Understanding Population Gaps and Using Explainable Machine Learning to Predict Risk of Falls for Senior Adults

Abstract

Senior adult falls are one of the leading causes of injury and death. The analysis of social determinants and healthcare utilization for senior patients with a history of falls is limited. Most healthcare outcome studies focus on risk factors for falls. There is a lack of studies on patients’ socioeconomic and demographic effects on healthcare utilization. With unequal utilization of healthcare adoption, clinical developments, such as clinical decision-making tools, cannot reach people equally, which limits the potential of improving healthcare coverage.To close the healthcare utilization gap, this dissertation includes two studies focusing on healthcare disparities and the use of innovative technology to provide solutions. 1) To understand the population gaps in healthcare adoption: the first study examines socioeconomic disparities and their impact on healthcare utilization for senior patients with a history of falls in southeast Wisconsin and the Milwaukee metropolitan area, one of the most racial segregated communities in the United States. This study found that disadvantaged social conditions, including under-insurance, residing far from a hospital, lower education, and lower income, were associated with a lower healthcare utilization rate. These findings indicate the need to address healthcare inequities to facilitate equitable care for all patients. 2) To improve the clinical decision support model using an interpretable machine learning framework: The clinical decision support model has been demonstrated to be an effective method to improve healthcare and reduce inequality; however, the mechanisms of the clinical decision support model are difficult to interpret for clinical practitioners. To address this problem, the second study discusses interpretable machine learning uses and applications. The second study discussed an example of machine learning-based text classification for clinical computed tomography reports, specifically for temporal bone fractures, one of the severe outcomes of senior adult falls. This study develops a solution to use an interpretable artificial intelligence framework and computer-based methodology to understand the mechanisms of clinical decision models to close the inequality gap for senior adults. Machine learning models were used to classify fractures based on text reports, and two methodologies were used for interpretation, which resulted in high interpretability, possibly improving the physicians’ trust in the clinical decision-making model. This study can assist physicians in using technology more effectively to aid decision-making and increase trust in computerized models. In addition, this study demonstrates machine learning classification models can process a large quantity of clinical texts to reduce the physician’s documentation processing workload, possibly leading to high-quality healthcare.Together, the two papers in this dissertation underscore the importance of addressing healthcare disparities. The study shows it is feasible to use machine learning, a computer-based technology, to form a pathway for better care for patients. By addressing disparities and leveraging technology effectively, it is possible that healthcare providers could provide equitable and high-quality care for all patients, regardless of socioeconomic status or diagnosis.