Geostatistical analysis of disease data: estimation of cancer mortality risk from empirical frequencies using Poisson kriging

Table 6 Performance comparison of alternative estimators: goodness of the model of uncertainty. Results obtained on average over 100 realizations generated under two different population size scenarios and 3 types of risk map (1 = observed, 2 = smooth, 3 = random). Poisson kriging was conducted with the semivariogram estimated from the underlying risk values (true γ_R(h)) or the simulated mortality rates. Bold numbers refer to best performances outside the use of observed rates and the ideal case where the true semivariogram of risk is known.

Estimators	WF population			BF population
BREAST CANCER	Scenario 1	Scenario 2	Scenario 3	Scenario 1	Scenario 2	Scenario 3
Observed rates	0.973	0.974	0.975	0.956	0.964	0.965
Population-weighted average	0.967	0.943	0.968	0.887	0.761	0.910
Global Empirical Bayes	0.847	0.843	0.974	0.908	0.787	0.804
Local Empirical Bayes	0.964	0.952	0.971	0.899	0.782	0.921
Poisson kriging (true γ_R(h))	0.971	0.972	0.947	0.716	0.842	0.887
Poisson kriging	0.965	0.913	0.940	0.761	0.927	0.803
CERVIX CANCER
Observed rates	0.937	0.939	0.970	0.935	0.935	0.922
Population-weighted average	0.933	0.901	0.933	0.852	0.743	0.888
Global Empirical Bayes	0.875	0.870	0.973	0.833	0.785	0.787
Local Empirical Bayes	0.960	0.920	0.962	0.875	0.771	0.901
Poisson kriging (true γ_R(h))	0.963	0.947	0.973	0.931	0.935	0.966
Poisson kriging	0.968	0.924	0.934	0.940	0.893	0.935

ISSN: 1476-072X