Machine learning techniques may help identify patients who are most likely to experience a major bleeding after PCI, according to a study published July 10 in JAMA Network Open.

Bobak J. Mortazavi, PhD, et al., used machine learning techniques to recreate two existing bleeding risk models – a 31-variable full PCI model and a simplified risk score model – from ACC's CathPCI Registry. The researchers then conducted data analyses to evaluate performance of the existing models and recreated models in a sample of more than 3 million procedures from the CathPCI Registry. The study’s primary outcome was in-hospital bleeding within 72 hours of the procedures.

Results showed that among all 3,316,465 PCI procedures, the post-PCI bleeding rate was 4.5 percent. The new model did not improve risk prediction when factors were included as dichotomous variables. However, risk prediction improved when the researchers did not limit the data to dichotomous variables and implemented machine learning techniques. The existing risk prediction model achieved a C statistic of 0.78, compared with 0.82 when machine learning techniques were implemented. In addition, the machine learning model identified an additional 3.7 percent of patients as high risk for bleeding who experienced a bleeding event and an additional 1.0 percent of patients as low risk who did not experience a bleeding event.

According to the researchers, the study demonstrates that machine learning techniques could lead to “improvements in predictive model performance.” They note that “a key to successful implementation is the use of variables in a way that does not reduce information,” concluding that the findings “lay the groundwork for future work in more advanced models with additional variable for further improved performance.”

In an accompanying editorial comment, Jennifer A. Rymer, MD, MBA, and Sunil V. Rao, MD, FACC, write that moving forward, additional research is needed to “test the potential of these models to reduce the risk-treatment paradox and improve outcomes,” concluding that the “rise of machine learning is an important step into a new era of risk prediction.”

Clinical Topics: Invasive Cardiovascular Angiography and Intervention

Keywords: Registries, Percutaneous Coronary Intervention, National Cardiovascular Data Registries, CathPCI Registry

< Back to Listings