Abstract: In recent years, with the deep integration of big data and medical technology, hybrid data with or without block-wise missing arise more commonly in medical care. Efficient dimensionality reduction and extraction of important predictive information for such data have also become a popular research topic. In this article, for hybrid data without missing and with block-wise missing, we proposed a kind of new component-based model based on the unified approach to multi-source principal component analysis and multi-set canonical correlation analysis. After obtaining scores by using the unified framework, component-based regression models are established. Asymptotic properties are established under some mild conditions. Simulations and real data analysis show the proposed method works well.

Key words and phrases: Alzheimer's disease, block-wise imputation, component-based regression, hybrid data, multi-set canonical correlation analysis, multi-source principal component analysis.