Over 80% of people with Alzheimer's disease or a related dementia are cared for in their home environments by family members. Family caregivers often report increased anxiety and depression, and many forego their own health needs as the demands of being a family caregiver are sustained over many years. It is also known that poor interactions between patient and caregiver increase the difficulty of providing care. Monitoring reactivity between patient and caregiver could signal when problematic interactions might occur. Just-in-time or even predictive recommendations in those moments could improve these interactions and reduce strain on caregivers. This project develops a monitoring, modeling, and interactive recommendation solution (for caregivers) for in-home dementia patient care that focuses on caregiver-patient relationships. This includes monitoring for mood and stress and analyzing the significance of monitoring those attributes to dementia patient care and subsequent behavior dynamics between patient and caregiver. In addition, novel and adaptive behavioral suggestions at the right moments aim at helping improve familial interactions related to caregiving, which over time should ameliorate the stressful effects of the patient's illness and decrease strain on caregivers. This work could also benefit nursing homes and assisted living facilities by improving care for their residents, and could be useful for other caregiving situations, including the care of children with emotional/behavioral challenges who are cared for at home by their families. Educational modules introduce both healthcare students and technology students to this multidisciplinary area of research.
The technical solution consists of a core set of statistical learning based techniques for automated generation of specialized modules required by in-home dementia patient care. Personalization, context, and stages of dementia all contribute to the need for specialized modules; and without new solutions for rapid and automatic generation of these specialized modules, progress in effective treatment and patient/caregiver relationship improvement will be very difficult and slow. There are three main technical components in the solution. The first obtains textual content and prosody from voice and uses machine learning techniques to create classification models. This approach not only monitors patients' behavior, but also caregivers', and infers the underlying dynamics of their interactions, such as changes in mood and stress. The second is the automated creation of classifiers and inference modules tailored to the particular patients and dementia conditions (such as different stages of dementia). The third is an adaptive recommendation system that closes the loop of an in-home behavior monitoring system. The main technical contribution is the quick and accurate development of personalized smart and connected health platforms and the potential for reduced medical costs.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.