Characteristics of ICU Admissions Following Pediatric Spinal Fusion: A Longitudinal Cohort Study

Introduction: Spinal fusion surgeries have increased in past decades resulting in variable PICU admissions. We aim to measure longitudinal changes in clinical characteristics and risks associated with PICU admission and length of stay (LOS). We hypothesize a predictive model will accurately identify patients with long PICU admissions.


Methods: We conducted analysis on data from the American College of Surgeons–National Surgical Quality Improvement Program Pediatric Spinal Fusion database from 2016-2022. We defined outcomes as PICU admission or PICU LOS >2 days. For each year, we measured associations with pre- and intra-operative clinical characteristics. We assessed multicollinearity and calculated odds ratios for characteristics associated with negative outcomes. We fit linear models to assess risk trends across years. We applied standard modeling techniques (e.g., temporal data splitting, normalization, multiple feature selection) to train, tune, and test a predictive model against a long PICU LOS.


Results: Of 39,663 included patients, 14,837 (37%) required PICU admission, including 5,639 (38%) with PICU LOS >2 days. From 2016-2022, clinical characteristics significantly associated with PICU outcome remained consistent (p < 0.001). Risks include long operative time (>270 minutes), neuromuscular or syndromic scoliosis, developmental delay, ventilator use, nutritional support, and structural pulmonary abnormalities, and an ostomy. Asthma and prior cardiac surgery were associated with admission but not increased stay. PICU admissions declined from 40% to 35%, but LOS increased. For all patients, prevalence of long operation time, long fusions, and syndromic scoliosis significantly increased. A trained XGBOOST model demonstrated reasonable recall (0.53), modest accuracy (0.38) and precision (0.28), but poor discrimination (AUC=0.64). Most important identified model features were low preoperative creatinine and presence of an ostomy.


Conclusions: We identified declining trends in PICU admission after pediatric spinal fusion with consistent predictors for both admission and long PICU LOS. An initial predictive model demonstrated poor to modest ability to prospectively stratify patients at risk for long PICU stays. Future work includes adding additional features and tuning models to optimize prediction.