Neighborhood-level confounding in epidemiologic studies: unavoidable challenges, uncertain solutions.