Upper GI bleeding common after acute MI, linked to poor prognosis
Click Here to Manage Email Alerts
In patients with acute MI, upper gastrointestinal bleeding occurred in 1.5% within 1 year and was associated with poor outcomes, researchers reported at the European Society of Cardiology Congress.
The researchers also employed logistic regression and machine learning models to predict which patients were at highest risk for upper gastrointestinal (GI) bleeding. The study was simultaneously published in European Heart Journal Cardiovascular Pharmacotherapy.
Upper GI bleeding “is of special interest because it can be prevented by available prophylaxis,” Philip Sarajlic, MD, research scientist in the translational cardiology team at Karolinska University Hospital and PhD student at the Karolinska Institute, Stockholm, said during a presentation. “It is unknown how prominent a role upper GI bleeding plays in compliance to medical therapy, and how this associates with ischemic outcomes. The incidence, associated cardiovascular outcomes and predictors of upper GI bleeding are not sufficiently understood.”
Sarajlic and colleagues analyzed 149,477 patients from the SWEDEHEART registry who had acute MI from January 2007 to June 2016.
As part of the study, the researchers compared traditional statistical methods with four machine learning models, Sarajlic said. “We picked the best-performing model and looked deeper into its mechanics to determine what variables were the most important ones to make the prediction of whether a patient will have a 1-year event of upper GI bleeding or not,” he said.
At 1 year, upper GI bleeding occurred in 1.5% of patients and was associated with elevated risk for all-cause mortality (adjusted HR = 2.86; 95% CI, 2.58-3.16), stroke (aHR = 1.8; 95% CI, 1.32-2.45) and major adverse CV events, defined as MI/stroke/all-cause death (aHR = 2; 95% CI, 1.81-2.2), but not recurrent MI (aHR = 1.17; 95% CI, 0.97-1.42), according to the researchers.
In those who had upper GI bleeding, the median age was 77 years and 39% were women. In those who did not, the median age was 71 years and 35% were women.
Sarajlic said when logistic regression was used, the following were identified as independent predictors of upper GI bleeding in this population: hemoglobin, age, smoking status, previous upper GI bleeding, antithrombotic treatment and gastroprotective treatment (P .001 for all).
The machine learning model with the best performance was the random forest model, Sarajlic said, noting that its most important elements were hemoglobin, age, systolic BP, blood glucose, gastroprotective treatment and corticosteroid treatment.
“We see that some variables are overlapping, but there are other variables that were not picked by the logistic regression model, such as systolic blood pressure, blood glucose and corticosteroid use,” Sarajlic said during the presentation.
Reference:
Collapse
Sarajlic P, et al. Latest Science in ACS. Presented at: European Society of Cardiology Congress; Aug. 27-30, 2021 (virtual meeting).
Click Here to Manage Email Alerts