Abstract

Background: The demographics of Canada's population have gradually become more aged, and there has been anecdotal evidence indicating increased care demands of individuals who are newly admitted to nursing homes in Ontario, Canada, over time. Longitudinal observations are critical to the surveillance and forecasting of nursing home care demands. However, there has been no formal investigation of these trends in Canada, partly due to incomplete resident-level data from Ontario's nursing homes before 2010.

Objectives: In this study, an algorithm was developed to identify incident admissions to nursing homes in Ontario using health administrative databases. The best-performing algorithm was then used to create incident-entry cohorts between 2000 and 2015 to examine their changing characteristics (e.g., age, functional capacity, cognition, the presence of chronic health conditions).

Methods: A cohort of incident nursing homes admissions in 2012 was first obtained from the Residence Assessment Instrument - Minimum Data Set (RAI-MDS), which was used as the reference standard. Twenty-four algorithms based on different combinations of physician billing and prescription drug dispensation claims were considered and evaluated against the RAI-MDS based on their sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) in determining nursing home entry. The best performing algorithm was then used to identify incident admissions among individuals aged 65 years and over between 2000 and 2015.

Results: The best performing algorithm used two physician claims, two prescription drug claims, or one physician claim and one prescription drug claim within 30 days of one another to identify nursing home entry. The sensitivity, specificity, PPV and NPV were 93.3%, 99.9%, 96.2% and 99.9%, respectively. Temporally, newly admitted residents were found to be increasingly dependent; for example, the proportion of those aged 85 years and older increased from 45.1% to 53.8%. The proportion of incoming residents with dementia also increased, from 42.3% to 54.1% over the same period. Overall, the proportion of individuals with multimorbidity (defined as more than one chronic health condition) grew from 91.6% to 96.9%. There were some differences in the functional status of individuals over time.

Conclusions: The validated algorithm demonstrated good performance and could be used to identify incident admissions to nursing homes in Ontario and, potentially, other jurisdictions that have similar prescription claims and physician billing data. Furthermore, the admission trends show that the new residents are older and have more chronic health conditions. These data are crucial for estimating the impact of population ageing on long-term care utilization.