METHODS
STUDY DESIGN, SETTING AND PARTICIPANTS
The ECMO NEXT study will be a multicenter, retrospective cohort study designed to evaluate real-world practices and outcomes of ECMO in critically ill patients in Japan. This protocol was drafted according to the Standardized Protocol Items: Recommendations for Observational Studies (SPIROS) statement to ensure methodological rigor and transparency (Supplementary Material 1)18). Detailed methods are described in Supplementary Material 2. The recruitment of participating sites and investigators was facilitated by a nationwide network of early career multidisciplinary critical care health professionals with additional support from institutions affiliated with the Japanese Society of Education for Physicians and Trainees in Intensive Care (JSEPTIC) Clinical Trials Group19),20). The participating sites included 12 university hospitals, one national hospital, and nine community hospitals. Currently, 22 centers have confirmed their participation (Fig. 1 and Supplementary Material 2), with the possibility of additional centers participating during the study period.
The cohort will include patients treated between January 1, 2018, and December 31, 2023. All patients aged ≥18 years who underwent ECMO management in the participating ICUs, defined as beds qualifying for ICU management fee 1–4 under Japan’s national health insurance system during this period, are eligible for inclusion. More specifically, patients who died before ICU admission or were weaned from ECMO before ICU admission are not eligible for enrollment. Additionally, ECMO cases managed exclusively in the operating room or catheterization laboratory without ICU admission will not be included in this study. Although currently the cohort has no exclusion criteria, these criteria may be defined exclusions may be defined in the future for specific research topics to support the targeted analyses. To ensure data accuracy and integrity, duplicate entries from inter-hospital transfers will be identified and manually excluded during data cleaning. Only the first ECMO session will be included as the starting point of observation for patients who underwent multiple ECMO sessions within the same hospitalization period. Each ECMO episode in separate hospitalizations will be treated independently.
KEY RESEARCH TOPICS PREDEFINED IN THIS COHORT
This study addresses six predefined research topics to investigate the critical aspects of ECMO management. The first theme focuses on investigating the incidence and clinical impact of fever after ECMO decannulation to elucidate its epidemiology and potential impact on patient outcomes. The second theme examines the epidemiology of infection and antimicrobial use during ECMO therapy, seeking to identify the patterns and factors associated with infection risk and treatment strategies. The third theme focuses on identifying the optimal ventilator settings during ECMO management to minimize ventilator-associated complications and improve patient outcomes. The fourth theme examines the predictors and outcomes of ECMO-associated AKI and electrolyte abnormalities, providing insight into the risk factors and long-term consequences of these complications. The fifth theme focuses on rehabilitation interventions during ECMO and evaluates their effectiveness in promoting recovery and mitigating complications. Finally, the sixth theme examines the epidemiology of veno-arterial (V-A) ECMO use in specific subpopulations, such as patients with toxicological emergencies and septic shock, to better understand its role and outcomes in these critical scenarios.
SAMPLE SIZE AND FEASIBILITY
This study aimed to provide descriptive information on the proportions of outcomes and confidence intervals to inform planning for future interventional trials. Because the prevalence of exposure and outcomes was largely unknown at the outset, an accurate estimation of the required sample size for specific causal associations was challenging21). Therefore, we sought to include the highest possible number of eligible patients during the study period. Preliminary screening suggested that over 2,000 cases were available from the participating sites. The minimum detectable odds ratios under various assumptions for this sample size are shown in Supplementary Material 2. The final number of enrolled patients may surpass the initial projections as additional sites may get included in the study.
ETHICAL APPROVAL AND REGISTRATION
The study protocol was approved by the Ethical Review Committee for Observational Studies of the Nagoya University Hospital (reference number: 2024-0295). Following this central review, approval to initiate the study implementation was obtained from the head of each participating institution. The requirement for individual informed consent was waived owing to the retrospective nature of data collection, and an opt-out method will be used to ensure compliance with ethical standards. The study protocols has been registered in the Japan Registry of Clinical Trials: jRCT1040240168 (https://jrct.niph.go.jp/latest-detail/jRCT1040240168).
DATA COLLECTION
Data were collected using an electronic case report form hosted on the Research Electronic Data Capture (REDCap) platform (REDCap version 13.11.4, Vanderbilt University) managed by Nagoya University22). Details and definitions of the variables collected are provided in the Supplementary Material 2.
The collected data included various parameters to ensure a comprehensive analysis. Baseline characteristics were collected, including age, sex, body mass index, comorbidities, and disease severity scores such as the RESP and SAVE scores, as well as other variables essential for defining and screening study participants in earch research topics23),24). Data related to ECMO implementation included the type of ECMO (V-V, V-A, and other configurations such as V-VA), duration, and key setting parameters. Clinical parameters, such as ventilator settings and vasopressor use before and after ECMO initiation, were also recorded to capture the clinical course.
Use of antimicrobial therapy during the ICU stay was meticulously recorded, detailing the type, dosage, and duration of antifungal, antiviral, and antibacterial agents. Additionally, data on microbiological cultures performed during the ICU stay were also collected, and pathogen identification was performed using antigen tests for Legionella, Streptococcus pneumoniae, SARS-CoV-2, influenza, and Clostridium species. Surveillance screening tests were excluded.
Daily laboratory test results were collected, and values closest to 8:00 AM were used for consistency. For patients in whom ECMO was decannulated, additional data were collected, including body temperature, white blood cell count, timing of fever resolution, and interventions to resolve the fever. Body temperature, including axillary and central measurements, was recorded from one day before to two days after ECMO decannulation.
Specific subpopulations, such as those receiving V-A ECMO for toxicological emergencies or septic shock, underwent tailored data collection items. For patients with plasma-free hemoglobin (PF-Hb) measurement, associated laboratory data were documented within 3 h before and after the PF-Hb measurement. Rehabilitation-related data were collected from 10 institutions, including physical, occupational, and speech therapy during the ICU stay. Information on these interventions’ initiation dates and duration was also recorded to assess their impact on recovery. To evaluate the representativeness of participating centers, we will collect data on hospital characteristics, ICU models, and ECMO case volume.
DEFINITIONS OF MAIN EXPOSURES
Fever following ECMO decannulation was defined as a new onset of body temperature ≥38°C within 48 h after ECMO removal. Sensitivity analyses included fever with leukocytosis and positive blood culture results. As mechanical ventilation generally continues after ECMO decannulation, the systemic inflammatory response syndrome criteria were considered inappropriate for this analysis.
Based on the medical records, infections were categorized as bloodstream infections (including catheter-related infections and infective endocarditis), respiratory infections (further subdivided into ventilator-associated and non-ventilator-associated lower respiratory tract infections), skin and soft tissue infections, abdominal infections, neurological infections, renal and urinary tract infections, and other infections25). Definitions were based on microbiologic criteria. For respiratory infections, endotracheal aspirates required ≥105 colony-forming units/mL, and bronchoalveolar lavage samples required ≥104 colony-forming units /mL26),27). Urinary tract infections were defined by ≥105 colony-forming units /mL for a single organism or ≥103 colony-forming units /mL with supporting evidence such as positive leukocyte esterase or nitrite tests, pyuria (≥10 white blood cells/mm3), or Gram stain evidence of microorganisms28),29). Significant pathogens for ventilator-associated lower respiratory tract infections and bloodstream infections, along with criteria for defining true positive cultures and exclusions, are detailed in Supplementary Material 225),30).
Ventilator settings were documented 6 h after ECMO initiation to capture early management decisions and assess their potential impact on patient outcomes. The incidence and severity of AKI were assessed using the KDIGO classification, based on serum creatinine levels measured daily for 7 d after ECMO initiation, and renal replacement therapy status31). Electrolyte alterations were identified using time-series blood laboratory data collected during the first three days of ECMO initiation and classified using a data-driven approach.
OUTCOMES
Outcomes were defined relative to specific time points (e.g., ECMO initiation or decannulation) tailored to each research topic to allow focused analyses addressing the clinical questions of the study. The primary outcome was weaning failure within 30 days of ECMO decannulation, which was a composite of death, ECMO reinitiation, or bridging to transplantation (left ventricular assist device or heart/lung transplantation). The outcome measures included in-hospital mortality and the longest available follow-up duration, as documented in the electronic health records of participating hospitals. Follow-up data were collected for patients who continued outpatient care at the hospital where ECMO was initiated. If a patient was transferred to another institution, further follow-up data were not obtained. Detailed records were maintained to ensure a comprehensive outcome assessments at the final follow-up points.
Functional outcomes were assessed using standardized scales, including the Johns Hopkins Safe Patient Handling Mobility, Barthel Index, ICU Mobility Scale, Medical Research Council Score, and Clinical Frailty Scale32–35). These assessments were performed at three specific time points: at ICU discharge,at 7 d after ICU discharge, and at hospital discharge, where available. Since this was a retrospective study, data for these assessments were extracted from available rehabilitation and electronic medical records, rather than collected according to a standardized protocol. Renal recovery was assessed at hospital discharge and defined by serum creatinine levels and dialysis dependency36).
COMPUTED TOMOGRAPHY (CT) IMAGING AND QUANTITATIVE ANALYSIS
CT images were collected from each participating institution to evaluate skeletal muscle wasting associated with ECMO management and investigate the impact and standardization of rehabilitation. The targeted imaging regions included the chest and lumbar-pelvic regions. CT scans were retrospectively analyzed based on available imaging data rather than a predefined imaging protocol. When available, CT was conducted at three key time points: before ECMO initiation, after ECMO decannulation, and as close as possible to hospital discharge. All CT images were exported in the DICOM format, anonymized, and assigned unique research identifiers to ensure confidentiality. The de-identified data were then submitted to a central analysis center for evaluation.
ImageJ (NIH, Bethesda, MD, USA) was used to analyze the CT images, focusing on skeletal muscles at the sixth thoracic and third lumbar vertebra levels. The T6 level emphasizes respiratory muscles, including the pectoralis major, pectoralis minor, serratus anterior, intercostal, subscapular, trapezius, and erector spinae muscles (Fig. 2A)37). The L3 level targets primary skeletal muscles such as the erector spinae, multifidus, psoas major, obliques, and rectus abdominis (Fig. 2B)38). Skeletal muscles were identified within a Hounsfield Unit (HU) range of –29 HU to 150 HU. The cross-sectional area at each level was the muscle mass [cm2], and the mean CT value within the region of interest was the muscle quality [HU]. The muscle mass area and mean CT values within the regions of interest were calculated to quantify the skeletal muscle changes.
PRIVACY PROTECTION, DATA MONITORING AND QUALITY CONTROL
All data were de-identified, and unique study-specific IDs were assigned before entry into the REDCap platform. A correspondence list linking these IDs to medical records was securely stored at each participating site to ensure confidentiality while providing traceability for data monitoring purposes. The data security was maintained using robust encryption and restricted access protocols. The study management office conducts systematic checks after data entry is completed at each site to maintain data quality. Any identified errors or inconsistencies were flagged, and queries were resolved in collaboration with the relevant centers. Standardized validation rules embedded in the electronic case report form minimized the risk of incomplete or inconsistent entries, ensuring the reliability and integrity of the data collected.
STATISTICAL ANALYSIS
Descriptive statistics will summarize patient characteristics, ECMO management details, and outcomes. Continuous variables will be presented as medians with interquartile ranges or means with standard deviations, depending on the data distribution. Categorical variables will be expressed as frequencies and percentages. Statistical tests such as the Wilcoxon rank-sum test for continuous variables and chi-squared test for categorical variables will be used for group comparisons. Kaplan–Meier survival curves will be used for time-to-event outcomes, with log-rank tests employed to compare survival distributions between groups.
Advanced statistical models will be selected for specific research topics based on their appropriateness for the research questions and data structure. These may include logistic regression, modified Poisson regression, and Cox proportional hazard models. The authors extensively discussed the selection of potential confounders for adjustment in these models, with expert input to ensure relevance. Variables identified as potential confounders were included during data collection to facilitate accurate and robust adjustments for subsequent analyses. The attributable risk of specific exposures, such as the impact of AKI or post-decannulation fever on mortality, will be assessed using propensity scores with inverse probability weighting methods39). This approach will balance the exposed and unexposed groups and control for measured confounders to provide reliable and unbiased estimates40).
The proportion of missing data will be documented, and appropriate imputation methods will be used to address missingness tailored to the analytical needs of each research question.
DISCUSSION
This study contributes significantly to the understanding and managing of ECMO in critically ill patients, particularly in the Japanese healthcare system. By collecting comprehensive data from 22 institutions, the ECMO NEXT study provided detailed insights into ECMO practices, outcomes, and associated complications. A distinguishing feature of this study is its comprehensive nature, which includes granular data on ICU management, infection control, ventilator settings, detailed renal function trajectories, and rehabilitation practices. These additions address the critical gaps in ECMO research and promote a more comprehensive understanding of patient care and outcomes.
A key strength of this study is its focus on underexplored areas, such as the incidence and impact of fever following ECMO decannulation, as well as the predictors and outcomes of ECMO-associated AKI and electrolyte abnormalities. The emphasis on rehabilitation practices and their role in recovery further broadens the scope of the study and highlights the potential of multidisciplinary approaches to improve patient outcomes. In addition, specific subpopulations, including patients receiving V-A ECMO for toxicological emergencies and septic shock, provide valuable insights into specialized applications of ECMO.
Despite its strengths, the study has limitations that need to be considered. The retrospective design, while allowing the analysis of a large dataset, may introduce bias owing to incomplete or variable data quality across the participating institutions. Standardized data collection and quality control measures reduce but do not eliminate these risk factors. One bias that could not be eliminated was the influence of unmeasured confounders, which may have affect the results. To address this issue, prospective observational studies are needed to collect more detailed variables or interventional studies to establish causal relationships. Additionally, the accurate diagnosis of ECMO-related infections requires standardized diagnostic criteria and specific laboratory testing, which were not uniformly applied across institutions in this retrospective study. Variability in diagnostic thresholds and testing availability may have introduced misclassification bias, potentially leading to an underestimation of the true impact of infections. Future studies with standardized infection surveillance protocols are needed to better characterize the epidemiology of ECMO-related infections. Furthermore, the focus of this study on Japanese healthcare settings may limit the generalizability and transportability of the findings to regions with different ECMO practices, healthcare systems, and patient demographics. More specifically, the participating centers were not randomly sampled, introducing a potential selection bias. In fact, the composition of the participating centers included a higher proportion of university hospitals, which may not fully represent the diversity of ECMO-providing institutions in Japan. Variability in management quality is also expected among the participating centers, possibly influenced by differences in clinical experience, level of intensivist coverage, and temporal changes before and after the COVID-19 pandemic. These factors may contribute to inconsistencies in the results. Expanding global initiatives such as the ELSO registry may help integrate broader data and improve the applicability of research findings worldwide. Rehabilitation-related data were collected from only 10 institutions, which may limit the generalizability of the findings regarding rehabilitation practices and their impact on recovery. Similarly, excluding pediatric patients leaves a significant gap in understanding ECMO outcomes in the younger population. Another limitation is the potential for missing data, particularly regarding long-term follow-up outcomes. To address this limitation, prospective studies are needed to complement the findings of this retrospective study and provide more robust and comprehensive follow-up data.
