A multivariate multilevel discrete-time hazard model for familial aggregation and co-aggregation of psychiatric disorders.

다국어 초록 (Multilingual Abstract)

The familial aggregation and comorbidity of psychiatric disorders is a public health concern studied by psychiatric epidemiologists. Offspring of affected parents are at elevated risk for psychopathology due to familial liability as well as individual liability for disorder. Childhood and adolescent psychopathology and its relationship with the onset and progression of substance use is an especially important issue. Children are appropriate targets of interventions to mitigate disorder onset and the severity of its course. Longitudinal studies of high-risk offspring elucidate the distribution, etiology and course of early-onset psychiatric disorders to inform intervention and prevention.
Many statistical models for familial aggregation have appeared in the genetic epidemiology and family study literature. Our aim in this manuscript is to offer a conceptualization of familial aggregation that differentiates variation in familial clustering from that of familial risk, and to develop a multilevel model that operationalizes this approach. Because the outcome of our analysis is the disease status of children who are observed until different ages and thus different points in the period of risk, we use a hazard model.
We apply our model to family study data collected by Dr. Kathleen Merikangas of the Genetic Epidemiology Research Unit at Yale University. The Yale Family Study high-risk component examined 203 children of 124 proband parents. Probands were ascertained from clinics and from the New Haven CT community as affected with anxiety and/or substance-related disorders or as healthy controls.
To analyze clustered duration data for patterns of familial aggregation and comorbidity, we propose a multivariate multilevel discrete-time hazard model. We apply the model to the reported ages of onset of anxiety disorder and alcohol use in the high-risk sample. We choose these outcomes and a set of related risk factors mainly for the purpose of giving a clear illustration of the modeling process. Although in this manuscript we may not necessarily provide a definitive answer to a substantive clinical question, we develop a tool that we offer to researchers in their quest to do so.