Systemic Drugs and Intravitreal Therapies for Preventing Progression of Non-Proliferative Diabetic Retinopathy to Vision-Threatening Complications: A Systematic Review and Baseline-Risk-Stratified Network Meta-Analysis of Randomized Controlled Trials

Non-proliferative diabetic retinopathy (NPDR) is the key window for preventing vision-threatening complications, yet systemic drugs and intravitreal therapies have never been compared within a single framework. Intravitreal trials (e.g., aflibercept in PANORAMA and DRCR Protocol W) mostly enrolled high-risk patients with short follow-up. Systemic drug trials (e.g., fenofibrate in FIELD, ACCORD Eye and LENS; RAS inhibitors) mostly enrolled lower-risk patients with long follow-up. Because of this, comparing relative effects or rankings directly may mislead clinical decisions. This systematic review and network meta-analysis (NMA) will include randomized controlled trials (follow-up ≥24 weeks) in adults with type 1 or type 2 diabetes and NPDR at baseline. Eligible trials compare any systemic drug (oral or subcutaneous, including Chinese patent medicines) or any ocular drug (intravitreal injection, sustained-release implant, eye drops) with placebo, sham, observation, usual care or another drug. The primary outcome is vision-threatening complications: PDR, anterior-segment neovascularization, CI-DME/CSME, need for DR treatment, or referable DR. The secondary outcome is DR severity worsening on a grading scale. We will search PubMed, Embase, Cochrane CENTRAL, Europe PMC, ClinicalTrials.gov, CNKI and Wanfang with no language restriction. Risk of bias will be assessed with RoB 2 and certainty of evidence with CINeMA. The primary analysis is a frequentist random-effects NMA on the risk-ratio scale (REML, P-scores, design-by-treatment inconsistency test). A Bayesian time-standardized NMA (cloglog link with follow-up offset, hazard ratios) with latent baseline-risk meta-regression will examine whether relative effects vary with baseline risk. Results will then be translated into 2-year absolute risk reductions, NNT and injections needed to prevent one event across ETDRS severity levels to derive treatment thresholds. We will also validate DR severity worsening as a trial-level surrogate endpoint using a Daniels–Hughes bivariate model, and run eight prespecified sensitivity analyses. Reporting follows PRISMA-NMA. All code (Python) and data will be made publicly available.