Retinal vascular features enhance short-term prediction of delivery in hypertensive disorders of pregnancy.

OBJECTIVE: Hypertensive disorders of pregnancy (HDP) are a significant global health concern requiring effective strategies for risk stratification. We hypothesized that integrating quantitative retinal vascular features with the soluble fms-like tyrosine kinase-1 (sFlt-1)/placental growth factor (PlGF) ratio and clinical data would significantly improve the prediction of delivery within 2 weeks due to HDP progression. METHOD: This prospective cohort study enrolled 54 hospitalized patients with HDP who underwent simultaneous sFlt-1/PlGF testing and fundus photography. Retinal vascular parameters (density, fractal dimension, tortuosity) were automatically quantified from images using the deep-learning-based AutoMorph pipeline, with separate analyses performed for arteries and veins. The primary outcome was delivery for maternal or fetal indications within 2 weeks. Predictive models were developed and compared using logistic regression with feature selection. RESULTS: Patients with an sFlt-1/PlGF ratio at least 38 exhibited significant retinal microvascular alterations, including reduced vessel density and fractal dimension, compared to those with a lower ratio. These changes were more pronounced in patients who progressed to preeclampsia. The full multimodal model, incorporating retinal features, the sFlt-1/PlGF ratio, and clinical data, achieved an area under the curve (AUC) of 0.85 [95% conflicts of interest (CI), 0.75-0.96)] for predicting delivery within 2 weeks. This significantly outperformed models based on the sFlt-1/PlGF ratio alone (AUC = 0.64) or the ratio combined with clinical data (AUC = 0.72). CONCLUSION: The integration of retinal vascular features with clinical and biochemical data may improve predictive performance of short-term risk stratification in HDP. This non-invasive, multimodal approach warrants validation in larger prospective studies to determine its potential for improving clinical decision-making.