Predicting the Next-Day Perceived and Physiological Stress of Pregnant Women by Using Machine Learning and Explainability: Algorithm Development and Validation.

BACKGROUND: Cognitive behavioral therapy-based interventions are effective in reducing prenatal stress, which can have severe adverse health effects on mothers and newborns if unaddressed. Predicting next-day physiological or perceived stress can help to inform and enable pre-emptive interventions for a likely physiologically and perceptibly stressful day. Machine learning models are useful tools that can be developed to predict next-day physiological and perceived stress by using data collected from the previous day. Such models can improve our understanding of the specific factors that predict physiological and perceived stress and allow researchers to develop systems that collect selected features for assessment in clinical trials to minimize the burden of data collection. OBJECTIVE: The aim of this study was to build and evaluate a machine-learned model that predicts next-day physiological and perceived stress by using sensor-based, ecological momentary assessment (EMA)-based, and intervention-based features and to explain the prediction results. METHODS: We enrolled pregnant women into a prospective proof-of-concept study and collected electrocardiography, EMA, and cognitive behavioral therapy intervention data over 12 weeks. We used the data to train and evaluate 6 machine learning models to predict next-day physiological and perceived stress. After selecting the best performing model, Shapley Additive Explanations were used to identify the feature importance and explainability of each feature. RESULTS: A total of 16 pregnant women enrolled in the study. Overall, 4157.18 hours of data were collected, and participants answered 2838 EMAs. After applying feature selection, 8 and 10 features were found to positively predict next-day physiological and perceived stress, respectively. A random forest classifier performed the best in predicting next-day physiological stress (F1 score of 0.84) and next-day perceived stress (F1 score of 0.74) by using all features. Although any subset of sensor-based, EMA-based, or intervention-based features could reliably predict next-day physiological stress, EMA-based features were necessary to predict next-day perceived stress. The analysis of explainability metrics showed that the prolonged duration of physiological stress was highly predictive of next-day physiological stress and that physiological stress and perceived stress were temporally divergent. CONCLUSIONS: In this study, we were able to build interpretable machine learning models to predict next-day physiological and perceived stress, and we identified unique features that were highly predictive of next-day stress that can help to reduce the burden of data collection.

Assessing recall of personal sun exposure by integrating UV dosimeter and self-reported data with a network flow framework.

BACKGROUND: Melanoma survivors often do not engage in adequate sun protection, leading to sunburn and increasing their risk of future melanomas. Melanoma survivors do not accurately recall the extent of sun exposure they have received, thus, they may be unaware of their personal UV exposure, and this lack of awareness may contribute towards failure to change behavior. As a means of determining behavioral accuracy of recall of sun exposure, this study compared subjective self-reports of time outdoors to an objective wearable sensor. Analysis of the meaningful discrepancies between the self-report and sensor measures of time outdoors was made possible by using a network flow algorithm to align sun exposure events recorded by both measures. Aligning the two measures provides the opportunity to more accurately evaluate false positive and false negative self-reports of behavior and understand participant tendencies to over- and under-report behavior. METHODS: 39 melanoma survivors wore an ultraviolet light (UV) sensor on their chest while outdoors for 10 consecutive summer days and provided an end-of-day subjective self-report of their behavior while outdoors. A Network Flow Alignment framework was used to align self-report and objective UV sensor data to correct misalignment. The frequency and time of day of under- and over-reporting were identified. FINDINGS: For the 269 days assessed, the proposed framework showed a significant increase in the Jaccard coefficient (i.e. a measure of similarity between self-report and UV sensor data) by 63.64% (p < .001), and significant reduction in false negative minutes by 34.43% (p < .001). Following alignment of the measures, under-reporting of sun exposure time occurred on 51% of the days analyzed and more participants tended to under-report than to over-report sun exposure time. Rates of under-reporting of sun exposure were highest for events that began from 12-1pm, and second-highest from 5-6pm. CONCLUSION: These discrepancies may reflect lack of accurate recall of sun exposure during times of peak sun intensity (10am-2pm) that could ultimately increase the risk of developing melanoma. This research provides technical contributions to the field of wearable computing, activity recognition, and identifies actionable times to improve participants' perception of their sun exposure.

Is More Always Better?: Discovering Incentivized mHealth Intervention Engagement Related to Health Behavior Trends.

Behavioral medicine is devoting increasing attention to the topic of participant engagement and its role in effective mobile health (mHealth) behavioral interventions. Several definitions of the term "engagement" have been proposed and discussed, especially in the context of digital health behavioral interventions. We consider that engagement refers to specific interaction and use patterns with the mHealth tools such as smartphone applications for intervention, whereas adherence refers to compliance with the directives of the health intervention, independent of the mHealth tools. Through our analysis of participant interaction and self-reported behavioral data in a college student health study with incentives, we demonstrate an example of measuring "effective engagement" as engagement behaviors that can be linked to the goals of the desired intervention. We demonstrate how clustering of one year of weekly health behavior self-reports generate four interpretable clusters related to participants' adherence to the desired health behaviors: healthy and steady, unhealthy and steady, decliners, and improvers. Based on the intervention goals of this study (health promotion and behavioral change), we show that not all app usage metrics are indicative of the desired outcomes that create effective engagement. As such, mHealth intervention design might consider eliciting not just more engagement or use overall, but rather, effective engagement defined by use patterns related to the desired behavioral outcome.