Our investigation had four main limitations. First, refractive error was measured infrequently, and by means of non-cycloplegic autorefraction. Much greater precision in determining the exact time of myopia onset has been obtained in previous studies using annual cycloplegic autorefraction assessments.
44,70 Noncycloplegic autorefraction has relatively high sensitivity and specificity for detecting myopia in older children, but is likely to have led to a progressively greater proportion of nonmyopic subjects being classified wrongly as myopic at earlier ages, due to the tendency for younger subjects to accommodate more during the test.
55 However, we chose to maintain a consistent threshold (≤−1.00 D) for classifying myopes, since this provided greater stringency in detecting, and thus excluding, children who already were myopic when we came to study incident myopia from the age of 11 years. In support of the validity of our myopia classification method, the risk of incident myopia was similar when we considered all subjects who were “nonmyopic” at age 11 years, and when we restricted our analysis to those categorized as “emmetropic/hyperopic” at this age. Second, we estimated time spent outdoors using a crude assessment method (parental questionnaire) administered at only a single time point. More frequent assessments,
44 and more precise, quantitative assessment methods
71,72 (Hewitt, A.J., et al.
IOVS 2011;52:ARVO E-Abstract 1190) would provide the opportunity to quantify better the full extent of the associationwith time spent outdoors. Inaccurate parental questionnaire responses, or changes in the amount of time individual children spent outdoors over the period from age 11 to 15 years, would each act to lessen our statistical power to detect an association between time outdoors and myopia development, and cause us to underestimate this variable's effect size. Furthermore, our analyses relied on the assumption that children's prospectively measured exposures would tend to be consistent throughout the 11- to 15-year age interval of particular interest, in other words that a child's behavior would track forward, or that any causal effect associated with an exposure would occur after a delay that was within our 4-year follow-up period. Third, it was apparent that the subjects attending the research clinic at age 15 years were not a random sample of those attending the research clinic at age 11 years. Although the differences between the 15-year clinic attendees and nonattendees were small, they nevertheless limit the extent to which our results can be generalized more widely. Importantly, there also was substantial loss-to-followup over the 15+ years of the ALSPAC study. Coupled with the high rate of missing data for key predictor variables, this meant, for example, that only 2845 (Figure S1; Online) of the approximately 14,000 children who were alive at 1 year of age were available for inclusion in the analysis of incident myopia from the age of 11 years (note that of these, only the 2542 children not already myopic at age 11 years actually could be included). Because the relationship between time spent outdoors and myopia may have differed in the children who regularly attended clinics compared to those never/rarely attending, this is a further reason for caution in extrapolating from the observed results to the general population. This limitation relating to the sampling of subjects also is likely to apply to the other cohorts that have been used to examine the relationship between time outdoors and myopia. Indeed, since visual/refractive development is only one component of the ALSPAC study, it may be that our results are more representative than those obtained from cohorts whose sole purpose has been to study refractive development. Fourth, our study had a high rate of missing data for the key variables, parental myopia (31%), physical activity (14%), and time outdoors (12%). If these data were
not missing-at-random, the fact that they were missing will have introduced bias into our results. This either could have strengthened or weakened the observed associations, depending on the relationships between data not missing-at-random compared to the data that were observed.