Smooth individual level covariates adjustment in disease mapping.

Area of Science:

• Biostatistics
• Spatial Epidemiology
• Statistical Modeling

Background:

• Disease mapping models often require both individual and area-level covariates.
• Existing conditional autoregressive (CAR) models like indiCAR accommodate these but assume log-linear individual covariate effects.
• Non-log-linear relationships between individual covariates and outcomes are common but not well-addressed in spatial regression.

Purpose of the Study:

• To introduce smooth-indiCAR, an extension of the CAR model accommodating both linear and nonlinear individual-level covariate effects.
• To adjust for group-level covariates and spatial correlation in disease rates.
• To provide a flexible framework for spatial regression in disease mapping.

Main Methods:

• The smooth-indiCAR model utilizes penalized splines to capture nonlinear effects of continuous individual-level covariates.
• A two-step estimation procedure separates individual and group-level covariate effect estimation.
• A distributed computing framework supports application to Big Data scenarios with numerous covariates.

Main Results:

• Simulation studies demonstrate that smooth-indiCAR yields reliable estimates for regression and random effect parameters.
• The method successfully models both linear and nonlinear covariate effects in spatial regression.
• Performance evaluation confirms the robustness of the proposed algorithm.

Conclusions:

• The smooth-indiCAR method provides a valuable advancement for spatial disease mapping, particularly when nonlinear individual-level covariate effects are present.
• It offers a statistically sound approach to handling complex covariate relationships in disease surveillance.
• The methodology was successfully illustrated using neutropenia admission data from New South Wales, Australia.