Predicting the onset of myopia in children by age, sex, and ethnicity: Results from the CLEERE Study

Abstract

Significance: Clinicians and researchers would benefit from being able to predict the onset of myopia for an individual child. This report provides a model for calculating the probability of myopia onset, year-by-year and cumulatively, based on results from the largest, most ethnically diverse study of myopia onset in the United States.

Purpose: This study aimed to model the probability of the onset of myopia in previously nonmyopic school-aged children.

Methods: Children aged 6 years to less than 14 years of age at baseline participating in the Collaborative Longitudinal Evaluation of Ethnicity and Refractive Error (CLEERE) Study who were nonmyopic and less hyperopic than +3.00 D (spherical equivalent) were followed up for 1 to 7 years through eighth grade. Annual measurements included cycloplegic autorefraction, keratometry, ultrasound axial dimensions, and parental report of children's near work and time spent in outdoor and/or sports activities. The onset of myopia was defined as the first visit with at least -0.75 D of myopia in each principal meridian. The predictive model was built using discrete time survival analysis and evaluated with C statistics.

Results: The model of the probability of the onset of myopia included cycloplegic spherical equivalent refractive error, the horizontal/vertical component of astigmatism (J0), age, sex, and race/ethnicity. Onset of myopia was more likely with lower amounts of hyperopia and less positive/more negative values of J0. Younger Asian American females had the highest eventual probability of onset, whereas older White males had the lowest. Model performance increased with older baseline age, with C statistics ranging from 0.83 at 6 years of age to 0.92 at 13 years.

Conclusions: The probability of the onset of myopia can be estimated for children in the major racial/ethnic groups within the United States on a year-by-year and cumulative basis up to age 14 years based on a simple set of refractive error and demographic variables.

Figure 1.

Model-derived probabilities at future ages (i + k) given the baseline data shown and baseline ages (i) of 7, 9, and 11 for Asian females (panels A-C) and white males (panels D-F). The probability of becoming myopic at a given age (i + k) given no previous myopia is shown by the filled squares ◼. The probability of remaining non-myopic by a given age (i + k) is shown by the filled circles . The probability of becoming myopic by a given age (i + k) is shown by the open circles .

Figure 2.

ROC analyses for myopia predictions year-by-year for three baseline ages: (A) 7 years; (B) 9 years; and (C) 11 years. Each line represents an additional year a child remained non-myopic. Model performance was best when predicting myopia onset within one year from the baseline age and decreased with each additional year a child remained non-myopic after the baseline age. The diagonal line represents a chance level of performance.