OBJECTIVE: To investigate clinical and spectral-domain optical coherence tomography (SD-OCT) biomarkers correlating with pre-injection visual acuity (VA), post-injection VA, and the likelihood of macular oedema (MO) regression following dexamethasone (DEX) implant injection in non-infectious uveitic (NIU) patients. METHODS: Patient data from Uveitis Services in Milan, Paris, and Berlin were analysed. Eligible participants were NIU patients aged >18 years with MO as the primary indication for DEX treatment. SD-OCT scans and clinical data were collected at the time of DEX injection (pre-injection visit) and after 3 months (post-injection visit). Multivariable regression models, adjusted for pre-injection VA and lens status, were employed to explore associations. MO regression was defined as the absence of intraretinal/subretinal fluid at the post-injection visit. RESULTS: Our analysis comprised data from 173 DEX treatments, encompassing 103 eyes from 80 patients, with 38 eyes (37%) receiving repeated DEX injections. The absence of the ellipsoid zone (EZ) layer and disorganisation of the inner retinal layers (DRIL) were associated with worse pre- (+0.19 LogMAR, 95% CI 0.01-0.38, p = 0.06, and +0.10 LogMAR, 95% CI 0.02-0.21, p = 0.01) and post-injection VA (+0.33 LogMAR, 95% CI 0.08-0.57, p = 0.01, and +0.17 LogMAR, 95% CI 0.01-0.32, p = 0.04). EZ disruption and DRIL increased significantly (p = 0.01 and p = 0.04), and the chance of gaining ≥5 letters declined in eyes undergoing repeated DEX (p = 0.002). The rate of MO regression after each DEX was 67%. Prolonged MO duration (OR = 0.75/each year, p = 0.02) was associated with reduced likelihood of MO regression. Subretinal fluid was associated with higher rate of MO regression (OR = 6.09, p = 0.01). CONCLUSION: Integrity of the inner and outer retina is associated with better visual response to DEX. Long-standing or recurrent MO is associated with less chance of both visual and anatomic response. Timely treatment is necessary to maximise the outcomes of MO in NIU patients.< Réduire