Detection of clusters of a rare disease over a large territory: performance of cluster detection methods

Table 3 Performance of cluster detection methods on one replicated dataset.

	Scan-c	Scan-e0	FleX	GA-1	Double	Mlink
p-value	< 0.0001	< 0.0001	< 0.0001	< 0.0001	< 0.0001	< 0.0001
No. LZ	18	18	13	13	8	13
True Positive LZ¹	10	11	8	9	6	7
Sensitivity²	0.91	1.00	0.73	0.82	0.55	0.64
PPV³	0.56	0.61	0.62	0.69	0.75	0.54
Cost⁴	9	7	8	6	7	10

Results for the sixth cluster scenario (compact cluster, E = 50.1 cases, covering 11 LZ), with a relative risk of 2.
Results based on 250 Monte Carlo replications. Scan-c: circular scan method, Scan-e 0: elliptic scan method with no penalty, FleX: unrestricted flexible scan method, GA-1: strongly penalized genetic algorithm, Double and Mlink: dynamic minimum spanning tree method with double and maximum link connections, respectively. No. LZ: number of living zones in the detected cluster. ¹ number of living zones in the intersection of the true and detected clusters. ² sensitivity: proportion of living zones in the true cluster that are correctly detected. ³ proportion of living zones in the detected cluster that are in the "true" cluster. ⁴ number of living zones that are either missed or erroneously detected.

ISSN: 1476-072X