Unequal Per-Grade Loss of Macular Ganglion Cell–Inner Plexiform Layer Thickness in Non-Proliferative Diabetic Retinopathy: An Indonesian Cross-Sectional Study

Background: Diabetic retinopathy is a neurovascular disease in which neurodegeneration accompanies microvascular injury. How ganglion cell–inner plexiform layer (GCL–IPL) loss is distributed across non-proliferative diabetic retinopathy (NPDR) severity is unsettled; Indonesian data are scarce. Objective: To quantify mean macular GCL–IPL thickness across mild, moderate and severe NPDR; test whether successive per-grade decrements are equal and scale-dependent; characterise discrimination between grades; and bound the influence of aggregate-data uncertainty and confounding. Methods: Cross-sectional analysis of 32 eyes of 29 consecutive adults with mild, moderate or severe NPDR without macular oedema at Dr. M. Djamil General Hospital, Padang (2023–2025). Mean macular GCL–IPL thickness was measured by spectral-domain optical coherence tomography. Only aggregate data existed, so sharp bounds came from exhaustively enumerating every eye-level dataset compatible with the published summaries and the reported normality test. Result: GCL–IPL thickness fell from 86.0 ± 3.63 μm (mild) to 71.0 ± 4.71 μm (moderate) and 63.8 ± 3.39 μm (severe); F(2, 29) = 86.288, p < 0.001, η² = 0.856 (95% CI 0.724–0.900). All pairwise contrasts were significant (Bonferroni p ≤ 0.0008). The per-grade steps were unequal: mild→moderate lost 15.0 μm against 7.2 μm for moderate→severe, a difference of 7.8 μm (95% CI 1.75–13.85, p = 0.013). That asymmetry is a property of the equal-interval grade score and disappears under ETDRS-level scoring (p = 0.759). Mild and severe eyes separated completely in every admissible dataset (area under the curve 1.000); moderate versus severe, 0.827–0.964. Conclusion: GCL–IPL thickness tracks NPDR severity steeply, but the apparent front-loading of loss is scale-dependent and must not be read as biology. The mild-grade mean lay within the published normative band — an interval built for individual eyes, not for group means — so these data show a gradient, not pre-clinical thinning.