Digital + health research

The growing ubiquity and ability of connected devices (e.g., smartphones and wearables) to collect day-to-day lived experiences of people has gotten a lot of attention from healthcare researchers. High-frequency continual usage of smartphones allows for objective quantification of personalized daily behavior using multimodal real-world data (RWD) streams.

Through this decentralized approach, researchers can also cast a wider net and access larger and more diverse portions of the population at a fraction of the cost of traditional brick-and-mortar clinical trials. One of the key promises of RWD is to help us learn a comprehensive picture of an individual’s unique lived experience. Ultimately, the goal is to know what works and for whom to guide the development of new patient-centric treatments.

The Global South also stands to benefit from this a great deal due to regional challenges - inadequate health systems and infrastructure, low doctor-patient ratios, and expensive healthcare. The rising penetration of smartphones makes digital health a potentially viable solution to these challenges.

Beyond the Technology

However, to expect diverse and representative use of technology for health may be delusionary, atleast in the short term. There are several real-world challenges - like the reluctance of people to participate and continue to engage and share their health-related data in decentralized research trials due to privacy concerns or the digital divide.

There is a risk for researchers to jump at the opportunity to have this wealth of data at arm’s reach and perhaps unintentionally overlook the need for stringent privacy policies. Recent misadventures in data sharing could seriously impact the willingness of people to participate in research that collects sensitive health-related data.

Digital divide is another major concern - the disparity between people who are open to adapting to new technology and those who stay on the periphery, typically older and less tech-savvy people. The divide could exclude the latter from equitable representation in digital health research, making the derived learnings from such studies all the more incomprehensible to them, and so leaving them behind as the world advances.

The digital divide can also be impacted by the difference in the rate of the population connected to the internet between the Global North and Global South. This poses other challenges like digital health tools not functioning properly due to lack of internet access, low data bandwidth, varying hardware specifications, etc.

So while going digital can prove to be an effective and scalable solution to address the massive supply and demand gap in healthcare in the Global South, we need to be more mindful in the design and development of decentralized studies to collect representative data that can inform the development of equitable health policies.

The interplay of such real-world factors can lead to bias that may limit the impact of digital health from seeking its full potential by leaving a significant chunk of the global population behind - i.e., the less tech-savvy, the reluctant, and the disconnected.

And bias can originate in the many stages of research, from unbalanced recruitment to differential long-term retention of the population in longitudinal studies. Underrepresentation can make resulting outcomes inapplicable to those who weren’t included in the trials and, therefore, biased.

A well-known phenomenon is algorithms learning bias from datasets derived from human beings’ actions and responses, which may be biased. The use of these algorithms in health tools and applications sets a dangerous precedent for the perpetuation of bias.

Our ongoing efforts to help address biases in digital (mental) health research

We think there is a critical and unmet demand to explore bias in digital health research to develop long-term equitable and inclusive technology-augmented solutions. We aim to quantify these biases and develop methods to correct them to generate robust, generalizable, and, most importantly, transparent insights. We also create strategies to improve cohort diversity and boost long-term engagement in decentralized studies by working closely with patients, families, and providers.

1. Our current projects build upon our past work generating data-driven insights on real-world engagement in large, fully remote observational studies assessing. Across multiple large-scale decentralized studies, we are interested in factors linked to differential recruitment, representation, and retention trends.
   Team - Ethan Kim, Sophia Li

2. We are also researching factors leading to biases in wearable data collection from real-world settings. As part of our wearable operations (WearOPS), we are developing an analytical workflow to assess potential bias in data collection due to differential recruitment and retention.
   Team - Ramzi Halabi, Ava Homiar, Ethan Kim, Calvin Herd

3. We are working with the Frontiers journal to curate the latest findings through a research topic series: Understanding Participant Engagement and Need for User-centered Assessment and Interventional Digital Tools. We want to engage the wider research community in sharing the latest findings on how to engage the target population, address their data sharing & privacy concerns, and improve long-term retention by developing user-centered digital tools.
   Team - Abhishek Pratap, Patricia Arean, Benjamin Nelson, Brenna Renn, Abigail Ortiz

If bias in digital health research is left unchecked, it can result in solutions that work for some but not all.

We aspire to inform the development and deployment of equitable and inclusive digital mental health solutions and work towards mending the gaping global health inequity.

Wearables in Mental Health Research & Care

Augmented with self-reported outcomes and episodic clinical data, physiological data from wearables could provide a more accurate picture of living with mental health illness (or issues).

There is growing interest to explore the feasibility and utility of using wearable tech to monitor individual physiological responses to various psychosocial settings as people go about their daily lives. Ongoing early research has demonstrated the potential of wearable tech in mental health research and care. Some examples include wrist devices that can alleviate anxiety, monitoring sleep, and motor data to predict crisis periods in people with depression^4-5. And by observing real-world behavior at scale, researchers could identify underlying transdiagnostic signatures (e.g sleep, stress, local environment, etc) related to mental health. Such risk factors could accumulate differentially over time in individuals leading to different lived experiences.

Our goal is to enable a transparent ecosystem for deploying wearbles in clinical research (and eventually care), to collect real-world end-points related to clinical outcomes of interest.

Wearable Operations in Real-World Settings - WearOPS

Deploying the devices (+ accompanying smartphone apps) from the research facility/clinic to participants and patients in the real world to collect high fidelity continual RWD involves many critical operational steps.

The WearOps effort at KCNI/CAMH aims to bring the crucial stakeholders (from manufacturers, researchers, privacy experts, clinicians to participants/patients, and their families) to develop an efficient and transparent system for enabling wearable research. The WearOPS overall process ranges from selecting and configuring fit-for-purpose devices to using machine learning/artificial intelligence methods to create robust and transparently validated digital biomarkers linked to clinical outcomes of interest in a target population.

For our initial research studies, we are focusing on the following key areas:

• Stakeholder experience: To ensure the smoothest possible deployment of devices to the participants and study management by research coordinators, we are developing device-specific onboarding materials. We are focusing to prioritize the participants’ user experience by providing a hybrid of on-site and remote onboarding options with on-demand support available to participants

• Privacy: Owing to the sensitive nature of the data being shared using multiple third-party vendors, we enforce strict regulations to assure the security and confidentiality of participants’ data. Only necessary data elements will be collected, and participant data will not be shared with third parties unless consented to by the participant and required for the study. Participant accounts will be linked to non-identifiable participant IDs to ensure confidentiality, and data access will only be available to the study team.

• Data management: Development of an automated workflow to download and process the lowest granularity of individual participant-level wearable data on a daily basis using a scalable time-series database. And as part of data curation, we will compute near-real-time compliance/engagement metrics for study participants. The system will use the wearable compliance metrics to triage participants for “human-in-the-loop” support.

• Bias / Engagement / Human-in-the-loop: We recognize technology-based data collection in naturalistic settings, especially in populations facing mental health problems, could lead to biases in collected data. We aim to address this transparently and during the study design and data collection phase. We are developing a data quality monitoring workflow. The workflow will assess the potential biases in data collection in near real-time due to differential recruitment and retention. The pipeline will also triage the participants to “human-in-the-loop” support. By bringing in a study coordinator, we aim to further understand and address participants’ concerns about not being able to share their data.

• Generating Actionable Insights: Upon wearable raw data acquisition and secure storage, transparent data validation is performed through quality assessment and control prior to any analysis. Once processed and validated, wearable sensor data is analyzed using signal processing and statistical techniques to compute data features and learn underlying patterns to serve as mental health digital biomarkers that aid in the understanding and treatment of mental illnesses. Ultimately we want to stream near-real-time data to understand significant drift in behavior at an individualized level that could trigger just-in-time contextual assessments and eventually interventions.

Upcoming Projects:

• Major Depressive Disorder (MDD): To assess if real-world behavior trajectories gathered through wearables could inform selection and tailoring of treatment in MDD as well as early prediction of an impending mental health crisis.

• Substance use disorder: Investigate the utility of moment-to-moment monitoring of physiological states in the detection of the frequency, timing, and intensity of urges and cravings of substance use, which are essential for monitoring progress through treatment programs.

• Understanding postural changes in bipolar disorder: Analyze and model potential postural dynamics patterns in bipolar disorder patients at different illness severity levels

• Assessment quality of sensor data collected in real-world settings: We are building a pipeline to help quantitatively assess sensor data collected in naturalistic settings for correctness, consistency, completeness, and compliance. The pipeline will also quantify potential biases regarding the sociodemographic and clinical characteristics of the target population.

This work has been supported in part by the CAMH Foundation and the Krembil Foundation.

References:

_{1. Heintzman, N. D. (2015). A digital ecosystem of diabetes data and technology: Services, systems, and tools enabled by wearables, sensors, and apps. Journal of Diabetes Science and Technology, 10(1), 35–41. https://doi.org/10.1177/1932296815622453}

_{2. Mason, H et al. (2022). Detection of COVID-19 using multimodal data from a wearable device: Results from the first TemPredict Study. Scientific Reports, 12(1), 1–15. https://doi.org/10.1038/s41598-022-07314-0}

_{3. Rubenstein, L. Z. (2006). Falls in older people: Epidemiology, risk factors and strategies for prevention - PubMed. Age and Ageing, 35 Suppl 2. https://doi.org/10.1093/ageing/afl084}

_{4. “Wearable” technology for mental health. (2020, February 27). Amelia Virtual Care. https://ameliavirtualcare.com/wearable-technology-for-mental-health/}

_{5. Lee, S., Kim, H., Park, M. J., & Jeon, H. J. (2021). Current advances in wearable devices and their sensors in patients with depression. Frontiers in Psychiatry, 0. https://doi.org/10.3389/fpsyt.2021.672347}