Construction and validation of risk prediction models for different subtypes of retinal vein occlusion

While prognostic models for retinal vein occlusion (RVO) exist, subtype-specific risk prediction tools for central retinal vein occlusion (CRVO) and branch retinal vein occlusion (BRVO) remain limited. This study aimed to construct and validate distinct CRVO and BRVO risk stratification nomograms.

We retrospectively analyzed electronic medical records from a tertiary hospital in Guangzhou (January 2010-November 2024). Non-RVO controls were matched 1:4 (CRVO) and 1:2 (BRVO) by sex and year of admission. The final cohorts included 630 patients (126 CRVO cases and 504 controls) and 813 patients (271 BRVO cases and 542 controls). Predictors encompassed clinical histories and laboratory indices. Multivariate regression identified independent risk factors, and model performance was evaluated using the area under the receiver operating characteristic curve (AUC), calibration plots, and decision curve analysis (DCA).

The CRVO-nom and BRVO-nom highlighted significant predictors, including the neutrophil-to-lymphocyte ratio (NLR). Additional risk factors for CRVO included high-density lipoprotein cholesterol (HDL-C), platelet distribution width (PDW), history of diabetes, cerebral infarction, and coronary artery disease (CAD). For BRVO, significant predictors included a history of hypertension, age, and body mass index (BMI). The AUC for CRVO-nom was 0.80 (95% CI: 0.73-0.87) in the training set and 0.77 (95% CI: 0.65-0.86) in the validation set, while BRVO-nom yielded an AUC of 0.95 (95 ​%CI: 0.91-0.97) in the training set and 0.95 (95% CI: 0.89-0.98) in the validation set.

CRVO and BRVO exhibit distinct risk profiles. The developed nomograms-CRVO-nom and BRVO-nom-provide subtype-specific risk stratification with robust discrimination and clinical applicability. An online Shiny calculator facilitates real-time risk estimation, enabling targeted prevention for high-risk populations.