Statistical Methods for Binary Outcomes Adjusting for Outcome Dependent Sampling in Longitudinal Studies with Nonignorable Dropout

Longitudinal clinical trials and cohort studies often collect clinical data paired with stored biospecimens. An increasing focus of biomedical research is aimed at leveraging these existing specimens to address new research questions. When a hypothesis of interest proposes to utilize costly, limited or difficult to obtain samples, it may not be possible or desirable to assay all samples. In these situations informed sampling strategies (ISS) can be used to minimize costs and preserve biospecimens by providing a framework to select a subset of subjects that is more informative than a simple random sample. The samples from selected subjects can be assayed and the resulting data can be analyzed in concert with an analytical correction. Dropout is common in longitudinal studies but existing ISS methods do not address nonignorable dropout. Ignoring cases where poor outcomes may influence the propensity to dropout could bias study results. We propose an expansion of current ISS frameworks to account for nonignorable dropout. Mixture models, commonly used to adjust for dropout, are modified to accommodate analysis of data from ISS designs. Methods are available in the BUILD R package.