|
Abstract:
|
At the early hypothesis generating stage of exploring relationships between dosage and response, usually only data from observational studies is available. Examples include establishing correlates of diseases with antibody levels in early vaccine development stage using data from various disease surveillance platforms. For such observational study data, the relationship between the observed dosage and the outcome is usually confounded by some covariates. To be able to identify and to adjust for such confounding effects is fundamental in establishing an accurate dose-response curve to guide future studies. Following a typical epidemiology practice in adjusting for confounding effects on odds ratio between exposure and outcome, we developed an innovative and systematic method called Covariate Adjusted Logit Model (CALM) to address this issue. Here, instead of obtaining a simple adjusted odds ratio, we attempt to obtain an adjusted response curve. The approach can be applied to different study designs, such as cohort studies, case-control designs, and even matched case-control designs. Inspired by all-or-nothing interpretation of vaccine effects, a Bayesian approach is developed to simplify the computation. Extensive simulation studies were conducted to demonstrate the accuracy and advantages of CALM.
|
