1426 - Trajectory Profile Clustering for Identifying Stroke Recovery Phenotypes in Longitudinal Data Sets

Objectives:
To demonstrate the utility of network science and machine learning tools (Trajectory Profile Clustering, “TPC”) to identify distinct stroke recovery phenotypes.

Design: We analyzed data from the NINDS tPA trial using the Trajectory Profile Clustering method to identify distinct stroke recovery patterns for 11 different neurological domains at 5 discrete time points.

Results: Our analysis identified 3 distinct stroke trajectory profiles that align with clinically relevant stroke syndromes, characterized both by distinct clusters of symptoms, as well as differing degrees of symptom severity.

Conclusions:
Stroke recovery phenotypes are characterized by impairments in multiple neurological domains including cognition, language, sensory, and motor functions. The fact that clinical recovery in these domains is tracked using a wide range of measures that are continuous, ordinal, interval and categorical in nature presents challenges for multivariate regression approaches. TPC offers an alternative approach using novel computational methods based in network science and machine learning. Our analysis using the TPC method identified 3 distinct stroke trajectory profiles that align with clinically relevant stroke syndromes, characterized both by distinct clusters of symptoms, as well as differing degrees of symptom severity. We demonstrate that trajectory profile clustering is an effective method for identifying clinically relevant recovery subtypes in multidimensional longitudinal datasets, and for early prediction of symptom progression subtypes in individual patients.