多維度結果依賴設計之長期追蹤資料在不同的相關矩陣中的統計分析

An outcome-dependent sampling (ODS) design is a cost-effective design in most epidemiological or large-cohort studies. The multivariate ODS (MODS) design is a further generalization for clustered or longitudinal data sampled under the ODS design. The use of the MODS design for data in which more than two outcome observations made on the same subject while considering various working correlation structures can benefit and improve the model estimation, which is, however, not discussed yet. In this grant, we considered an ODS scheme for multivariate longitudinal or clustered data with a higher dimension and established the model under different types of the working correlation structure. A semiparametric empirical likelihood approach was developed for the proposed design under commonly-used working correlation structures - independent, exchangeable, and first-order autoregressive. We also set up a likelihood-based selection criterion for choosing the most appropriate working correlation structure. Through extensive simulation studies. we evaluated the proposed estimators and compared its efficiency with other competing approaches. We also successfully applied the proposed approach to analyze the dental restoration data collected from the University of Iowa College of Dentistry's Geriatric and Special Needs (SPEC) Clinic.