The vision of patient-centric, personalized, precision medicine and wellness will be fully realized only when an individual?s self-care and clinical decision making are informed by a rich, predictive model of that individual?s health status. The evolution and dissemination of mobile technology has created unprecedented opportunities for highly detailed and personalized data collection in a far more granular, unobtrusive, and even affordable way; these data include activity levels, location patterns, sleep, consumption, and communication and social interaction. However, turning this potential into practice requires that we develop the algorithms and methodologies to transform these raw data into actionable information. The research will develop novel and generalizable techniques to derive robust measures relevant to individual health and clinical decision making. The team will develop and evaluate tools that convert raw human-activity data into clinically actionable behavioral biomarkers. This demands creative uses of the underlying technical capabilities (i.e., passive data capture, data analysis and machine learning, data visualization, user experience), as well as rigorous understanding of the underlying health condition and management (i.e. functional health measures, achievable and optimal health outcomes, patient challenges in adherence, risks and benefits associated with medication and other aspects of treatment, and clinical decision making). The approach has broad applicability across disease management (e.g., auto-immune, gastrointestinal, depression, cognitive decline, and neurologic disorders), but also calls for tailoring to specific conditions and individuals. Therefore, we will conduct this initial work in a specific context, that of chronic pain management for three prominent conditions: rheumatoid arthritis, osteoarthritis, and lower back pain. The behavioral biomarkers associated with our initial target domain, pain management, center around: (i) decline in activity levels; (ii) increase in stress; (iii) decrease in sleep quality; (iv) drop in function, e.g., reduction in travel distance or inability to go to work. The effectiveness of passive sensing capabilities of the mobile phone to track sleep, changes in activity level, stress, social isolation, geographic location and several other indicators that are likely antecedents or symptoms of pain interference has been demonstrated previously.
While behavioral biomarkers rely extensively on passively captured data streams (such as activity, location, communication, application usage and audio), there remain important cases in which self-report data is required to augment or clarify passively collected data. However, the standardized patient survey instruments that assess relevant symptoms and behavior are not suitable for use on a daily basis because of length, question design, or both. Further, traditional forms of self report are often intrusive, burdensome, and suffer high rates of attrition. A new approach, contextual recall, aims to mitigate the issues related to self-report through three key mechanisms: optimizing the delivery of prompts, providing the user with key contextual cues to improve recall, and employing visual input techniques as an alternative to long-form measures that do not scale well to frequent mobile self-reports. The approach to personalizing disease management is intentionally scalable in terms of affordability and accessibility. Passive data collection requires no user attention, and contextual recall is a form of self-report designed for busy individuals with a range of demands and constraints on their time, as well as potential literacy and numeracy constraints. The clinician-facing components of this approach are also designed to work in resource-constrained clinical settings where clinicians are under particular time pressure. The team will recruit patients and clinicians from typically underserved communities to engage in the participatory design process. The overall contributions of this work will include development and evaluation of: (1) software techniques to combine and transform passively monitored and self-reported data streams into clinically meaningful, actionable, and personalized indicators, which we call behavioral biomarkers; (2) contextual recall that allows the collection of highly granular and contextually specific self-report data to enhance passively captured data with information from the patient perspective, while balancing the tension faced in balancing recall bias and usability; and (3) a methodology that systematizes the collaboration with clinical domain experts to develop and integrate behavioral biomarkers into clinical decision making for specific diseases. We will create and evaluate a modular and extensible suite of analytics and user interaction techniques designed to facilitate iterative implementation and evaluation. These modules will themselves be a contribution, but equally important will be the evaluation of the overall approach of behavioral biomarkers as a driver of precision medicine.
