Topographical data analysis to identify high-density clusters in stroke patients undergoing post-acute rehabilitation

During acute stroke rehabilitation, the recovery of motor and cognitive function is highly variable: while some patients regain function, others do not.

Our objective was to identify data-driven subgroups of stroke patients undergoing acute rehabilitation using topological data analysis (TDA), compare TDA with K-means clustering, and to assess inter-group demographic and clinical differences among the subgroups.

This is a secondary data analysis of clinical, functional outcomes, and demographic data collected from 339 stroke patients undergoing acute rehabilitation post-stroke. We identified stroke recovery sub-groups using TDA on the point cloud, persistent homology, and finally, density clustering. We assessed inter-group differences in demographic and clinical characteristics using one-way ANOVA, Kruskal-Wallis, or χ² tests.

TDA revealed three high-density clusters among 137 subjects in the point cloud- poor-recoverers (G1(n = 34)), intermediate-recoverers (G2 (n = 88)) and good-recoverers (G3(n = 15)).Significant differences across clusters were observed for amantadine use (p = .009), number of stroke risk factors (p = .047), creatinine (p = .015), length of stay (p < .001), discharge destination (p < .001), FIM motor, FIM cognition, and FIM total on admission and discharge (all p < .001), and motor, cognition, and total MRFS scores (all p < .001).

This study revealed that in addition to functional status on admission, stroke risk factors are associated with recovery outcomes. Future studies using TDA to analyze omic data, including clinical, biological, and sociodemographic factors, will accelerate the development of personalized treatment plans in post-acute stroke rehabilitation patients.