Clinical Research & Data, AMI Cardiogenic Shock
Data-Driven Point-of-Care Risk Model in Patients with AMICS
Ole Helgestad, PhD, discusses an automated variable selection method for predicting 30-day mortality in patients with acute myocardial infarction and cardiogenic shock (AMICS) receiving acute percutaneous coronary intervention (PCI). Dr. Helgestad, a postdoctoral fellow at Odense University Hospital in Odense, Denmark, is the first author on a paper titled, “Data-driven point-of-care risk model in patients with acute myocardial infarction and cardiogenic shock,” published in European Heart Journal: Acute Cardiovascular Care in August 2021.
“Prognostication in acute myocardial infarction and cardiogenic shock is difficult,” states Dr. Helgestad. Current prognosis models have restrictions or limitations related to how they stratify patients into low, medium, and high risk. “And today there’s a trend moving towards personalized, individualized medicine and part of that is to provide personalized prognostication,” Dr. Helgestad explains. “And another key issue in cardiogenic shock is that time is of the essence. So, you need a readily available model that can give you an individualized idea of prognosis in a short moment of time.”
Dr. Helgestad notes that our theoretical considerations and assumptions about variables and their association with outcome may be imperfect and incomplete. “And that brings in the idea of machine learning.” He explains that with machine learning you provide the algorithm with data and the algorithm looks at the data, creating a risk model that is data-driven rather than theory-driven.
Dr. Helgestad describes the methods used in his paper, explaining that they used a form of machine learning called LASSO (least absolute shrinkage and selection operator) that automates variable selection in a logistic regression model. He discusses patient selection for model development and validation.
This model is in the early development phase and Dr. Helgestad hopes to validate the model in a different geographic area in the future. “I definitely believe that this is a good way to proceed and to use the tremendous amount of data that are being collected,” he states. He also hopes that others will use their data to validate and update the models and that in the future the models can assist in identifying suitable candidates for mechanical circulatory support.
His take-home message: “A personalized pre-PCI prognostication in patients with acute myocardial infarction and shock is feasible using readily available information that are based on point-of-care available data.”