The goal of the present study was to determine whether it is possible to predict the future development of myopia in young children accurately enough that targeted interventions could practically be used to prevent or slow myopic axial elongation of the globe. The answer to this depends on several factors, including the accuracy, cost, and difficulty of the screening tests, and the prevalence of myopia, an important determinant of positive predictive value. Our results suggest that a predictive algorithm based on results of inexpensive, simple, and noninvasive testing (gender, A-scan ultrasonography, keratometry, and measurements of height and visual acuity) can achieve an acceptable positive predictive value in the range of 75%, at least in a setting such as Singapore, where myopia prevalence is high. It seems likely that such screening and prevention programs would generally be undertaken in settings with a similarly high burden of refractive error. In view of this fact, we chose to test the model in the Singapore cohort, rather than develop it based on Singapore data and test it in the Xiamen cohort, which had lower myopia prevalence.
The practicality of strategies for myopia prevention in children rests in part on the efficacy and safety of available treatments. Present modalities have clear limitations, including rebound progression after cessation of medical treatments such as atropine
13 and limited success of refractive strategies such as the use of progressive addition lenses,
23 –25 contact lenses,
20 –22 and deferred prescription of corrective spectacles.
26,27 Nonetheless, research is ongoing on the efficacy of compounds such as selective γ-aminobutyric acid type C antagonists in retarding axial elongation in animal models
44,45 ; the question of whether screening accuracy will limit the practical applications of such advances in the prevention of human myopia remains relevant.
The values for sensitivity, specificity, and AUC in the Singapore test cohort are somewhat lower than those reported by Zadnik et al.
39 for a model using baseline refractive error to predict onset of myopia ≤ −0.75 D within the Orinda Longitudinal Study of Myopia (sensitivity, 0.867; specificity, 0.733; AUC, 0.880). Jones-Jordan et al.
38 reported values comparable to ours (sensitivity, 0.625; specificity, 0.819) for a model using myopia at baseline and parental history of myopia to predict the onset of myopia ≤ −0.75 D in the Collaborative Longitudinal Evaluation of Ethnicity and Refractive Error Study. Unlike these studies, the present report used one cohort (Xiamen) to develop a model and another (Singapore) to test it. Although our model was somewhat more complex than that reported by Zadnik and colleagues,
39 our analyses (
Table 5) suggest that removal of any of the variables from the model reduced its performance as measured by the AUC. Applying Zadnik's model to the Xiamen data set yielded an AUC of 0.711, considerably reduced from a value of 0.974 for our original model. Unlike the Zadnik model, the baseline spherical equivalent refractive error was not an important predictor in the present study. Our results indicating that female gender is associated with myopia risk are consistent with other studies in Chinese populations.
30,31
Significantly increased risk for development of glaucoma,
10 cataract,
9 and retinal detachment
8 appears most pronounced at higher powers of myopic refractive error. Thus, in practice, there will likely be the greatest interest in early prediction and prevention of high myopia. The model in the present study was constrained by the absence of myopia in this range in the Xiamen cohort on which calculations were based, and thus focused on the ascertainment of myopia < −0.75 D. Children reaching this level of myopia by 11 years of age might be expected to subsequently progress to higher, more visually significant levels. Still, future prospective studies using comparable protocols in matched cohorts with greater prevalence of high myopia are needed to develop and test models predicting the onset of high myopia. Such models will likely benefit from higher sensitivity and specificity, in that there will be less overlap of baseline biometric characteristics between normal children and those who will later develop high myopia. However, the positive predictive value will likely be reduced by the lower prevalence of high compared with low myopia. It is unknown at present how the balance of these two influences will affect predictive value, and thus the practical utility, of models for high myopia.
Parental myopia appears in other studies
30,33,38 to be an important determinant of children's myopia progression, and thus might improve the accuracy of predictive models; this would, however, come at the cost of added complexity in the screening process, which might render the collection of such data impractical in many settings. In many parts of China, for example, a history of glasses wear, as has been used as a simple proxy for parental myopia in some studies, will not be useful in view of low rates of spectacle wear among myopic persons. The same may be true for detailed assessment of near work,
40,46 –48 outdoor activity,
49 and genetic makeup,
34 also previously reported as risk factors for myopia and myopia progression.
The strengths of this study include the validation of our model by replication in a Singapore cohort, in which data were gathered using the identical protocol and equipment as those used in the original Xiamen children. In each of these two quite different locations, a high degree of follow-up was achieved over 3 years. Both the settings and the screening modalities used are relevant to the practical problem of myopia screening: Singapore and China have very high burdens of myopic refractive error and all the tests used are widely available, inexpensive, and noninvasive. The high rate of success in data collection within the original studies indicates that screening of all children at entry to elementary school, for example, could be practical in a setting with high myopia prevalence such as Singapore or urban China.
Weaknesses of the study include the complexity of the model. The use of squared and cubed terms and interaction variables means that risk of myopia does not vary in an intuitively clear way with biometric factors such as axial length and corneal curvature. However, the purpose of this model was to predict future myopia in an eye as accurately as possible, and not to better understand risk factors for myopia. An additional limitation noted earlier is that a model predicting onset of higher degrees of myopia, not practical with the Xiamen cohort, would have been more relevant to the screening problems facing program planners hoping to decrease the visual burden of myopia. Finally, generalizability of the model is limited by the relatively narrow range of ages at baseline in the two samples.
Nonetheless, despite its limitations, this report is among the first of which we are aware to explore the potentially important question of whether myopia prevention strategies, as they improve, can be delivered accurately in a targeted fashion in China. It is hoped that further work on such models will be joined with a better understanding of the biochemistry of axial elongation to yield practical and cost-effective strategies for the prevention of myopia in severely affected populations.
Supported in part by The Joint Shantou International Eye Center of Shantou University and The Chinese University of Hong Kong.