Enhancing preoperative risk stratification: an interpretable machine learning model for ICU admission in spinal...

Area of Science:

• Neurosurgery
• Oncology
• Data Science

Background:

• Spinal metastases present significant clinical challenges, often necessitating complex surgical interventions.
• Postoperative intensive care unit (ICU) admissions are a major concern following metastatic spinal tumor surgery.
• Predicting unplanned ICU admissions is crucial for optimizing patient care and resource allocation.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting 30-day unplanned ICU admission after metastatic spinal tumor surgery.
• To identify key clinical features associated with increased risk of ICU admission.
• To compare the performance of various ML algorithms for this predictive task.

Main Methods:

• A multicenter study involving 642 patients with metastatic spinal disease.
• Data from 525 patients were used for model derivation and internal validation, with an independent cohort of 117 patients for external validation.
• Six ML algorithms were trained using 11 selected clinical features, with performance evaluated using metrics like AUC, accuracy, and Brier score.

Main Results:

• The KNN algorithm demonstrated superior predictive performance in the derivation cohort (AUC: 0.884) and external validation (AUC: 0.834).
• Patients admitted to the ICU exhibited higher comorbidity burdens, elevated inflammatory markers, and impaired renal function.
• The KNN model showed significantly better discriminative ability than the ANN model in external validation (P<0.001).

Conclusions:

• Validated ML models, particularly the KNN model, can effectively predict ICU admissions following spinal metastasis surgery.
• The KNN model shows strong potential for clinical implementation to aid in risk stratification and patient management.
• Accurate prediction of ICU admission can improve surgical outcomes and healthcare resource utilization.