What is the clinical evidence from this study?

Study design: Observational. Population: Acute chest pain (n=9519). Intervention: Convolutional neural network (CNN-MB) diagnostic model vs. ESC 0 h algorithm. Primary outcome: Proportion of patients safely ruled out for 30-day AMI or death (sensitivity >99%, NPV >99.5%).

February 2, 2023Open Access

Machine learning for early prediction of acute myocardial infarction or death in acute chest pain patients using electrocardiogram and blood tests at presentation

PCPontus Olsson de CapretzLund University ABAnders BjörkelundLund University JBJonas BjörkLinköping University

Key Result

A convolutional neural network model using ECG and initial blood tests safely ruled out 30-day acute myocardial infarction or death in 55.0% of emergency department chest pain patients, compared to 47.2% using the ESC 0 h algorithm.

Study Design

Type

Observational (n=9,519)

Multicenter

Yes

Structured PICO

Does a machine learning model using ECG and initial blood tests improve the early rule-out and rule-in of 30-day AMI or death in ED chest pain patients compared to the ESC 0 h algorithm?

Population

9,519 adult emergency department patients presenting with acute chest pain, evaluated with ECG and blood tests, followed for 30 days for acute myocardial infarction or death.

Exposure

Machine learning models (Logistic regression, Artificial Neural Network, Convolutional Neural Network on raw ECG or median beat) using age, sex, ECG, and first blood tests (hs-cTnT, glucose, creatinine, hemoglobin) at presentation

Comparator

The 0 h arm of the European Society of Cardiology (ESC) 0/1 h algorithm (hs-cTnT < 5 ng/L for rule-out and ≥ 52 ng/L for rule-in)

Outcome

Acute myocardial infarction (AMI) or all-cause death within 30 days including the index visitcomposite

A machine learning model incorporating age, sex, ECG, and initial blood tests can safely rule out or rule in 30-day AMI or death in a larger proportion of ED chest pain patients than the standard ESC 0 h algorithm.

Main Result

Absolute Event Rate: 55% vs 47.2%

Limitations

Risk of incorporation bias since troponin and ECG were part of both the input data and diagnostic criteria for AMI.
Few patients with 30-day AMI or death in the tuning and testing groups, making the models conservative.
Use of ICD codes for AMI diagnoses without adjudication may have led to mislabeling.
Lack of data on time from chest pain onset to hs-cTnT sampling.
Results from a single region (Sweden) may limit generalizability to other populations.

Abstract

AIMS: In the present study, we aimed to evaluate the performance of machine learning (ML) models for identification of acute myocardial infarction (AMI) or death within 30 days among emergency department (ED) chest pain patients. METHODS AND RESULTS: Using data from 9519 consecutive ED chest pain patients, we created ML models based on logistic regression or artificial neural networks. Model inputs included sex, age, ECG and the first blood tests at patient presentation: High sensitivity TnT (hs-cTnT), glucose, creatinine, and hemoglobin. For a safe rule-out, the models were adapted to achieve a sensitivity > 99% and a negative predictive value (NPV) > 99.5% for 30-day AMI/death. For rule-in, we set the models to achieve a specificity > 90% and a positive predictive value (PPV) of > 70%. The models were also compared with the 0 h arm of the European Society of Cardiology algorithm (ESC 0 h); An initial hs-cTnT < 5 ng/L for rule-out and ≥ 52 ng/L for rule-in. A convolutional neural network was the best model and identified 55% of the patients for rule-out and 5.3% for rule-in, while maintaining the required sensitivity, specificity, NPV and PPV levels. ESC 0 h failed to reach these performance levels. DISCUSSION: An ML model based on age, sex, ECG and blood tests at ED arrival can identify six out of ten chest pain patients for safe early rule-out or rule-in with no need for serial blood tests. Future studies should attempt to improve these ML models further, e.g. by including additional input data.

KI fragen

Bookmark

View Full Paper