Skip to main content
. 2018 Nov 6;15(11):e1002674. doi: 10.1371/journal.pmed.1002674

Fig 4. Algorithm performance in external validation.

Fig 4

(A–F) Our algorithm achieved stable performance for high myopia detection in DGC (the AUC ranged from 0.768 to 0.969 for 10 years), the HZD (the AUC ranged from 0.773 to 0.968 for 10 years), the PYD (the AUC ranged from 0.854 to 0.951 for 5 years), the HZC (the AUC ranged from 0.822 to 0.941 for 6 years), the HPD (the AUC ranged from 0.802 to 0.976 for 8 years), the HKC (the AUC ranged from 0.897 to 0.929 for 3 years), and the LHD (the AUC ranged from 0.888 to 0.916 for 2 years). (G) Our algorithm provided excellent performance at 19 time points (43.18%, 19/44) and good performance at 22 time points (50.0%, 22/44). (H) A clinically acceptable prediction of the refraction value was provided at the majority of the time points examined (the MAE ranged from 0.201 to 0.494 for 3 years, 0.354 to 0.731 for 5 years, and 0.508 to 0.879 for 8 years). AUC, area under the curve; D, dioptres; DGC, Dongguan Guangming Ophthalmic Hospital; HKC, Xiuying Optometry Department in Haikou City; HPD, Huangpu Optometry Department; HZC, Optometry Centre in Huizhou City; HZD, Haizhu Optometry Department; LHD, Haikou Longhua Optometry Department; MAE, mean absolute error; PYD, Panyu Optometry Department.