Authors:

Saloni Jajoo

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

Although percutaneous coronary intervention (PCI) can save lives by restoring blood flow to the myocardium, it is associated with post-interventional complications such as bleeding, myocardial infarction, stent thrombosis, and stroke. This study aims to use machine learning to stratify patients with regard to complication risk based on routine clinical data available prior to PCI. The study makes use of a large registry dataset from the COAP quality improvement initiative in the state of Washington comprising data on demographics, clinical conditions, procedures, and outcomes. Data from 91,000 patients who underwent PCI between 2000 and 2023 were preprocessed to account for missing values via mean imputation, to standardize and categorize data, and to balance class distribution using Synthetic Minority Oversampling Technique (SMOTE). A Column Transformer, an effective preprocessor for both numerical and categorical data, was used. We develop and validate individual XGBoost-based prediction models for ten different postinterventional complications including distal coronary artery perforation, stroke, and stent thrombosis, using a random 60-20- 20 train-validation-test split and a 5-fold cross validation scheme. To maximize performance, model hyperparameters are adjusted using randomized search. An independent test set is used to validate the final model, which is trained using optimized parameters on the whole training dataset. We report model performance using confusion matrices and metrics including precision, recall, and F1-score. Our preliminary results indicate that XGBoost-based machine learning models can predict distal coronary artery perforation after PCI (precision = 0.75, recall = 0.62, F1-score = 0.68). In summary, this work aims to enhance patient care by highlighting the potential of machine learning in preinterventional risk stratification prior to PCI and initiating steps towards their clinical translation. Keywords: Percutaneous Coronary Intervention; Machine Learning; Biomedical Engineering

Source:

Purdue University / 2024

Topics:

No topics listed

Co-authors:

Saloni Jajoo