Junko Fukui Innes

Topic

Predicting Transition of the Next Level of Care for Aging Population with Home Support Services

School of Health Information Science

Date & location

• Monday, December 15, 2025
• 9:00 A.M.
• Virtual Defence

Examining Committee

Supervisory Committee

• Dr. Abdul Rousdari, School of Health Information Science, University of Victoria (Supervisor)
• Dr. Karen Courtney, School of Health Information Science, UVic (Member)
• Dr. Mary Ellen Purkis, School of Nursing, UVic (Outside Member)

External Examiner

• Dr. Riccardo Bellazzi, Department of Electrical, Computer and Biomedical Engineering, University of Pavia

Chair of Oral Examination

• Dr. Oliver Schmidtke, Department of History, UVic

Abstract

Background: With the rapid increase in the aging population and longer life expectancy, it is important to consider how to support older adults to live at home healthily and independently as long as they can. Home support provides recipients with assistance with activities of daily living at home and plays an important role as part of the integrated health care system, particularly for older adults. The increasing demand for home support has been challenging

The objective of this research was to predict the transition to a higher level of care for home support recipients. Identifying the home support recipients who are at risk of transitioning to a higher level of care can lead to more effective resource planning and preventative care to delay the deterioration of their condition.

The study aimed to predict which older home support recipients will transition to a higher level of care and determine the factors that have a significant effect on higher care transition. The overarching research questions were:

• What key factors are associated with a higher-level care transition within a certain period after beginning to receive home support?
• Can data science models reliably predict such transitions based on routinely collected administrative data?

In particular, the analyses were conducted to investigate the following:

1. Identify the risk of transitioning to a higher level of care for hospitalized home support recipients at the hospital discharge;
2. Identify the influential factors that were associated with higher care level transitions for home support recipients; and
3. Identify the probabilities and risks of transitioning to a higher level of care from home support

Methods: This study was based on 26,333 home support referrals of 19,517 home support recipients, including 8,609 home support referrals for recipients who were 65 years old or older at a regional health authority in British Columbia, Canada between November 1, 2019 and March 31, 2023. The datasets were prepared by linking data from multiple services within the health care system, including home and community care, acute care, and long-term care. Several analytic methods were used to develop models, including survival analysis, binomial and multi-level machine learning classifiers based on random forest and Extreme Gradient Boosting, and a multi-state Markov model. Cross validation, and sensitivity analysis were performed to ensure the results were reliable.

Results:

1) Identify the risk of transitioning to a higher level of care for hospitalized home support recipients at the hospital discharge.

Two types of survival analysis were used in this section. The Kaplan-Meier model indicated that fifty percent of the home support recipients moved on to a higher level of care within two years of their home support referral starting. In addition, the Cox-proportional hazard showed that age at the start of the home support referral, emergency department visits, and average duration of each home support visit, were significant factors to determine the risk of transitioning to a higher level of care. In particular, the risk increased by 2% for every additional year of the cohort’s age at the start of the referral. Conversely, the risk of transitioning to a higher level of care decreased by 0.97% if the cohort had had at least one emergency department visit in the previous 12 months compared with those with no emergency department visits, and decreased by 8.68% if the cohort had averaged over one hour for each home support visit, compared with those whose average home support duration per visit was less than one hour.

2) Identify the influential factors that were associated with higher care level transitions for home support recipients.

The factors that affect home support recipient transitions to a higher level of care within one or two years from the start of a home support referral were identified based on binary and multi-level machine learning classification models called Random Forest and Extreme Gradient Boosting. The binary classification models were developed based on the referrals which remained in home support and those which transitioned to a higher level of care (hospital, residential care/assisted living or end of life care). The multi-level classification model included five groups: “no service required”, remained in home support, hospitalization, moved to residential care/assisted living, and received end of life care. For the high-level care transition vs. home support (binary model) in year one, the top three features were urgent hospital admission, home support prescribed service – medication assistance, and home support prescribed service – other. Hospital admission, home support prescribed – personal hygiene, and professional home care - case management were the top three for year two.

For the multi-level transition model, Extreme Gradient Boosting models performed better than Random Forest model across the years. For the multi-level transition Year 1 Model, the top three features of each referral destination were as follows. For no service required, they were urgent hospitalization, HS duration per day, and urgent hospital admission prior to the referral. For home support, they were prescribed home care – case management, emergency department visits during the referral and home support prescribed - respite. For hospitalization, urgent hospitalization, professional home care -case management, and elective hospital admission were the top three factors. The top three factors for residential care/assisted living were urgent hospitalization, home support prescribed - medication assistance and professional home care-case management. For end of life, professional home care - nursing, home support prescribed - personal hygiene, and professional home care-case management was the top three factors.

For the multi-level transition Year 2 Model, the top three actors for NSR were: urgent hospitalization, HC case management, and professional home care - assisted living. For home support they were home support prescribed– personal hygiene, professional home care case management and home support perceived service - respite. For hospitalization they were urgent hospitalization, professional home care - case management, and professional home care-nursing. For residential care/assisted living they were home support duration per visit, home support prescribed–medication, and professional home care - case management. For end of life they were professional home care nursing, professional home care - case management and home support prescribed–other.

3) Identify the probabilities and risks of transitioning to a higher level of care from home support.

Based on the important factors that were identified by the multi-level machine learning models, a multi-state Markov model was developed to determine the probability and the harzard ration of transitioning from home support to a higher level of care. The multi-state Markov model included five discharge dispositions: “no service required”, home support, hospital, residential care/assisted living and end of life care. The probability of staying in home support declined over time: at the end of the first year, 46%, and by the end of the second year, 21% of the home support referrals were still in home support. The hazard ratio indicated that at least one professional home care - case management visit in six months was associated with a lower risk of the referrals ending in hospitalization, and end of life, but increased the risk of ending in residential care/assisted living. At least one professional home care - nursing visit was associated with a lower chance of transitioning to residential care/assisted living and “no service required” by about 0.8 times, but increased the risk of end-of-life care by 2.06 times. Urgent hospitalization during the six-month period was associated with a higher risk of the referrals ending in hospital, RC/AC and end of life care. Compared with the reference group, more than ti0 minutes of each home support service visit was associated with a 28–44% lower risk of leaving home support. Having home support prespecified cleaning service twice a day was associated with a lower risk of moving to residential care/assisted living and end of life care by 0.67 and 0.53 times compared with the baseline (no service). Conversely, having home support prescribed service medication assistance and continence assistance increased the risk of transitioning out from home support.

Discussion and Conclusion: The study took a rigorous, innovative approach using data science and advanced analytics to develop predictive models - based on survival analysis, machine learning and MSM - for home support recipients’ higher-level care transitions and identified factors that influence the risk of the transitions.

In general, the risk of transitioning to a higher level of care increased as time passed. The study also revealed the importance of a regular visit from the professional home care-case management. Professional home care - case management was associated with a reduced risk of transitioning out from home support or hospitalization and with an increased chance of the home support recipients staying home longer with the positive effect of the case manager visits increasing over time. However, the study also discovered an association between the increasing risk of transitioning to residential care/assisted living with more frequent professional home care – case management visits and home support visit duration.

Furthermore, increased frequency of any type of home support prescribed services was associated with an increased risk for home support recipients to move on to a higher level of care. In particular, increased frequency of medication assistance was linked to an increased risk for home support recipients to move to residential care/assisted living.

Home support duration per visit was between 30 minutes and 2 hours was associated with a reduction in the risk of transitioning to hospital, residential care/assisted living and end of life care, and an increase in the chances of staying in home support or moving to “no service required”. However, the home support duration per day did not show a significant hazard ration, except for residential care/assisted living when the daily duration is between 1 and 2 hours. This indicates that even if a home support recipient receives a total home support of over one hour per day via multiple visits, it does not help the recipients to stay in home support longer if each visit is a short visit.

This study shows how routine administrative data can be turned into computable, probability-based insights on transitions from Home Support to a higher level of care. It also generated actionable signals for planning and case management. using both a multistate model and a machine-learning model. With the movement of Canadian health care data connection and standardization and the recent shift of home-based health care delivery, the study can be positioned as a foundation in digital workflows to support risk-informed, coordinated care at home. It can be a part of a care management plan and readily re-trained as service models.

One of the limitations of this study is that because of the limited availability and quality of data, the study could address only up to two years since the home support referral started. Since the Kaplan-Meier analysis showed that 50% of home support referrals were expected to be open for about four years, further studies using a longer timeline are needed.

Through this study, it was evident that the research in the area of home support needs to be continued. This study identified a lack of standardization of terms, methods and measures of home support related research and standardization in this area internationally. Standardization is needed to enhance the monitoring of aging populations’ health and wellness (Fukui Innes et al., 2024). In addition, further research specifically on home support areas are needed to investigate the characteristics of home support recipients and factors that are associated with their physical and mental conditions. In order to understand that, clinical information such as Resident Assessment Instrument and medical diagnostics, need to be included as a part of the analysis. Although, as discussed, there were some limitations due to dataset availability and quality, the model outputs provided very important information. This study contributes to the area of home support research by identifying predictors of higher-level care transitions and estimating transition probabilities, information that can be used to guide case management, resource allocation, and policy planning.