Development and External Validation of a Machine Learning-Enhanced Nomogram to Stratify Risk of Early-Onset Cognitive Impairment Following Pediatric Posterior Fossa Surgery: A Multicenter Retrospective Study.
AI interpretation is pending for this paper.
Open original publication →What the AI sees
Not AI summarized yet.
Research significance
Pending deeper interpretation.
Source abstract
INTRODUCTION: Posterior fossa tumor (PFT) survivors are prone to develop cerebellar cognitive affective syndrome (CCAS) as a major neurocognitive late effect, profoundly impacting quality of life, psychosocial adaptation, and long-term survivorship. Reliable early predictive tools for this survivorship outcome remain lacking. OBJECTIVE: This study aimed to construct a predictive model to enable early risk stratification for medium-term cognitive dysfunction-a critical survivorship outcome-in children after PFT surgery, thereby supporting timely survivorship care planning and personalized psychosocial intervention. METHODS: A multicenter retrospective study was conducted with two cohorts: a derivation cohort in 3 institutions from 2015 to 2020 and an external validation cohort in an independent institution from 2021 to 2023. Preoperative multimodal data such as fMRI, DTI, tumor characteristics, and factors were collected. Medium-term cognitive dysfunction (primary outcome) was diagnosed via standardized neuropsychological scales at 12 months postoperatively. Random forest (RF) and LASSO regression were used for model construction. Performance was evaluated by AUC, accuracy, sensitivity, specificity, calibration curves (Hosmer-Lemeshow test), and clinical decision curve analysis (DCA). A nomogram was established for clinical application. RESULTS: In the derivation cohort, 172 (28.7%) children developed cognitive dysfunction. LASSO regression identified 8 key predictors: preoperative SCP FA value, tumor volume ≥ 15 cm3, midline tumor location, suboccipital midline approach, radiotherapy, preoperative fMRI prefrontal abnormal activation, age < 6 years, and medulloblastoma histology. The RF model outperformed LASSO (derivation AUC = 0.887, 95% CI:0.87-0.93 vs. LASSO AUC = 0.776, 95% CI:0.75-0.80) and maintained stable performance in external validation (AUC = 0.840). Calibration curves showed good consistency between predicted and actual probabilities in both derivation (χ2 = 8.32, P = 0.401) and external validation (χ2 = 7.89, P = 0.32) cohorts. CONCLUSION: The RF-based model and its derived nomogram effectively predict medium-term cognitive dysfunction in pediatric PFT survivors, with robust performance in external validation. This tool supports early survivorship risk stratification and personalized psychosocial intervention for pediatric PFT survivors, advancing precision care in psycho-oncology.