TY - JOUR TI - Estimating the relationship between a transient effect and the onset of an acute event DO - https://doi.org/doi:10.7282/T3S1848V PY - 2015 AB - The case-crossover design was first published in 1991 as an epidemiological method to estimate the transient effect of an exposure on an acute event in research where primary data collection is conducted. Since the inception of the case-crossover design, the quality and availability of data warehouses has become standard. Health care providers and insurers have migrated from recording routinely collected patient information on paper to using electronic health records which are stored in data warehouses. This development has enabled researchers to observe the same acute events and exposures of interest in the traditional case-crossover paradigm at any time the patient is in care without expending the resources associated with primary data collection. Recent epidemiological studies have implemented the case-crossover design in situations where the data necessary for a retrospective cohort design are readily available. The case-crossover design's main appeal is that it implicitly controls for time-invariant characteristics of each patient in the study, measured or unobserved, by utilizing conditional logistic regression. In a retrospective cohort, an investigator typically would choose between using a Cox Proportional Hazard Model or a longitudinal logistic regression model. Since researchers also are interested in studying the transient effect of an exposure on subsequent acute events in an observational setting, and since developments in health information technology have provided researchers with more plentiful and detailed data than were available when the case-crossover design originally was proposed, researchers can now select from variety of methods. This thesis shows how the case-crossover design compares to a time-dependent covariate analysis in a cohort setting, and provides recommendations when one design preferable over the other. This thesis makes an important connection between the two designs, and proposes that the principle of lagged covariates can be applied in the case-crossover design. Furthermore, this thesis also proposes a two parameter, geometric lag estimation method which can describe a non-linear, deteriorating effect within the case-crossover design setting. KW - Public Health KW - Epidemiology--Statistical methods KW - Epidemiology--Methodology KW - Medical records LA - eng ER -