Photo Credit: Silver
The following is a summary of “Artificial Intelligence-Based Early Prediction of Acute Respiratory Failure in the Emergency Department Using Biosignal and Clinical Data,” published in the February 2025 issue of Pulmonology by Han et al.
The early identification of patients at risk for acute respiratory failure (ARF) was recognized as critical for developing preventive strategies, and the potential of analyzing biosignals with artificial intelligence (AI) to reveal hidden information and variability within time series was explored.
Researchers conducted a retrospective study to investigate the use of AI in analyzing biosignals for early identification of patients at risk for ARF.
They developed an AI model using convolutional recurrent neural networks for biosignal feature extraction and sequence modeling. The model was internally validated with data from 5,284 admissions [1,085 (20.5%) positive for ARF] and externally validated with data from 144 admissions [7 (4.9%) positive for ARF] from a different institution. The ARF was defined as the use of advanced respiratory support devices.
The results showed that the AI model predicted ARF with an area under the receiver operating characteristic curve (AUROC) of 0.840 in internal validation and 0.743 in external validation. It outperformed the Modified Early Warning Score (MEWS) and XGBoost models using only clinical variables. The AI model also demonstrated a high predictive ability for mortality with an AUROC of 0.809. A 10% increase in AI prediction scores corresponded to a 1.44-fold increase in ARF risk and a 1.42-fold increase in mortality risk, even after adjusting for MEWS and demographic factors.
Investigators concluded that the AI model demonstrated high predictive accuracy, associations with clinical outcomes, and the potential to promptly aid in triage decisions, advancing disease detection and prediction through biosignal analysis.