|
Abstract:
|
An excess number of zeros is common in applications involving count data. For example, in the study of optimal trough serum immunoglobulin G (IgG) level against serious bacterial infections (SBIs) in patients with primary immunodeficiency diseases (PIDD), SBIs are rare events. PIDD patients are susceptible to recurrent infections because of immune defects and are often treated with weekly or monthly infusions of IgG. Although a minimum target serum IgG trough level of 500 mg/dL is often used as guidance when deciding on the amount of IgG needed for a patient, the level for optimal protection against SBIs is not yet established and is likely to be subject specific. In this work, we investigate the problem of individualizing IgG amount using data from multiple studies. We propose a Bayesian approach that incorporates prior information and considers the features seen in the data: heterogeneity in the data source, repeated measures, and excess zeros. In simulation studies, we compare the Bayesian approach to a frequentist approach for fitting of repeated measure of zero-inflated count data.
|
