NCDR Study Finds Machine Learning Could Improve AKI Risk Prediction After PCI
The use of machine learning and data-driven approaches could be more accurate at predicting acute kidney injury (AKI) in PCI patients than the current risk prediction model used in ACC's CathPCI Registry, according to a study published Nov. 27 in PLoS Medicine.
Chenxi Huang, PhD, et al., used the original cohort – 947,091 patients – and variables that guided development of the CathPCI Registry's current AKI risk prediction model to create a series of new models with machine learning techniques, and compared performance. In the original patient cohort, there were 69,826 (7.4 percent) AKI events.
In comparisons to the original model, the new model derived from machine learning reclassified 42,167 patients whose AKI risk was underestimated with the original model and 61,388 whose risk was overestimated. The researchers also validated the new models using a new cohort of 970,869 patients. Results showed that the machine learning model was more accurate than the traditional one in several variables.
According to the authors, the study demonstrates the potential of machine learning to improve risk prediction and identify patients who could benefit from strategies to minimize risk. Further research should evaluate feasibility of integrating the machine learning model into clinical care, the researchers noted.
Clinical Topics: Invasive Cardiovascular Angiography and Intervention
Keywords: Acute Kidney Injury, Risk, Registries, National Cardiovascular Data Registries, Percutaneous Coronary Intervention, CathPCI Registry
< Back to Listings