Towards Personalised Mood Prediction and Explanation for Depression from Biophysical Data

Digital health applications using Artificial Intelligence (AI) are a promising opportunity to address the widening gap between available resources and mental health needs globally. Increasingly, passively acquired data from wearables are augmented with carefully selected active data from depressed individuals to develop Machine Learning (ML) models of depression based on mood scores. However, most ML models are black box in nature, and hence the outputs are not explainable. Depression is also multimodal, and the reasons for depression may vary significantly between individuals. Explainable and personalised models will thus be beneficial to clinicians to determine the main features that lead to a decline in the mood state of a depressed individual, thus enabling suitable personalised therapy. This is currently lacking. Therefore, this study presents a methodology for developing personalised and accurate Deep Learning (DL)-based predictive mood models for depression, along with novel methods for identifying the key facets that lead to the exacerbation of depressive symptoms. We illustrate our approach by using an existing multimodal dataset containing longitudinal Ecological Momentary Assessments of depression, lifestyle data from wearables and neurocognitive assessments for 14 mild to moderately depressed participants over one month. We develop classification- and regression-based DL models to predict participants' current mood scores-a discrete score given to a participant based on the severity of their depressive symptoms. The models are trained inside eight different evolutionary-algorithm-based optimisation schemes that optimise the model parameters for a maximum predictive performance. A five-fold cross-validation scheme is used to verify the DL model's predictive performance against 10 classical ML-based models, with a model error as low as 6% for some participants. We use the best model from the optimisation process to extract indicators, using SHAP, ALE and Anchors from explainable AI literature to explain why certain predictions are made and how they affect mood. These feature insights can assist health professionals in incorporating personalised interventions into a depressed individual's treatment regimen.

Physical Activity Pattern of Adults With Metabolic Syndrome Risk Factors: Time-Series Cluster Analysis

Background

Physical activity plays a crucial role in maintaining a healthy lifestyle, and wrist-worn wearables, such as smartwatches and smart bands, have become popular tools for measuring activity levels in daily life. However, studies on physical activity using wearable devices have limitations; for example, these studies often rely on a single device model or use improper clustering methods to analyze the wearable data that are extracted from wearable devices.

Objective

This study aimed to identify methods suitable for analyzing wearable data and determining daily physical activity patterns. This study also explored the association between these physical activity patterns and health risk factors.

Methods

People aged >30 years who had metabolic syndrome risk factors and were using their own wrist-worn devices were included in this study. We collected personal health data through a web-based survey and measured physical activity levels using wrist-worn wearables over the course of 1 week. The Time-Series Anytime Density Peak (TADPole) clustering method, which is a novel time-series method proposed recently, was used to identify the physical activity patterns of study participants. Additionally, we defined physical activity pattern groups based on the similarity of physical activity patterns between weekdays and weekends. We used the χ2 or Fisher exact test for categorical variables and the 2-tailed t test for numerical variables to find significant differences between physical activity pattern groups. Logistic regression models were used to analyze the relationship between activity patterns and health risk factors.

Results

A total of 47 participants were included in the analysis, generating a total of 329 person-days of data. We identified 2 different types of physical activity patterns (early bird pattern and night owl pattern) for weekdays and weekends. The physical activity levels of early birds were less than that of night owls on both weekdays and weekends. Additionally, participants were categorized into stable and shifting groups based on the similarity of physical activity patterns between weekdays and weekends. The physical activity pattern groups showed significant differences depending on age (P=.004) and daily energy expenditure (P<.001 for weekdays; P=.003 for weekends). Logistic regression analysis revealed a significant association between older age (≥40 y) and shifting physical activity patterns (odds ratio 8.68, 95% CI 1.95-48.85; P=.007).

Conclusions

This study overcomes the limitations of previous studies by using various models of wrist-worn wearables and a novel time-series clustering method. Our findings suggested that age significantly influenced physical activity patterns. It also suggests a potential role of the TADPole clustering method in the analysis of large and multidimensional data, such as wearable data.

Predicting Hospital Readmission among Patients with Sepsis Using Clinical and Wearable Data

Sepsis is a life-threatening condition that occurs due to a dysregulated host response to infection. Recent data demonstrate that patients with sepsis have a significantly higher readmission risk than other common conditions, such as heart failure, pneumonia and myocardial infarction and associated economic burden. Prior studies have demonstrated an association between a patient's physical activity levels and readmission risk. In this study, we show that distribution of activity level prior and post-discharge among patients with sepsis are predictive of unplanned rehospitalization in 90 days (P-value<1e-3). Our preliminary results indicate that integrating Fitbit data with clinical measurements may improve model performance on predicting 90 days readmission.Clinical relevance Sepsis, Activity level, Hospital readmission, Wearable data.

Estimating circadian phase in elementary school children: leveraging advances in physiologically informed models of circadian entrainment and wearable devices

STUDY OBJECTIVES

Examine the ability of a physiologically based mathematical model of human circadian rhythms to predict circadian phase, as measured by salivary dim light melatonin onset (DLMO), in children compared to other proxy measurements of circadian phase (bedtime, sleep midpoint, and wake time).

METHODS

As part of an ongoing clinical trial, a sample of 29 elementary school children (mean age: 7.4 ± .97 years) completed 7 days of wrist actigraphy before a lab visit to assess DLMO. Hourly salivary melatonin samples were collected under dim light conditions (<5 lx). Data from actigraphy were used to generate predictions of circadian phase using both a physiologically based circadian limit cycle oscillator mathematical model (Hannay model), and published regression equations that utilize average sleep onset, midpoint, and offset to predict DLMO. Agreement of proxy predictions with measured DLMO were assessed and compared.

RESULTS

DLMO predictions using the Hannay model outperformed DLMO predictions based on children's sleep/wake parameters with a Lin's Concordance Correlation Coefficient (LinCCC) of 0.79 compared to 0.41-0.59 for sleep/wake parameters. The mean absolute error was 31 min for the Hannay model compared to 35-38 min for the sleep/wake variables.

CONCLUSION

Our findings suggest that sleep/wake behaviors were weak proxies of DLMO phase in children, but mathematical models using data collected from wearable data can be used to improve the accuracy of those predictions. Additional research is needed to better adapt these adult models for use in children.

CLINICAL TRIAL

The i Heart Rhythm Project: Healthy Sleep and Behavioral Rhythms for Obesity Prevention https://clinicaltrials.gov/ct2/show/NCT04445740.