Causal Inference for Case-Control Studies

摘要: Author(s): Rose, Sherri | Advisor(s): van der Laan, Mark J. Abstract: Case-control study designs are frequently used in public health and medical research to assess potential risk factors for disease. These particularly attractive investigators researching rare diseases, as they able sample known cases of disease, vs. following a large number subjects waiting disease onset relatively small individuals. The data-generating experiment case-control involves an additional complexity called biased sampling. That is, one assumes the underlying that randomly samples unit from target population, measures baseline characteristics, assigns exposure, final binary outcome, but conditional probability distribution, given value outcome. One still desires causal effect exposure on outcome population.The targeted maximum likelihood estimator treatment based such studies is presented. Our proposed case-control-weighted relies knowledge true prevalence probability, or reasonable estimate this eliminate bias sampling design. We use weights, our weighting scheme successfully maps random into method sampling.Individually matched commonly implemented field health. While matching intended confounding, main potentiall\italicg benefit gain efficiency. investigate effects designs. also compare unmatched effort determine which design yields most information about effect. In many practical situations where parameter interest, researchers may be better served using design.We consider two-stage designs, including so-called nested studies, takes population completes measurements each subject first stage. second stage drawing subsample original sample, collecting data subsample. This structure can viewed missing full-data collected study. propose inverse-probability-of-censoring-weighted Two-stage common prediction questions. present analysis super learner Kaiser Permanente database generate function mortality prediction.