Ä¢¹½tv

Humanized data, predictive modeling, and a cohesive digital strategy can aid the transition

The  last May as the first consumer wearable digital health technology (DHT) to cross over the clinical barrier, when the U.S. Food & Drug Administration approved its atrial fibrillation history feature as a medical device development tool, clearing it for use in clinical trials.

Smart watches, fitness trackers, and other commercial wearables are increasingly being considered for noninvasive monitoring and reporting on physical activity, health, and wellness for a broad range of use cases. By incorporating novel, data-driven approaches, wearable DHTs have the potential to provide valuable insights and predictive capabilities, particularly in the healthcare space, including improving early disease detection, enhancing personalized patient outcomes, and informing optimized interventions and therapeutics. These data-driven approaches can be further enhanced through application of evolving artificial intelligence and machine learning techniques, affording stronger predictive capabilities and greater insights via novel data aggregation and modeling methods.

Such approaches afford unique opportunities to augment existing clinical care, which is mainly provided through in-person visits. DHTs can improve patient compliance with treatment regimens by enabling caregivers to remotely monitor the progression of symptoms and associated patient activities and health metrics on a more regular cadence. Wearable DHTs that measure clinically validated patient-centered endpoints can unlock opportunities for both patient care and for clinical trials and observational studies.

By offering more frequent patient health outcome measurements in naturalistic, contextualized, patient-centric environments, DHTs can provide novel data to create and implement personalized treatment plans, which can similarly evolve in line with patient symptoms, recovery, or rehabilitation. This can result in direct improvements to patient quality of care and patient-specific health and wellness outcomes.

Despite these benefits, few wearable DHTs have made the successful  from the consumer products space into the clinical arena, which is forecasted to reach . Why do wearable technologies continue to grow and proliferate in the commercial sector while a much smaller number currently offer viable clinical solutions? 

By helping stakeholders create strong experimental designs, predictive models that leverage humanized data have the potential to maximize the probability of success when transitioning from the consumer electronics space to the clinical and digital health domains. 

One key reason for this disconnect is that the data-driven outcomes and metrics provided for the general consumer may not necessarily be directly relevant in the clinical context. More specifically, the metrics used by DHTs in the consumer space, like measurements for average heart rate, may not immediately translate to the demand for more complex patient outcome requirements in the clinical domain, like heart rhythm measures for atrial fibrillation.

The accuracy and reliability of wearable-derived data in the clinical space may need to be higher than what's required commercially, particularly considering such data may be used to detect serious health outcomes and consequences. For example, certain conditions and their severity may be informed by slight, yet significant changes in a patient's resting heart rate. If commercial solutions are unable to accurately and reliably detect these changes, be it acutely or over time, then it may have an adverse impact on critical health outcomes or derived intervention approaches.

Contextualization and humanization of wearable data that accounts for individual user physiology, the activities of the person, and associated environmental and behavioral determinants, can address these potential limitations and provide a more actionable means through which consumer products can transition into clinical domains.

Evaluating and contextualizing wearable-derived data outcomes

Often, early research into wearable-derived health and wellness outcomes focuses solely on the wearable device — not the people using it, their surroundings, or the types of activities they perform. As a result, the reliability and subsequent utility of the data can be compromised. Researchers may view participants as idealized, homogeneous users, dismissing participant-driven variability in data as noise in their ensuing analyses and interpretations. This can lead to variability in data that stems from inherent and potentially important individual user differences (arising for example, from concomitantly different demographics, experiences, socio-economic status, behaviors, mental states, etc.) being dismissed or considered erroneous, misinforming valid and potentially important measured outcomes.

Understanding and contextualizing the environment where data are being collected, as well as the user activities being performed within that environment, will also enhance wearable solution data and how it is interpreted and implemented. Understanding whether a user is performing an activity while in pain or discomfort, for instance, may provide critical information when evaluating and interpreting their step counts. Similarly, determining whether an increase in heart rate measured by a wearable device is due to exercise or the onset of an illness speaks directly to how the data is assessed and a potential intervention selected. When trying to draw conclusions from any wearable-derived data set, it's important to know the additional real-world factors impacting the data so that they can be analyzed and interpreted in the appropriate context. 

To gather high-quality data, a systematic review of the tasks a DHT will monitor and environmental features that may influence the behavior and physiology of individual participants can help reduce difficulties in their experience that could lead to non-compliance. Additionally, it's important to contextualize factors relating to an individual's perceptions (e.g., pain, motivation, fatigue, etc.), which can be captured and analyzed using a variety of qualitative and quantitative methods. For example, surveys and interviews can be used to identify different sub-types of users (persona modeling); audio and video recordings can be combined with language, gesture, and activity recognition algorithms; and social media posts can be used to understand a participant's emotions and social interactions. 

Humanizing the data by understanding what people will do or how they'll interact with products in real-world settings also makes for more reliable and compliant data, affording deeper insights. Specifically, humanized data accounts for the complex web of individual, task, and environmental determinants of behavior over time. By incorporating and analyzing these data in parallel, personalized outcomes and targeted, use-case specific interventions may be possible. 

Other benefits of data humanization include:

• Improved participant management (e.g., enhances participant retention and compliance, reducing the number of required participants)
• Reduced costs through remote testing and monitoring, which is lower in cost than in-person visits 
• Improved efficiency/study management (only collecting information that is pertinent to the problem — not over-collecting data)
• Accurate verification and validation of data to assure that analyses consider key, real-world contexts

Using humanized data helps build predictive models for human health and wellness and associated behavioral outcomes by ensuring the models are grounded in real-world contexts.

Crossing the clinical trial barrier

Using humanized data helps build predictive models for human health and wellness and associated behavioral outcomes by ensuring the models are grounded in real-world contexts, making them not only more reliable, but more informative. By helping stakeholders create strong experimental designs, predictive models have the potential to maximize the probability of success when transitioning from the consumer space to the clinical and digital health domains. Moreover, using such models to aggregate and assess large, multifaceted data streams with high computational efficiency makes more personalized, data-driven and patient-centric insights and resultant outcomes possible. Even so, capturing the human experience can require complex models that may be harder to interpret in clinical settings, which can create challenges in transitioning consumer products to clinical DHTs. 

Barriers

When implementing wearable solutions and associated humanized data models in clinical trials, potential barriers and challenges may include the following.

• Paucity of standards: Do the algorithms work as intended? Are they safe? How can their safety and effectiveness be appropriately validated, analytically and clinically?
• User acceptance, adherence, and compliance: Will patient populations want to use the device? Are devices suitable for the target demographic? Has adequate education and familiarization been undertaken?
• Trial facilitator expertise and experience: Does the team conducting the clinical trial have appropriate experience and training to successfully implement the technology solution, educate patients, and support necessary onboarding and troubleshooting requirements?

Overcoming barriers to implementing wearable solutions and humanized data models in clinical settings requires deep expertise in model construction, training, and interpretation. Employing AI and ML algorithms to train and improve these models can enable stakeholders to anticipate the challenges that may arise in the clinical stage and proactively address them with informed experimental designs. 

Moreover, it's necessary to have a cohesive digital strategy in place that considers clinical user-case requirements, the targeted user/patient demographic, and the prioritized health and wellness endpoints — drawing in parallel on innovative AI and ML approaches to generate more granular, real-world patient insights. Such an approach can enhance participant compliance and technology adherence, as well as improve operational efficiency in clinical research efforts. Collectively, such an approach can evaluate and potentially optimize the success of health interventions and resulting patient outcomes using fewer participants at substantially lower implementation costs.