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

Fig 3. Algorithm performance in internal validation.

Fig 3

(A and B) Our algorithm provided high-precision prediction for detecting high myopia in the cross-validation (the AUC ranged from 0.903 to 0.958 for 3 years, 0.886 to 0.889 for 5 years, and 0.862 to 0.888 for 8 years) and OOB tests (the AUC ranged from 0.934 to 0.986 for 3 years, 0.875 to 0.901 for 5 years, and 0.852 to 0.874 for 8 years). (C) AUC values more than 0.9 indicated excellent performance, and AUC values from 0.8 to 0.9 indicated good performance. Our algorithm provided excellent performance in the first 3 years and provided good performance in all 10 years. (D) MAE within 0.75 dioptres was considered a clinically acceptable prediction. Meanwhile, our algorithm achieved clinically acceptable prediction of the refraction value at time points after the initial baseline visit (the MAE ranged from 0.253 to 0.395 for 3 years, 0.394 to 0.496 for 5 years, and 0.503 to 0.799 for 8 years). AUC, area under the curve; D, dioptres; MAE, mean absolute error; OOB, out-of-bag.