Machine learning-driven health profiling and multidimensional trajectory analysis in first-ever ischaemic stroke:...

Ischaemic stroke, the most prevalent stroke subtype, imposes a significant long-term disease burden. However, patients with first-ever stroke exhibit substantial individual variability in poststroke health trajectories, manifesting heterogeneous clinical presentations. We therefore started with the overall health of patients in order to delineate heterogeneous clusters characterised by distinct demographic profiles, clinical features and behavioural determinants and elucidate shared longitudinal trajectories in the temporal development of adverse health outcomes.

We designed a multicentre, cross-sectional and longitudinal study focusing on patients with first-ever ischaemic stroke. We will employ patient self-reported outcomes and objective measurements to comprehensively evaluate patients' health status from a multidimensional perspective. Following baseline assessments, participants will undergo follow-up evaluations at 1 month, 3 months and 6 months post inclusion. The primary objective is twofold: (1) to identify distinct patient clusters with heterogeneous multidimensional health profiles using the k-prototype clustering algorithm and (2) to characterise synergistic trajectories of core health attributes within the largest cluster through parallel process latent class growth modelling. By combining cross-sectional and longitudinal analyses, this phased study should elucidate static heterogeneity and dynamic recovery patterns following a first-ever ischaemic stroke.

The project conforms to the ethical principles enshrined in the Declaration of Helsinki (2013 amendment) and all local ethical guidelines. The ethics committee at the University of South China approved the study (approval no. 2024 NHHL023). The ethics committee of Gansu Provincial Hospital approved the study (approval no. 2025-023). The ethics committee of the Central Hospital of Shaoyang approved the study (approval no.KY-2025-12). The findings will be published and presented at conferences for widespread dissemination.

ChiCTR2500098442.