD and a random forest algorithm. Predicted MI risk obtained through internal cross-validation was next used to separate individuals at high and low MI risk(high MI-risk>0.5 and low MI-risk=<0.5). We investigated potential survival rate differences between these by completing Kaplan-Meier curves and a Log-rank test. We assessed D_f utility period by comparing in time windows of three years the MI risk accuracy of our model with the one from a reproduction of an established risk model named SCORE, based only on epidemiological variables

Results : There is a significant survival rate difference(Log-rank Test P-value<10⁵⁰) between high vs low MI risk groups. Longitudinal analysis on the timeline of these MI events suggest that our model yields the greatest accuracy during the first 5 years since the ophthalmic examination(Accuracy_0-2=0.749, Accuracy_3-5=0.787, and Accuracy_6-8=0.732). Time-matched comparisons between our MI risk predictions and SCORE-based estimations shows that our approach achieves a higher accuracy during the first 5 years(Accuracy-SCORE_0-2=0.697, Accuracy-SCORE _3-5= 0.734). Afterwards, the accuracy between both MI risk models is similar(Accuracy- SCORE_6-8=0.708).

Conclusions : This study discloses a promising holistic strategy that can prevent MI incidence and triage those with an elevated hazard. Our project highlights the value of easily accessible vascular imaging phenotypes and their promising application in personalised medicine.

This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.