The triglyceride-glucose index as a biomarker of diabetic retinopathy: A systematic review and meta-analysis

Adilet Amirashov¹Aigul Balmukhanova²Arip Koishybayev¹Denis Petrachkov³Dana Koishybayeva¹Altynay Balmukhanova⁴Nadiar Maratovich Mussin¹Amin Tamadon^1*

• ¹West Kazakhstan Marat Ospanov State Medical University, Aktobe, Fars province, Kazakhstan
• ²Kaspijskij Universitet, Almaty, Kazakhstan
• ³FGBNU Naucno-issledovatel'skij institut glaznyh boleznej imeni M M Krasnova, Moscow, Russia
• ⁴Al-Farabi Kazakh National University, Almaty, Kazakhstan

Background: The triglyceride-glucose (TyG) index, a surrogate marker of insulin resistance, has been linked to various diabetic complications. However, its association with diabetic retinopathy (DR) remains inconsistent. We conducted a systematic review and meta-analysis to evaluate the relationship between TyG index levels and the risk of DR. Methods: We searched PubMed, Scopus, and Web of Science from inception to July 2025 for observational studies reporting the association between TyG index and DR in adults with type 1 or type 2 diabetes. Two reviewers independently screened studies, extracted data, and assessed risk of bias using the Agency for Healthcare Research and Quality (AHRQ) checklist and Newcastle-Ottawa Scale. Pooled odds ratios (ORs) with 95% confidence intervals (CIs) were calculated using a random-effects model. Heterogeneity was evaluated with the I² statistic. Publication bias was assessed via funnel plots and Egger's test. Subgroup and meta-regression analyses were conducted to explore heterogeneity. Results: Sixteen studies with a total of 33,436 participants were included. The pooled OR for the association between higher TyG index and DR was 1.89 (95% CI: 1.27–2.82) when TyG was treated as a categorical variable, and 1.57 (95% CI: 1.25–1.98) when treated as continuous. Significant heterogeneity was observed (I² > 87%). Subgroup analyses revealed stronger associations in studies with smaller sample sizes and higher male proportions. Meta-regression showed that male proportion accounted for 48.71% of the heterogeneity. In categorical analyses, funnel-plot asymmetry and Egger's test indicated small-study effects; after trim-and-fill adjustment the pooled effect attenuated and was no longer significant, suggesting sensitivity to publication bias. Conclusions: While higher TyG levels correlate with DR—particularly when modeled continuously—the signal is heterogeneity-and bias-sensitive in categorical analyses. Our moderator analyses newly indicate a sex-composition effect, and the current lack of harmonized clinical TyG thresholds limits immediate translation.