Sullivan SS, Bo W, Li CS, Xu W, Chang YP. Predicting Hospice Transitions in Dementia Caregiving Dyads: An Exploratory Machine Learning Approach.
Innov Aging 2022;
6:igac051. [PMID:
36452051 PMCID:
PMC9701063 DOI:
10.1093/geroni/igac051]
[Citation(s) in RCA: 1] [Impact Index Per Article: 0.5] [Reference Citation Analysis] [Abstract] [Key Words] [Grants] [Track Full Text] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 03/29/2022] [Indexed: 10/19/2023] Open
Abstract
Background and Objectives
Hospice programs assist people with serious illness and their caregivers with aging in place, avoiding unnecessary hospitalizations, and remaining at home through the end-of-life. While evidence is emerging of the myriad of factors influencing end-of-life care transitions among persons living with dementia, current research is primarily cross- sectional and does not account for the effect that changes over time have on hospice care uptake, access, and equity within dyads.
Research Design and Methods
Secondary data analysis linking the National Health and Aging Trends Study to the National Study of Caregiving investigating important social determinants of health and quality-of-life factors of persons living with dementia and their primary caregivers (n = 117) on hospice utilization over 3 years (2015-2018). We employ cutting-edge machine learning approaches (correlation matrix analysis, principal component analysis, random forest [RF], and information gain ratio [IGR]).
Results
IGR indicators of hospice use include persons living with dementia having diabetes, a regular physician, a good memory rating, not relying on food stamps, not having chewing or swallowing problems, and whether health prevents them from enjoying life (accuracy = 0.685; sensitivity = 0.824; specificity = 0.537; area under the curve (AUC) = 0.743). RF indicates primary caregivers' age, and the person living with dementia's income, census division, number of days help provided by caregiver per month, and whether health prevents them from enjoying life predicts hospice use (accuracy = 0.624; sensitivity = 0.713; specificity = 0.557; AUC = 0.703).
Discussion and Implications
Our exploratory models create a starting point for the future development of precision health approaches that may be integrated into learning health systems that prompt providers with actionable information about who may benefit from discussions around serious illness goals-for-care. Future work is necessary to investigate those not considered in this study-that is, persons living with dementia who do not use hospice care so additional insights can be gathered around barriers to care.
Collapse