Using Decision Curve Analysis to Evaluate Common Strategies for Myopia Screening in School-Aged Children

Purpose: To evaluate common strategies for screening myopia. Methods: A total of 2,248 children aged 6 to 12 years from five randomly selected primary schools were included for the screening. Enrolled study participants underwent distant uncorrected visual acuity (UCVA, Standard Logarithmic Visual Acuity E Chart) and non-cycloplegic auto-refraction (NCAR, Topcon KR-8800). Among them, 1,639 children (72.9%) accepted cycloplegic auto-refraction. Taking rejection of cycloplegia into account, receiver operating characteristic curves were drawn to compare the accuracies of the four strategies (I, Cycloplegic auto-refraction; II, NCAR; III, UCVA; IV, Combination of UCVA and NCAR). Decision curve analysis (DCA) was used to compare net benefits. Tenfold cross-validation was used for statistical analyses. Results: For myopia (spherical equivalent refraction, SE <= -0.5D) screening, the mean sensitivities were 73.79% (SD: 5.40%), 85.57% (6.84%), 59.71% (13.49%), and 85.06% (6.68%) for Strategy I to IV; with mean specificities of 100% (0%), 87.43% (4.27%), 89.74% (10.25%), and 88.65% (5.07%), respectively. For screening early myopia (SE <= -0.5D and >=-1.0D), the mean sensitivities were 73.44% (7.69%), 82.39% (5.32%), 54.27% (14.58%), and 81.76% (9.60%) for Strategy I to IV; with mean specificities of 100% (0%), 79.13% (4.86%), 85.48% (9.86%), and 81.17% (4.16%). Based on DCA, the net benefits of Strategy IV were the highest, with the probability thresholds ranging from 12% to 50%, after adjusting the TestHarms. For early myopia, the net benefits of Strategy IV were the highest with the probability threshold ranging from 5% to 34%. Conclusion: Combination of UCVA and NCAR produced the highest net benefits for myopia screening.