Abstract:
|
Large, administratively collected datasets present many challenges for prospective analyses. Retrospective analyses, under nested case control designs, can avoid many of these issues, but do not use all of the information available in the larger cohort. They are further limited by the kinds of analyses that can be conducted and the population parameters that can be estimated. We propose examining the nested case control from a designed based perspective. We view the full cohort as a fixed finite population and the sampled cases and controls as an unequal probability weighted sample. We approach analyses of the case-control sample as an attempt to predict the results of prospective analyses on the full cohort, in other words producing designed-based predictions of prospective model-based estimates. Importantly, by approaching the problem from this direction, it allows for prospective analyses on large cohorts that are not traditionally possible. We apply our approach to a nested case control study of the effect of opioid exposure on community-acquired pneumonia.
|