Personalized physiotherapy effect modeling with continuous causal machine learning.
4 affiliations
- 1Vrije Universiteit Amsterdam, Department of Computer Science, Amsterdam, Netherlands.
- 2Amsterdam UMC Location Vrije Universiteit Amsterdam, Department of General Practice, Amsterdam, Netherlands.
- 3Amsterdam Public Health, Aging & Later Life, Amsterdam, Netherlands.
- 4School of Physiotherapy, Faculty of Health, Dalhousie University, Halifax, NS, Canada.
Abstract
Introduction
Estimating how physiotherapy (PT) intensity influences functional independence is central to personalized care planning for older adults. However, most existing causal inference approaches treat PT as a binary intervention, which requires collapsing a naturally continuous measure of treatment intensity and limits the ability to characterize nuanced dose-response relationships.Methods
To address this limitation, we use continuous causal inference to estimate individualized continuous dose-response curves for PT. This approach enables granular assessment of treatment effects across therapy intensities and supports the identification of optimal treatment doses. The model is trained on Canadian Community Care data and externally validated using an independent Canadian Home Care cohort.Results
Across factual, proxy counterfactual, and external validation settings, our approach generally achieved lower prediction error on activities of daily living outcomes, as measured by root mean squared error (RMSE), than standard baselines. In particular, we attained factual and proxy counterfactual RMSEs of 3.79 and 3.89, respectively, compared with 4.09 and 6.49 for linear regression, while also demonstrating strong external generalization.Conclusion
Continuous causal modeling of individualized dose-response relationships provides a principled approach to estimating physiotherapy intensity effects on functional independence, with direct implications for personalized rehabilitation planning in older adults.
Related
- Estimating Optimal Dynamic Treatment Regimes for Personalized Education: A Tutorial and Applications with Machine Learning
- Prevalence, associated factors, and machine learning classification of low back pain among postmenopausal women attending rehabilitation and physiotherapy facilities in Bangladesh
- A novel machine learning-based framework for integrating scheduling and control in continuous multiproduct chemical process operations
- A Legendre-Based Modified Extreme Learning Machine for Convection-Diffusion Equations: Steady-State, Continuous-Time, and Discrete-Time Formulations
- Measurement Error and Nonlinear Confounding: Structural Equation Modeling Versus Double Machine Learning
BibTeX
@article{remmerden2026personalized,
title = {{Personalized physiotherapy effect modeling with continuous causal machine learning.}},
author = {Louk van Remmerden and Eline Kooijmans and Shujian Yu and Karlijn Joling and Emiel Hoogendijk and Caitlin McArthur and Mark Hoogendoorn and Hein van Hout},
journal = {Frontiers in digital health},
year = {2026},
month = oct,
doi = {10.3389/fdgth.2026.1793661},
url = {https://europepmc.org/article/MED/43011093},
note = {CC BY},
}
RIS
TY - JOUR
TI - Personalized physiotherapy effect modeling with continuous causal machine learning.
AU - Louk van Remmerden
AU - Eline Kooijmans
AU - Shujian Yu
AU - Karlijn Joling
AU - Emiel Hoogendijk
AU - Caitlin McArthur
AU - Mark Hoogendoorn
AU - Hein van Hout
PY - 2026
DA - 2026/10/27
JO - Frontiers in digital health
DO - 10.3389/fdgth.2026.1793661
UR - https://europepmc.org/article/MED/43011093
AB - Introduction Estimating how physiotherapy (PT) intensity influences functional independence is central to personalized care planning for older adults. However, most existing causal inference approaches treat PT as a binary intervention, which requires collapsing a naturally continuous measure of treatment intensity and limits the ability to characterize nuanced dose-response relationships.Methods To address this limitation, we use continuous causal inference to estimate individualized continuous dose-response curves for PT. This approach enables granular assessment of treatment effects across therapy intensities and supports the identification of optimal treatment doses. The model is trained on Canadian Community Care data and externally validated using an independent Canadian Home Care cohort.Results Across factual, proxy counterfactual, and external validation settings, our approach generally achieved lower prediction error on activities of daily living outcomes, as measured by root mean squared error (RMSE), than standard baselines. In particular, we attained factual and proxy counterfactual RMSEs of 3.79 and 3.89, respectively, compared with 4.09 and 6.49 for linear regression, while also demonstrating strong external generalization.Conclusion Continuous causal modeling of individualized dose-response relationships provides a principled approach to estimating physiotherapy intensity effects on functional independence, with direct implications for personalized rehabilitation planning in older adults.
SN - 2673-253X
ER -