We were interested to read the paper by Fotouhi and colleagues, on the prevalence of refractive errors in schoolchildren in Iran,1 in which the authors used cycloplegic autorefraction for children aged 7–15 years and non-cycloplegic autorefraction for children aged 15–18 years. This paper illustrates the well-recognised need for cycloplegia in order to obtain accurate refractive estimates in children up to at least age 12.2 However, the use of cycloplegia may reduce compliance and may not be possible in some situations—for example, studies of general child development such as the 1958 and 1970 national cohort studies in the UK. In such situations, other measures such as visual acuity may be used to infer the presence, but not the type of refractive error.3

We have faced this problem when participating with a birth cohort study (Avon Longitudinal Study of Parents and Children; ALSPAC4). Rather than collect no refractive data, we have examined the children with an autorefractor (a Cannon R50) without cycloplegia and conducted a nested comparison with refraction under cycloplegia, to assess the most appropriate interpretation and uses for the non-cycloplegic data. We included 7-year old children with acuity worse than 0.2 logMAR (6/9), despite the use of a pinhole (n = 414), and we compared the non-cycloplegic autorefractions with cycloplegic retinoscopy by an experienced optometrist in the 345 (83.3%) who agreed to the use of cycloplegic drops. Cycloplegia was induced using 1–2 drops of 1% cyclopentolate in each eye. After 20 min, the optometrist carried out the examination if the reflex was stable or added more drops if needed.

We assessed the data for the right eyes, with Bland–Altman plots and Receiver Operator Characteristic (ROC) curves (where different cut-off points are used to predict the presence or absence of particular refractive errors). The Bland–Altman analyses confirmed the expected bias towards “over-minussing” the spherical refractive error, increasing as the amount of hypermetropia increased (p<0.001). Estimation of astigmatic error showed no such bias, and was on average moderately accurate, but with large differences between methods for some individuals: the mean (SD) difference between methods was –0.13 (0.53) D, with the range of differences –3.00 to +1.00 D.

Table 1 illustrates the results obtained from ROC curves, with sensitivities shown for two arbitrarily set levels of specificity: ⩾95% and ⩾99%. The sensitivities of the autorefractor are best when screening for hypermetropia. If the target level of mean spherical equivalent (MSE) hypermetropia +2.00 is used as a cut-off point to identify affected children, then nearly all children identified will have hypermetropia at least that severe (⩾99% specificity), and 71% true cases will be detected. The sensitivities for myopia, astigmatism and anisometropia are worse, particularly at 99% specificity. Repeatability data are shown in table 2 and are comparable with other devices in the literature.

From these analyses, we suggest that these autorefractor data can be used to identify hypermetropic children with reasonable accuracy, although any prevalence data will underestimate the true value. The data for myopia, astigmatism and anisometropia are not accurate enough to use for prevalence estimates. Instead they could be better used to identify subgroups of children “enriched” with children who truly have the refractive error in question, as well as some who do not, eg, myopes and pseudomyopes, in risk-factor analyses.

Thus, with appropriate use, non-cycloplegic data may help the vision science community to gain useful information from general developmental studies such as ALSPAC and thus facilitate wider perspectives on how visual development and function impact on (and are impacted by) other aspects of a child’s life, as an adjunct to dedicated eye studies such as that by Fotouhi et al.1