Abstract 4142676: Advancing towards the development of an AI application for angina diagnosis through medical interviews

Background: Interview techniques for angina pectoris in patients with chest pain have demonstrated a low positive probability of 35.7% (PROMISE study) when using the Diamond-Forrester approach, which evaluates pre-test probability (PTP) based on classic chest symptoms, age, and gender. Therefore, it is beneficial to use artificial intelligence (AI) to create a more precise medical interview system by combining patient clinical data. Our objective is to create such a system. Methods: A medical interview is taken from patients presenting with chest pain as a chief complaint. Then, coronary artery examinations (CAG, contrast-enhanced CT, stress tests) will be conducted to definitively diagnose angina. The results of these definitive diagnoses will be combined with the patients' baseline data to develop an AI algorithm. The input attributes used for the interview are listed in Table 1. A multilayer perceptron (MLP) was applied to predict patients diagnosed with angina pectoris, and these models were validated using 10-fold cross-validation. Results: There were a total of 315 patients, of whom 135 were diagnosed with angina pectoris. Additionally, patient information including atherosclerosis risk factors such as diabetes (102 cases), dyslipidemia (216 cases), hypertension (225 cases), and smoking status (never: 141 cases, past: 124 cases, current: 23 cases) were added as attributes and analyzed. Using only medical interviews and setting the threshold at 80% (where the AI determined the probability of angina to be 80% or higher), the prediction accuracy was 62.6%, with a sensitivity of 30.8% and a specificity of 87.7%. The precision was 66.6%. Additionally, when the four atherosclerosis risk factors were included, the prediction accuracy increased to 69.9%. Notably, the precision reached 79.9%. (Figure1) Conclusion: Our AI system based on interview data demonstrated high accuracy for diagnosis of angina while it was improved by adding the attribute of the four atherosclerosis risk factors. Further investigation is needed to complete a highly accurate AI-based application by increasing the number of samples.