Integrating NLP and Clinical Data for AI-Driven Chest Pain Triage in the Emergency Department
Bibliographic record
Abstract
<h3>Context</h3> Emergency Department (ED) crowding is a growing challenge that undermines patient outcomes and strains healthcare systems. The ED functions as a critical entry point for urgent, unscheduled care. In the United States and Canada, ED visits have increased by 13% in the past six years, are projected to rise by 30% in the next 25 years. <1% of patients present with life-threatening conditions requiring immediate intervention, 10–20% have serious conditions but may wait hours before treatment begins. <h3>Objective</h3> Develop and evaluate the feasibility of an AI-based triage system, operating alongside nursing staff, to streamline care for chest pain patients. By standardizing triage and initiating diagnostics early, the system aims to enable faster disposition decisions during the initial assessment. <h3>Study Design and Analysis</h3> We built NLP-driven models to predict patient risk categories during initial triage. Inputs included vital signs, patient self-reported narratives, and pre-existing EHR data. Additional EHR-derived features diagnostic results, observations, treatments, ED diagnoses, disposition, and time stamps were retrospectively annotated and integrated into training. Models were evaluated using structured data alone and in combination with unstructured text. <h3>Setting or Dataset</h3> Data consisted of 2,424 EHR cases with structured fields, free-text notes, and patient questionnaires. Multiple experimental configurations were tested for optimal performance. <h3>Population Studied</h3> The dataset was stratified by age and sex to ensure balanced and unbiased representation of all risk groups. <h3>Intervention</h3> A machine learning model predicted chest pain diagnostic categories—unstable angina, potentially cardiac, low-risk, and non-cardiac—to improve triage speed and accuracy. <h3>Outcome Measures</h3> Performance was assessed via precision, sensitivity, F1-score, and accuracy, benchmarked against human expert annotations. Feature importance analysis identified high-value variables for interpretability. <h3>Results</h3> The model achieved weighted precision, recall, and F1-scores of 0.96, specificity of 100%, and overall accuracy of 0.96. Macro-average F1 was 0.94, with balanced performance across all categories. <h3>Conclusion</h3> Combining clinical and linguistic features improved model generalizability and interpretability, enabling accurate prediction of cardiac risk levels. This approach demonstrates potential to expedite chest pain triage and enhance data-driven decision-making in the ED.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.006 | 0.003 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".