A clinician-in-the-loop smart home to support health monitoring and intervention for chronic conditions

点击翻译按钮获取中文摘要

PROJECT SUMMARY/ABSTRACT The world’s population is aging and the increasing number of older adults with chronic health conditions is a challenge our society must address. While the idea of smart environments is now a reality, there remain gaps in our knowledge about how to scale smart homes technologies for use in complex settings and to use machine learning (ML) and activity learning technologies to design automated health assessment and intervention strategies. The long-term objective of the parent project is to improve human health and positively impact health care delivery by developing smart environments that aid with health monitoring and intervention. The primary objective of this supplement project is to extend the design of our ‘clinician-in-the-loop’ smart home to include computational methods for recognizing and reporting on bias in ML algorithms. Bias in ML algorithms is a public health hazard. A secondary objective is to add to the scarce body of knowledge regarding ML techniques for assessing and quantifying algorithm bias. In this supplement study, we will demonstrate how to operationalize bioethics principles such as autonomy (the right to make informed decisions about care and right to be informed about any associated risks) into actionable design requirements to close the gap between principles and practice. We will then share these with the ML healthcare application (ML- HCA) research community by creating a new open source online course containing 3 modules. This supplement proposal builds upon our extensive experience in the CASAS lab at Washington State University and prior collaboration on the development of smart home design, activity recognition, and the use of these technologies for functional health monitoring, assessment, and intervention. This supplement effort will allow us to expand our sample size and enhance the robustness of pervasive-computing-based behavior monitoring and machine learning to further advance the quality of health monitoring, including bias recognition. We will offer new advances in computational methods for anti-bias algorithm development using data from smart homes and smart watches that are monitoring the daily activities of underrepresented persons. Our course modules will support knowledge development regarding how to operationalize bioethics principles for practice at each stage of ML-HCA design. In the online course we will also address the importance and impact of multidisciplinary teams in designing for transparency, fairness, and non-discriminatory ML-HCA applications such as the health smart home. Given that ML models are known to introduce bias and risk to members of minority groups and given our national conversation on equity, fundamental human rights, and the risks associated with living as a minority person, it is critical that health technology-human interface with underrepresented persons is improved. Anti-bias ML training techniques that could be broadly applied in the development of ML-HCAs would be of significant value to minority individuals and families and all of society.