This study implemented Subtype and Stage Inference (SuStaIn) on standard preoperative neuropsychological assessments of patients considered for temporal lobe resection. The analytic goal was to model cognitive function as a continuous, network-level process that captures both pathological burden and compensatory reserve, rather than relying on conventional static, threshold-based neuropsychological classifications.
Using this progression-based computational framework, the investigators identified three distinct latent cognitive trajectories in the preoperative data: Verbal, Naming, and Visual. Each trajectory represents a different pattern of cognitive impairment progression and is proposed to map onto differing underlying network pathology and reserve capacity.
The Verbal subtype mirrored the classic mesial temporal pathology commonly associated with temporal lobe epilepsy. In this group, postoperative decline aligned with the functional adequacy of the remaining hippocampal tissue: the degree of postoperative verbal memory change tracked with how much hippocampal function was preserved. This indicates that in the Verbal trajectory, traditional mesial temporal mechanisms remain central to predicting memory outcome after resection.
The Naming subtype represented a neocortical-predominant or 'Temporal Plus' phenotype. Patients assigned to this trajectory had lower rates of hippocampal sclerosis but nonetheless demonstrated severe postoperative vulnerability in verbal memory. The authors attribute this vulnerability to un-reorganized frontotemporal language networks prior to surgery, suggesting that neocortical involvement and dysfunctional language network reorganization can drive postoperative cognitive decline independent of classic hippocampal pathology.
The Visual trajectory was defined by progressive visuospatial decline. The study reports distinct sex-specific reserve profiles within this subtype and links it to poorer visual recall outcomes after temporal lobe resection. This suggests that visuospatial network involvement and differing reserve dynamics between sexes may be important determinants of visual memory outcome.
A key finding was that these progression-based trajectories outperformed conventional static neuropsychological classifications in predicting 12-month postoperative neuropsychological outcomes on unseen test data. The approach operates using routine preoperative evaluations without requiring additional testing. By modeling progression and stage, the method disentangles measures of pathological burden from indices of network reserve, yielding biomarkers that were both scalable and biologically interpretable according to the authors.
Because the framework can be applied directly to standard clinical neuropsychological data, it has potential utility for personalized risk counselling and for informing surgical planning that accounts for network-level pathology and reserve. The authors propose these subtype and stage biomarkers could guide more tailored discussions about cognitive risk and may support planning approaches that consider distributed neocortical involvement as well as mesial temporal integrity.
The study was conducted retrospectively on clinically acquired data and received approval from the Health Research Authority and London Bloomsbury Research Ethics Committee (REC reference: 20/LO/0149). The authors state that all patients had the opportunity to opt out of research and that relevant ethical and consent procedures were followed.
All data produced are available upon reasonable request to the authors. Code and inference scripts will be made available online at the project GitHub repository: https://github.com/lbinding/TLE_Neuropsych_SuStaIn. Funding sources declared include the Wellcome Trust and the NIHR. The authors declared no competing interests.
Note: This report summarizes findings from a medRxiv preprint. The manuscript is a preprint and has not been certified by peer review; the authors note it should not be used to guide clinical practice until formally reviewed and validated.