Avian GIS models signal human risk for West Nile virus in Mississippi

Cooke, William H; Grala, Katarzyna; Wallis, Robert C

doi:10.1186/1476-072X-5-36

International Journal of Health Geographics

Table 4 Summary table; results comparison and validation.

From: Avian GIS models signal human risk for West Nile virus in Mississippi

Model	Variables/weights	Validation WNV data	Majority Risk				Mean Risk				Median Risk
			0	1	# hc	val.	0	1	# hc	val.	0	1	# hc	val.
			difference		% correct		difference		% correct		difference		% correct
Landscape -base	Road den. – 0.4 Stream den. – 0.3 Slope % – 0.2 NDVI – 0.1	Human cases in 2002 and 2003	5.4 1.0	6.4	65 62.5	104	5.5 0.7	6.2	65 62.5	104	5.5 0.7	6.2	67 64.4	104
Summer 2002	Land. base – 0.7 P-E – 0.3	Human cases in Sum. 2002	5.3 1.0	6.3	47 61.8	76	5.4 0.8	6.2	47 61.8	76	5.4 0.9	6.3	46 60.5	76
	Land. base – 0.8 P-E – 0.2		5.4 0.9	6.3	45 59.2	76	5.5 0.7	6.2	50 65.8	76	5.5 0.7	6.2	49 64.5	76
	Land. base – 0.9 P-E – 0.1		5.4 0.8	6.2	46 60.5	76	5.5 0.6	6.1	48 63.2	76	5.5 0.7	6.2	50 65.8	76
	Land. base – 1.0 P-E – 0.0		5.5 0.8	6.3	44 57.9	76	5.6 0.5	6.1	44 57.9	76	5.5 0.3	5.8	46 60.5	76
Fall 2002	Land. base – 0.7 P-E – 0.3	Human cases in Fall 2002	5.5 0.4	5.9	20 60.6	33	5.5 0.3	5.8	16 48.5	33	5.5 0.3	5.8	16 48.5	33
	Land. base – 0.8 P-E – 0.2		5.5 0.5	6.0	20 60.6	33	5.6 0.2	5.8	16 48.5	33	5.6 0.2	5.8	16 48.5	33
	Land. base – 0.9 P-E – 0.1		5.6 0.5	6.1	18 54.5	33	5.6 0.3	5.9	15 45.5	33	5.6 0.3	5.9	15 45.5	33
	Land. base – 1.0 P-E – 0.0		5.6 0.5	6.1	19 57.6	33	5.6 0.3	5.9	17 51.5	33	5.7 0.3	6.0	18 54.5	33
Summer 2003	Land. base – 0.7 P-E – 0.3	Human cases in Sum. 2003	5.4 1.4	6.8	20 66.7	30	5.4 1.3	6.7	21 70.0	30	5.5 1.2	6.7	21 70.0	30
	Land. base – 0.8 P-E – 0.2		5.4 1.4	6.8	23 76.7	30	5.5 1.2	6.7	25 83.3	30	5.5 1.2	6.7	24 80.0	30
	Land. base – 0.9 P-E – 0.1		5.5 1.4	6.8	22 73.3	30	5.5 1.2	6.7	25 83.3	30	5.6 1.2	6.8	25 83.3	30
	Land. base – 1.0 P-E – 0.0		5.6 1.3	6.9	24 80.0	30	5.6 1.1	6.7	24 80.0	30	5.6 1.2	6.8	25 83.3	30
Fall 2003	Land. base – 0.7 P-E – 0.3	Human cases in Fall 2003	5.6 1.0	6.6	16 59.3	27	5.7 0.7	6.4	16 59.3	27	5.7 0.8	6.5	16 59.3	27
	Land. base – 0.8 P-E – 0.2		5.6 1.1	6.7	18 66.7	27	5.6 1.0	6.6	18 66.7	27	5.6 1.0	6.6	18 66.7	27
	Land. base – 0.9 P-E – 0.1		5.5 1.4	6.9	24 88.9	27	5.6 1.1	6.7	23 85.2	27	5.6 1.2	6.8	23 85.2	27
	Land. base – 1.0 P-E – 0.0		5.6 1.4	7.0	24 88.9	27	5.6 1.2	6.8	25 92.6	27	5.6 1.3	6.9	26 96.3	27
	Average % correct for all models				66.8				61.3				66.9

The risk estimates by zip code were calculated using a zonal function. Majority, mean and median measures of risk were calculated and recorded for zip codes of WNV human occurrence (1) and zip codes of non-occurrence (0). Difference between the two categories of zip codes (0 versus 1) was determined and for all models. Estimated risk was higher for zip codes with at least one human case than for zip codes where human cases were not recorded. Modeling results were validated with human infection data (val.) for appropriate corresponding season. Number of human cases (#hc) in the high-risk category was determined. The high-risk category included zip codes of the combined top two risk classes (out of five) defined using quantile classification method. For each model percentage correct (% correct) was calculated to determine the measure of central tendency that works best for the validation. Examination of average % correct for all models indicated that the median (66.9%) and majority (66.8%) measures reflect the actual WNV risk better than the mean (61.3%).

Back to article page

ISSN: 1476-072X

Contact us

Submission enquiries: journalsubmissions@springernature.com