Multidimensional preoperative cognitive phenotypes predict postoperative cognition in glioblastoma

Purpose

Glioblastoma patients suffer from cognitive deficits across domains. We used principal component analysis and gaussian mixture modelling to identify preoperative cognitive phenotypes which are at risk of postoperative cognitive impairment.

Methods

44 patients underwent neuropsychological screening before and after surgery (within 1 week and 4–6 weeks post-surgery). Patient scores were standardised to a normative reference population and compared using one-sided z-scores. Reliable change indices were calculated with correction for hospitalization effects using a cohort of biopsy patients. Gaussian mixture models were fit after principal component analysis of standardised preoperative scores. Linear mixed effects models were performed to assess predictors of postoperative principal component scores using demographic, clinical and imaging variables.

Results

Three principal components accounted for 55% of the variance in cognitive scores: The first component loaded on Language, Executive Function and Perception (25%). The second, on Prospective and Retrospective Memory (15%). The third component on Recognition Memory and Working Memory (15%). The optimal model identified four preoperative cognitive phenotypes: Minimal (n = 23), Mild (n = 12), Moderate (n = 5) and Severe (n = 4) cognition. Post-surgery, preoperative cognition limited recovery and few patients returned to their preoperative baseline. Moderate and Severe phenotypes were significantly associated with worse postoperative Language, Executive Function as well as Prospective and Retrospective Memory.

Conclusion

We identified four preoperative cognitive phenotypes which stratify patients into those at risk of postoperative cognitive deficit. Better preoperative cognition was associated with improved postoperative cognition.