Spatially Explicit Nonlinear Models for Explaining the Occurrence of Infectious Zoonotic Diseases
Table 3
Model summary statistics for spatially explicit models describing the occurrence of medically diagnosed cases of Lyme disease and Rocky Mountain spotted fever for the 2000–09 study period within Tennessee.
Model type
Lyme disease (LD)
Rocky mountain spotted fever (RMSF)
GBT
SLR
NNET
CART
GBT
SLR
NNET
CART
Model performance
Misclassification rate
0.232*
0.272
0.288
0.296
0.304
0.312
0.288*
0.296
Average square error
0.187
0.182
0.253
0.206
0.230
0.210
0.232
0.213
ROC
0.789
0.812
0.688
0.674
0.702
0.727
0.696
0.712
PPV
83.7%
75.0%
77.1%
85.7%
69.8%
68.5%
72.5%
70.4%
Sensitivity
66.1%
67.7%
59.7%
48.4%
62.7%
62.7%
62.7%
64.4%
Specificity
87.3%
77.8%
82.5%
92.1%
75.8%
74.2%
78.8%
75.8%
Input variables**
Land cover
Forested wetland
+/−
−
+
+/−
Nonforested wetland
+/−
−
Pasture/grassland
+
+/−
+/−
+
Upland deciduous forest
+/−
+/−
−
Urban/developed
+
+
+
+
+/−
Wetland type
PUBHh
−
Geographic
Distance to river
+/−
Demographic
Population counts
+
+
+
+
+
+
+
+/−
Median income
+
+
+/−
+
Clinical
Lyme Dis. cooccurrence
+
+
+
+
RMSF cooccurrence
+
+
+
+
*Best model chosen using lowest misclassification rate on validation dataset.
**Denotes that aggregations were made at 1.6 and 8 km where applicable. Variables missing from this table indicate nonsignificance across all models, and plus and minus signs indicate direction of relationship.
