{
  "abstract": "Background Timely recognition of ST-elevation myocardial infarction (STEMI) and rapid activation of the primary percutaneous coronary intervention (PPCI) pathway are critical to optimise outcomes. In routine practice, delays often arise from diagnostic uncertainty at the point of ECG interpretation, while a substantial proportion of PPCI referrals are ultimately non-STEMI or non-cardiac, leading to unnecessary catheterisation laboratory activation and resource use. Artificial intelligence (AI)-based electrocardiographic (ECG) interpretation tools may help standardise STEMI recognition, reduce inappropriate activations, and shorten treatment delays.Aim To evaluate, in a retrospective cohort from a high-volume tertiary PPCI centre, the diagnostic accuracy of the Queen of Hearts (QOH) AI ECG Application (Powerful Medical Cardio, Slovakia) for classifying STEMI/STEMI-equivalent versus non-STEMI ECGs among patients referred for PPCI, and to explore its potential role in supporting PPCI activation decisions.Methods We performed a single-centre retrospective cohort study of all consecutive PPCI referrals to the Essex Cardiothoracic Centre between 1 and 30 June 2020. To facilitate complete outcome ascertainment, analyses were restricted to patients with residential postcodes within the Mid and South Essex NHS Trust catchment. Patients were eligible if a pre-transfer 12-lead ECG and sufficient outcome data were available. A consultant cardiologist, blinded to the AI output, assigned a reference diagnosis (STEMI vs non-STEMI/non-cardiac) using a hierarchical approach based on: (i) evidence of an acutely occluded culprit artery on coronary angiography, where performed (preferred standard), or failing this (ii) serial high-sensitivity troponin measurements and the final discharge diagnosis. Cases without a retrievable ECG or adequate outcome data were excluded. Exported ECGs were anonymised and processed through the QOH application, which classified each trace as STEMI, STEMI-equivalent or no STEMI ( figure 1). For the primary analysis, STEMI and STEMI-equivalent outputs were considered test positive. Diagnostic performance metrics (sensitivity, specificity, positive and negative predictive values, and overall accuracy) were calculated by comparing AI classifications with the adjudicated reference diagnosis.Results Baseline characteristics are shown in table 1. The cohort comprised 131 consecutive PPCI referrals (mean age 68 ±17.9 years, 70.2% male); cardiovascular risk factors were modestly prevalent, and an occluded culprit vessel was identified in 24/41 STEMI cases in the LAD, 15/41 in the RCA and 2/41 in the LCx. Overall, 41 patients (31.3%) had a final diagnosis of STEMI and 90 (68.7%) were classified as non-STEMI or non-cardiac.The QOH AI correctly identified 39 of 41 STEMI cases (sensitivity 95.1%), and 81 of 90 non-STEMI cases (specificity 90.0%), yielding a positive predictive value (PPV) of 81.3% (39/48), a negative predictive value (NPV) of 97.6% (81/83) and an overall diagnostic accuracy of 91.6% (121/130) (table 2). Receiver-operating-characteristic analysis demonstrated excellent discrimination, with an area under the curve of 0.93 (figure 2). Among the two false-negative cases, one involved in-stent restenosis and the other acute stent thrombosis. The nine false-positive AI classifications included four NSTEMI presentations, one patient with LBBB and a history of Takotsubo cardiomyopathy, one aortic dissection and three non-cardiac presentations.During the same period, PPCI co-ordinator assessment correctly identified 40/41 STEMI and 84/90 non-STEMI cases, corresponding to a sensitivity of 97.6%, specificity of 93.3%, PPV of 87.0%, NPV of 98.8% and overall accuracy of 94.7% (124/131).Discussion and Conclusion In this single-centre retrospective PPCI cohort, the QOH AI ECG application demonstrated excellent diagnostic performance for STEMI detection, achieving sensitivity and specificity above 90% and a very high NPV, despite operating without access to clinical information. Its performance was comparable to that of an experienced PPCI co-ordinator, suggesting that AI-based ECG interpretation could serve as an adjunctive, objective and rapid decision-support tool rather than a replacement for current practice. Such a tool has the potential to standardise STEMI triage, reduce inappropriate catheterisation laboratory activations and free PPCI co-ordinators to focus on other critical clinical tasks, with particular promise in pre-hospital paramedic settings where expert interpretation may be limited. However, these findings are based on a relatively small, single-centre, retrospective analysis, and prospective validation in larger, diverse populations is required before routine implementation.Abstract 368 Figure 1Queen of hearts™ app: possible diagnostic outcomesAbstract 368 Table 1Baseline characteristics of the study populationCharacteristicOverall (N=131)Age, years68.0 ± 17.9Male sex, n (%)92 (70.2)Hypertension, n (%)17 (13.0)Hyperlipidaemia, n (%)12 (9.2)Diabetes mellitus, n (%)17 (13.0)Current smoker, n (%)6 (4.6)Peak troponin, median [IQR]*102 [19-1129]Culprit vessel among STEMI (n=41), n (%): – Left anterior descending (LAD)24 (58.5) – Right coronary artery (RCA)15 (36.6) – Left circumflex (LCx)2 (4.9)*Peak troponin values shown as median [interquartile range]Abstract 368 Table 2Diagnostic performance of Queen of Hearts AI vs PPCI co-ordinator (N=131)ParameterQueen of Hearts AIPPCI co-ordinatorTrue positives (TP)39 / 4140 / 41False positives (FP)96True negatives (TN)81 / 9084 / 90False negatives (FN)21Sensitivity, %95.1 (39 / 41)97.6 (40 / 41)Specificity, %90.0 (81 / 90)93.3 (84 / 90)Diagnostic accuracy, %91.6 (120 / 131)94.7 (124 / 131)Positive predictive value, %81.3 (39 / 48)87.0 (40 / 46)Negative predictive value, %97.6 (81 / 83)98.8 (84 / 85)Values are shown as % (n/N). TP, true positive; FP, false positive; TN, true negative; FN, false negative; PPV, positive predictive value; NPV, negative predictive value; PPCI, primary percutaneous coronary intervention.Abstract 368 Figure 2ROC curve of queen of hearts algorithm for STEMI diagnosis",
  "authors": [
    {
      "affiliations": [
        "Essex Cardiothoracic Centre, Essex, United Kingdom"
      ],
      "name": "Vaishnav Manoj"
    },
    {
      "affiliations": [
        "Essex Cardiothoracic Centre, Essex, United Kingdom"
      ],
      "name": "Haroun Butt"
    },
    {
      "affiliations": [
        "Essex Cardiothoracic Centre, Essex, United Kingdom"
      ],
      "name": "Uzma Sajjad"
    },
    {
      "affiliations": [
        "Essex Cardiothoracic Centre, Essex, United Kingdom"
      ],
      "name": "Abdalla Ibrahim"
    },
    {
      "affiliations": [
        "Essex Cardiothoracic Centre, Essex, United Kingdom"
      ],
      "name": "Christopher Cook"
    },
    {
      "affiliations": [
        "Essex Cardiothoracic Centre, Essex, United Kingdom"
      ],
      "name": "Ozan Demir"
    },
    {
      "affiliations": [
        "Essex Cardiothoracic Centre, Essex, United Kingdom"
      ],
      "name": "Gerald Clesham"
    },
    {
      "affiliations": [
        "Essex Cardiothoracic Centre, Essex, United Kingdom"
      ],
      "name": "John R Davies"
    },
    {
      "affiliations": [
        "Essex Cardiothoracic Centre, Essex, United Kingdom"
      ],
      "name": "Thomas R Keeble"
    }
  ],
  "title": "368 Diagnostic accuracy of the queen of hearts AI ECG application for STEMI Triage: a retrospective PPCI cohort study",
  "uid": "1db2f5ec-c75f-5e75-a7f1-ecc146cf6a60"
}