A Repeated Block Perturbation Subsampling for Large-Scale Longitudinal Data

Area of Science:

• Biostatistics
• Health Informatics
• Data Science

Background:

• Large-scale longitudinal data are increasingly prevalent in healthcare research.
• Mobile health (mHealth) applications generate substantial longitudinal datasets.
• Analyzing these large datasets presents significant computational challenges.

Purpose of the Study:

• To propose a novel subsampling algorithm for analyzing large-scale longitudinal mHealth data.
• To develop a method that provides consistent point and variance estimators.
• To address the analytical challenges posed by big data in mHealth.

Main Methods:

• A repeated block perturbation subsampling algorithm was developed.
• The algorithm is based on generalized estimating equations.
• Asymptotic properties of the subsampling estimators were established.

Main Results:

• The proposed method yields consistent point and variance estimators.
• Asymptotic properties of the subsampling estimators were theoretically established.
• Simulations and real mHealth data analyses demonstrated the method's performance.

Conclusions:

• The novel subsampling algorithm is effective for analyzing large-scale longitudinal mHealth data.
• The method offers a computationally efficient approach to statistical inference.
• This work contributes to the robust analysis of big data in digital health.