Appendix 1 of Guggenheim, Mol Vis 2015; 21:621-632.

Autorefraction without cycloplegia is inaccurate in children. The level of inaccuracy depends on the type of instrument (e.g.,
open-field versus closed-field), the age and refractive error of the subjects, other subject-specific factors, and random
sources of variation [33,35-38]. In general, the degree of inaccuracy is worse in younger compared to older children, and typically causes hyperopia to
be under-estimated and myopia to be overestimated. A convenient and widely-used, albeit simplistic, method to quantify the
inaccuracy of non-cycloplegic autorefraction is to assess the mean ± standard deviation of the difference between the non-cycloplegic
measurement and a gold standard measurement (e.g., cycloplegic autorefraction or subjective refraction). This approach assumes
that there is a systematic measurement error coupled with a random measurement error: it therefore ignores the effects of
the child’s age and refractive error. Reports of non-cycloplegic autorefraction measurement error in population-representative
samples vary widely, e.g., −0.36±0.41 D for 13-year-old children in the BATS & TEST cohorts [35] and −1.23±0.97 D for children aged 7–18 years-old in the Shunyi district RESC study [38] reflecting the age range and refractive error distribution of the respective samples. As reported previously [32] we compared non-cycloplegic autorefraction measurements obtained at ALSPAC research clinics against subjective refraction
findings recorded by the participant’s own optometrist, for optometrist eye examinations performed within 6 months of the
15-year ALSPAC research clinic visit (n=346). The mean difference between the two measures was −0.22±0.84 D. Although this
sample is not representative of the ALSPAC cohort, it does at least provide an indication of the degree of measurement error.
New information provided by ALSPAC allowed us to exclude results for 13 of the above subjects whose clinic record indicated
that an “error” occurred during autorefraction, and also to examine separately the findings for subjects who were confirmed
as *not* wearing contact lenses during the autorefraction measurement (Figure 1 and Figure 2); typically, this ALSPAC record either confirmed that contact lenses were not worn, or was left incomplete by the clinical
assessor. The measurement error of non-cycloplegic autorefraction better approximated a normal distribution in subjects confirmed
as not wearing contact lenses (contact lens wear excluded: standard deviation of difference=0.60 D, n=108; contact lens wear
*not* excluded: standard deviation of difference=0.95 D, n=225) suggesting that outlier autorefraction readings – likely due to
a handful of subjects not removing their contact lenses before autorefraction and thus erroneously appearing to be emmetropic
by autorefraction – had inflated the standard deviation of the subjective refraction versus non-cycloplegic autorefraction
comparison (Figure 2). The greater the degree of measurement error noise, the greater the underestimation of SNP-heritability (Figure 3), therefore we chose to use a normal distribution with a standard deviation of 0.50 D from which to simulate measurement
error noise, to provide a conservative adjusted estimate of SNP-heritability. To access the data, click or select the words
“Appendix 1.”