A road toward effective vision screening

Identification of children at risk for amblyopia is an essential first step toward the delivery of timely treatment. The road to a superior vision screening strategy comprises careful scrutiny of the available instruments and protocols as well as continued methodological improvements and refinement. The study of Rowatt and colleagues, whose work is reported in this issue of the Journal,1 takes us on that road. In an attempt to improve vision screening, the authors made two modifications to the manufacturer’s recommended methodology: they adjusted the SureSight (Welch Allyn Medical Products, Skaneateles Falls, NY) instrument’s failure criteria, and they used the MTI PhotoScreener (PhotoScreener Inc., Palm Springs, FL) instrument to achieve their desired lower referral rate and higher positive predictive value.

Children failing screening with the SureSight with a cutoff set to achieve 90% specificity and 63% sensitivity underwent a second vision screening with the MTI PhotoScreener. Those with a crescent in any portion of the photoscreening photograph failed screening. This is an excellent method to identify children with hyperopia.2

A different modification improves the identification of those with supra-threshold astigmatic error. The authors find that a cutoff for the SureSight at >2.20 D identifies children with >1.50 D of astigmatism by eye examination. This higher cutoff improves the referral rate while maintaining the positive-predictive value and thus higher specificity. Figure 1 of Rowatt et al1 has far more points below the line with a slope of 1 through the origin than above, suggesting that the SureSight software tends to overestimate the magnitude of the astigmatic error. The SureSight algorithm for calculating astigmatism and likely other refractive errors should be refined.

The Rowatt report pointed out one difference between their study and the VIP report3: “the VIP study did not evaluate children under age 3.” The two studies also had a different means of patient recruitment, making a direct comparison of the results difficult. The participants in the VIP study were from a cohort failing an acuity-based Head Start vision screening program. In the Rowatt et al study,1 the subjects were more cross-sectional. Acuity screening identifies children with a functional deficit. Vision screening by autorefraction (or photoscreening) is based on suprathreshold anatomic features. A population selected from functional criteria may screen positive more easily by functional methodologies, such as acuity screening, and vice versa with an anatomic test. For example, a child with oblique astigmatism may be more prone to be screened positive by a functional rather than an anatomic screening procedure. Therefore, an anatomic screening test may not perform identically on a cohort identified using a functional approach as compared with a cross-sectional sample.

The power of the Rowatt study would be still greater if a direct assessment of the sensitivity and specificity were possible. A clinical examination of 5% to 10% of the children in the study would give us a good estimate of the specificity. Furthermore, the examination of a fraction of the children failing screening by certified examiners would address the authors’ concern regarding the standardization of the eye examination. The number of subjects receiving gold standard examination need not be daunting to render the results of the study much more powerful.

Rowatt et al1 showed us that the SureSight algorithm yields greater astigmatic error than actual and that the results can be refined by the adjunctive use of the MTI PhotoScreener for detection of other refractive errors. These necessary refinements present an opportunity for improvement of SureSight. While the authors conclude that the SureSight “can be used successfully” in vision screening with incorporation of cutoff for higher specificity, I am cautiously optimistic that the instrument’s algorithm for the autorefraction output can be refined, obviating the need to adjust the instrument cutoff. When only one of the two modifications is adopted without the adjunctive use of the MTI PhotoScreener, how adequately the SureSight will perform is not known. I believe the SureSight should be used in select regional vision screenings, where a subset of children passing and nearly all children failing screening are examined. An effectively functioning vision screening methodology after such field testing can be even more fully embraced. The lower positive-predictive value in children younger than 3 years old suggests that another technology may be necessary to fill this gap.4 On the road to effective vision screening, we move forward by iterative steps; the methodologies are scrutinized, the limitations are embraced as new opportunities, and the door is kept open for new modalities.