Assessing the Impact of Context Inference Error and Partial Observability on RL Methods for Just-In-Time Adaptive Interventions.

Just-in-Time Adaptive Interventions (JITAIs) are a class of personalized health interventions developed within the behavioral science community. JITAIs aim to provide the right type and amount of support by iteratively selecting a sequence of intervention options from a pre-defined set of components in response to each individual's time varying state. In this work, we explore the application of reinforcement learning methods to the problem of learning intervention option selection policies. We study the effect of context inference error and partial observability on the ability to learn effective policies. Our results show that the propagation of uncertainty from context inferences is critical to improving intervention efficacy as context uncertainty increases, while policy gradient algorithms can provide remarkable robustness to partially observed behavioral state information.

Advancing Behavioral Intervention and Theory Development for Mobile Health: The HeartSteps II Protocol.

Background: Recent advances in mobile and wearable technologies have led to new forms of interventions, called "Just-in-Time Adaptive Interventions" (JITAI). JITAIs interact with the individual at the most appropriate time and provide the most appropriate support depending on the continuously acquired Intensive Longitudinal Data (ILD) on participant physiology, behavior, and contexts. These advances raise an important question: How do we model these data to better understand and intervene on health behaviors? The HeartSteps II study, described here, is a Micro-Randomized Trial (MRT) intended to advance both intervention development and theory-building enabled by the new generation of mobile and wearable technology. Methods: The study involves a year-long deployment of HeartSteps, a JITAI for physical activity and sedentary behavior, with 96 sedentary, overweight, but otherwise healthy adults. The central purpose is twofold: (1) to support the development of modeling approaches for operationalizing dynamic, mathematically rigorous theories of health behavior; and (2) to serve as a testbed for the development of learning algorithms that JITAIs can use to individualize intervention provision in real time at multiple timescales. Discussion and Conclusions: We outline an innovative modeling paradigm to model and use ILD in real- or near-time to individually tailor JITIAs.

BayesLDM: A Domain-specific Modeling Language for Probabilistic Modeling of Longitudinal Data.

In this paper we present BayesLDM, a library for Bayesian longitudinal data modeling consisting of a high-level modeling language with specific features for modeling complex multivariate time series data coupled with a compiler that can produce optimized probabilistic program code for performing inference in the specified model. BayesLDM supports modeling of Bayesian network models with a specific focus on the efficient, declarative specification of dynamic Bayesian Networks (DBNs). The BayesLDM compiler combines a model specification with inspection of available data and outputs code for performing Bayesian inference for unknown model parameters while simultaneously handling missing data. These capabilities have the potential to significantly accelerate iterative modeling workflows in domains that involve the analysis of complex longitudinal data by abstracting away the process of producing computationally efficient probabilistic inference code. We describe the BayesLDM system components, evaluate the efficiency of representation and inference optimizations and provide an illustrative example of the application of the system to analyzing heterogeneous and partially observed mobile health data.

Out-of-Sample Extension for Dimensionality Reduction of Noisy Time Series.

This paper proposes an out-of-sample extension framework for a global manifold learning algorithm (Isomap) that uses temporal information in out-of-sample points in order to make the embedding more robust to noise and artifacts. Given a set of noise-free training data and its embedding, the proposed framework extends the embedding for a noisy time series. This is achieved by adding a spatio-temporal compactness term to the optimization objective of the embedding. To the best of our knowledge, this is the first method for out-of-sample extension of manifold embeddings that leverages timing information available for the extension set. Experimental results demonstrate that our out-of-sample extension algorithm renders a more robust and accurate embedding of sequentially ordered image data in the presence of various noise and artifacts when compared with other timing-aware embeddings. Additionally, we show that an out-of-sample extension framework based on the proposed algorithm outperforms the state of the art in eye-gaze estimation.

Learning Time Series Detection Models from Temporally Imprecise Labels.

In this paper, we consider the problem of learning time series detection models from temporally imprecise labels. In this problem, the data consist of a set of input time series, and supervision is provided by a sequence of noisy time stamps corresponding to the occurrence of positive class events. Such temporally imprecise labels occur in areas like mobile health research when human annotators are tasked with labeling the occurrence of very short duration events. We propose a general learning framework for this problem that can accommodate different base classifiers and noise models. We present results on real mobile health data showing that the proposed framework significantly outperforms a number of alternatives including assuming that the label time stamps are noise-free, transforming the problem into the multiple instance learning framework, and learning on labels that were manually aligned.

Learning Tree-Structured Detection Cascades for Heterogeneous Networks of Embedded Devices.

In this paper, we present a new approach to learning cascaded classifiers for use in computing environments that involve networks of heterogeneous and resource-constrained, low-power embedded compute and sensing nodes. We present a generalization of the classical linear detection cascade to the case of tree-structured cascades where different branches of the tree execute on different physical compute nodes in the network. Different nodes have access to different features, as well as access to potentially different computation and energy resources. We concentrate on the problem of jointly learning the parameters for all of the classifiers in the cascade given a fixed cascade architecture and a known set of costs required to carry out the computation at each node. To accomplish the objective of joint learning of all detectors, we propose a novel approach to combining classifier outputs during training that better matches the hard cascade setting in which the learned system will be deployed. This work is motivated by research in the area of mobile health where energy efficient real time detectors integrating information from multiple wireless on-body sensors and a smart phone are needed for real-time monitoring and the delivery of just-in-time adaptive interventions. We evaluate our framework on mobile sensor-based human activity recognition and mobile health detector learning problems.

Impact of a Collective Intelligence Tailored Messaging System on Smoking Cessation: The Perspect Randomized Experiment.

BACKGROUND: Outside health care, content tailoring is driven algorithmically using machine learning compared to the rule-based approach used in current implementations of computer-tailored health communication (CTHC) systems. A special class of machine learning systems ("recommender systems") are used to select messages by combining the collective intelligence of their users (ie, the observed and inferred preferences of users as they interact with the system) and their user profiles. However, this approach has not been adequately tested for CTHC. OBJECTIVE: Our aim was to compare, in a randomized experiment, a standard, evidence-based, rule-based CTHC (standard CTHC) to a novel machine learning CTHC: Patient Experience Recommender System for Persuasive Communication Tailoring (PERSPeCT). We hypothesized that PERSPeCT will select messages of higher influence than our standard CTHC system. This standard CTHC was proven effective in motivating smoking cessation in a prior randomized trial of 900 smokers (OR 1.70, 95% CI 1.03-2.81). METHODS: PERSPeCT is an innovative hybrid machine learning recommender system that selects and sends motivational messages using algorithms that learn from message ratings from 846 previous participants (explicit feedback), and the prior explicit ratings of each individual participant. Current smokers (N=120) aged 18 years or older, English speaking, with Internet access were eligible to participate. These smokers were randomized to receive either PERSPeCT (intervention, n=74) or standard CTHC tailored messages (n=46). The study was conducted between October 2014 and January 2015. By randomization, we compared daily message ratings (mean of smoker ratings each day). At 30 days, we assessed the intervention's perceived influence, 30-day cessation, and changes in readiness to quit from baseline. RESULTS: The proportion of days when smokers agreed/strongly agreed (daily rating ≥4) that the messages influenced them to quit was significantly higher for PERSPeCT (73%, 23/30) than standard CTHC (44%, 14/30, P=.02). Among less educated smokers (n=49), this difference was even more pronounced for days strongly agree (intervention: 77%, 23/30; comparison: 23%, 7/30, P<.001). There was no significant difference in the frequency which PERSPeCT randomized smokers agreed or strongly agreed that the intervention influenced them to quit smoking (P=.07) and use nicotine replacement therapy (P=.09). Among those who completed follow-up, 36% (20/55) of PERSPeCT smokers and 32% (11/34) of the standard CTHC group stopped smoking for one day or longer (P=.70). CONCLUSIONS: Compared to standard CTHC with proven effectiveness, PERSPeCT outperformed in terms of influence ratings and resulted in similar cessation rates. CLINICALTRIAL: Clinicaltrials.gov NCT02200432; https://clinicaltrials.gov/ct2/show/NCT02200432 (Archived by WebCite at http://www.webcitation.org/6lEJY1KEd).

Collective-Intelligence Recommender Systems: Advancing Computer Tailoring for Health Behavior Change Into the 21st Century.

BACKGROUND: What is the next frontier for computer-tailored health communication (CTHC) research? In current CTHC systems, study designers who have expertise in behavioral theory and mapping theory into CTHC systems select the variables and develop the rules that specify how the content should be tailored, based on their knowledge of the targeted population, the literature, and health behavior theories. In collective-intelligence recommender systems (hereafter recommender systems) used by Web 2.0 companies (eg, Netflix and Amazon), machine learning algorithms combine user profiles and continuous feedback ratings of content (from themselves and other users) to empirically tailor content. Augmenting current theory-based CTHC with empirical recommender systems could be evaluated as the next frontier for CTHC. OBJECTIVE: The objective of our study was to uncover barriers and challenges to using recommender systems in health promotion. METHODS: We conducted a focused literature review, interviewed subject experts (n=8), and synthesized the results. RESULTS: We describe (1) limitations of current CTHC systems, (2) advantages of incorporating recommender systems to move CTHC forward, and (3) challenges to incorporating recommender systems into CTHC. Based on the evidence presented, we propose a future research agenda for CTHC systems. CONCLUSIONS: We promote discussion of ways to move CTHC into the 21st century by incorporation of recommender systems.

Domain Adaptation Methods for Improving Lab-to-field Generalization of Cocaine Detection using Wearable ECG.

Mobile health research on illicit drug use detection typically involves a two-stage study design where data to learn detectors is first collected in lab-based trials, followed by a deployment to subjects in a free-living environment to assess detector performance. While recent work has demonstrated the feasibility of wearable sensors for illicit drug use detection in the lab setting, several key problems can limit lab-to-field generalization performance. For example, lab-based data collection often has low ecological validity, the ground-truth event labels collected in the lab may not be available at the same level of temporal granularity in the field, and there can be significant variability between subjects. In this paper, we present domain adaptation methods for assessing and mitigating potential sources of performance loss in lab-to-field generalization and apply them to the problem of cocaine use detection from wearable electrocardiogram sensor data.

Hierarchical Span-Based Conditional Random Fields for Labeling and Segmenting Events in Wearable Sensor Data Streams.

The field of mobile health (mHealth) has the potential to yield new insights into health and behavior through the analysis of continuously recorded data from wearable health and activity sensors. In this paper, we present a hierarchical span-based conditional random field model for the key problem of jointly detecting discrete events in such sensor data streams and segmenting these events into high-level activity sessions. Our model includes higher-order cardinality factors and inter-event duration factors to capture domain-specific structure in the label space. We show that our model supports exact MAP inference in quadratic time via dynamic programming, which we leverage to perform learning in the structured support vector machine framework. We apply the model to the problems of smoking and eating detection using four real data sets. Our results show statistically significant improvements in segmentation performance relative to a hierarchical pairwise CRF.

Towards collaborative filtering recommender systems for tailored health communications.

The goal of computer tailored health communications (CTHC) is to promote healthy behaviors by sending messages tailored to individual patients. Current CTHC systems collect baseline patient "profiles" and then use expert-written, rule-based systems to target messages to subsets of patients. Our main interest in this work is the study of collaborative filtering-based CTHC systems that can learn to tailor future message selections to individual patients based explicit feedback about past message selections. This paper reports the results of a study designed to collect explicit feedback (ratings) regarding four aspects of messages from 100 subjects in the smoking cessation support domain. Our results show that most users have positive opinions of most messages and that the ratings for all four aspects of the messages are highly correlated with each other. Finally, we conduct a range of rating prediction experiments comparing several different model variations. Our results show that predicting future ratings based on each user's past ratings contributes the most to predictive accuracy.