A Natural Language Processing Model Based on Anamnesis Reports to Predict Acute Myocardial Infarction Risk in...

This work is focused on predicting the risk of acute myocardial infarction (AMI) by analysing textual data from anamnesis reports (patient's personal medical history) to improve risk stratification in emergency departments. A natural language processing (NLP) model was fine-tuned using pre-trained algorithms and the European Society of Cardiology (ESC) 0-1h algorithm was employed to evaluate two troponin measurements, determining the endpoint as either rule-out or rule-in for AMI. The analysed dataset comprised patients with chest pain treated at the emergency department of the Hospital Clínico Universitario de Valencia (Spain).The results demonstrated that the model achieved an impressive accuracy of 0.975, with a sensitivity of 1.0 and a specificity of 0.95 when incorporating the first troponin measurement alongside the text from anamnesis reports. Models relying solely on textual data without troponin information showed limited clinical applicability.Clinical Relevance- This work underscores the significant potential of natural language processing (NLP) as a diagnostic support tool, particularly considering the substantial volume of patient information stored in textual format. The proposed model facilitates more precise identification of high-risk patients, enabling timely clinical interventions and optimizing resource allocation in emergency departments. Furthermore, the integration of a critical biomarker, such as the initial troponin measurement, enhances the reliability of the model, ensuring greater robustness and applicability for its implementation in routine clinical practice.