Design-based spatial modelling of malaria incidence among children in Nigeria using latent Gaussian model

National data on malaria are mostly individual-level datasets, collected through complex survey designs, e.g., Malaria Indicator Survey. However, models on malaria in the literature often use count data at geographical level, and scarcely account for complex survey design. The impact of complex sampling designs on data analysis results could lead to bias parameter estimates and nonrepresentativeness of the population. This study presented a hierarchical design-based spatial latent Gaussian model (DBSLGM) that accounted for individual-level factors for malaria incidence. DBSLGM was fitted to the 2021 Nigeria Malaria Indicator Survey (NMIS) data. Missing values were handled by multiple imputation methods and variance inflation factors was used to assess multicollinearity among covariates. Inference was done via integrated nested Laplace approximation (INLA). DBSLGM’s structured linear predictor followed a Gaussian prior, spatial effects followed the BYM2 prior, and the design information, primary sampling unit and strata effects both followed an independent Gaussian distribution. Relative risks, exceedance probabilties and clusters were extracted from the fitted model on imputed dataset. Posterior predictive checks, and comparisons diagnostics were used to assess performance of the model. Residual Moran’s I test was done on the fitted DBSLGM, and showed no evidence of remaining residual spatial autocorrelation. Posterior odd ratios were reported for risk factors on malaria incidence. Sensitivity analyses done showed the robustness of the model in several cases. Results of DBSLGM on 2021 NMIS data emphasised sensitivity over specificity, nevertheless, the incorporation of survey weights, multiple imputation and spatial random effects provided estimates that were more representative of the target population and more appropriate for national-level inference. DBSLGM framework affords policy-makers an intuitive answer to the question of estimates of the probability that malaria risk in a state is elevated and exceeds the intervention threshold.