ECMO in cardiogenic shock is increasingly visible in national practice, but comparing outcomes across support strategies remains difficult. Kochar and colleagues examined Medicare claims for patients with myocardial infarction-related cardiogenic shock (MI-CS), describing device use and testing several approaches to compare 30-day mortality. Their most important conclusion is methodological: the available claims data could not support valid causal estimates of ECMO’s effect on survival.
What the National Study Examined
This retrospective observational analysis used Medicare fee-for-service inpatient claims. Trends in mechanical circulatory support were assessed from 2016 through 2023, while baseline characteristics and comparative effectiveness analyses used data from 2016 through 2022. Patients had a primary diagnosis of myocardial infarction and a secondary diagnosis of cardiogenic shock, with at least one year of prior claims required for the principal cohort.
The baseline comparison included 102,718 patients: 1,165 received ECMO, 10,064 Impella, 24,631 an intra-aortic balloon pump (IABP), and 66,858 medical management alone as their most intensive therapy. In this study, ECMO refers to venoarterial support. These groups reflect clinical treatment selection rather than randomized assignment.
A more restricted comparative analysis included 23,658 patients with complete linked hospital information at hospitals offering all four treatment strategies and meeting the study’s volume criteria. That subset contained 723 ECMO patients. Keeping these populations separate is important when interpreting the mortality percentages reported in different sections of the paper.
ECMO in Cardiogenic Shock: Use Increased but Remained Uncommon
ECMO use rose from 0.8% to 1.8% of MI-CS admissions between 2016 and 2023. Over the same period, the percentage of hospitals using ECMO increased from 9.2% to 17.1%. Impella use increased from 5.7% to 12.4%, while IABP use fell from 28.0% to 19.6%.
These findings describe changing practice, not proof that broader device use improved outcomes. The Medicare population also differs from younger patients commonly enrolled in shock trials. Even within this cohort, ECMO recipients were younger on average than those receiving the other strategies: 68.8 years, compared with 72.9 for Impella, 73.7 for IABP, and 75.6 for medical management.
ECMO recipients had more acute features, including higher rates of ST-segment elevation myocardial infarction, cardiac arrest on presentation, and intubation. At the same time, they had lower rates of several chronic conditions. This combination illustrates how clinicians may select patients with greater immediate physiological compromise but fewer barriers to recovery, producing groups that are difficult to compare fairly.
Mortality Findings Need Their Analytical Context
In the broad baseline cohort, unadjusted 30-day mortality was approximately 69% with ECMO, 60% with Impella, 39% with IABP, and 46% with medical management. Those percentages are observed outcomes in different patient groups. They do not establish that ECMO increased mortality or that the lower-mortality strategies would have produced better outcomes in the same patients.
| Analysis in the restricted cohort | ECMO 30-day mortality | Impella 30-day mortality | Interpretation |
|---|---|---|---|
| Observed event rates | 67.9% | 58.5% | Unadjusted outcomes after clinical treatment selection. |
| Inverse probability treatment weighting | 68.6% | 57.4% | Adjusts for measured baseline differences; important imbalances remained. |
| G-formula | 70.2% | 55.9% | Uses outcome modeling and standardization; missing clinical detail remains a limitation. |
| Seven-day grace period | 53.6% | 53.5% | Addresses treatment timing differently, but cannot fully account for evolving shock severity. |
The estimates above come from Table 3 of the source. They are not interchangeable treatment-effect estimates. In particular, nearly identical ECMO and Impella estimates in the grace-period analysis do not prove equivalence, noninferiority, or equal benefit.
Why More Sophisticated Adjustment Was Not Enough
Inverse probability weighting attempts to create comparable groups using recorded characteristics. The G-formula estimates outcomes under modeled treatment strategies using a different set of assumptions. Both methods depend on the information available and on assumptions that cannot be repaired simply by increasing the number of records.
Cardiogenic shock changes over hours. Claims do not adequately capture the severity and progression that prompt escalation to ECMO, such as evolving perfusion failure or worsening hemodynamics. A device may therefore appear associated with poor survival because it was selected for patients already deteriorating, even after adjustment for baseline diagnoses.
Timing adds another problem. When patients are classified by a treatment received after admission, they must survive long enough to receive that treatment. This can introduce immortal time bias. The seven-day grace-period approach was designed to address delayed initiation more explicitly, but it still lacked sufficiently detailed time-varying clinical variables. Admission also served as time zero, rather than the actual onset of shock.
The investigators explored a hospital-preference instrumental variable approach, but insufficient variation in ECMO use meant the necessary relevance assumption was not satisfied. Their transparent decision not to present unsupported causal conclusions is a key contribution of the work.
Combined Support and Relevance for Perfusion Teams
Among the 723 ECMO recipients in the restricted comparative cohort, 68.5% also received at least one other support device. That finding is relevant to teams managing combinations of VA-ECMO, IABP, and microaxial support, but the claims could not reliably explain why a second device was added.
For example, Impella placement might represent planned ventricular unloading or an initial support strategy followed by rescue escalation to ECMO. Device sequence alone cannot distinguish those intentions. The study therefore cannot determine the best unloading device, optimal timing, or whether combined support caused a survival advantage.
For perfusion and shock teams, the practical implication is to interpret device outcome reports alongside patient selection, shock trajectory, and treatment timing. More informative registries would capture the clinical state before and during support, the reason for escalation, and complications that administrative claims describe incompletely. This paper does not evaluate a particular circuit, flow target, anticoagulation protocol, or bedside management algorithm.
Limitations and Disclosures
The study is valuable for national utilization patterns and for demonstrating the limits of claims-based comparative research. It remains observational, limited to Medicare fee-for-service patients, and vulnerable to residual confounding and incomplete measurement of shock severity. Mortality and rehospitalization were emphasized because other outcomes, including bleeding and vascular complications, were not well captured. The restricted analysis also represents a selected subset of patients and hospitals.
Funding came from the FDA’s Center for Devices and Radiological Health through a contract and subcontract supporting the project. Several authors report research, consulting, or other industry relationships; Kapur reports employment at Abiomed. The original article provides the full individual disclosures.
Original Article
Kochar A, Lalani C, Song Y, et al. National Use Patterns and Comparative Effectiveness of Extracorporeal Membrane Oxygenation in Myocardial Infarction-Related Cardiogenic Shock. Journal of the American Heart Association. 2026;15:e049443. doi:10.1161/JAHA.126.049443.
This independently written educational summary accompanies the unchanged source PDF, available under the article’s Creative Commons Attribution-NonCommercial-NoDerivs license. The cover and short visual animation are editorial illustrations, not a technical circuit or treatment guide.
