Two-step spatiotemporal anomaly detection corrected for lag reporting time with application to real-time dengue surveillance in Thailand

Table 1 Comparison of model-based cluster detection methods with and without nowcasting for reporting lags under evaluation metrics and likelihood assumptions. The bold numbers represent the highest value in each category

Likelihood	Delay	Cluster	Evaluation metric
model	correction	detection	Se	Sp	NPV	PPV	Acc
		EXC	0.8723	0.6123	0.8394	0.6531	0.7221
		CPO	0.8241	0.2324	0.6895	0.4863	0.5237
	Yes	PIT	0.4025	0.5579	0.5291	0.4308	0.4874
		DIC	0.8611	0.2313	0.6671	0.4822	0.5172
		WAIC	0.8661	0.2326	0.6929	0.4869	0.5247
Poisson		EXC	0.1135	0.9833	0.5716	0.8502	0.5885
		CPO	0.7063	0.4141	0.7201	0.5336	0.5921
	No	PIT	0.4146	0.5804	0.5439	0.4511	0.5051
		DIC	0.7562	0.4672	0.6975	0.5413	0.5984
		WAIC	0.7778	0.4689	0.7175	0.5491	0.6092
		EXC	0.8611	0.5021	0.8291	0.6296	0.6916
		CPO	0.8477	0.1951	0.6467	0.5116	0.5325
	Yes	PIT	0.3862	0.4337	0.4779	0.4403	0.4618
		DIC	0.8589	0.1928	0.6464	0.5112	0.5319
Generalized		WAIC	0.8577	0.1951	0.6466	0.5116	0.5325
Poisson		EXC	0.0161	0.9889	0.5716	0.8502	0.5885
		CPO	0.7771	0.4281	0.6691	0.5634	0.5981
	No	PIT	0.3129	0.6369	0.4939	0.4501	0.4791
		DIC	0.7777	0.4286	0.6699	0.5638	0.5987
		WAIC	0.7779	0.4281	0.6691	0.5634	0.5981

ISSN: 1471-2288