Department of Biostatistics Seminar/Workshop Series

The Chicken or the Egg: Looking for Causation in Observational Studies

Pingsheng Wu, PhD

Assistant in Biostatistics, Department of Biostatistics, Vanderbilt University School of Medicine

Wednesday, February 9, 1:30-2:30pm, MRBIII Conference Room 1220

Asthma is the most common chronic disease of childhood. Over the past several decades the prevalence of asthma has risen 100 percent worldwide. Efforts to find effective and targeted primary and secondary asthma prevention measures are needed as currently none exist. Infant respiratory viral infection is one modifiable environmental factor that we hypothesize is in the causal pathway for asthma. Respiratory viral infections are associated with asthma in two ways: 1) They are the primary cause of acute asthma disease exacerbations. 2) Children who experienced severe respiratory viral infection as infants are 2-4 fold more likely to develop asthma by school age compared with children who did not experience severe respiratory viral infection. Whether infant viral infection causes asthma, or merely serves as a marker for those genetically predisposed to develop disease, is unknown. This is an important question since if the association is causal, prevention of asthma might be possible. As children can not ethically be randomized during infancy to viral infection or not, observational studies are the only alternative to answer such a question. This is one of several examples where appropriate statistical methods are needed to analyze the data without randomization to infer causality. Propensity score method has been widely used in observational studies to achieve pseudo-randomization among treatment groups, and has been proven to be an effective tool for reducing bias and balancing the distribution of confounders between comparison groups. Currently, application of propensity score methods is mostly limited to dichotomous treatments. This talk will cover the application of propensity scores to the above question as one example, as well as the planned evaluation of propensity scores to handle multiple level exposure, with and without ordering, and continuous exposure thus providing researchers an effective tool for biomedical research.