Applying Contemporary Machine Learning Approaches to Nutrition Care Real-World Evidence: Findings From the National Quality Improvement Data Set.

Using real-world data from the Academy of Nutrition and Dietetics Health Informatics Infrastructure, we use state-of-the-art clustering techniques to identify 2 phenotypes characterizing the episodes of nutrition care observed in the National Quality Improvement (NQI) registry data set. The 2 phenotypes identified from recorded Nutrition Care Process data in the NQI exhibit a strong correspondence with the clinical expertise of registered dietitian nutritionists. For one of these phenotypes, it was possible to implement state-of-the-art classification techniques to predict the nutrition problem-resolution status of an episode of care. Prediction results show that the assessment of nutrition history, number of recorded visits in the episode, and use of nutrition counseling interventions were significantly and positively correlated with problem resolution. Meanwhile, evaluations of nutrition history that were not within the desired ranges were significantly and negatively correlated with problem resolution. Finally, we assess the usefulness of the current NQI data set and data model for supporting the application of contemporary machine learning methods to the data set. We also suggest ways of enhancing the NQI since registered dietitian nutritionists are encouraged to continue to contribute patient cases in this and other registry nutrition studies.

Citizen science to further precision medicine: from vision to implementation.

The active involvement of citizen scientists in setting research agendas, partnering with academic investigators to conduct research, analyzing and disseminating results, and implementing learnings from research can improve both processes and outcomes. Adopting a citizen science approach to the practice of precision medicine in clinical care and research will require healthcare providers, researchers, and institutions to address a number of technical, organizational, and citizen scientist collaboration issues. Some changes can be made with relative ease, while others will necessitate cultural shifts, redistribution of power, recommitment to shared goals, and improved communication. This perspective, based on a workshop held at the 2018 AMIA Annual Symposium, identifies current barriers and needed changes to facilitate broad adoption of a citizen science-based approach in healthcare.

Learning Interpretable Behavioral Engagement for Care Management.

The health outcomes of high-need patients can be substantially influenced by the degree of patient engagement in their own care. The role of care managers (CMs) includes enrolling patients and keeping them sufficiently engaged in care programs, so that patients complete assigned goals leading to improvement in their health outcomes. Here, we present a data-driven behavioral engagement scoring (BES) pipeline that can compute the patients' engagement level with regards to their interest in: (1) enrolling into a relevant care program, and (2) completing program goals. This score is leveraged to predict a patient's propensity to respond to CMs' actions. Using real-world care management data, we show that the BES pipeline successfully predicts patient engagement and provides interpretable insights to CMs, using prototypical patient cases as a point of reference, without sacrificing prediction performance.

Efficient goal attainment and engagement in a care manager system using unstructured notes.

OBJECTIVE: To improve efficient goal attainment of patients by analyzing the unstructured text in care manager (CM) notes (CMNs). Our task is to determine whether the goal assigned by the CM can be achieved in a timely manner. MATERIALS AND METHODS: Our data consists of CM structured and unstructured records from a private firm in Orlando, FL. The CM data is based on phone interactions between the CM and the patient. A portion of the data has been manually annotated to indicate engagement. We present 2 machine learning classifiers: an engagement model and a goal attainment model. RESULTS: We can successfully distinguish automatically between engagement and lack of engagement. Subsequently, incorporating engagement and features from textual information from the unstructured notes significantly improves goal attainment classification. DISCUSSION: Two key challenges in this task were the time-consuming annotation effort for engagement classification and the limited amount of data for the more difficult goal attainment class (specifically, for people who take a long time to achieve their goals). We successfully explore domain adaptation and transfer learning techniques to improve performance on the under-represented classes. We also explore the value of using features from unstructured notes to improve the model and interpretability. CONCLUSIONS: Unstructured CMNs can be used to improve accuracy of our classification models for predicting patient self-management goal attainment. This work can be used to help identify patients who may require special attention from CMs to improve engagement in self-management.

Modeling the Personas of Primary Care Communication Modality Usage: Experiences from the R-Health Direct Primary Care Model.

Recent advancements in mediated communication technologies enable unfettered communication between patients and their providers. We explore patients' utilization of different direct and mediated communication modalities in a Direct Primary Care (DPC) clinic and the patient characteristics that predict their communication behavior patterns. Based on this knowledge, we developed 2 patient personas that explicate the nuances of patients who tend to prefer visiting the clinic in person versus those who use mediated modalities more often. We hope this study may inform future work in understanding and supporting patient-provider communication in a new technical environment. The results suggest that patients and their health team alike may be incentivized to voluntarily adopt and utilize multi-modality communication in a DPC setting.

Quantile Coarsening Analysis of High-Volume Wearable Activity Data in a Longitudinal Observational Study.

Owing to advances in sensor technologies on wearable devices, it is feasible to measure physical activity of an individual continuously over a long period. These devices afford opportunities to understand individual behaviors, which may then provide a basis for tailored behavior interventions. The large volume of data however poses challenges in data management and analysis. We propose a novel quantile coarsening analysis (QCA) of daily physical activity data, with a goal to reduce the volume of data while preserving key information. We applied QCA to a longitudinal study of 79 healthy participants whose step counts were monitored for up to 1 year by a Fitbit device, performed cluster analysis of daily activity, and identified individual activity signature or pattern in terms of the clusters identified. Using 21,393 time series of daily physical activity, we identified eight clusters. Employment and partner status were each associated with 5 of the 8 clusters. Using less than 2% of the original data, QCA provides accurate approximation of the mean physical activity, forms meaningful activity patterns associated with individual characteristics, and is a versatile tool for dimension reduction of densely sampled data.

Learning to Personalize from Practice: A Real World Evidence Approach of Care Plan Personalization based on Differential Patient Behavioral Responses in Care Management Records.

Recent studies documented the importance of individuality and heterogeneity in care planning. In practice, varying behavioral responses are revealed in patients' care management (CM) records. However, today's care programs are structured around population-level evidence. What if care managers can take advantage of the revealed behavioral response for personalization? The goal of this study is thus to quantify behavioral response from CM records for informing individual-level intervention decisions. We present a Behavioral Response Inference Framework (BRIeF) for understanding differential behavioral responses that are key to effective care planning. We analyze CM records from a healthcare network over a 14-month period and obtain a set of 2,416 intervention-goal attainment records. Promising results demonstrate that the individual-level care planning strategies that are learned from practice by BRIeF, outperform population-level strategies, yielding significantly more accurate intervention recommendations for goal attainment. To our knowledge, this is the first study of learning practice-based evidence from CM records for care planning, suggesting that increased patient behavioral understanding could potentially benefit augmented intelligence for care management decision support.

Are Nomothetic or Ideographic Approaches Superior in Predicting Daily Exercise Behaviors?

OBJECTIVES: The understanding of how stress influences health behavior can provide insights into developing healthy lifestyle interventions. This understanding is traditionally attained through observational studies that examine associations at a population level. This nomothetic approach, however, is fundamentally limited by the fact that the environment- person milieu that constitutes stress exposure and experience can vary substantially between individuals, and the modifiable elements of these exposures and experiences are individual-specific. With recent advances in smartphone and sensing technologies, it is now possible to conduct idiographic assessment in users' own environment, leveraging the full-range observations of actions and experiences that result in differential response to naturally occurring events. The aim of this paper is to explore the hypothesis that an ideographic N-of-1 model can better capture an individual's stress- behavior pathway (or the lack thereof) and provide useful person-specific predictors of exercise behavior. METHODS: This paper used the data collected in an observational study in 79 participants who were followed for up to a 1-year period, wherein their physical activity was continuously and objectively monitored by actigraphy and their stress experience was recorded via ecological momentary assessment on a mobile app. In addition, our analyses considered exogenous and environmental variables retrieved from public archive such as day in a week, daylight time, temperature and precipitation. Leveraging the multiple data sources, we developed prediction algorithms for exercise behavior using random forest and classification tree techniques using a nomothetic approach and an N-of-1 approach. The two approaches were compared based on classification errors in predicting personalized exercise behavior. RESULTS: Eight factors were selected by random forest for the nomothetic decision model, which was used to predict whether a participant would exercise on a particular day. The predictors included previous exercise behavior, emotional factors (e.g., midday stress), external factors such as weather (e.g., temperature), and self-determination factors (e.g., expectation of exercise). The nomothetic model yielded an average classification error of 36%. The ideographic N-of-1 models used on average about two predictors for each individual, and had an average classification error of 25%, which represented an improvement of 11 percentage points. CONCLUSIONS: Compared to the traditional one-size-fits-all, nomothetic model that generalizes population-evidence for individuals, the proposed N-of-1 model can better capture the individual difference in their stressbehavior pathways. In this paper, we demonstrate it is feasible to perform personalized exercise behavior prediction, mainly made possible by mobile health technology and machine learning analytics.

A First Step Towards Behavioral Coaching for Managing Stress: A Case Study on Optimal Policy Estimation with Multi-stage Threshold Q-learning.

Psychological stress is a major contributor to the adoption of unhealthy behaviors, which in turn accounts for 41% of global cardiovascular disease burden. While the proliferation of mobile health apps has offered promise to stress management, these apps do not provide micro-level feedback with regard to how to adjust one's behaviors to achieve a desired health outcome. In this paper, we formulate the task of multi-stage stress management as a sequential decision-making problem and explore the application of reinforcement learning to provide micro-level feedback for stress reduction. Specifically, we incorporate a multi-stage threshold selection into Q-learning to derive an interpretable form of a recommendation policy for behavioral coaching. We apply this method on an observational dataset that contains Fitbit ActiGraph measurements and self-reported stress levels. The estimated policy is then used to understand how exercise patterns may affect users' psychological stress levels and to perform coaching more effectively.

Making Sense of Patient-Generated Health Data for Interpretable Patient-Centered Care: The Transition from "More" to "Better".

The rise of health consumers and the accumulation of patient-generated health data (PGHD) have brought the patient to the centerstage of precision health and behavioral science. In this positional paper we outline an interpretability-aware framework of PGHD, an important but often overlooked dimension in health services. The aim is two-fold: First, it helps generate practice-based evidence for population health management; second, it improves individual care with adaptive interventions. However, how do we check if the evidence generated from PGHD is reliable? Are the evidence directly deployable in realworld applications? How to adapt behavioral interventions for each individual patient at the touchpoint given individual patients' needs? These questions commonly require better interpretability of PGHD-derived patient insights. Yet the definitions of interpretability are often underspecified. In the position paper, we outline an interpretability-aware framework to handle model properties and techniques that affect interpretability in the patient-centered care process. Throughout the positional paper, we contend that making sense of PGHD systematically in such an interpretability-aware framework is preferrable, because it improves on the trustworthiness of PGHD-derived insights and the consequent applications such as person-centered comparative effectiveness in patient-centered care.

Progressive testing of health self-efficacy and literacy for personalized engagement.

Patient engagement can be enhanced through continuous monitoring of patient' health knowledge and self-efficacy levels across different co-morbid conditions and key aspects in health improvement and recovery. While self-reported test batteries and computerized instruments have been designed to perform evaluation of patient literacy and self-efficacy, they are not practical to be used in scenarios where concurrent tests are needed to understand the change over time. In this paper we propose an adaptive system that can progressively compose the most pertinent test for each user based on the provisional estimates of a patient's latent trait being measured. This requires modeling not only the latent health literacy and self-efficacy levels of patients and the difficulty and discriminating factors of test items, but also the temporal dependency among concurrent tests. The goal is to account for changes over the course of patient engagement history as the basis for devising personalized patient plans.

Development of temporal context-based feature abstractions for enabling monitoring and managing of interventions.

Disease self-management programs and intervention/care plan monitoring are often unable to systematically leverage patient-generated information, especially those requiring interpretation of the temporal contexts of the measurement. While existing techniques help in capturing and storing the relevant data, their ability to determine appropriate metrics most sensitive to that individual is limited or non-existent. This is attributable to the lack of unifying models for enabling such interpretations and the non-trivial process required to generate meaningful feature abstractions to support individualized prognosis. To address these issues, a data-driven approach designed to identify the right abstractions for key features relevant to personalization and monitoring of care is discussed.

Dynamic and accretive composition of patient engagement instruments for personalized plan generation.

Patient engagement is important to help patients become more informed and active in managing their health. Effective patient engagement demands short, yet valid instruments for measuring self-efficacy in various care dimensions. However, the static instruments are often too lengthy to be effective for assessment purposes. Furthermore, these tests could neither account for the dynamicity of measurements over time, nor differentiate care dimensions that are more critical to certain sub-populations. To remedy these disadvantages, we devise a dynamic instrument composition approach that can model the measurement of patient self-efficacy over time and iteratively select critical care dimensions and appropriate assessment questions based on dynamic user categorization. The dynamically composed instruments are expected to guide patients through self-management reinforcement cycles within or across care dimensions, while tightly integrated into clinical workflow and standard care processes.

From clinical requirement to personalized wellness decision support: a data-driven framework for computer-supported guideline refinement.

Personalized wellness decision support has gained significant attention, owing to the shift to a patient-centric paradigm in healthcare domains, and the consequent availability of a wealth of patient-related data. Despite the success of data-driven analytics in improving practice outcome, there is a gap towards their deployment in guideline-based practice. In this paper we report on findings related to computer-supported guideline refinement, which maps a patient's guideline requirements to personalized recommendations that suit the patient's current context. In particular, we present a novel data-driven personalization framework, casting the mapping task as a statistical decision problem in search of a solution to maximize expected utility. The proposed framework is well suited to produce personalized recommendations based on not only clinical factors but contextual factors that reflect individual differences in non-clinical settings. We then describe its implementation within the guideline-based clinical decision support system and discuss opportunities and challenges looking forward.

Personalized adherence activity recognition via model-driven sensor data assessment.

Creation of a personalized adherence feedback loop is crucial for initiating and sustaining health behavior change. However, self reports are not sufficient to measure actual adherence. Recording and recognizing personal activities in a ubiquitous environment has thus emerged as a promising solution. In this work, we present a model-driven sensor data assessment mechanism capable of identifying high level adherence-related activity patterns from low level signals. The proposed intelligent sensing algorithm can learn from a population-based training data set and adapt quickly to an individual's exercise patterns using the acquired personal data. Upon the recognition of each activity, the system can further provide personalized feedback such as exercise coaching, fitness planning, and abnormal event detection. The resulted system demonstrates the feasibility of a portable real-time personalized adherence feedback system that could be used for advanced healthcare services.