Individual Data Protected Integrative Regression Analysis of High-Dimensional Heterogeneous Data

Evidence-based decision making often relies on meta-analyzing multiple studies, which enables more precise estimation and investigation of generalizability. Integrative analysis heterogeneous studies is, however, highly challenging in the ultra high dimensional setting. The challenge is even pronounced when individual level data cannot be shared across known as DataSHIELD constraint (Wolfson et al., 2010). Under sparse regression models that are assumed to similar yet not identical we propose this paper a novel integrative procedure for data-Shielding High-dimensional Regression (SHIR). SHIR protects through summary-statistics-based integrating procedure, accommodates between study heterogeneity both covariate distribution model parameters, attains consistent variable selection. Theoretically, statistically efficient than existing distributed approaches integrate debiased LASSO estimators from local sites. Furthermore, error incurred by aggregating derived negligible compared statistical minimax rate shown asymptotically equivalent ideal estimator obtained sharing all data. finite-sample performance our method studied with via extensive simulation settings. We further illustrate utility derive phenotyping algorithms coronary artery disease using electronic health records chronic cohorts.