A new machine learning assessment of the ECLS-SHOCK study found there is a “low likelihood” that any subpopulation of infarct-related cardiogenic shock (CS) patients will respond to extracorporeal life support (ECLS). The findings were published Monday online in a research letter to JACC: Cardiovascular Interventions, authored by Karl-Patrik Kresoja, MD, from Johannes Gutenberg University, Germany, together with colleagues. The letter’s authors noted that, despite hopes that ECLS would improve outcomes for infarct-related CS — a condition associated with high 30-day mortality rates — the ECLS-SHOCK trial found no difference in 30-day mortality between ECLS and control groups. No positive effects were observed with ECLS for any of the predefined subgroups of the study. Given ongoing debate on whether certain subpopulations with infarct-related CS may still stand to benefit from ECLS, the researchers embarked on a new analysis using tree-based machine learning via extreme gradient boosting (XGBoost), trained to predict all-cause mortality at 30 days. The ECLS-SHOCK study included 417 patients aged 18 to 80 years with acute myocardial infarction complicated by severe CS and who were scheduled for revascularization. Patients were randomized 1:1 to receive standard care with or without ECLS. Severe CS was defined as a systolic blood pressure of <90 mm Hg for > 30 minutes or requiring vasopressors to maintain a systolic pressure > 90 mm Hg, an arterial lactate lebel > 3 mmol/L, and signs of impaired organ perfusion with at least 1 of the following: altered mental status, cold or clammy skin and limbs or urine output <30 mL/h. In the current analysis, the dataset was split into training (80%) and internal validation (20%), and the model included 25 clinical variables, including sex, coronary artery diseases extent, ST-segment elevation myocardial infarction and resuscitation before admission. Baseline characteristics between ECLS and control groups were similar. “To account for the effect of ECLS-associated complications, moderate or severe bleeding complications, occurring in 69 (17%) patients, and peripheral vascular complications warranting intervention, occurring in 31 (7.4%) patients, were included in the XGBoost model,” the researchers wrote. The model predicted mortality with an area under the curve (AUC) of 0.80 (95% confidence interval [CI]: 0.71-0.90), with a sensitivity of 79% (95% CI: 68%-91%) and a specificity of 74% (95% CI: 61%-88%) at a Youden-derived cutoff. The primary all-cause mortality outcome at 30 days occurred in 202 (48%) patients overall, 100 from the ECLS group and 102 from control. “Neither ECLS therapy nor its complications were among the most influential factors of the model, with a relative importance of 0.16% for ECLS therapy, 0.27% for peripheral vascular complications, and 0.83% for bleeding complications,” the researchers wrote. The most influential factors of the model included lactate levels (relative importance = 14.88%), C-reactive protein (relative importance = 11.54%), glucose levels (relative importance = 10.63%), and age (relative importance = 10.14%). The negligible effect on outcomes for ECLS patients in the study highlights a “low likelihood” of a responder group, the researchers said. “The neutral outcomes observed with ECLS may be explained by the multifaceted nature of CS, in which organ failure and, in cases of prior resuscitation, anoxic brain injury plays a more critical role in determining patient outcomes than hemodynamic stabilization alone,” the researchers speculated. “Supporting this, the study found that bleeding and vascular access site complications had minimal influence on outcomes, contrasting with previous reports. Instead, systemic factors like lactate levels, C-reactive protein, and renal function emerged as key determinants, underscoring that CS impacts the entire organism beyond its hemodynamic implications.” Source: Kresoja KP, Zeymer U, Thevathasan T, et al. Research Letter: The Influence of Extracorporeal Life Support on Patients in Cardiogenic Shock Assessed by Machine Learning, A Machine Learning Subanalysis From the ECLS-SHOCK Trial. JACC: Cardiovasc Interv 2024; DOI: 10.1016/j.jcin.2024.10.043. Image Credit: Vitalii Vodolazskyi – stock.adobe.com