ACM on Interactive, Mobile, Wearable, and Ubiquitous Technologies (IMWUT)
September 7, 2022
behavioral intervention, human-centered computing, risk prediction, smoking cessation, ubiquitous and mobile computing design and evaluation methods, wearable sensors
Passive detection of risk factors (that may influence unhealthy or adverse behaviors) via wearable and mobile sensors has created new opportunities to improve the effectiveness of behavioral interventions. A key goal is to find opportune moments for intervention by passively detecting rising risk of an imminent adverse behavior. But, it has been difficult due to substantial noise in the data collected by sensors in the natural environment and a lack of reliable label assignment of low- and high-risk states to the continuous stream of sensor data. In this paper, we propose an event-based encoding of sensor data to reduce the effect of noises and then present an approach to efficiently model the historical influence of recent and past sensor-derived contexts on the likelihood of an adverse behavior. Next, to circumvent the lack of any confirmed negative labels (i.e., time periods with no high-risk moment), and only a few positive labels (i.e., detected adverse behavior), we propose a new loss function. We use 1,012 days of sensor and self-report data collected from 92 participants in a smoking cessation field study to train deep learning models to produce a continuous risk estimate for the likelihood of an impending smoking lapse. The risk dynamics produced by the model show that risk peaks an average of 44 minutes before a lapse. Simulations on field study data show that using our model can create intervention opportunities for 85% of lapses with 5.5 interventions per day.
We present a model for identifying ideal moments for intervention by passively detecting risk of an imminent adverse behavior.
CP5 is recruiting daily smokers interested in quitting who wear the AutoSense chestband and MotionSense wrist sensors for one week prior to quitting and two weeks post-quit. They also carry a study-provided smartphone. From the sensor data, CP5 is obtaining sensor-derived biomarkers of stress, geoexposure, and smoking, and EMA’s provide additional assessment of emotions and context. CP5 will work with Aims 1, 2, and 3 of TR&D1. An initial model for SRC under development by CP5 does not explicitly account for uncertainty in sensor-derived biomarkers or in EMA-based assessments resulting from missingness. A push activity will utilize uncertainty models developed in Aim 1 of TR&D1 to represent and quantify uncertainty in sensor- and EMA-derived biomarkers. In a pull activity, we will jointly characterize the resulting uncertainty in the model for SRC. This will enable researchers to take measurement of uncertainty into account when analyzing the linkage between SRC and distinct emotions, EMA-derived context, sensor-derived states of stress and craving, and geographic context. In addition, uncertainty representation for the derived SRC biomarker can inform downstream analysis of the link between SRC and lapse risk. CP5 will work with TR&D1 Aim 2 to identify timing triggers from affect, context, SRC, and other lapse risks from the collected data, to guide the delivery of temporally-precise interventions. CP5 will utilize tools developed in Aim 2 for identifying specific temporal patterns that predict future lapse risk. Analysis of the effectiveness of these triggers on collected data will lead to a pull activity in which the identified temporal patterns are distilled into a composite risk score which can be computed from sensor data in real-time and characterizes the time-varying risk level. The resulting time triggers and risk scores will be utilized in the MRT which will be deployed by CP1. CP5 will utilize methods from TR&D1 Aim 3 for modeling the dynamic relationships between positive and negative discrete emotions, and biomarkers for stress, craving, and context that inform the temporal variations in SRC and lapse risk. Initially, the models will describe the lead-lag relationships between covariates. When sufficient sensor data has been collected, a pull activity with Aim 3 of TR&D1 will jointly develop deep feature learning methods based on variational autoencoders that can learn nonlinear temporal features to improve the accuracy in predicting the time-varying risk of lapse. This can help discover the optimal content and timing for temporally-precise interventions. The iterations between the two teams in testing the initial model, revising the model, and testing again will eventually lead to methods for all three aims that will be directly applicable to smoking cessation research using mHealth biomarkers.
CP, Emotional Context, Smoking Cessation, Stress, TR&D1