Evaluation techniques for mHealth outcome measures using patient generated health data

PROJECT SUMMARY This proposal investigates statistical models for developing mobile health (mHealth) measures using patient generated health data (PGHD) with high complexity and temporality. The emergence of mHealth technologies and computational tools are rapidly expanding their use in research and clinical settings, and engaging patients in self-management. mHealth technology further allows integration of multifarious data streams to improve outcome measurement and prediction to aid clinical decision making. To maximize their actionability, however, there is a need to investigate novel approaches for design, development and evaluation of mHealth-based measures. We ground our investigation in chronic pelvic pain (CPP) as the disease model, a prevalent, complex disorder with high societal burden and quality of life (QoL) impact. There is substantial heterogeneity between patients and day-to-day variations in how CPP unfolds. Therefore, mHealth methods are particularly valuable for capturing the complex disease scenarios. There are no CPP-specific self-reported measures to assess disease status or treatment response. We propose to investigate models that can handle the inherent challenges of PGHD to derive ecologically valid and actionable self-tracking measures for patient outcomes in health settings. The Specific Aims are: Specific Aim 1. Investigate “critical windows of tracking” for mHealth-based disease outcome measurement. We will enroll 90 participants undergoing 12 weeks of physical therapy treatment for their CPP to use a mHealth app for tracking their symptoms, daily function, and medications. We will triangulate these data with clinician assessments and passive data on sleep and activity to build distributed lag models (DLMs) to identify predictors that can be used for outcome monitoring. Specific Specific Aim 2. Investigate a functional data analytic framework grounded in CPP to develop self- tracking pain and QoL measures. We will enroll 180 CPP patients to track their disease symptoms through a mHealth app and wear activity monitors for 3 months. Through a series of supervised and unsupervised models leveraging functional data analytic methods, we will identify variables to inform the design of the composite pain and QoL measures. Aim 2a. Design and develop a multidimensional self-tracking pain measure. We will build estimation models where the unit of observation is a set of curves (i.e., pain location, severity, type) over time, leveraging functional data analytic methods. Aim 2b. Design and develop a flexible self-tracking QoL measure. We will assess the relative predictive ability of individual items on CPP symptoms to derive a CPP-specific QoL measure that can be used at the day- vs week-level. Exploratory Aim 2: We will assess disease specificity of the models by comparing output from a non-CPP control group. Flexible, non- parametric data approaches allow maximizing the features of the available mHealth technology, which can aid in robust models to inform design of mHealth-based disease measures. Proposed work addresses the gap in mHealth evidence-base to improve the application and translation of efficacious mHealth assessments.