Age-based prediction of non-routine discharge after anterior cervical discectomy and fusion using machine learning

Purpose: To examine factors influencing non-routine discharge in ACDF patients stratified by age utilizing machine learning. Methods: A cohort of 219,380 weighted ACDF cases from the National Inpatient Sample (NIS) database spanning 2016–2020 was divided into three age groups: 50–64, 65–79, and 80 + years. Eight supervised machine learning models predicted non-routine discharge based on patient characteristics, including age, length of stay (LOS), and comorbidities. Chi-square and t-tests compared outcomes. After Bonferroni correction, significance was set at P < 0.004. Results: Across all age groups, several patient-specific factors were associated with non-routine discharge. In the 50–64 group, deficiency anemias (1.1% vs. 0.6%, P < 0.001), paralysis (1.2% vs. 0.1%, P < 0.001), and race (Black: 15.4% vs. 10.0%, P < 0.001) were significant predictors. For 65–79, heart failure (1.2% vs. 0.5%, P < 0.001) and dementia (0.5% vs. 0.1%, P < 0.001) increased risk. In the 80 + group, racial disparities persisted. Machine learning models—especially AdaBoost and Gradient Boosting—demonstrated strong predictive performance, with AUCs exceeding 80% for the 65–79 and 80 + cohorts. Prolonged LOS was also significantly associated with non-routine discharge across all age groups, with patients staying over twice as long on average (all P < 0.001). Conclusion: Non-routine discharge after ACDF is influenced by patient-specific factors. Strategies targeting older patients with complex comorbidities could help reduce adverse outcomes.