Machine learning models for prediction of adverse events after percutaneous coronary intervention
Problem
Clinicians need accurate prediction of major adverse events (acute kidney injury, bleeding, in-hospital mortality) after percutaneous coronary intervention to improve decisions.
Why it matters
Better prediction of peri-PCI adverse events directly improves clinical decisions and patient-specific interventions.
What the AI did
Developed logistic regression and XGBoost machine learning models on a prospective multicenter Japanese PCI registry (2008–2020) and compared them against NCDR-CathPCI risk scores.
Results
NCDR-CathPCI scores achieved C-statistics of 0.82 (AKI), 0.76 (bleeding), 0.95 (mortality); XGBoost modestly improved discrimination for AKI and bleeding (abstract truncated before full figures).
Impact
Peer-reviewed study showing ML modestly improved prediction of AKI and bleeding over established risk scores.
Record
| Venue | Scientific Reports |
|---|---|
| Published | 2022 |
| Work type | article |
| Authors | Nozomi Niimi, Yasuyuki Shiraishi, Mitsuaki Sawano, Nobuhiro Ikemura, Taku Inohara, Ikuko Ueda, Keiichi Fukuda, Shun Kohsaka |
| Citations | 48 (OpenAlex, at import time) |
| References | 35 |
| Model or method | XGBoost (gradient descent boosting) and logistic regression |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41598-022-10346-1 |
| Open access | gold |
| Keywords | machine learning, XGBoost, percutaneous coronary intervention, adverse events, prediction, C-statistic |
Importance rating
real but narrower advance
Peri-PCI adverse event prediction aids clinical decisions for coronary artery disease patients, but abstract reports only modest improvement over adequate existing risk scores, limiting added impact.