Research & Innovation
Publications
A data-driven approach to define the stages and patterns of neurological recovery after traumatic spinal cord injury
PMU Authors
Wanda Lauth, Iris Leister
All Authors
Wanda Lauth, Abel Torres-Espin, Angel Arevalo-Martin, Ganiel Garcia-Ovejero, Orpheus Mach, Doris Maier, Rainer Abel, Norbert Weidner, Yorck-Bernhard B Kalke, Jiri Kriz, Armin Curt, , Catherine Jutzeler, Iris Leister
Journal association
Journal of Neurology
Abstract
Background Recovery after spinal cord injury (SCI) follows a complex and variable trajectory, yet the field lacks clear,
data-driven definitions of the temporal stages of recovery. This study aimed to model the trajectory of recovery after SCI
and define distinct post-injury phases based on real-world clinical data.
Methods We analyzed longitudinal data from 4407 individuals with traumatic SCI enrolled in the European Multicenter
Study about Spinal Cord Injury (EMSCI). Neurological improvement was quantified using the Integrated Neurological
Change Score (INCS). Functional recovery was assessed using the Spinal Cord Independence Measure (SCIM) and a derived
measure of hand motor function based on motor scores of the C6, C8, and T1 myotomes. Longitudinal recovery trajectories
were modeled using random forest regression, locally estimated scatterplot smoothing (LOESS), and neural networks. Lastly,
latent class mixed models (LCMM) were employed to identify subgroups with distinct recovery profiles.
Results Across all models, neurological recovery followed a four-stage sigmoidal pattern: an acute phase (up to ∼2 weeks
post-injury) with minimal change; a transition phase (∼2 to 9–11 weeks) marked by substantial neurological change; a pre
chronic phase (∼10 weeks to 7–8 months) characterized by ongoing but non-linear recovery; and a chronic phase (beyond
∼7–8 months) in which recovery plateaued. Neurological improvement consistently preceded functional gains captured by
SCIM and hand motor strength. Latent class analysis identified four distinct recovery profiles: class 1 (marked recovery),
classes 2 and 3 (moderate recovery), and class 4 (minimal to no recovery). Notably, despite significant variability in the
magnitude of lower extremity recovery across classes, the timing of the transition phase was remarkably consistent, aligning
with the population-level window of approximately 10–11 weeks post-injury.
Conclusions The majority of neurological and functional improvement occurs within the first ∼10 weeks post-injury. These
findings align with the concept of a period of heightened neuroplasticity and provide a data-driven framework for timing
assessments and interventions in SCI recovery
Keywords
deep learning, Spinal Cord Injuries, MODELING ANALYSIS