Clinical Research Directory

Developing Intelligent Wearable Algorithms

Sponsor: Linnaeus University

Summary

Physical activity (PA) is one of the few behaviors that individuals can change on their own, incurring minimal costs while simultaneously yielding significant health benefits. Over the past decade, new methods have been developed to measure both physical activity and associated health outcomes, such as blood pressure. Notably, there has been an explosive development of so-called wearables, including smartwatches and activity trackers. Wearables are equipped with multiple sensors that measure various aspects of PA, such as steps and heart rate, as well as cardiovascular health indicators like blood pressure and oxygen saturation. Therefore, wearables can be viewed as Swiss army knives with many tools in one instrument. They are highly popular in the fitness industry, but their role in healthcare is appropriately limited. However, most wearables on the market have several disadvantages that make them unsuitable for use, even among healthy individuals. Several studies have revealed that they do not produce reliable or valid data for metrics like pulse, steps, and PA-related energy expenditure. Furthermore, they are primarily designed for the fitness market, not for use within healthcare systems or as support for behavior change, and they have not been transparently evaluated. Additionally, the algorithms translating signals from sensors into interpretable outcomes are often trade secrets. Worse still, they are updated and modified at irregular intervals, making it challenging to compare outcomes over time. Other significant limitations include questionable patient confidentiality, as data is often uploaded to companies\' cloud services. While research monitors are more flexible and transparent compared to commercial wearables, they lack essential features for daily use that are crucial in healthcare environments, such as the ability to communicate with the user. Currently, both commercial and research monitors cannot assess PA on an individual level, as they only utilize a limited portion of the rich data collected. Therefore, it is not surprising that their implementation in clinical care remains a challenge. Given the plethora of new products entering the market without documented validity, it is crucial to provide consumers, patients, healthcare professionals, and researchers with a transparent, evidence-based wearable. Against this backdrop, an interdisciplinary research group with the ambitious goal of developing and testing a high-functioning wearable tailored for use in healthcare-an e-physiotherapist (as opposed to commercial wearables targeting the fitness market-an \"e-personal trainer\") have been formed. In this project, the focus is on measuring PA, blood pressure, and energy consumption, as they represent some of the most significant risk factors for mortality and morbidity, namely inactivity, hypertension, and obesity. The overall goal of this project is to develop and validate AI-based algorithms for individually measuring various aspects of physical activity (PA), heart rate, energy expenditure, and blood pressure in laboratory settings as well as in everyday conditions. These algorithms represent a significant advancement compared to previous methods. In the case of PA metrics from accelerometry, current approaches rely on cut-points (threshold values) to define the intensity of PA. These cut-points are absolute, and individual variations in biology and biomechanics increase the risk of serious misclassification. To estimate intensity using heart rate, it is well-known that both resting heart rate and maximum heart rate are relative, requiring individual calibration for accurate measurements-essential even for accelerometry if one aims to measure PA on an individual level, a step not commonly taken today. Furthermore, heart rate is influenced by factors beyond PA, such as emotions and medication. To address these issues, combining information from accelerometry (biomechanics) and heart rate (physiological response), enhancing the ability to identify individual intensity and energy expenditure of PA. In this project, artificial intelligence (AI) and machine learning (ML) will be employed to analyze the collected data and predict the intensity of PA. If the proposed method demonstrates the ability to measure PA and blood pressure at an individual level, the project will proceed. Our intention is to use AI/ML to combine PA information with blood pressure data, creating a self-learning system capable of suggesting an appropriate dose of PA to optimize blood pressure. This approach has not been studied yet, likely due to the complexity of obtaining and analyzing these data. However, the technology, processing power, and analysis tools are now available, making it timely to investigate its feasibility.

Official title: Design of an Intelligent Wearable to Assess Physical Activity and Health Related Outcomes - the DIWAH Study

Key Details

Conditions

Interventions

Locations (1)