Validation of a novel Bayesian predictive algorithm for detection of carbon dioxide retention using retrospective neonatal ICU data

Objective: To validate a novel Bayesian prediction algorithm (IVCO2 index) to calculate the probability of CO₂ retention in neonates using existing medical device outputs. Study design: A retrospective validation study from two level IV NICUs between September 2021 and May 2023. The algorithm calculated probabilities of PaCO₂ exceeding 50 mmHg (IVCO2_50) and 60 mmHg (IVCO2_60) using multimodal physiologic data. Performance was assessed through ROC analysis, range utilization, and resolution/limitation analysis. Results: Among 180 included neonates, 1092 arterial blood gas measurements were analyzed. IVCO2_50 and IVCO2_60 demonstrated excellent discriminatory performance (AUC 0.87, 95% CI 0.85–0.89 and AUC 0.90, 95% CI 0.68–0.93, respectively). The risk of elevated PaCO₂ scaled linearly with increasing index quartiles. Minimum scores (<1) showed >6-fold reduction in hypercapnia risk, while maximum scores (>99) demonstrated >3-fold reduction in normocapnia risk. Conclusion: The IVCO2 index accurately predicts CO₂ retention in neonates, offering potential for early detection of ventilation inadequacy without additional invasive monitoring.