Adrian Hill


Screening decisions between colour normal and colour defective appear to be a simple choice between two Cartesian options, in reality they are represented by a continuum of probabilities ranging from 0 to 1. This is because none of the pseudo-isochromatic screening tests have a perfect correlation with a validating anomaloscope performance. The consequences are inevitable misclassifications of false positive or false negative from screening test performance. Sensitivity and specificity data will be presented as receiver operator characteristic curves for several colour vision tests from two independent studies(1000 adults and 440 children). An index of discrimination has been derived for each of the tests using the signal detection statistic d' and its associated confidence limits. Bayes' theory has been applied to the screening performance of each test to determine both the positive and negative predictive power of alternative tests results. Two features become evident from this analysis: (1) screening decisions for colour normal or colour defective cannot be achieved with comparable high confidence for a single test score cut-off criterion, and (2) the screening for colour vision deficiency in populations where the incidence of the condition is very low (e.g., females) may give a result in which the most rational decision is one of "colour normal" even if all the plates in the test were failed. There are important practical implications of these two findings for colour vision test users. Firstly, they either need to state the confidence with which they want to make inferences and decisions from individual test performance (this may require setting two cut-off criteria on the same test, one for decisions of normality, the other for decisions of colour deficiency) or consider the probabilities associated with a given test outcome. Secondly, the screening for colour vision deficiency in populations where its presence is rare (e.g., females) has such poor pr


Year: 1992

Program Number: 2:45 pm

Author Affiliation: n/a

Co-Authors: n/a

Co-Author Affiliation: n/a

Room: Ireland B