Machine Learning for Enhanced Prognostication: Predicting 30-Day Outcomes Following Posterior Fossa Decompression Surgery in a Pediatric Cohort with Chiari Malformations

Introduction: Congenital Chiari Malformation (CM) affects 0.1% of the population. Pediatric CM surgery with posterior fossa decompression (PFD) has substantial short-term unfavorable outcomes. Given the financial expenses and patient well-being, prediction methods that can quantify the likelihood of such adverse outcomes are essential. Our study uses machine learning (ML) algorithms to predict 30-day readmissions and reoperations after posterior fossa decompression (PFD) in juvenile CM patients.

Methods: The American College of Surgeons National Surgical Quality Improvement Program-Pediatrics (ACS NSQIP-P) Database was used for this retrospective investigation. Patients undergoing PFD (CPT code: 61343) between 2012 and 2021 were eligible. Patients undergoing tumor or vascular surgery were excluded. The major research outcomes were 30-day unplanned readmission and reoperation. The length of hospital stay, 30-day problems, discharge disposition, and 30-day mortality are also examined. We used multivariable logistic regression. We also used k-nearest neighbors, decision tree, Gaussian naive Bayes, and Random Forest. The models were trained and tested using 80:20 data. Synthetic Minority Over-Sampling Technique addressed class disparity. 10-fold cross-validation internally validated model performance. F1 score, calibration plots, and AUC-ROC were used to evaluate model performance. 2 sided p < 0.05 was used to determine statistical significance.

Results: The sample comprises 7,106 pediatric PFD patients. The median age was 9.2 (IQR: 4.7-14.2). Patients were predominantly women (56%). The 30-day readmission and reoperation rates were 7.5% and 3.4%. CNS (32%) and wound (30%) issues were the most common reasons for 30-day readmission, whereas wound modifications and fluid or blood evacuation (61%), followed by CSF diversion (28%), were the most common reasons for reoperation. Reoperations and 30-day readmissions were best predicted by random forest classifiers (AUC=0.960 and 0.990).

Conclusion : Using a large-scale nationwide data set, ML models for the prediction of both 30-day readmissions and reoperations were developed and achieved high accuracy. This highlights the utility of ML in risk stratification and surgical decision-making for pediatric CMs.