Beyond Resident Typologies: Machine Learning for Population Health States and Trajectories in Swiss Nursing Homes
Dr. Ania Mikos (UZH Healthy Longevity Center),
Dr. med. Florian Riese (University Hospital of Psychiatry Zurich, Department of Old Age Psychiatry and Psychotherapy),
Prof. Dr. Cöltekin Arzu (Institute of Interactive Technologies, Fachhochschule Nordwestschweiz)
Nursing home residents are among the most heterogeneous and clinically complex older populations, with diverse combinations of functional dependence, cognitive impairment, behavioral symptoms, multimorbidity, and care needs. Yet this complexity is commonly summarized using individual diagnoses, clinical scales, or resident typologies that assign people to discrete groups. While useful, these approaches may obscure an important feature of real-world populations: residents do not necessarily fall neatly into distinct types. How we represent this heterogeneity matters because different approaches may reveal or obscure important patterns in residents' health and how it changes over time.
Our preliminary work using nationwide Swiss nursing home assessment data has created a multidimensional “health landscape” in which residents are positioned according to similarities across functional, cognitive, and behavioral characteristics. The landscape reveals recognizable population profiles, but also substantial overlap between them. Rather than treating this “fuzziness” as a limitation of classification, we will examine it explicitly by considering not only which profile best represents each resident, but also how strongly residents correspond to different profiles. This allows us to retain meaningful differences between residents that may be lost when a complex population is divided into distinct groups.
With this PRC Seed Grant, we will investigate how multidimensional health heterogeneity among Swiss nursing home residents can best be represented and how it changes over time. The project addresses three related aspects of population health heterogeneity:
1. Population health structure: Determine how best to represent population heterogeneity by comparing discrete and graded representations of resident profiles within a continuous population health landscape and examining the extent of overlap between profiles.
2. Population health dynamics: Characterize transitions between population health states over time by examining how individual residents move within the multidimensional health landscape.
3. Future applications: Explore whether alternative representations of population health states provide meaningful information about residents’ care needs and outcomes, establishing a basis for future research on individual trajectories and population-level care requirements.
Together, these analyses will provide a framework for understanding heterogeneity at both the individual and population levels. Characterizing the health landscape and movement within it will show which patterns of health are common in the nursing home population, how clearly different profiles can be distinguished, where residents fall between them, and how these patterns change over time.
The seed project will thereby establish the methodological and empirical foundation for larger externally funded research. Future work can build on this framework to investigate whether differences in residents’ health states and trajectories are associated with subsequent outcomes and care needs, and how changes at the individual level translate into changes in the composition and needs of nursing home populations over time. Ultimately, this could support models of future long-term care needs and planning.