<![CDATA[Background: Cytomegalovirus retinitis (CMVR) is a leading preventable cause of AIDS-related blindness where systematic screening is absent, and World Health Organization guidance triggers ophthalmic assessment below a CD4 count of 50–100 cells/µL, a range not previously calibrated in an Indonesian population. Objective: To determine the diagnostic accuracy of the CD4 count against adjudicated cytomegalovirus retinitis, and to derive an optimal CD4 threshold for triggering ophthalmic screening, in Indonesian adults with advanced HIV disease. Methods: This prospective single-gate diagnostic accuracy study, reported to STARD 2015, enrolled 51 consecutive adults with HIV-1 infection and a CD4 count below 100 cells/µL at Dr. Mohammad Hoesin Central General Hospital, Palembang, from January 2024 to January 2025. The index test was the CD4 count by flow cytometry; the reference standard was masked adjudication by two retina specialists using dilated ophthalmoscopy, fundus photography and optical coherence tomography (Cohen's κ = 0.91). Because only aggregated summaries were released, estimands the data do not identify are reported as sharp bounds. Results: CMVR was present in 17 of 51 participants (33.3%, 95% CI 22.0–47.0); the median CD4 count was 17.0 cells/µL (IQR 11.0–27.0) versus 48.0 (30.0–72.0). The area under the ROC curve was 0.863 (95% CI 0.741–0.985). At the Youden-derived threshold of 32.5 cells/µL apparent sensitivity was 88.2% (95% CI 65.7–96.7) and specificity 91.2% (77.0–97.0); after correction for in-sample threshold selection these became 80.5% and 93.8%, and across two admissible reconstructions sensitivity ranged from 76.5% to 88.2%. CD4 rose monotonically across fulminant, indolent and frosted-branch phenotypes (bounded Kruskal–Wallis p = 0.001–0.013). Conclusion: A CD4 count of approximately 30–35 cells/µL, with an unconditional visual-symptom over-ride, prioritises dilated retinal examination within the World Health Organization stratum in Indonesian HIV services.]]>
