Log-linear, logistic model fitting and local score statistics for cluster detection with covariate adjustments

Area of Science:

• Spatial statistics
• Biostatistics
• Epidemiology

Background:

• Traditional p-value computation for spatial scan statistics relies on Monte Carlo simulations.
• Adjustments for covariate effects are standard but can lead to overly conservative estimates when covariates are geographically unbalanced.
• This conservatism results in a loss of statistical power for detecting spatial clusters.

Purpose of the Study:

• To address the issue of overly conservative p-value estimates in spatial scan statistics when covariates are geographically unbalanced.
• To propose an alternative statistical procedure that improves power and accuracy.
• To explore extensions for multiple or continuous covariates.

Main Methods:

• Utilizing local score statistics as an alternative to standard Monte Carlo simulations.
• Fitting parameters using log-linear or logistic models.
• Developing extensions for handling multiple and continuous covariates.

Main Results:

• The proposed method using local score statistics effectively addresses the conservativeness of standard Monte Carlo p-values.
• This approach leads to increased statistical power in detecting spatial clusters.
• The method is shown to be robust even with geographically unbalanced covariates.

Conclusions:

• Local score statistics offer a more powerful and accurate approach for p-value computation in spatial scan statistics with unbalanced covariates.
• This method mitigates issues of confounding between covariates and location.
• The procedure is adaptable for complex covariate scenarios, including multiple and continuous variables.