Integrating bibliometric analysis and external validation of the SAVE score in VA-ECMO for refractory cardiogenic shock: insights from global trends and the MIMIC-IV database

Introduction

Cardiogenic shock is a life-threatening condition characterized by severe cardiac pump failure, leading to inadequate tissue perfusion and multi-organ dysfunction (1,2). Despite advances in medical and mechanical support, mortality rates remain high, particularly among patients with acute myocardial infarction (AMI), with in-hospital mortality exceeding 40% in this group (3,4). Although the incidence has declined slightly in recent years, cardiogenic shock continues to pose a significant global healthcare burden, resulting in substantial costs and intensive resource utilization (5).

Extracorporeal membrane oxygenation (ECMO), especially veno-arterial extracorporeal membrane oxygenation (VA-ECMO), has become an essential rescue therapy for refractory cardiogenic shock (6). ECMO offers temporary circulatory and respiratory support, enabling myocardial recovery or serving as a bridge to definitive therapies such as heart transplantation or ventricular assist device (VAD) implantation (7). However, optimal patient selection, timing of initiation, and management of ECMO-related complications—such as bleeding, infection, and thromboembolism—remain major challenges (8,9). Given these complexities and the growing use of ECMO, there is a critical need to clarify evolving global trends, identify research hotspots, and refine risk stratification and prognostic assessment.

One of the most widely used prognostic tools in this context is the Survival After Veno-Arterial ECMO (SAVE) score (10), which helps predict survival in patients receiving VA-ECMO for refractory cardiogenic shock (11). However, the generalizability and calibration of the SAVE score across diverse populations and real-world clinical datasets, such as the Medical Information Mart for Intensive Care IV (MIMIC-IV) database, have not been fully established.

Bibliometric analysis offers a systematic approach to mapping the knowledge landscape, revealing publication trends, collaboration networks, and emerging themes in ECMO research (12). Nevertheless, comprehensive bibliometric analyses that integrate traditional literature mapping with external validation of prognostic models using large-scale clinical databases remain scarce. To bridge this gap, our study utilizes global bibliometric trends to identify critical clinical needs—specifically the urgent mandate for ‘predicting survival’—and directly addresses this need by executing a rigorous external validation and recalibration of the SAVE score using the MIMIC-IV database. By integrating these approaches, we aim to provide new insights into research trends, key contributors, and the practical utility of prognostic tools, thereby informing future research directions and optimizing clinical management of cardiogenic shock. We present this article in accordance with the BIBLIO reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-0964/rc).

Methods

Literature search and selection

A comprehensive literature search was performed utilizing the Web of Science Core Collection (WoSCC), which is widely recognized as one of the most appropriate databases for bibliometric analysis. The focus of the search was on publications pertaining to ECMO within the context of cardiogenic shock from 1988 to 2024. Conducted on July 11, 2024, the search utilized the following strategy: “(TS = (refractory OR intractable OR rebellious OR incurable OR immedicable)) AND (TS = (“cardiogenic shock” OR “cardiac shock” OR “acute cardiogenic shock”)) AND (TS = (“extracorporeal membrane oxygenation” OR ECMO))”. This search was restricted to peer-reviewed original articles to ensure the inclusion of only high-quality publications. The data were sourced from public databases, and since the research did not involve human subjects, no ethical consent was necessary.

Bibliometric analysis and visualization

Two independent reviewers systematically extracted key data from the included publications, gathering information such as titles, keywords, publication dates, countries/regions, authors, institutions, journals, total citations (TC), and H-index. The extracted data were entered into Microsoft Excel 365 for organization and analysis. Additionally, bibliometric visualization and analysis were conducted using VOSviewer 1.6.20 (Leiden University, Netherlands), CiteSpace 6.3.R1, and R 4.3.3. The characteristics of all included publications were described using the Web of Science database, while key metrics such as the H-index were employed to assess the scientific impact of authors and institutions (13,14). Journal impact factors (IF) were obtained from the latest version of Journal Citation Reports (JCR). Microsoft Excel 365 was also utilized to compute bibliometric indicators and generate models predicting future publication trends. VOSviewer facilitated the mapping of institutional and author collaborations, co-authorship networks, and co-citation relationships, thereby providing insights into the academic network within this field. Furthermore, keyword co-occurrence analysis was employed to identify emerging research trends and hotspots. The visualization enabled cluster analysis, grouping keywords based on their co-occurrence and color-coding them to reflect the timing of their emergence in the literature.

External validation and recalibration of the SAVE score using the MIMIC-IV database

The primary hypothesis of this study is that, within the MIMIC-IV cohort, the SAVE score will demonstrate moderate discrimination in predicting in-hospital mortality among patients with refractory cardiogenic shock receiving VA-ECMO; however, its calibration may be suboptimal, meaning that predicted mortality risks may not consistently align with observed outcomes. The secondary hypothesis is that alternative contemporary prognostic tools, such as the prEdictioN of Cardiogenic shock OUtcome foR AMI patients salvaGed by VA-ECMO (ENCOURAGE) or Sequential Organ Failure Assessment (SOFA) scores, may demonstrate different transportability and performance metrics within this same clinical population.

To test these hypotheses, we conducted a single-center, retrospective cohort study utilizing the MIMIC-IV (v2.2) database. Eligible patients were adults aged 18 years or older who received VA-ECMO support during their ICU stay, with cardiogenic shock as the primary indication for ECMO. Patients were excluded if they received veno-venous extracorporeal membrane oxygenation (VV-ECMO) or veno-arterio-venous extracorporeal membrane oxygenation (VAV-ECMO), underwent ECMO for out-of-hospital cardiac arrest [extracorporeal cardiopulmonary resuscitation (E-CPR)] without return of spontaneous circulation, or if more than two core variables required for SAVE score calculation were missing. Applying these criteria, we identified a cohort of 101 adult patients with refractory cardiogenic shock treated with VA-ECMO. This sample size meets the widely accepted minimum for external validation studies (n≥100), thereby providing sufficient statistical power for subsequent analyses.

For each patient, all variables required to calculate the SAVE score were extracted from the MIMIC-IV database. To ensure complete transportability and minimize bias, all clinical and laboratory parameters were strictly derived from the last documented values within the 24 hours immediately preceding the initiation of VA-ECMO, consistent with the original model development guidelines. The primary endpoint was in-hospital mortality. Preliminary assessment of the dataset indicated excellent data completeness, with greater than 95% availability for all essential SAVE score components, thereby supporting the reliability and validity of the planned analyses.

Statistical analysis

All statistical analyses were performed using Python 3.10 and R 4.3.2. Continuous variables were expressed as mean ± standard deviation (SD) or median [interquartile range (IQR)] as appropriate; categorical variables were summarized as frequency (percentage). For variables with <5% missingness, multiple imputation was performed to minimize bias and preserve analytical integrity. The discrimination of the model was evaluated using the C-statistic [area under the receiver operating characteristic curve (AUC-ROC)] with 95% confidence intervals (CIs), while calibration was assessed by calibration plots and the Hosmer-Lemeshow goodness-of-fit test. Data extraction was executed using Navicat Premium (version 17.2.8) with structured query language (SQL). Statistical significance was defined as a two-sided P value <0.05 for all analyses.

Results

Overview of publications

The search was confined to publications in the English language and included only peer-reviewed articles (Figure 1A). This bibliometric analysis assessed a total of 571 publications pertaining to ECMO in cases of cardiogenic shock, with contributions from 4,013 authors across 161 journals, and encompassing a cumulative total of 8,829 cited references. The average annual growth rate of publications was determined to be 9.47% (Figure 1B). The peak in annual publications was observed in 2021, with a total of 73 articles. An analysis of publication activity from 1988 to 2024 reveals a consistent upward trajectory, which can be categorized into three distinct phases: a slow-growth phase from 1988 to 2007, a moderate growth phase from 2008 to 2015, and a recent phase characterized by accelerated growth from 2016 to 2023. The exponential growth curve depicted in the figure illustrates this pronounced increase, particularly in recent years. Nevertheless, despite the apex in 2021, there is an observable slight decline in annual publications in 2023 and 2024, as indicated by the blue line representing yearly output. The predictive equation for the cumulative publication trend is expressed as y=0.8351e^0.1811x (R²=0.9836), suggesting a robust exponential growth trajectory over time and reflecting sustained interest and output within this research domain (Figure 1C).

Figure 1 Overview of publications. (A) Flowchart of data screening process. (B) Comprehensive overview of the bibliometric analysis. (C) Annual growth of publications on ECMO research in cardiogenic shock. ECMO, extracorporeal membrane oxygenation.

Analysis of highly cited literatures

The three most cited articles include two from the European Heart Journal (IF =37.6) and the European Journal of Heart Failure (IF =16.9). The first article, titled “Predicting Survival After ECMO for Refractory Cardiogenic Shock: The Survival After Veno-Arterial-ECMO (SAVE) Score”, has received 574 citations (15). The second article, “Advanced Heart Failure: A Position Statement of the Heart Failure Association of the European Society of Cardiology”, has garnered 500 citations (16). Both articles focus on advancements in survival and prognosis for heart failure. The third article, published in Critical Care Medicine (IF =7.7), is titled “Outcomes and Long-Term Quality of Life of Patients Supported by Extracorporeal Membrane Oxygenation for Refractory Cardiogenic Shock”, with 480 citations. This article primarily addresses the effectiveness of ECMO in treating refractory cardiogenic shock (17) (Table S1).

Analysis of countries

Among all countries, the USA led in the number of publications on ECMO related to cardiogenic shock, with a total of 129 publications. France ranked second with 96 publications, followed by China with 72. In terms of total publications (TP), the USA again topped the list with 615 documents, followed by France with 547 and Germany with 273. Furthermore, the USA also led in TC, accumulating 4,453 citations, while France and China recorded 3,162 and 1,914 citations, respectively.

When assessing the average number of citations per publication, Australia ranked first with an average of 79.4 citations per document, closely followed by Spain with an average of 78.0 and Canada with 48.7. Regarding international collaboration, both the Netherlands and Australia demonstrated notable levels of multiple-country publications (MCP), achieving an MCP ratio of 50%. Austria and Switzerland also exhibited strong international collaboration, each attaining an MCP ratio of 30%. In contrast, countries such as China and South Korea displayed a greater focus on single-country research output, with MCP ratios of 4.2% (72 and 24 total documents, respectively) (Table S2, Figure 2A). The collaboration network map illustrates the USA as the central node in global research collaborations on ECMO related to cardiogenic shock, with a total of 115 collaborations among the 48 involved countries. Germany ranks second with 85 collaborations, closely followed by Italy with 81 collaborations (Figure 2B).

Figure 2 Global distribution and collaboration in ECMO research for cardiogenic shock. (A) Distribution of corresponding authors’ publications by country. (B) Visualization map depicting the collaboration among different countries. The collaborative relationships between countries, with nodes representing countries, the size of nodes indicating publication count, and the thickness of links showing the strength of co-authorship collaborations. (C) Top 10 institutions by article count and rank. (D) Visualization map depicting the collaboration among different institutions. ECMO, extracorporeal membrane oxygenation.

Analysis of institutions

The institution with the highest publication volume is Assistance Publique Hôpitaux Paris (APHP), which has produced a total of 154 articles. It is closely followed by the Institut National de la Santé et de la Recherche Médicale (INSERM) with 114 articles and Sorbonne Université, contributing 109 articles. Other significant contributors include Hôpital Universitaire Pitié-Salpêtrière with 84 articles, Université Paris Cité with 76 articles, and Columbia University with 73 articles (Figure 2C).

The collaboration network among institutions demonstrates a robust international presence, with Harvard Medical School leading global collaborations at 82 articles, followed by Brigham & Women’s Hospital with 69 articles and the University of Pennsylvania with 59 articles. The network is organized into distinct clusters: the green cluster, which encompasses major U.S. institutions such as Mayo Clinic and Tufts Medical Center, indicates strong intra-national collaboration within the USA. The blue cluster, led by European institutions including Sorbonne Université and Université de Lorraine, reflects significant collaborative efforts across Europe. The red cluster highlights collaborations among Japanese institutions, including Osaka University and Kyushu University, underscoring strong regional connections. These clusters illustrate clear geographical collaboration patterns, with the USA and Japan exhibiting prominent intra-national partnerships, while European institutions engage more frequently in both regional and international collaborations (Figure 2D).

Analysis of journals

The 20 most productive journals collectively published 267 papers, which represent 46.76% of all retrieved publications (Table S3). The ASAIO Journal occupies the leading position with 38 TP and the highest TC amounting to 760. Following this, Perfusion-UK has published 24 articles, while Artificial Organs ranks third with 21 publications. In terms of the IF in 2023, Intensive Care Medicine ranks highest with an IF of 27.1, followed by Critical Care with an IF of 8.8 and Critical Care Medicine with an IF of 7.7. Most of these journals are categorized within the Q1 quartile of the JCR rankings, underscoring their significant contributions to the fields of ECMO and cardiogenic shock research. In relation to the H-index, the ASAIO Journal again ranks highest with an H-index of 14, closely followed by Critical Care and the Journal of Thoracic and Cardiovascular Surgery, both of which have an H-index of 13.

The co-occurrence networks of journals encompass 161 publications with at least one occurrence. The three leading journals exhibiting the highest total link strength within these networks are Critical Care Medicine [204], Annals of Thoracic Surgery [195], and ASAIO Journal [192] (Figure 3A). The coupling networks comprise 78 publications with a minimum of two couplings, with the top three journals being ASAIO Journal [6,905], Perfusion-UK [4,785], and the Journal of Heart and Lung Transplantation [4,397] (Figure 3B).

Figure 3 Network analyses of journals in ECMO research for cardiogenic shock. (A) The co-occurrence networks of journals. The frequency with which journals are cited together within the same articles reflects thematic or topical connections between the research they publish. (B) The coupling networks of journals. The extent to which journals are linked is based on common references cited in their articles, indicating a shared intellectual foundation or research focus. (C) Author collaboration network in ECMO research for cardiogenic shock. ECMO, extracorporeal membrane oxygenation.

Analysis of authors

Table S4 presents a summary of the top 20 most prolific authors, detailing their years of publication, total number of publications (ranked), TC (ranked), and H-index. Collectively, these authors have produced 310 articles that have accrued a total of 21,384 citations. Alain Combes occupies the leading position with 30 publications, followed by Guillaume Lebreton with 25 and Pascal Leprince with 23. In terms of TC, Combes ranks first with 2,735 citations, followed by Leprince with 2,114 and Matthieu Schmidt with 1,839. Combes also maintains the highest H-index, with a score of 20, while Leprince and Schmidt follow closely with H-indices of 18 and 16, respectively.

To examine collaboration among authors, we constructed an author collaboration network utilizing VOSviewer (Figure 3C). Among the 270 authors engaged in international collaborations with at least three articles, Alain Combes demonstrates the highest number of collaborative efforts at 212, followed by Guillaume Lebreton with 165 and Charles-Edouard Luyt with 160.

Analysis of keywords and burst keywords

The analysis conducted using VOSviewer identified 124 keywords, each occurring at least five times. The diagram highlights core keywords such as “life support” and “outcomes”, which are represented by larger nodes, indicating their high frequency and strong associations with other keywords. A significant network of terms relevant to critical care, including “cardiac arrest”, “refractory cardiogenic shock”, and “acute myocardial infarction”, underscores the importance of life-saving interventions in this field. The color gradient of the keywords reveals that more recent terms, such as “ventricular assist device” and “membrane oxygenation” (depicted in yellow), have garnered increased attention in recent years. In contrast, older terms such as “cardiac surgery” and “circulatory support” (shown in blue) were more prevalent in earlier research studies (Figure 4A). Figure 4B illustrates the top 20 keywords exhibiting the most significant citation bursts from 1988 to 2024. The red line segments denote specific periods of citation bursts, whereas the blue lines indicate the broader temporal relevance of these keywords. “Experience” demonstrates the highest burst strength at 10.69, followed by “bridge” at 11.85 and “ventricular assist device” at 6.45. Initial bursts were documented for keywords such as “circulatory support” (1992–2014), “experience” (1995–2017), and “failure” (2003–2010). Recent bursts, extending through 2024, have been observed for keywords such as “hospital cardiac arrest”, “guidelines”, and “veno-arterial extracorporeal membrane oxygenation”, suggesting a sustained research interest in these domains.

Figure 4 Keyword analysis in ECMO research for cardiogenic shock. (A) Keyword co-occurrence network, illustrating major research themes and their interconnections within the field. (B) Citation burst analysis of keywords, highlighting emerging and declining research trends over time. ECMO, extracorporeal membrane oxygenation.

External validation of SAVE score in the MIMIC-IV VA-ECMO cohort

A total of 101 adult patients with refractory cardiogenic shock treated with VA-ECMO were included in the MIMIC-IV validation cohort, with an in-hospital mortality of 54.5% (n=55). Baseline characteristics are summarized in Table S5. Compared to survivors, non-survivors were less likely to be male (54.5% vs. 82.6%, P=0.005) and had significantly higher lactate concentrations at admission [8.10 (IQR, 4.77–12.22) vs. 3.90 (IQR, 1.90–6.30) mmol/L, P<0.001]. There were no statistically significant differences in age, SAVE score, or other key clinical variables between groups.

Comparative performance of SAVE, ENCOURAGE, and SOFA scores

To rigorously contextualize the prognostic value of the SAVE score, we performed a head-to-head comparison with the ENCOURAGE and SOFA scoring systems within the validation cohort. The baseline comparison revealed that non-survivors presented with significantly higher ENCOURAGE scores compared to survivors [12.0 (IQR, 8.0–18.0) vs. 7.5 (IQR, 5.0–10.0), P<0.001] (Table S5). Regarding predictive metrics, the ENCOURAGE score achieved a substantially higher discrimination with an AUC-ROC of 0.716 (95% CI: 0.617–0.815), outperforming both the SAVE score (AUC: 0.578, 95% CI: 0.460–0.695) and the SOFA score (AUC: 0.597, 95% CI: 0.483–0.711) (Table S6). Furthermore, calibration plots (Figure 5A) and the Hosmer-Lemeshow test confirmed that while the SAVE (χ²=25.54, P=0.001) and SOFA (χ²=17.47, P=0.03) scores were severely miscalibrated, the ENCOURAGE score exhibited robust calibration accuracy (χ²=10.27, P=0.25). Decision curve analysis (Figure 5B) systematically corroborated these findings, demonstrating that the ENCOURAGE score yields the highest net clinical benefit across a broad spectrum of threshold probabilities.

Figure 5 External validation and comparative performance of the three prognostic scoring systems in the MIMIC-IV cohort. (A) Calibration curve for the SAVE score (coral line), ENCOURAGE score (blue line), and SOFA score (green line) in predicting in-hospital mortality. The dashed diagonal line indicates perfect calibration. The ENCOURAGE score demonstrating robust calibration accuracy (P=0.25), whereas both the SAVE (P=0.001) and SOFA (P=0.03) scores show significant systematic miscalibration. (B) Decision curve analysis for the three scoring systems compared with the “treat all” (dashed grey line) and “treat none” (solid black line) strategies, revealing that the ENCOURAGE score yields the highest net clinical benefit across a wide range of threshold probabilities. AMI, acute myocardial infarction; ECMO, extracorporeal membrane oxygenation; ENCOURAGE, prEdictioN of Cardiogenic shock OUtcome foR AMI patients salvaGed by VA-ECMO; MIMIC-IV, Medical Information Mart for Intensive Care IV; SAVE, Survival After Veno-Arterial ECMO; SOFA, Sequential Organ Failure Assessment; VA-ECMO, veno-arterial ECMO.

Discussion

The bibliometric analysis provides a comprehensive overview of global research trends and developments in ECMO for cardiogenic shock from 1988 to 2024. The analysis highlights the transition towards optimizing ECMO management strategies, improving patient outcomes, and establishing standardized guidelines, which reflects the escalating complexity and significance of ECMO in the treatment of cardiogenic shock. Importantly, by integrating real-world data from the MIMIC-IV database, our study further demonstrates that while established risk prediction tools such as the SAVE score are widely used, their performance in diverse clinical populations may be limited. In our external validation cohort, the SAVE score showed only modest discriminative ability and significant miscalibration, underscoring the necessity of ongoing model validation and adaptation in real-world practice. Together, these findings emphasize the need for both continued research into ECMO management and rigorous external validation of prognostic tools to optimize outcomes for patients with refractory cardiogenic shock.

Geographical distribution, institutional contributions, and key intellectual networks

The global landscape of ECMO research for refractory cardiogenic shock from 1988 to 2024 is predominantly shaped by a highly consolidated network of specific countries, institutions, and clinical investigators. Geographically, the field is led by the USA, France, and China, with the USA demonstrating the most profound academic footprint in terms of both cumulative publication volume (129 articles) and absolute citation impact (4,453 citations) (18). While major French research bodies like APHP and INSERM emerge as frontline institutional contributors, international collaborative dynamics vary markedly; countries such as the Netherlands and Australia exhibit robust multi-country integration with exceptionally high MCP ratios, whereas China and South Korea remain largely focused on localized, single-country research outputs. This intellectual output is disseminated through targeted critical care platforms, among which the ASAIO Journal maintains leading positions in both TP and TC. The journal’s historical focus heavily reflects contemporary technical domains, specifically expanding on clinical outcomes, specialized cannulation configurations, and left ventricular unloading strategies to mitigate complications and improve patient prognosis (19-22). At the individual level, this research trajectory is anchored by prolific investigators such as Alain Combes, Pascal Leprince, and Guillaume Lebreton. Combes, in particular, maintains the highest H-index (20) and citation metrics within the domain, cementing an international collaborative network that drives the ongoing transition toward standardized protocol optimization, multidisciplinary critical care paradigms, and contemporary global risk-stratification practices (23,24).

Research hotspots and frontiers

Before 2016, research on ECMO in cardiogenic shock primarily focused on keywords such as “membrane oxygenation”, “support”, and “cardiac arrest”. This emphasis indicates that early studies aimed to understand the fundamental life-support mechanisms of ECMO, particularly in managing critical conditions like cardiac arrest and circulatory failure. During this period, researchers investigated how ECMO could provide emergency life support for patients experiencing refractory cardiogenic shock and acute cardiac arrest. Li and colleagues (25) conducted a retrospective observational study that found early lactate dynamics, especially lactate clearance, are strongly associated with in-hospital mortality in postcardiotomy patients receiving ECMO support and can predict successful weaning from ECMO. Pöss and colleagues (26) performed a first-in-man study demonstrating that the i-cor VA-ECMO assist device effectively treats patients with refractory cardiogenic shock, resulting in significant hemodynamic stabilization and reductions in both vasopressor and lactate levels. Loforte and his team (27) concluded that veno-arterial ECMO support using RotaFlow and CentriMag systems benefits patients with refractory cardiogenic shock, identifying blood lactate level, CK-MB relative index, and PRBC transfusions as significant predictors of mortality. Despite these advancements, a notable gap during this period was the lack of research on long-term outcomes. Furthermore, the optimization of patient selection criteria for ECMO use remained underdeveloped, which limited its broader application across diverse clinical settings.

Between 2016 and 2020, the research focus shifted towards keywords like “ventricular assist device”, “acute myocardial infarction”, and “outcomes”. This change reflected a growing emphasis on refining the clinical application of ECMO, particularly for patients experiencing AMI and those in need of ventricular support. During this period, greater attention was directed towards improving survival rates, understanding ECMO-related complications, and incorporating advanced mechanical circulatory support devices such as VADs. Fernandes and colleagues (28) conducted a perioperative case study, concluding that argatroban can be used as an anticoagulant during left ventricle assist device (LVAD) implantation in patients with heparin-induced thrombocytopenia. However, achieving adequate anticoagulation and managing complications such as clot formation and severe coagulopathy can be challenging. Gennari and colleagues (29) presented a case study indicating that late thrombosis of a transcatheter aortic valve requires proactive management, as reactive measures may lead to severe complications, including myocardial ischemia and cardiogenic shock. Ariza-Solé and colleagues (30) conducted a retrospective observational study demonstrating that the use of short-term ventricular assist devices (ST-VADs) has increased in high-risk post-infarction ventricular septal rupture (PIVSR) patients. They suggested that early preoperative VA-ECMO should be considered, as it may enhance outcomes in extremely high-risk cases. Despite these advancements in expanding indications for ECMO and managing outcomes, challenges remained in reducing ECMO-related complications such as thromboembolism and bleeding, as well as in developing standardized protocols for ECMO weaning and long-term patient management.

Based on the analysis of keyword bursts, future research trends in ECMO for cardiogenic shock after 2020 can be summarized as focusing on several key areas. Since 2020, terms such as “predicting survival”, “guidelines”, and “veno-arterial extracorporeal membrane oxygenation” have gained prominence. This trend indicates that recent and ongoing research increasingly seeks to improve patient outcomes by developing models and criteria to predict survival in ECMO-treated patients, particularly those on veno-arterial ECMO for cardiogenic shock. Mo and colleagues (31) conducted a retrospective analysis, concluding that predictive models based on BP neural networks, random forests, and decision trees show strong predictive performance for the prognosis of elderly patients with cardiogenic shock after ECMO. Notably, the random forest model exhibited the highest accuracy and clinical application value. Khodaee and colleagues (32) conducted a computational study demonstrating that patient-specific vascular anatomy significantly affects the hemodynamics of ECMO support, particularly influencing the location and extent of the watershed region in the aorta, which has ramifications for differential hypoxia and thromboembolic risk. Furthermore, the emergence of “guidelines” reflects efforts to standardize ECMO use across institutions, ensuring consistency in protocols, indications, and patient management strategies.

From 2021 to 2024, there has been a growing focus on hospital cardiac arrest, indicating increasing attention to ECMO’s use in emergency situations within hospital settings. ECMO is being evaluated more frequently as a life-saving measure for patients experiencing in-hospital cardiac arrest, especially when traditional resuscitation efforts fail. Research aims to optimize the timing of ECMO initiation, reduce complications, and improve survival rates. Benseghir and colleagues (33) conducted a retrospective analysis, finding that male gender, type of intervention, and occurrence of cardiac arrest were significantly associated with higher mortality in patients receiving ECMO for post-cardiotomy refractory cardiogenic shock. Additionally, Yang and colleagues (34) performed a national study with propensity-matched analysis, concluding that among ST-segment elevation myocardial infarction (STEMI) patients complicated by cardiogenic shock undergoing primary percutaneous coronary intervention (PCI), ECMO was associated with better in-hospital survival compared to intra-aortic balloon pump (IABP).

In addition to the bibliometric perspective, our study provides a crucial real-world validation of the SAVE score using the MIMIC-IV database, thereby bridging the gap between bibliometric trends and actual clinical outcomes. The MIMIC-IV cohort analysis revealed that the SAVE score demonstrated only limited discrimination for predicting in-hospital mortality among patients with refractory cardiogenic shock undergoing VA-ECMO. This is notably lower than the moderate to good discrimination reported in some original development and single-center validation studies, suggesting that the SAVE score may not generalize well to diverse, real-world North American populations represented in MIMIC-IV. Moreover, the calibration curve showed significant miscalibration, with the model systematically over- or underestimating actual risk across different risk strata. Decision curve analysis further indicated that the net clinical benefit of the SAVE score was limited in this cohort, with no clear advantage over “treat all” or “treat none” strategies across a range of threshold probabilities.

Our expanding comparative analysis yielded a critical clinical insight: while the SAVE score demonstrated limited prognostic accuracy, the ENCOURAGE score demonstrated superior discrimination and excellent calibration within the same MIMIC-IV cohort. This disparity in transportability is likely rooted in the specific structural variables of each tool. The ENCOURAGE score weighs metabolic and shock severity markers heavily, such as lactate kinetics and intensive organ parameters, which align perfectly with the high proportion of profound metabolic derangements observed in our non-survivor cohort (e.g., markedly elevated baseline lactate levels). This indicates that although general ECMO tools like the SAVE score require cautious application or localized adaptation, alternative contemporary scoring frameworks like ENCOURAGE might be inherently more optimal for real-world risk stratification in specific refractory cardiogenic shock sub-populations.

The evolving paradigm of VA-ECMO management in cardiogenic shock

The transition of keyword clusters over the decades in our bibliometric analysis is not merely bibliographical; it mirrors a profound paradigm shift in real-world critical care medicine. Modern refractory cardiogenic shock management has evolved from a reactive, rescue-centric application of VA-ECMO toward highly proactive, structured, and system-wide protocols. A cornerstone of this evolution is the widespread implementation of specialized “shock-team” models. These multidisciplinary collaborative frameworks ensure rapid, standardized patient screening and optimal timing for ECMO initiation, mitigating the delays and unmeasured selection biases often native to retrospective cohorts like the one examined in this study (35-37). Parallel to institutional system optimization, the mechanical management of the supported heart itself has undergone rigorous refinement, particularly regarding left ventricular unloading approaches. Unloading the left ventricular during VA-ECMO is vital to counteract increased afterload, prevent pulmonary edema, and promote myocardial recovery. In our MIMIC-IV cohort, a substantial proportion of patients (68.3%) received concomitant IABP therapy as an unloading strategy, reflecting traditional combination protocols. However, contemporary trends highlighted in our burst keyword analysis indicate an accelerating shift toward more potent active venting strategies, such as the “ECPELLA” approach (combining VA-ECMO with the Impella axial flow pump), which provides superior hemodynamic stabilization and active left ventricular decompression compared to IABP alone (38,39). Ultimately, optimizing these advanced mechanical strategies relies heavily on precision-risk stratification. Our comparative findings demonstrate that general, static scores like the SAVE score may underperform in localized North American settings, whereas specific tools like the ENCOURAGE score—which is finely tuned to the acute metabolic and organ dysfunction dynamics of specific shock sub-populations—offer markedly superior discrimination (AUC 0.716) and calibration (40). Moving forward, the field must transcend historical descriptive epidemiology; future efforts should focus on integrating multi-center clinical networks and machine learning frameworks to dynamically risk-stratify patients, thereby tailoring both the timing of initiation and specific unloading modalities to the individual patient’s precise physiological phenotype (41-43).

Strengths and limitations

These findings highlight several important considerations for both clinical practice and future research. The necessity of external validation using large, heterogeneous databases such as MIMIC-IV is underscored, as risk prediction tools like the SAVE score may perform differently in new populations or settings due to differences in patient characteristics, management protocols, and case-mix. Our results emphasize that risk scores derived from highly selected populations may require recalibration or even redevelopment for optimal utility in real-world scenarios. The robust data integrity and sufficient sample size of the MIMIC-IV cohort affirm the critical value of open-access critical care datasets for model assessment and hypothesis generation in ECMO research. However, limitations such as the retrospective, single-center design and the potential for unmeasured confounding must be acknowledged for our MIMIC-IV analysis. Similarly, from a bibliometric perspective, our study is the first to comprehensively investigate global research trends and key focuses on ECMO for cardiogenic shock, leveraging a multi-year analysis to identify relevant literature. Nevertheless, as with other bibliometric studies, our analysis is subject to potential biases, including reliance on citation counts, which may not fully capture clinical impact, and the exclusion of non-English publications, which could limit comprehensiveness. By integrating large-scale bibliometric analysis with MIMIC-IV-based external validation, our study not only maps the research landscape but also provides a pragmatic assessment of clinical tools in contemporary practice. However, these findings should be interpreted with caution due to the single-center nature and modest sample size (n=101) of our validation cohort, which may limit the statistical robustness of the calibration and discrimination curves. Rather than suggesting a broad invalidation of the SAVE score, our results imply that its transportability may vary across different healthcare systems and case-mixes, underscoring the value of local recalibration. This dual approach highlights the need for ongoing model evaluation and adaptation as patient populations and standards of care evolve, supporting the advancement of precision medicine in the management of refractory cardiogenic shock.

Conclusions

Our combined bibliometric analysis and external validation using the MIMIC-IV database highlight both the progress and ongoing challenges in ECMO management for cardiogenic shock. Although our findings suggest limited discrimination and potential miscalibration of the SAVE score within this specific cohort, these results should be interpreted with caution given the single-center nature and modest sample size (n=101) of the validation set. Future research should not only focus on optimizing patient outcomes and minimizing complications but also prioritize the rigorous external validation and recalibration of predictive models in diverse, real-world populations. Key priorities include the development of standardized guidelines for ECMO use, refinement of patient selection criteria, and the improvement of survival prediction tools that are generalizable across different clinical settings. Continued efforts to address common complications—such as thromboembolism, bleeding, and infection—remain essential. Furthermore, expanding collaborative, multicenter research and leveraging large, open-access critical care datasets like MIMIC-IV will be vital for advancing evidence-based ECMO management and supporting precision medicine in the care of patients with refractory cardiogenic shock.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the BIBLIO reporting checklist. Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-0964/rc

Peer Review File: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-0964/prf

Funding: The study was supported by the Program of Yunnan Applied Basic Research Project-Kunming Medical University Union Foundation (No. 202501AY070001-234).

Conflicts of Interest: Both authors have completed the ICMJE uniform disclosure form (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-0964/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Laghlam D, Benghanem S, Ortuno S, et al. Management of cardiogenic shock: a narrative review. Ann Intensive Care 2024;14:45. [Crossref] [PubMed]
2. Shirakabe A, Matsushita M, Shibata Y, et al. Organ dysfunction, injury, and failure in cardiogenic shock. J Intensive Care 2023;11:26. [Crossref] [PubMed]
3. Martínez León A, Bazal Chacón P, Herrador Galindo L, et al. Review of Advancements in Managing Cardiogenic Shock: From Emergency Care Protocols to Long-Term Therapeutic Strategies. J Clin Med 2024;13:4841. [Crossref] [PubMed]
4. Osman M, Syed M, Patibandla S, et al. Fifteen-Year Trends in Incidence of Cardiogenic Shock Hospitalization and In-Hospital Mortality in the United States. J Am Heart Assoc 2021;10:e021061. [Crossref] [PubMed]
5. Senman B, Jentzer JC, Barnett CF, et al. Need for a Cardiogenic Shock Team Collaborative-Promoting a Team-Based Model of Care to Improve Outcomes and Identify Best Practices. J Am Heart Assoc 2024;13:e031979. [Crossref] [PubMed]
6. Geetha S 3rd, Verma N, Chakole V. A Comprehensive Review of Extra Corporeal Membrane Oxygenation: The Lifeline in Critical Moments. Cureus 2024;16:e53275. [Crossref] [PubMed]
7. Mastoris I, Tonna JE, Hu J, et al. Use of Extracorporeal Membrane Oxygenation as Bridge to Replacement Therapies in Cardiogenic Shock: Insights From the Extracorporeal Life Support Organization. Circ Heart Fail 2022;15:e008777. [Crossref] [PubMed]
8. Perry T, Brown T, Misfeldt A, et al. Extracorporeal Membrane Oxygenation in Congenital Heart Disease. Children (Basel) 2022;9:380. [Crossref] [PubMed]
9. Rajsic S, Breitkopf R, Rugg C, et al. Thrombotic Events Develop in 1 Out of 5 Patients Receiving ECMO Support: An 11-Year Referral Centre Experience. J Clin Med 2023;12:1082. [Crossref] [PubMed]
10. Jung JH, Kwon HJ, Kim TJ, et al. Immunohistochemical expression of p53, Bcl-2, and Ki-67 proteins in traditional serrated adenomas of colon. Korean J Gastroenterol 2013;62:336-43. [Crossref] [PubMed]
11. Leng A, Bachina P, Liu O, et al. Enhancing Survival Prediction After Venoarterial Extracorporeal Membrane Oxygenation Using Machine Learning. ASAIO J 2026;72:376-83. [Crossref] [PubMed]
12. Jangid H, Kumar D, Kumar G, et al. Bibliometric Examination of Global Scientific Research about Carbapenem-Resistant Acinetobacter Baumannii (CRAB). Antibiotics (Basel) 2023;12:1593. [Crossref] [PubMed]
13. Bertoli-Barsotti L, Lando T. A theoretical model of the relationship between the h-index and other simple citation indicators. Scientometrics 2017;111:1415-48. [Crossref] [PubMed]
14. Hirsch JE. An index to quantify an individual's scientific research output. Proc Natl Acad Sci U S A 2005;102:16569-72. [Crossref] [PubMed]
15. Schmidt M, Burrell A, Roberts L, et al. Predicting survival after ECMO for refractory cardiogenic shock: the survival after veno-arterial-ECMO (SAVE)-score. Eur Heart J 2015;36:2246-56. [Crossref] [PubMed]
16. Crespo-Leiro MG, Metra M, Lund LH, et al. Advanced heart failure: a position statement of the Heart Failure Association of the European Society of Cardiology. Eur J Heart Fail 2018;20:1505-35. [Crossref] [PubMed]
17. Combes A, Leprince P, Luyt CE, et al. Outcomes and long-term quality-of-life of patients supported by extracorporeal membrane oxygenation for refractory cardiogenic shock. Crit Care Med 2008;36:1404-11. [Crossref] [PubMed]
18. Moyon Q, Combes A, Stern J, et al. Validation of predictive models for 90-day survival in immunocompromised patients with cardiogenic shock on venoarterial extracorporeal membrane oxygenation. Intensive Care Med 2024;50:1172-4. [Crossref] [PubMed]
19. Guglin M. Randomized Clinical Trials on Venoarterial Extracorporeal Membrane Oxygenation in Cardiogenic Shock: Why Are They All Negative? ASAIO J 2025;71:e136-40. [Crossref] [PubMed]
20. Kalra A, Kang JK, Wilcox C, et al. Pulse Pressure and Acute Brain Injury in Venoarterial Extracorporeal Membrane Oxygenation: An Extracorporeal Life Support Organization Registry Analysis. ASAIO J 2025;71:99-108. [Crossref] [PubMed]
21. Shugh SB, Tume SC, Bansal N, et al. Transcatheter Axial Pump Use in Pediatric Patients on Veno-Arterial Extracorporeal Membrane Oxygenation: An ACTION Collaborative Experience. ASAIO J 2024;70:602-8. [Crossref] [PubMed]
22. Cappelli J, Emling J, Edwards A, et al. Direct Apical Cannulation With Protek Duo Rapid Deployment Cannula via Mini Thoracotomy for Ambulatory Venoarterial-Extracorporeal Membrane Oxygenation. ASAIO J 2024;70:565-9. [Crossref] [PubMed]
23. Danial P, Zamorano C, Carillion A, et al. Incidence and outcomes of prosthetic valve thrombosis during peripheral extracorporeal membrane oxygenation. Eur J Cardiothorac Surg 2024;66:ezae321. [Crossref] [PubMed]
24. Moyon Q, Triboulet F, Reuter J, et al. Venoarterial extracorporeal membrane oxygenation in immunocompromised patients with cardiogenic shock: a cohort study and propensity-weighted analysis. Intensive Care Med 2024;50:406-17. [Crossref] [PubMed]
25. Li CL, Wang H, Jia M, et al. The early dynamic behavior of lactate is linked to mortality in postcardiotomy patients with extracorporeal membrane oxygenation support: A retrospective observational study. J Thorac Cardiovasc Surg 2015;149:1445-50. [Crossref] [PubMed]
26. Pöss J, Kriechbaum S, Ewen S, et al. First-in-man analysis of the i-cor assist device in patients with cardiogenic shock. Eur Heart J Acute Cardiovasc Care 2015;4:475-81. [Crossref] [PubMed]
27. Loforte A, Pilato E, Martin Suarez S, et al. RotaFlow and CentriMag extracorporeal membrane oxygenation support systems as treatment strategies for refractory cardiogenic shock. J Card Surg 2015;30:201-8. [Crossref] [PubMed]
28. Fernandes P, O'Neil M, Del Valle S, et al. A 24-hour perioperative case study on argatroban use for left ventricle assist device insertion during cardiopulmonary bypass and veno-arterial extracorporeal membrane oxygenation. Perfusion 2019;34:337-44. [Crossref] [PubMed]
29. Gennari M, Polvani G, Pepi M, et al. Late thrombosis of a Transcatheter aortic valve: the border between a proactive and reactive management. J Cardiothorac Surg 2018;13:126. [Crossref] [PubMed]
30. Ariza-Solé A, Sánchez-Salado JC, Sbraga F, et al. The role of perioperative cardiorespiratory support in post infarction ventricular septal rupture-related cardiogenic shock. Eur Heart J Acute Cardiovasc Care 2020;9:128-37. [Crossref] [PubMed]
31. Mo Z, Lu Z, Tang X, et al. Construction and evaluation of prognostic models of ECMO in elderly patients with cardiogenic shock based on BP neural network, random forest, and decision tree. Am J Transl Res 2023;15:4639-48.
32. Khodaee F, Nezami FR, Zampell BA, et al. Effect of anatomical variation on extracorporeal membrane oxygenation circulatory support: A computational study. Comput Biol Med 2022;141:105178. [Crossref] [PubMed]
33. Benseghir Y, Sebestyan A, Durand M, et al. ECMO for post cardiotomy refractory cardiogenic shock: Experience of the cardiac surgery department of the Grenoble Alpes University Hospital. Ann Cardiol Angeiol (Paris) 2021;70:63-7. [Crossref] [PubMed]
34. Yang H, Luo L, Song Y, et al. ECMO versus IABP for patients with STEMI complicated by cardiogenic shock undergoing primary PCI: a Chinese National Study and propensity-matched analysis. Hellenic J Cardiol 2026;88:4-12. [Crossref] [PubMed]
35. Tehrani BN, Truesdell AG, Psotka MA, et al. A Standardized and Comprehensive Approach to the Management of Cardiogenic Shock. JACC Heart Fail 2020;8:879-91. [Crossref] [PubMed]
36. Tehrani BN, Truesdell AG, Sherwood MW, et al. Standardized Team-Based Care for Cardiogenic Shock. J Am Coll Cardiol 2019;73:1659-69. [Crossref] [PubMed]
37. Truesdell AG, Tehrani B, Singh R, et al. 'Combat' Approach to Cardiogenic Shock. Interv Cardiol 2018;13:81-6. [Crossref] [PubMed]
38. Schrage B, Becher PM, Bernhardt A, et al. Left Ventricular Unloading Is Associated With Lower Mortality in Patients With Cardiogenic Shock Treated With Venoarterial Extracorporeal Membrane Oxygenation: Results From an International, Multicenter Cohort Study. Circulation 2020;142:2095-106. [Crossref] [PubMed]
39. Gandhi KD, Moras EC, Niroula S, et al. Left Ventricular Unloading With Impella Versus IABP in Patients With VA-ECMO: A Systematic Review and Meta-Analysis. Am J Cardiol 2023;208:53-9. [Crossref] [PubMed]
40. Schrutka L, Rohmann F, Binder C, et al. Discriminatory power of scoring systems for outcome prediction in patients with extracorporeal membrane oxygenation following cardiovascular surgery†. Eur J Cardiothorac Surg 2019;56:534-40. [Crossref] [PubMed]
41. Gao H, Huang X, Zhou K, et al. Development and validation of a machine learning model for predicting mortality risk in veno-arterial extracorporeal membrane oxygenation patients. Sci Rep 2025;15:41581. [Crossref] [PubMed]
42. Braun J, Sahli SD, Spahn DR, et al. Predicting Survival for Veno-Arterial ECMO Using Conditional Inference Trees-A Multicenter Study. J Clin Med 2023;12:6243. [Crossref] [PubMed]
43. Kalra A, Bachina P, Shou BL, et al. Predicting Acute Brain Injury in Venoarterial Extracorporeal Membrane Oxygenation Patients with Tree-Based Machine Learning: Analysis of the Extracorporeal Life Support Organization Registry. Res Sq 2024;rs.3.rs-3848514.