Life Adapt Inc. CEO Larry Holder and CRO Diane Cook recently developed an innovative approach to assess cognitive health using wearable sensor data, even in the presence of data gaps. This study is significant in the context of aging populations and the increasing prevalence of cognitive health issues.
Background and Objective
The research acknowledges the challenges in health assessment due to the limited interaction time between patients and healthcare providers and the constraints of conventional diagnostic tests. Recognizing the potential of continuous wearable sensor data in predicting clinical measures, the study focuses on addressing the challenge posed by inevitable data collection interruptions. The objective is to adapt a data generation algorithm, specifically a bidirectional time series generative adversarial network, to impute missing sensor readings, allowing for the creation of digital behavior markers that can predict clinical health measures.
Methodology
The methodology involves creating a generative model to impute multivariate time series data, preserving temporal dynamics and inter-feature dynamics. The process includes generating a behavior profile from the complete time series, labeled with activity categories, from which digital behavior markers are extracted and mapped to predicted clinical measures. The research validates this approach using continuous smartwatch data from 14 participants, showing promising results in reconstructing omitted data and in predicting clinical measures from the reconstructed complete data.
Results and Implications
The results demonstrate the viability of using wearable sensor data to gain insights into a person’s health in natural settings. The study achieves an average normalized mean absolute error of 0.0197 in reconstructing omitted data and presents correlations ranging from r=0.1230 to r=0.7623 in predicting clinical measures from the reconstructed data. These findings indicate that wearable sensor data, even with gaps, can be effectively used to assess health conditions in real-world settings.
The study’s implications are significant for the field of health monitoring and assessment, especially for older adults and individuals with cognitive impairments. By leveraging wearable technology and advanced data imputation methods, the research contributes to the development of more accurate, non-invasive, and continuous health monitoring systems. This approach can potentially enhance early detection and intervention for cognitive health issues, improving care quality and reducing healthcare costs.
In conclusion, this research marks a crucial step in the intersection of wearable technology, data science, and healthcare, offering a novel method for assessing cognitive health using sensor data. The approach addresses critical challenges in data collection and analysis, paving the way for more sophisticated health monitoring tools in everyday settings.
The research appeared in the article “Automated Cognitive Health Assessment Using Partially-Complete Time Series Sensor Data” published in the Methods of Information in Medicine.