Machine Learning for Predicting Outcomes in Cardiac Arrest Patients
Author Information
Author(s): Lee Sijin, Lee Kwang-Sig, Park Sang-Hyun, Lee Sung Woo, Kim Su Jin
Primary Institution: Korea University Anam Hospital
Hypothesis
This study aims to identify in-hospital variables and predictors of favorable neurological outcomes in patients who underwent out-of-hospital cardiac arrest (OHCA) and to develop decision support systems for various subgroups.
Conclusion
The study identified key determinants of favorable neurological outcomes in OHCA patients and demonstrated the effectiveness of a random forest model as a decision support system for stratified subgroups.
Supporting Evidence
- 19.6% of participants had good neurological outcomes (CPC 1–2).
- The major determinants of outcomes included in-hospital CPR duration and post-ROSC pH.
- Machine learning models outperformed traditional logistic regression in predicting outcomes.
- Patients were divided into six subgroups based on prehospital ROSC and first monitored rhythm.
Takeaway
Doctors can use a computer program to help decide how well patients who had a heart attack outside the hospital might recover, based on their specific situations.
Methodology
The study used machine learning and multicenter registry data from the Korean Cardiac Arrest Research Consortium, analyzing 2679 patients with out-of-hospital cardiac arrest.
Potential Biases
Potential selection bias due to exclusion of certain patient groups and variability in outcomes across different medical institutions.
Limitations
The study did not consider the etiologies of OHCA outcomes, included outcomes as a binary variable, and had a median missing rate of 13% across variables.
Participant Demographics
The median age of participants was 65 years, with 67.7% being male and 71.1% experiencing witnessed arrest.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website