Impact of confidence number on the screening accuracy of the retinomax autorefractor.

PURPOSE: To assess the impact of Retinomax reading confidence number on screening accuracy and to determine whether repeated testing to achieve a higher confidence number improves screening accuracy in preschool children.
METHODS: Lay and nurse screeners trained in the use of the Retinomax Autorefractor screened 1452 children enrolled in the Vision in Preschoolers (VIP) Phase II Study. All children also received a comprehensive eye examination. Using statistical comparison of correlated proportions, we compared sensitivity and specificity for detecting any VIP-targeted condition and conditions grouped by severity and by type (amblyopia, strabismus, significant refractive error, and unexplained decreased visual acuity) among three groups of children who had confidence numbers below, at or above the manufacturer's suggested confidence number of 8. The reading with the highest confidence number for each eye was used in the analysis. Each child's confidence number group was defined based on the lower confidence number of the pair of readings for the two eyes. Among the 771 (53.1%) children who had repeated testing either by lay or nurse screeners because of a low confidence number (<8) for one or both eyes in the initial testing, the same analyses were also conducted to compare results between the initial reading with confidence number<8 and repeated test reading with the highest confidence number in the same child. These analyses were based on the failure criteria associated with 90% specificity for detecting any VIP condition in VIP Phase II. We also examined the association between ocular conditions and confidence number. Hochberg procedure was used to adjust the p value for multiple comparisons.
RESULTS: A lower confidence number category was associated with higher sensitivity (0.78, 0.65, and 0.61 for<8, 8, >8, respectively, p=0.04) but much lower specificity (0.64, 0.89, and 0.93, p<0.0001) of detecting any VIP-targeted condition. Through repeated testing, 87% of readings that initially had a confidence number below 8 reached 8 or above, and the increased confidence number that resulted from repeated testing was associated with significantly higher specificity (0.81 vs. 0.86, p=0.002) and a nonsignificant change (by -0.04 to 0.03) in sensitivities. Children with any VIP-targeted condition, significant refractive error, hyperopia, astigmatism, or myopia were more likely to have a low confidence number.
CONCLUSIONS: A higher confidence number obtained during Retinomax Autorefractor screening is associated with better screening accuracy. Repeated testing to reach the manufacturer's recommended minimum value is worthwhile in preschool vision screening with the Retinomax. Failure to achieve manufacturer's recommended minimum value through repeated testing should be a factor considered in referring children for a comprehensive eye examination.