STATISTICAL METHODS FOR SPARSE DEPENDENT DATA

描述（申请人摘要）：医学或流行病学研究

DESCRIPTION (Applicant's abstract): In medical or epidemiological investigations of complex diseases, often it is desired to assess the aggregation of disease within clusters. For example, a finding of familial aggregation might suggest a genetic component in the etiology of the disease. Unfortunately, such investigations are hampered by the many confounding factors associated with complex diseases, such as demographic, cultural and socioeconomic factors. The study of disease aggregation, with adjustment for many confounding factors, gives rise to sparse dependent data. New statistical methods are necessary to analyze such data. The long-term objective of this research is to develop novel statistical methods that are suitable for sparse dependent data. Special attention is given to genetic epidemiological studies of familial aggregation of disease. The specific aims are to: (1) develop estimating functions that provide inferences for sparse, dependent binary data with proper adjustment for mode of ascertainment of the cluster; (2) develop the theory and application of a general conditional estimating function approach that is valid for various types of sparse dependent data, e.g., discrete data, continuous data or age of onset data; (3) develop approximate likelihood methods to accompany these estimating functions that provide better confidence intervals than the usual Wald confidence interval; (4) evaluate the robustness and efficiency of these methods compared to random-effects methods that require more modeling assumptions; and (5) use the novel statistical methods to reanalyze three data sets involving the aggregation of schizophrenia, obsessive-compulsive disorder, and hypertension, respectively. This research will provide new, more powerful methods to assess aggregation of disease within clusters with proper adjustment for many confounding factors and ascertainment bias.