Table 4. Diagnostic Tests for Cut Points of a Screening Tool for Diabetic Retinopathy for Use in Low-Income Communities, by Misclassification-Cost Ratio and Various Scenarios of Diabetic Retinopathy Prevalence, Mexico, 2014–2016.
| Misclassification Cost Ratiob | Predictive Probit Model (n = 939)a
|
||||
|---|---|---|---|---|---|
| Sensitivity, % | Specificity, % | Positive Predictive Value, % | Negative Predictive Value, % | z Cut Point | |
| Diabetic retinopathy prevalence of 31.7% (observed) | |||||
| 1 | 56.4 | 83.0 | 60.7 | 80.4 | −0.046 |
| 4 | 82.9 | 61.9 | 50.3 | 88.6 | −0.640 |
| 10 | 96.6 | 28.7 | 38.7 | 94.9 | −1.209 |
| Diabetic retinopathy prevalence of 35.0% | |||||
| 1 | 60.1 | 81.1 | 63.2 | 79.1 | −0.121 |
| 4 | 90.9 | 45.9 | 47.5 | 90.4 | −1.017 |
| 10 | 96.6 | 28.7 | 42.2 | 94.1 | −1.209 |
| Diabetic retinopathy prevalence of 40.0% | |||||
| 1 | 67.8 | 76.4 | 65.7 | 78.1 | −0.305 |
| 4 | 90.9 | 45.9 | 52.8 | 88.4 | −1.017 |
| 10 | 96.6 | 28.7 | 47.5 | 92.8 | −1.209 |
| Diabetic retinopathy prevalence of 45.0% | |||||
| 1 | 71.5 | 74.0 | 69.2 | 76.0 | −0.374 |
| 4 | 96.0 | 31.7 | 53.5 | 90.6 | −1.190 |
| 10 | 96.6 | 28.7 | 52.6 | 91.3 | −1.209 |
a
Multivariate probit model with any grade of diabetic retinopathy (grade 1, grade 2, or grade 3) as dependent variable according to Revised English Diabetic Eye Screening Program Grading System (25). Estimated coefficients from the multivariate probit model are shown in Table 2.
b
Misclassification-cost ratio = cost of classification of false negatives divided by cost of classification of false positives. Ratios of 1, 4, and 10 were used, assuming that false-negative classification of a person receiving diabetic retinopathy screening would generate greater health costs than would a false-positive classification.