Predicting Principal Stratum Membership in Randomized Environmental Interventions

Principal Stratification;Environmental Interventions;Asthma;Clinical Trial;Biostatistics

Freeland, Katherine ElizabethPeng, Roger D. ;

Johns Hopkins University

关键词: Principal Stratification;    Environmental Interventions;    Asthma;    Clinical Trial;    Biostatistics;   

Others  :  https://jscholarship.library.jhu.edu/bitstream/handle/1774.2/38132/FREELAND-THESIS-2015.pdf?sequence=1&isAllowed=y

瑞士|英语

来源: JOHNS HOPKINS DSpace Repository

【 摘 要 】

Environmental interventions targeted at reducing indoor allergens and pollutants have shown promise as a method for improving respiratory health outcomes in children by reducing exposure in the home. However, in these interventions, it is difficult to determine the effect of reduced exposure, a post-randomization variable, on the respiratory outcomes. Using principal stratification, a framework for calculating principal effects (i.e. effects within a stratum), we are able to measure the effect of reduced exposure on respiratory outcomes. These principal effects allow for the comparison of treatment effects for those who would and would nothave seen a reduction in allergen (or pollutant) levels. With the exposure reduction variable, we can identify principal strata membership for some individuals in the control and treatment groups. However, the observed data only allow us partial identification of strata for other individuals. We develop a resampling based estimator that incorporates the uncertainty from fitting a model (of which the ;;true’form is unknown) to predict likely stratum membership, classifying individuals based on this estimated probability, as well as finite sampling uncertainty. This estimation procedure allows the model form to change to best fit the resampled data at hand, reflecting the true uncertainty in this process.