Bidirectional association between obstructive sleep apnea and diabetic retinopathy: a severity-stratified systematic review and meta-analysis with GRADE assessment.

purposeTo synthesise the bidirectional evidence on the association between obstructive sleep apnea (OSA) and diabetic retinopathy (DR), separating studies framing OSA as the exposure for DR (Arm 1) from studies framing DR as the exposure for OSA (Arm 2).

methodsSystematic review and meta-analysis of observational studies (PROSPERO: CRD420251135427). PubMed, Scopus, Web of Science, and Google Scholar were searched from inception to July 13, 2025, supplemented by structured re-screening of five prior systematic reviews and forward and backward citation searching. Eligible studies enrolled individuals with diabetes mellitus and reported binary OSA (or, in Arm 2, DR) exposure with an unexposed comparator. Within each arm, exposure and outcome were severity-stratified; unadjusted and adjusted odds ratios (ORs) were pooled separately using random-effects models in MetaAnalysisOnline.com. Risk of bias was assessed with the Newcastle-Ottawa Scale (NOS) and certainty of evidence with the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) framework.

resultsTwenty-six studies met eligibility criteria. In Arm 1 (OSA → DR), the unadjusted pooled OR for any DR with OSA classified as a binary exposure was 2.41 (95% confidence interval [CI] 1.53-3.80; 10 studies; very low certainty), but attenuated to 1.03 (95% CI 0.64-1.66; 2 studies; low certainty) in confounder-adjusted analyses. OSA-severity-stratified Arm 1 estimates for mild, moderate, and severe OSA were directionally consistent but not statistically significant in either unadjusted or adjusted pooling. DR-stage-specific Arm 1 analyses (background DR, mild/moderate/severe non-proliferative DR [NPDR], proliferative DR [PDR]) were directionally aligned with the overall Arm 1 finding but imprecise. In Arm 2 (DR → OSA), pooled estimates showed inverse or null associations across DR severity strata; severe OSA as an outcome of DR yielded a pooled OR of 1.24 (95% CI 0.92-1.68; 4 studies; low certainty).

conclusionsThe OSA-DR association is sensitive to the direction of the framed exposure-outcome relationship, to confounder adjustment, and to severity classification. Substantial attenuation of the unadjusted Arm 1 signal after adjustment, together with the inverse or null Arm 2 pattern, indicates that previously reported pooled estimates were influenced by confounding and by directional misclassification. Polysomnography-confirmed prospective cohorts with uniform DR grading and individual-participant-level adjustment for body mass index, diabetes duration, and metabolic comorbidities are needed to clarify the direction and magnitude of the relationship.