We initially compared ROC curves for each of the five corneal zones against the sum of all five zones using the presence or absence of aqueous, evaporative, or mixed dry eye as the disease marker. In all cases, the sum of zones was significantly better than individual zones at predicting the presence of dry eye disease, with
P values less than 0.05 for all individual zones when compared with the sum of zones (
Table 2). We next repeated this analysis using pairs of zones instead of single zones to compare against the sum of all zones. In the case of aqueous disease, the combination of inferior and superior corneal zones was able to detect the disease as efficiently as using information from all five zones (
P = 0.130;
Table 3), whereas the inferior/central zone combination predicted evaporative disease (
P = 0.059;
Table 3). Mixed dry eye disease was predicted by several zone pairs, including superior/inferior (
P = 0.565), inferior/nasal (
P = 0.088), and inferior/central (
P = 0.106), with similar efficiency to five zones (
Table 3). ROC curves for aqueous, evaporative, and mixed dry eye are shown in
Figure 2 for each of the effective zone pairs, the sum of five zones, and an ineffective zone pair.
Review of the corneal staining patterns for each type of dry eye disease (see
Fig. 1) indicated that the predictive zone pairs were not simply those with the highest mean staining, although in all cases the inferior zone was required to predict the presence of a dry eye subtype. Review of the data (
Table 3) show that at the ideal points of the ROC, the thresholds for the Baylor scores with optimal sensitivity and specificity differ between the scores from zone pairs and those from sum of the five zones. We found that a Baylor score greater than 1 for the superior/inferior zone pair predicted aqueous disease with comparable efficacy to five zones. Similarly, the combination of inferior/central zones with a Baylor score greater than 1 predicted evaporative dry eye, whereas superior/inferior, inferior/nasal, and inferior/central zone pairs predicted dry eye at Baylor score cutoffs of greater than 2, greater than 3, and greater than 2, respectively. The actual sensitivities and specificities attained by staining grades in zone pairs were surprisingly high. In some instances, the sensitivity achieved by the zone pairs could be as high as 87% and the specificity as high as 84%.