Background/Aims To identify sociodemographic and clinical factors associated with best-corrected visual acuity (BCVA) restoration at the visit after lost to follow-up (LTFU) in patients with myopic choroidal neovascularization (mCNV) treated with antivascular endothelial growth factor intravitreal injections (IVI). Methods A multicentre, retrospective cohort study was conducted in China of mCNV patients receiving injections who were LTFU >6 months. Data were collected from baseline visit, pre-LTFU, initial post-LTFU visit and 6 months post-LTFU, and the final visit. Logistic regression analyses assessed associations between sociodemographic/clinical factors and BCVA restoration. Results The study included 1155 LTFU patients with mCNV (mean (SD) age: 54 (16) years; 61.6% female), with 66.6% achieving BCVA restoration at the initial post-LTFU visit after LTFU (mean(SD) duration: 315 (154) days. Patients with BCVA restoration were younger (p=0.004), more often from eastern China (p=0.022), and had shorter LTFU duration (p=0.013). Stratification by pre-LTFU BCVA showed that patients with BCVA ≥20/40 had lower restoration odds (OR 0.65, 95% CI 0.44 to 0.96; p=0.032) compared with those with BCVA 280 µm had higher BCVA restoration odds (OR 1.98, 95% CI 1.02 to 3.92; p=0.046). Conclusions Better pre-LTFU BCVA (≥20/40) increased post-LTFU deterioration risk (associated with progressive CST thickening), while elevated post-LTFU CST (>280 µm) predicted higher BCVA restoration odds in patients with sustained follow-up, supporting CST-guided monitoring and early intervention for high-risk groups.
Diabetic retinopathy (DR) remains a leading cause of blindness globally, driving the rapid development of automated diagnostic systems leveraging deep learning. Recent research demonstrates that multimodal deep learning models that integrate retinal images, electronic health records (EHR), and clinical text data can surpass single‐modality approaches in DR screening accuracy. In this survey, we systematically review advances from the last 3 years, highlighting key models and methodologies, as well as datasets and evaluation metrics. We find multimodal approaches consistently outperform single‐modality models, and often by a clear margin, and identify data scarcity, domain adaptation, and interpretability as the three main hurdles ahead.