Standalone Smartphone Photorefraction: High Accuracy for Timely Myopia Screening

Background Recent smartphone-based health technologies emphasize engineering, but evidence on standalone devices, without attachments, for detecting uncorrected refractive errors remains limited. Validation is essential for scalable vision screening in resource-limited settings. This study was conducted to validate smartphone-based eccentric photorefraction against open-field autorefraction and assess comparable performance under non-cycloplegic (dry) and cycloplegic (wet) conditions. Methods This diagnostic study (2022–2024), baseline for the Smartphone AI Refraction System (STARS), enrolled 948 Chinese children and adults (aged 6–43 years; mean [SD], 19.8 [9.03] years; spherical equivalent, − 12.50 to + 6.88 D) with clear media, excluding presbyopia, tropia, or nystagmus. This study was conducted at the Hong Kong Polytechnic University Optometry Research Clinic using convenience sampling. Participants underwent open-field autorefraction and smartphone photorefraction at 1 m distance under non-cycloplegic and cycloplegic conditions. We evaluated the refractive error agreement with intraclass correlation coefficient (ICC) and Bland-Altman analysis. Diagnostic performance for myopia worse than − 2.00 D (2021 AAPOS criteria) was assessed using sensitivity, specificity, and diagnostic odds ratio (DOR). Results We analysed 5,350 non-cycloplegic and 4,919 cycloplegic eye photographs. Non-cycloplegic photorefraction demonstrated higher sensitivity (96%; 95% CI, 96–97) and specificity (83%; 95% CI, 81–85) for detecting significant myopia compared with cycloplegic photorefraction (91% (95% CI, 90–92); 79% (95% CI, 77–81)). ICCs were 0.77 (non-cycloplegic) and 0.67 (cycloplegic). Non-cycloplegic photorefraction also showed higher DOR than cycloplegic photorefraction. Conclusions Standalone smartphone photorefraction without external attachments offers high accuracy, particularly under noncycloplegic conditions, and supports large-scale refractive errors screening in underserved areas.