Methodological Issues Regarding the ECMO Mortality Prediction Model: A Critical Appraisal [Letter]
Journal of Inflammation Research · 2026-08 · Volume 19 · 1-3
DOI: 10.2147/jir.s643302 ↗Kurz gefasst
Leserbrief in der Zeitschrift Journal of Inflammation Research, 2026-08. Verfasst von Qingqing Shan. Beteiligte Einrichtung: Department of Respiration, Chengdu Integrated TCM and Western Medicine Hospital, Chengdu, 610095, People's Republic of China. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 3.0).
Abstract (Originalsprache)
Membrane Oxygenation: A Machine Learning-Based Nomogram Study", recently published in the Journal of Inflammation Research. 1 The authors are to be commended for their efforts to develop a practical risk stratification tool for ECMO patients using machine learning approaches.The identification of novel composite biomarkers such as MWR and AAR serves as an innovative direction worthy of further study.However, we wish to raise several methodological concerns that merit close consideration, as they may affect the clinical meaningfulness and generalizability of the proposed nomogram.First and foremost, the study combined VA-ECMO (76.23%) and VV-ECMO (23.77%) patients within a single prediction model.This is a critical issue from both somatic and clinical standpoints.VA-ECMO provides predominantly hemodynamic support for patients with cardiogenic shock, whereas VV-ECMO mainly provides respiratory support for patients with refractory hypoxemia.2 The pathologic mechanisms underlying lactate elevation differ substantially between these two modalities: in VA-ECMO, hyperlactatemia primarily reflects inadequate cardiac output and peripheral hypoperfusion, while in VV-ECMO, it more commonly indicates severe tissue hypoxia from underlying respiratory failure.3 Furthermore, coagulation profiles-including APTT responses-vary considerably between VA-ECMO (associated with consumption coagulopathy and heparin-induced effects) and VV-ECMO (where pulmonary-derived coagulation abnormalities predominate).4 The authors themselves acknowledge in the Results section that "VA-ECMO and VV-ECMO patients differ in etiology, immune status, metabolism, and coagulation."Despite this recognition, the decision to pool these two distinct populations for model development appears inconsistent with established clinical reasoning.A stratified modeling approach or at least a sensitivity analysis examining the model's performance separately in each ECMO subtype would have substantially enhanced the clinical relevance of this work.Second, we observed an unusual pattern in the model results metrics that warrants explanation.The logistic regression model demonstrated a training set AUC of 0.765 (95% CI: 0.667-0.863)but achieved a substantially higher test set AUC of 0.891 (95% CI: 0.761-1.000).In conventional model development, test set performance typically falls below or approximates training set performance, as models are optimized on training data.This inversion of performance measures-particularly the wide confidence interval in the test set (spanning from 0.761 to 1.000)-strongly suggests that the test cohort of only 24 patients may be too small to provide stable performance estimates, and the observed superiority may reflect random sampling variation as opposed to genuine model applicability.The authors did not explain this surprising finding, nor did they conduct a sensitivity analysis with alternative train-test splits to examine the robustness of their results.Third, the candidate variables included in this study were limited to baseline laboratory parameters obtained within 24 hours before ECMO initiation.However, emerging evidence indicates that dynamic changes during ECMO supportparticularly lactate clearance at 24, 48, and 72 hours-are more powerful predictors of mortality than single baseline
Autor:innen und Institutionen
- Department of Respiration, Chengdu Integrated TCM and Western Medicine Hospital, Chengdu, 610095, People's Republic of China.
ORCID-Profile: Qingqing Shan
Zitieren
APA
Shan, Q. (2026). Methodological Issues Regarding the ECMO Mortality Prediction Model: A Critical Appraisal [Letter]. Journal of Inflammation Research, Volume 19, 1-3. https://doi.org/10.2147/jir.s643302
Vancouver
Shan Q. Methodological Issues Regarding the ECMO Mortality Prediction Model: A Critical Appraisal [Letter]. Journal of Inflammation Research. 2026;Volume 19:1-3. doi:10.2147/jir.s643302
BibTeX
@article{Shan2026,
author = {Shan, Qingqing},
title = {Methodological Issues Regarding the ECMO Mortality Prediction Model: A Critical Appraisal [Letter]},
journal = {Journal of Inflammation Research},
year = {2026},
volume = {Volume 19},
pages = {1--3},
doi = {10.2147/jir.s643302}
}
EUVIMED