March 22, 2021Journal of the American College of Emergency Physicians OpenOpen Access

Machine learning compared with rule‐in/rule‐out algorithms and logistic regression to predict acute myocardial infarction based on troponin T concentrations

Q: What is the clinical evidence from this study?

Study design: Cross-Sectional. Population: Acute myocardial infarction (AMI) among patients presenting with chest pain (n=5695). Intervention: Artificial neural network (ANN) machine learning algorithm vs. European Society of Cardiology (ESC) 0/1-hour and 0/3-hour algorithms and logistic regression. Primary outcome: Size of the intermediate risk group where AMI could not be ruled in or out (Relative decrease 9.2%, 95% CI 4.4% to 13.8%).

View Full Paper

Ask AI

Bookmark

Key result

An artificial neural network reduced the intermediate risk group for acute myocardial infarction by a relative 9.2% (from 24.5% to 22.2%) compared to standard ESC algorithms.

Why the study?

Computerized decision-support tools may improve diagnosis of AMI among patients presenting with chest pain at the ED, leading authors to assess machine learning algorithms using paired hs-cTnT concentrations with varying sampling times, age, and sex.

Does an artificial neural network machine learning algorithm reduce the size of the intermediate risk group compared to standard ESC algorithms and logistic regression in ED patients with chest pain?

Population

5695 chest pain patients at 2 hospitals in Sweden

Comparison

Artificial neural network vs European guideline-recommended 0/1- and 0/3-hour algorithms vs logistic regression

Design

Retrospective register-based cross-sectional diagnostic study

Authors

ABAnders BjörkelundLund University MOMattias OhlssonLund University JFJakob Lundager ForbergGeneral Cardiology

Discussion

Member takes

Implication

ML algorithms with hs-cTnT may aid AMI diagnosis in ED chest pain; leaves open need for prospective validation before clinical adoption.

Study Design

Type

Cross-Sectional (n=5,695)

Multicenter

Yes

Structured PICO

Does an artificial neural network machine learning algorithm reduce the size of the intermediate risk group compared to standard ESC algorithms and logistic regression in ED patients with chest pain?

Population

5,695 adult patients presenting to the emergency department with chest pain who had two serial high-sensitivity cardiac troponin T samples drawn.

Exposure

Artificial neural network (ANN) machine learning algorithm using age, sex, and two serial hs-cTnT measurements with their rate of change.

Comparator

European Society of Cardiology (ESC) guideline-recommended 0/1-hour and 0/3-hour algorithms, and a logistic regression model.

Outcome

Size of the intermediate risk group where acute myocardial infarction (AMI) could not be ruled in or out, while holding sensitivity (rule-out) and specificity (rule-in) constant.surrogate

Main Result

Effect estimate: Relative decrease 9.2% (95% CI 4.4% to 13.8%)

Absolute Event Rate: 22.2% vs 24.5%

Machine learning algorithms using serial hs-cTnT, age, and sex can modestly reduce the proportion of chest pain patients classified as intermediate risk for AMI compared to standard ESC algorithms.

Limitations

The algorithms were trained and tested on a selected subsample with 2 hs-cTnT analyzed and are therefore not necessarily generalizable to settings where indications for ordering a second sample are different.
Results may not generalize to settings where the prevalence of AMI differs at entry to the ED presentation.
Used routine care diagnoses, which may contain misclassifications.
Both ANN and logistic regression were highly dependent on the choice of probability thresholds.
ANN would generally require retraining on new populations whenever deployed across hospitals.
Did not extend the logistic regression model to incorporate interaction terms between the input variables.
Trained and tested on a selected subsample with 2 hs-cTnT analyzed, which may not generalize to different ordering indications
May not generalize to settings where the prevalence of AMI differs
Used routine care diagnoses rather than adjudicated outcomes
Highly dependent on the choice of probability thresholds
ANN would generally require retraining on new populations whenever deployed across hospitals

Cite This Study

Björkelund et al. (2021) conducted a cross-sectional in Acute myocardial infarction (AMI) among patients presenting with chest pain (n=5,695). Artificial neural network (ANN) machine learning algorithm vs. European Society of Cardiology (ESC) 0/1-hour and 0/3-hour algorithms and logistic regression was evaluated on Size of the intermediate risk group where AMI could not be ruled in or out (Relative decrease 9.2%, 95% CI 4.4% to 13.8%). An artificial neural network reduced the intermediate risk group for acute myocardial infarction by a relative 9.2% (from 24.5% to 22.2%) compared to standard ESC algorithms.

synapsesocial.com/papers/6a7b295402ab7a73dd033ab6 https://doi.org/10.1002/emp2.12363

View Full Paper

Ask AI

Bookmark

Machine learning compared with rule‐in/rule‐out algorithms and logistic regression to predict acute myocardial infarction based on troponin T concentrations

Discussion

Member takes

Implication

Study Design

Structured PICO

Main Result

Limitations

Cite This Study

Machine learning compared with rule‐in/rule‐out algorithms and logistic regression to predict acute myocardial infarction based on troponin T concentrations

Discussion

Member takes

Implication

Study Design

Structured PICO

Main Result

Limitations

Cite This Study

Also Consider

Also Consider