Challenges in Using mHealth Data From Smartphones and Wearable Devices to Predict Depression Symptom Severity: Retrospective Analysis

A template and tutorial for preregistering studies using passive smartphone measures

Passive smartphone measures hold significant potential and are increasingly employed in psychological and biomedical research to capture an individual's behavior. These measures involve the near-continuous and unobtrusive collection of data from smartphones without requiring active input from participants. For example, GPS sensors are used to determine the (social) context of a person, and accelerometers to measure movement. However, utilizing passive smartphone measures presents methodological challenges during data collection and analysis. Researchers must make multiple decisions when working with such measures, which can result in different conclusions. Unfortunately, the transparency of these decision-making processes is often lacking. The implementation of open science practices is only beginning to emerge in digital phenotyping studies and varies widely across studies. Well-intentioned researchers may fail to report on some decisions due to the variety of choices that must be made. To address this issue and enhance reproducibility in digital phenotyping studies, we propose the adoption of preregistration as a way forward. Although there have been some attempts to preregister digital phenotyping studies, a template for registering such studies is currently missing. This could be problematic due to the high level of complexity that requires a well-structured template. Therefore, our objective was to develop a preregistration template that is easy to use and understandable for researchers. Additionally, we explain this template and provide resources to assist researchers in making informed decisions regarding data collection, cleaning, and analysis. Overall, we aim to make researchers' choices explicit, enhance transparency, and elevate the standards for studies utilizing passive smartphone measures.

Longitudinal Assessment of Seasonal Impacts and Depression Associations on Circadian Rhythm Using Multimodal Wearable Sensing: Retrospective Analysis

BACKGROUND

Previous mobile health (mHealth) studies have revealed significant links between depression and circadian rhythm features measured via wearables. However, the comprehensive impact of seasonal variations was not fully considered in these studies, potentially biasing interpretations in real-world settings.

OBJECTIVE

This study aims to explore the associations between depression severity and wearable-measured circadian rhythms while accounting for seasonal impacts.

METHODS

Data were sourced from a large longitudinal mHealth study, wherein participants' depression severity was assessed biweekly using the 8-item Patient Health Questionnaire (PHQ-8), and participants' behaviors, including sleep, step count, and heart rate (HR), were tracked via Fitbit devices for up to 2 years. We extracted 12 circadian rhythm features from the 14-day Fitbit data preceding each PHQ-8 assessment, including cosinor variables, such as HR peak timing (HR acrophase), and nonparametric features, such as the onset of the most active continuous 10-hour period (M10 onset). To investigate the association between depression severity and circadian rhythms while also assessing the seasonal impacts, we used three nested linear mixed-effects models for each circadian rhythm feature: (1) incorporating the PHQ-8 score as an independent variable, (2) adding seasonality, and (3) adding an interaction term between season and the PHQ-8 score.

RESULTS

Analyzing 10,018 PHQ-8 records alongside Fitbit data from 543 participants (n=414, 76.2% female; median age 48, IQR 32-58 years), we found that after adjusting for seasonal effects, higher PHQ-8 scores were associated with reduced daily steps (β=-93.61, P<.001), increased sleep variability (β=0.96, P<.001), and delayed circadian rhythms (ie, sleep onset: β=0.55, P=.001; sleep offset: β=1.12, P<.001; M10 onset: β=0.73, P=.003; HR acrophase: β=0.71, P=.001). Notably, the negative association with daily steps was more pronounced in spring (β of PHQ-8 × spring = -31.51, P=.002) and summer (β of PHQ-8 × summer = -42.61, P<.001) compared with winter. Additionally, the significant correlation with delayed M10 onset was observed solely in summer (β of PHQ-8 × summer = 1.06, P=.008). Moreover, compared with winter, participants experienced a shorter sleep duration by 16.6 minutes, an increase in daily steps by 394.5, a delay in M10 onset by 20.5 minutes, and a delay in HR peak time by 67.9 minutes during summer.

CONCLUSIONS

Our findings highlight significant seasonal influences on human circadian rhythms and their associations with depression, underscoring the importance of considering seasonal variations in mHealth research for real-world applications. This study also indicates the potential of wearable-measured circadian rhythms as digital biomarkers for depression.

Framework for Ranking Machine Learning Predictions of Limited, Multimodal, and Longitudinal Behavioral Passive Sensing Data: Combining User-Agnostic and Personalized Modeling

BACKGROUND

Passive mobile sensing provides opportunities for measuring and monitoring health status in the wild and outside of clinics. However, longitudinal, multimodal mobile sensor data can be small, noisy, and incomplete. This makes processing, modeling, and prediction of these data challenging. The small size of the data set restricts it from being modeled using complex deep learning networks. The current state of the art (SOTA) tackles small sensor data sets following a singular modeling paradigm based on traditional machine learning (ML) algorithms. These opt for either a user-agnostic modeling approach, making the model susceptible to a larger degree of noise, or a personalized approach, where training on individual data alludes to a more limited data set, giving rise to overfitting, therefore, ultimately, having to seek a trade-off by choosing 1 of the 2 modeling approaches to reach predictions.

OBJECTIVE

The objective of this study was to filter, rank, and output the best predictions for small, multimodal, longitudinal sensor data using a framework that is designed to tackle data sets that are limited in size (particularly targeting health studies that use passive multimodal sensors) and that combines both user agnostic and personalized approaches, along with a combination of ranking strategies to filter predictions.

METHODS

In this paper, we introduced a novel ranking framework for longitudinal multimodal sensors (FLMS) to address challenges encountered in health studies involving passive multimodal sensors. Using the FLMS, we (1) built a tensor-based aggregation and ranking strategy for final interpretation, (2) processed various combinations of sensor fusions, and (3) balanced user-agnostic and personalized modeling approaches with appropriate cross-validation strategies. The performance of the FLMS was validated with the help of a real data set of adolescents diagnosed with major depressive disorder for the prediction of change in depression in the adolescent participants.

RESULTS

Predictions output by the proposed FLMS achieved a 7% increase in accuracy and a 13% increase in recall for the real data set. Experiments with existing SOTA ML algorithms showed an 11% increase in accuracy for the depression data set and how overfitting and sparsity were handled.

CONCLUSIONS

The FLMS aims to fill the gap that currently exists when modeling passive sensor data with a small number of data points. It achieves this through leveraging both user-agnostic and personalized modeling techniques in tandem with an effective ranking strategy to filter predictions.

From smartphone data to clinically relevant predictions: A systematic review of digital phenotyping methods in depression

BACKGROUND

Smartphone-based digital phenotyping enables potentially clinically relevant information to be collected as individuals go about their day. This could improve monitoring and interventions for people with Major Depressive Disorder (MDD). The aim of this systematic review was to investigate current digital phenotyping features and methods used in MDD.

METHODS

We searched PubMed, PsycINFO, Embase, Scopus and Web of Science (10/11/2023) for articles including: (1) MDD population, (2) smartphone-based features, (3) validated ratings. Risk of bias was assessed using several sources. Studies were compared within analysis goals (correlating features with depression, predicting symptom severity, diagnosis, mood state/episode, other). Twenty-four studies (9801 participants) were included.

RESULTS

Studies achieved moderate performance. Common themes included challenges from complex and missing data (leading to a risk of bias), and a lack of external validation.

DISCUSSION

Studies made progress towards relating digital phenotypes to clinical variables, often focusing on time-averaged features. Methods investigating temporal dynamics more directly may be beneficial for patient monitoring. European Research Council consolidator grant: 101001118, Prospero: CRD42022346264, Open Science Framework: https://osf.io/s7ay4.

Biomarker-Guided Tailored Therapy in Major Depression

This chapter provides a comprehensive examination of a broad range of biomarkers used for the diagnosis and prediction of treatment outcomes in major depressive disorder (MDD). Genetic, epigenetic, serum, cerebrospinal fluid (CSF), and neuroimaging biomarkers are analyzed in depth, as well as the integration of new technologies such as digital phenotyping and machine learning. The intricate interplay between biological and psychological elements is emphasized as essential for tailoring MDD management strategies. In addition, the evolving link between psychotherapy and biomarkers is explored to uncover potential associations that shed light on treatment response. This analysis underscores the importance of individualized approaches in the treatment of MDD that integrate advanced biological insights into clinical practice to improve patient outcomes.