BACKGROUND
Cognitive behavioral therapy (CBT)-based interventions are effective in reducing prenatal stress, which can have severe adverse health effects on mother and newborn if unaddressed. Predicting next-day physiologic or perceived stress can help to inform and enable preemptive interventions for a likely physiologically and/or perceptibly stressful day. Machine learning models are useful tools that can be developed to predict next-day physiologic and perceived stress using data collected the previous day. Such models can improve our understanding of the specific factors that predict physiologic and perceived stress and will also allow researchers to develop systems that collect selected features for assessment for clinical trials in order to minimize the burden of data collection.
OBJECTIVE
To build and evaluate a machine-learned model that predicts next-day physiologic and perceived stress 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 CBT intervention data over 12 weeks. We used the data to train and evaluate six machine learning models to predict next-day physiologic and perceived stress. After selecting the best performing model, SHapley Additive exPlanations (SHAP) were used to identify 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 physiologic and perceived stress, respectively. A random forest classifier performed the best in predicting next-day physiologic (F1-score 0.84) and next-day perceived stress (F1-score 0.74) using all features. While any subset of sensor-based, EMA-based, and/or intervention-based features could reliably predict next-day physiologic stress, EMA-based features were necessary to predict next-day perceived stress. Analysis of explainability metrics showed that prolonged duration of physiologic stress was highly predictive of next-day physiologic stress and that physiologic stress and perceived stress were temporally divergent.
CONCLUSIONS
In this study we were able to build interpretable machine learning models to predict next-day physiologic and perceived stress, and we identify unique features that were highly predictive of next-day stress that can help reduce the burden of data collection.