Recent advancements in mobile health (mHealth) technology and the ubiquity of wearable devices and smartphones have expanded a market for digital health and have emerged as innovative tools for data collection on individualized behavior. Heterogeneous levels of device usage across users and across days within a single user may result in different degrees of underestimation in passive sensing data, subsequently introducing biases if analyzed without addressing this issue. In this work, we propose an unsupervised 2-Stage Pre-processing Algorithm for Passively Sensed mHealth Data (2SpamH) algorithm that uses device usage variables to infer the quality of passive sensing data from mobile devices. This article provides a series of simulation studies to show the utility of the proposed algorithm compared to existing methods. Application to a real clinical dataset is also illustrated.
