Approximate Inference in Disease Mapping with Sparse Log Gaussian Process Priors