Deep Learning-Based Retinal Biomarker Captures Cardiovascular Risk in Hypertensive Abnormal Dippers Beyond Conventional Metrics

Introduction 

Among hypertensive subtypes, abnormal dippers — including non-dippers and risers — are characterized by blunted nocturnal blood pressure (BP) declines and are associated with poor prognosis. However, limited data exist on their relationship with deep learning-derived retinal biomarkers for cardiovascular risk.

Hypothesis 

We hypothesized that hypertensive individuals with abnormal dipping patterns would have a higher risk of cardiovascular disease (CVD) than normal dippers and that a retinal deep learning-based CVD risk score (Dr.Noon CVD score) would reflect higher cardiovascular burden from nocturnal hypertension compared to traditional risk markers.

Goals

This study aimed to compare carotid intima-media thickness (CIMT), coronary artery calcification (CAC), and a deep learning-based retinal biomarker between normal and abnormal dippers.

Methods

We analyzed 901 hypertensive patients from a 10-year prospective cohort at Severance Hospital, categorizing them as normal (>10% nocturnal drop, N = 463) or abnormal dippers (<10%, N = 438). Group comparisons of clinical variables were performed using chi-squared or nonparametric tests. CVD incidence was assessed by survival analysis. CIMT, CAC, and Dr.Noon CVD scores were compared using Wilcoxon rank-sum tests. Multivariate analyses adjusted for age, gender, and BMI.

Results

Over 10 years, abnormal dippers had a higher cumulative incidence (p = 0.027) and adjusted hazard ratio of CVD events (HR 1.93, 95% CI 1.09-3.39). Age, gender, and BMI did not differ between groups (p > 0.05). Daytime BP was higher in normal dippers (135/82 vs. 132/80 mmHg; p < 0.006), while nighttime BP was higher in abnormal dippers (113/69 vs. 128/76 mmHg; p < 0.001). CIMT did not differ (p = 0.297), and CAC showed a borderline difference (p = 0.068). The Dr.Noon CVD score was higher in abnormal dippers (p = 0.006). In multivariate linear regression, abnormal dippers had a 2.08-point higher Dr.Noon CVD score compared to normal dippers (p < 0.001). Each one-year increase in age was associated with a 0.59-point increase (p < 0.001), suggesting abnormal dipping is comparable to around three years of aging regarding CVD risk.

Conclusion

Abnormal nocturnal BP patterns are associated with elevated cardiovascular risk. Traditional risk markers did not fully capture this difference, whereas the deep learning-based retinal biomarker reflected the increased risk, likely due to its sensitivity to microvascular changes from nocturnal hypertension.