Sensing the care: Advancing Unobtrusive Sensing Solutions to Support Informal Caregivers of Older Adults with Cognitive Impairment
Nikita Sharma is a PhD student in the department Psychology, Health & Technology. Promotors are prof.dr. J.E.W.C. van Gemert-Pijnen from the faculty of Behavioural, Management of Social Sciences, prof.dr. P.J.M. Havinga from the faculty of Electrical Engineering, Mathematics and Computer Science and prof.dr. H. Oinas-Kukkonen from the University of Oulu, Finland.
Older adults (65 years and above) make up a growing proportion of the world’s population which is anticipated to increase further in the coming decades. With age, older adults become fragile and susceptible to chronic diseases such as chronic obstructive pulmonary disease (COPD), diabetes, heart disease, Alzheimer’s or other cognitive impairments. As a result, they require a wide range of care and support services to maintain their quality of life including medical assistance, aid in performing daily selfcare activities, and social and emotional support from their caregivers. However, the scarcity of professional care workers and care homes, coupled with the desire of older adults to age in place, puts a substantial burden on informal caregivers. This burden negatively affects their physical, mental, and social well-being. Numerous sensing solutions have been developed to assist caregivers by monitoring the physical and physiological activities of older adults. These solutions include tactile (wearables), vision-based, and radio-frequency (RF)-based sensing systems. However, it has been observed that many of these solutions are not ideal for older adult care, specifically with cognitive impairments. Wearables may lead to feelings of stigmatization or may be easily forgotten to be worn. Vision-based systems require the person being monitored to remain within the line of sight, raising concerns about privacy and ethics. While RF-based sensing systems, such as Wi-Fi channel state information (CSI) or mmWave radar, possess the potential to mitigate these limitations by providing unobtrusive monitoring capabilities, their integration into older adult care remains limited. This limitation can be attributed to the gap between theoretical understanding (pertaining to their technological development) and real-world applications (by considering user’s needs). As a result, only a few commercially available unobtrusive sensing solutions (USSs) exist to support older adults and their caregivers.
In that regard, the overarching aim of this thesis was to develop and evaluate the USSs for in-home monitoring of older adults with cognitive impairment (OwCI) who live alone in their own houses to ease the support of their informal caregivers. Chapter 1 of this thesis establishes the contextual background for the utilization of USSs in general older adult care. It provides an overview of the stakeholders involved in the older adult care ecosystem and the technologies that can support them. Through this exposition, the chapter attempts to create a basic foundation for comprehending the implementation opportunities and challenges when incorporating USSs in older adult care. Based on identified implementation opportunities and challenges, the research aims and corresponding research questions pursued in this study were outlined. These aims and questions serve as a guide for the subsequent chapters, ensuring a cohesive and purposeful exploration of the topic at hand.
The criteria for using an appropriate technology become more stringent when considering the context of older adult care. While real-time monitoring is a crucial aspect of in-home older adult care, it is of paramount importance that this monitoring should be unobtrusive i.e., it should not require care recipients to contact or interact with sensing devices. Additionally, it is financially advantageous to utilize ubiquitous technology having the capability to monitor larger spaces (at least the house/apartment of the care recipient). In Chapter 2, a scoping review was conducted to identify the state-of-the-art studies utilizing unobtrusive sensing technologies for human activity recognition (HAR). Among the identified technologies, RF-based sensing technologies, specifically Wi-Fi CSI-based sensing, emerged as a promising option owing to its ubiquitous and unobtrusive nature. However, to implement Wi-Fi CSI in real-world older adult care scenarios, it is crucial to thoroughly identify the associated potential challenges from the perspective of both technology and users.
Despite the availability of numerous CSI datasets in the literature, the lack of datasets specifically tailored to activities related to OwCI care in realistic settings was evident. Among other behaviors exhibited by OwCI, agitation is common during the advanced stages of cognitive impairment. In that regard, the Wi-Gitation dataset was meticulously collected and evaluated in Chapter 3. Wi-Gitation captures agitation-related physical activities, both full-body and fine-grained, within a one-bedroom apartment. This dataset was evaluated utilizing different analysis approaches, namely the mixed-data analysis (where training and testing occur on the same group of participants) and the leave-one-out analysis (where training is conducted on one group of participants and testing is performed on a different participant). These analyses not only demonstrated the potential of Wi-Fi CSI for HAR but also highlighted its limited generalizability across different domains such as participants and locations. To further explore the generalizability issue caused due to data mismatch, in Chapter 4, more detailed exploratory analyses on the Wi-Gitation dataset were conducted. These exploratory analyses confirmed the person-wise and location-wise data mismatch, with person-wise data mismatch being more dominant compared to location-wise data mismatch in realistic scenarios. These results indicate the need for further advancement in CSI-based HAR algorithms facilitating domain invariance or adaptation. In Chapter 5, the InSSeqTra framework was proposed to enhance the domain adaptation and evaluated for person-wise and location-wise data mismatch. InSSeqTra uses multiple (available) CSI datasets for pre-training a base model in selective sequential order followed by using task domain data (Wi-Gitation) for fine-tuning. The use of multiple CSI datasets in selective sequential order for pre-training helps the base model in gathering CSI features closer to the Wi-Gitation dataset. Thus, significantly enhancing the overall performance of CSI-based HAR.
In this thesis, alongside technological advancements, a parallel focus was given to identifying the needs and requirements of informal caregivers of OwCI from USSs. Understanding these needs and requirements was deemed crucial for the development of a caregiver-centric interaction platform that can effectively deliver real-time (in case of emergencies) monitoring information to caregivers. Chapter 6 presents a mixed-method study with informal caregivers of OwCI to highlight their current experiences, future expectations, and perceived usefulness of such advanced solutions. Informal caregivers expect centralized and empathetic care approaches in the care infrastructure. They also suggested the development of sensor-based care solutions that can be adaptable to care needs while ensuring the privacy and safety of the care recipients. The study also highlighted the information communication (IC) needs and requirements for communicating the information obtained from the sensing system to informal caregivers in different care scenarios namely, Fall, Nocturnal Unrest, Agitation, and Normal daily life. The IC needs were found to vary depending on a multitude of care related and personal factors. These findings served as a foundation for eliciting design features for designing the interaction platform in accordance with the persuasive system design (PSD) principles. Based on the findings obtained in Chapter 6, a low-fidelity (Lo-Fi) prototype of the interaction platform was prototyped and evaluated for its conceptual workflow and use of PSD functionalities with informal caregivers in Chapter 7. Overall, informal caregivers were found positive regarding the Lo-Fi prototype of the interaction platform, particularly valuing the personalization feature. They suggested the need for strategic IC agreements between informal and formal caregivers as a pre-condition for the successful implementation of USSs in OwCI care.
Lastly, Chapter 8 extends the discussion on the findings of this thesis. From a technological standpoint, the exploration of CSI for HAR revealed the significant challenge posed by data mismatch in realistic monitoring scenarios, where multiple factors (such as persons and locations) influence the performance of CSI much more adversely as compared to controlled laboratory settings. Furthermore, the development of the InSSeqTra framework not only addresses the issue of data mismatch but also advances research in CSI-based HAR. From the perspective of informal caregivers, USSs have proven highly valuable in objective decision-making, optimizing care, and providing peace of mind from a distance. However, care scenarios for older adults are dynamic and subject to rapid changes depending on the type of illness, co-morbidities, and personal situation of care recipients and their informal caregivers. Therefore, it necessitates adaptability in the design and development of USSs to effectively address diverse care needs arising from evolving care circumstances. Furthermore, the strengths and limitations of the studies conducted in this thesis, the implication of results obtained for research and policies, and the direction for future works were provided to enhance the practical and pragmatic usage of the research findings as well as to encourage further innovations for OwCI care.
