Volume 10, Number 4—April 2004
Research
Predicting Geographic Variation in Cutaneous Leishmaniasis, Colombia
Table 3
Diagnostic statistics of predictive models for presence/absence of ACL transmissiona,b
Type of model | Predictors used | Accuracy measures |
|||||
---|---|---|---|---|---|---|---|
AUC | Maximum κ | Sensitivity (%) | Specificity (%) | PPV (%) | NPV (%) | ||
Single model for whole country |
Elevation |
0.66 |
0.23 |
59.9 |
65.8 |
41.3 |
80.3 |
Land cover |
0.70 |
0.28 |
53.4 |
75.7 |
46.9 |
80.2 |
|
All |
0.72 |
0.34 |
55.3 |
79.3 |
51.8 |
81.6 |
|
Combination of zonal models |
Elevation |
0.70 |
0.28 |
53.7 |
75.2 |
46.5 |
80.2 |
Land cover |
0.82 |
0.46 |
67.0 |
80.6 |
58.1 |
85.9 |
|
All | 0.84 | 0.54 | 62.8 | 89.1 | 69.8 | 85.6 |
aACL, American cutaneous leishmaniasis; AUC, area under receiver-operator curve; PPV, positive predictive value; NPV, negative predictive value. Sensitivity, specificity, PPV, and NPV are calculated at the probability threshold that gives the highest value of kappa.
bFor all comparisons of observations against predictions, χ2 > 79.2, df =1, p < 0.0001. κ values are given by (proportion correct – Proportion expected)/(1- proportion expected), where proportion correct = (a + d)/n, and proportion expected = (a + b) x (a + c) + (c + d) x (b + d)/n2. a = true positive predictions, b = false positive, c = false negative, d = true negative, n = total.