Cycloplegic refraction is the gold standard for epidemiological studies

In the early days of ophthalmology, there was considerable debate about the importance of cycloplegia for accurate refraction. Duke-Elder's Practice of Refraction (10th edition, Abrams 1993) summarizes what became the general consensus on the importance of cycloplegia for accurate refraction.

Since then, many studies on children over the years have shown that lack of cycloplegia is associated with some overestimation of the prevalence of myopia, and marked errors in the estimation of the prevalence of emmetropia and hyperopia (for example Zhao et al. 2004; Choong et al. 2006; Fotedar et al. 2007). The amount of error appears to differ between individuals, as well as on the type of cycloplegic refractive error, and thus there is no reliable way of adjusting non-cycloplegic refractions to approximate cycloplegic refractions. Non-cycloplegic refraction is thus problematic in epidemiological studies, particularly in studies addressing questions of associations, because inconsistent variations in the dependent variable (be it refractive category or spherical equivalent refraction (SER)) can obscure significant associations.

As ophthalmological research spread beyond its origins in populations of European origin, another problem has emerged – that of ensuring that cycloplegia is adequate. Since its introduction in 1951, cyclopentolate has largely replaced atropine and homatropine in epidemiological studies, although tropicamide, a somewhat weaker agent (Smith 1976; Lovasik 1986; Egashira et al. 1993), is sometimes used. Careful choice of cycloplegic regime can be important, as it is well known that it is more difficult to achieve complete cycloplegia in children with darker irises (Kleinstein et al. 1999; Fan et al. 2004), because of the sequestration of cycloplegic agents by iris pigment.

Particular attention therefore needs to be paid to using appropriate cycloplegic regimes in populations of different ethnic origin, such as those of southern rather than northern European origin. The problem is particularly acute for populations which are not of European origin, where much darker irises are the general rule, and where very accurate measurements are required. The literature in this area is far from systematic, but in general, we recommend the use of 1% cyclopentolate, and we note that in populations with darker irises such as the Chinese population, three or even four drops may be required to achieve satisfactory cycloplegia for most participants (Lan et al. 2013).

We recently surveyed the most recent 100 papers in PubMed using the search term ‘prevalence and myopia and children’, finding that, consistent with the importance of cycloplegia in children, most studies used cycloplegic refraction. Studies on children which do not use cycloplegia need to be critically reviewed to ensure that the non-cycloplegic approach is appropriate to the study goals, and any data of this kind that are published need to be carefully interpreted. One approach to critical evaluation of non-cycloplegic data is to determine whether the same conclusions can be obtained using axial length, and even better, the AL/CR ratio as surrogate measures, as they correlate highly with SER (Grosvenor & Scott 1994; Goss & Jackson 1995; Hashemi et al. 2013), but are not subject to errors associated with lack of cycloplegia. We stress, however, that cycloplegic refraction remains the definitive measure. It does require considerable effort to obtain high participation rates in epidemiological studies which use cycloplegia, but it is possible to achieve high participation, as the pioneering studies on children in Taiwan (Lin et al. 2004) and the RESC series on refraction in over 40 000 children (Maul et al. 2000; Negrel et al. 2000; Zhao et al. 2000, 2002, 2004; Dandona et al. 2002; Ellwein 2002; Murthy et al. 2002; Naidoo et al. 2003; He et al. 2004, 2007; Goh et al. 2005) show. Studies which do not use cycloplegia should not be encouraged, and we suggest that relying on surrogate measures should only be used as a backup, when cycloplegia is genuinely impossible.

The literature survey also revealed that over 20% of the papers did not clearly state whether cycloplegia was used, or did not clearly specify the cycloplegia regime in the abstract. Given its importance, we recommend that clear specification of the regime for cycloplegia in the abstract should be standard practice for authors and journals.

This consensus around the importance of cycloplegia in children left undefined at what age cycloplegia was no longer needed. It has often been implicitly assumed that cycloplegia is not necessary beyond childhood or early adulthood, possibly because of the evidence suggesting that accommodation was very high in neonates and declined significantly during adolescence (Duane 1922). However, this picture has been overturned by recent studies of changes in accommodation with age which have overcome the limitations of subjective push-up estimates, using minus lenses combined with measurement of accommodation with a Grand Seiko autorefractor (Anderson et al. 2008). Using this approach, it is clear that the amplitude of accommodation hardly declines from the age of 3 until after the age of 20–30, is still appreciable at the age of 40, and only reaches minimal levels at the age of 50. This corresponds to what we know clinically about the age of onset of presbyopia and the need for reading glasses.

Combined with logistical issues, the idea that accommodation declines rapidly in older children and young adults probably contributed to the fact that none of the classical studies of refraction in older adults, such as the Beaver Dam Eye study (Wang et al. 1994), the Blue Mountains Eye Study (Attebo et al. 1999) and the Tanjong Pagar Eye Study (Wong et al. 2000), used cycloplegia. Fortunately, their high age cut-offs have protected them from significant errors. This is not true for major national population surveys such as the US NHANES (Sperduto et al. 1983; Wong et al. 2000; Vitale et al. 2008, 2009) and Korean NHANES (Kim et al. 2013), which cover children and young adults as well as older adults. In addition, a number of population-based studies on older adults in Europe have followed the non-cycloplegic model (Gudmundsdottir et al. 2005; Bertelsen et al. 2013; Verhoeven et al. 2013; Mirshahi et al. 2014; Wolfram et al. 2014), but some have extended the age range downwards into age ranges where lack of cycloplegia may be a significant problem.

To the best of our knowledge, only the Tehran Eye Study has systematically compared cycloplegic and non-cycloplegic refractions in a population-based survey across a range of ages (Hashemi et al. 2004; Fotouhi et al. 2012). The mean difference by age for different refractive classes (myopic, emmetropic and hyperopic) is shown in Fig. 1. Very similar results have been reported from a retrospective study of differences between cycloplegic and non-cycloplegic refractions from a large optometry practice (Schallhorn et al. 2014, ARVO Abstracts #2734). For comparative purposes, we have reproduced figure 3 from Anderson et al. (2008) as Fig. 2. The parallel between the changes in accommodative amplitude with age and the changes in the difference between cycloplegic and non-cycloplegic refraction for those with cycloplegic hyperopia with age is striking.

The gap between cycloplegic and non-cycloplegic refractions was high for hyperopes (mean 0.7–0.8 D) up to the age of 30, and then declined progressively, but was still appreciable at the age of 40 and minimal only by around the age of 50. This study showed that over the age range from 20 to 50, the greatest errors in SER were observed for those with hyperopic refractions. Errors were much smaller for myopic refractions, as is the case in children and adolescents. It also pointed out that population estimates of mean SER were particularly prone to error and suggested that the need for cycloplegia could only be ignored, in most circumstances, in studies on adults over the age of 50.

Some studies do not appear to support this idea. In particular, Krantz et al. (2010) compared cycloplegic and non-cycloplegic refractions in the Beaver Dam Offspring Study for those aged 22–84, with the overall conclusion stating that ‘cycloplegia may not be necessary in studies of refraction in adults’. However, the data reported in this paper are very similar to those obtained in the Tehran Eye Study, and in the text, there were many caveats to this conclusion. The study noted that differences between cycloplegic and non-cycloplegic refractions were not uniform across categories of age and refractive status. In participants younger than 50 years, differences between cycloplegic and non-cycloplegic measures were as high as 1–2 D, with particular problems for hyperopic refractions. The more detailed conclusions state that ‘when hyperopia and refractive errors are not primary end-points, or the study is conducted in a setting where the risks of cycloplegia may outweigh the greater accuracy, the error in measuring participants with hyperopia may be acceptable. If investigators are primarily interested in reporting the prevalence of hyperopia among young adults, cycloplegia may be necessary. However, if a study targets older populations, the expected number of young participants with hyperopia is low, or the study is measuring the prevalence of myopia, cycloplegia may not be necessary’. These caveats mean that the results of these two studies converge on the conclusion that measurement of non-cycloplegic refractions in those younger than 50 is problematic and that cycloplegic refraction should be considered as the gold standard, at least up to this age.

Krantz et al. (2010) have made an important point by stressing that how crucial cycloplegia depends on the study aim. If the only important outcome measure is the prevalence of myopia, then cycloplegia may not be crucial, although there may be some overestimation, even in populations up to the age of 50. But, without cycloplegia, measurements of other refractive categories as well as SER are unreliable, which will prevent full exploitation of the results of large epidemiological studies. Studies of associations with risk factors are likely to be significantly affected.

The conclusion that cycloplegia is necessary in young adults has been supported in some other studies on those under 50 (Jorge et al. 2007), but conflicts with the conclusions of a recent paper which suggests that cycloplegia is required for adolescents, but not for young adults (Sanfilippo et al. 2014). Given the otherwise consistent and strong evidence that cycloplegia is required in young adults, the weak cycloplegia (1% tropicamide) used in the young adults in this study, compared to the 1% cyclopentolate used in the adolescents, may explain the discrepancy. A recent review of the literature on ophthalmic diseases suggests that there is a major gap in coverage, with few studies covering young and middle-aged adults (Forward et al. 2012). In filling this gap, it needs to be clearly recognized that cycloplegia is necessary for accurate refraction.

Even for estimating the prevalence of myopia, cycloplegia may sometimes be essential. This situation occurs in paediatric populations, where the prevalence of myopia is very low (Giordano et al. 2009; Dirani et al. 2010; MEPEDS Study Group 2010; Lan et al. 2013; Wen et al. 2013). Under these conditions, even small errors in the estimation of myopia can distort the study outcomes considerably, and under these conditions, particularly rigorous attention to cycloplegia is essential to ensure that a low prevalence of pseudomyopia is not confused with a low prevalence of genuine myopia. Similar problems may occur in developing countries, where the prevalence of myopia is often very low (Pokharel et al. 2000; Dandona et al. 2002; Murthy et al. 2002; Naidoo et al. 2003; Gao et al. 2012).

Cycloplegia may also be particularly essential when the important study outcomes concern longitudinal change. In some circumstances, cycloplegia may even be important in those over 50. For example, while it is generally accepted that there are longitudinal hyperopic shifts in refraction with age (Brown et al. 1999), there is little direct longitudinal evidence of such changes, and as far as we are aware, none that are based on cycloplegic refraction. As the magnitude of the longitudinal shifts is small, even small errors associated with non-cycloplegic refractions can undermine the validity of the results.

However, it should be recognized that when myopia is the major end-point measure, non-cycloplegic refractions may give useful results. Logically, increased variability and error in the estimation of myopia would be expected to make it harder to find associations, but it would not, except in exceptional circumstances, be expected to lead to false associations. An example of this comes from the work of Jonas and colleagues on myopia in children in the Beijing Eye Study, in which non-cycloplegic refractions were used to establish a relationship between the amount of time that children spend outdoors and the prevalence of myopia (Guo et al. 2013). In this case, the surrogate measures AL and AL/CR were used to confirm the associations. This outcome does not however justify abandoning cycloplegia as the gold standard, because the conclusions based on the more problematic methodology were rendered plausible because of previous studies which did use gold standard techniques (Rose et al. 2008a,b).

In summary, the current evidence suggests that cycloplegic refraction should be considered as the gold standard for epidemiological studies of refraction, not only in children but in adults up to the age of 50. This is particularly important in studies on associations with risk factors, because use of non-cycloplegic refraction leads to errors in the estimation of refractive categories, and in estimation of SER. The need for cycloplegia is less crucial if study aims are narrowly defined around the prevalence of myopia, but this is likely to be only rarely the case. Even when the prevalence of myopia is the only study outcome, particular attention needs to be paid to cycloplegia, and the adequacy of cycloplegia, if the prevalence of myopia is low, as it often is in paediatric populations, or in developing countries, particularly in populations which are not of European ancestry.