In-Hospital Mortality Prediction of Patients Requiring Extracorporeal Membrane Oxygenation Using Composite Lactate Metrics and Weight of Evidence Modeling

BACKGROUND: Mortality prognostication in adult patients requiring extracorporeal membrane oxygenation (ECMO) is not accurate or established. We hypothesized that composite lactate-based metrics and machine-learning (ML) models would improve in-hospital mortality prognostication compared to lactate alone in this population.

STUDY DESIGN: We conducted a retrospective study of adult patients supported with ECMO at a Level I trauma center from 2022-2024 (N=104). Composite metrics were generated by multiplying lactate by acid-base markers and comparing performance to lactate after Benjamini-Hochberg correction to control the false discovery rate (FDR). For ML models, 3 multilayer perceptron (MLP) architectures were developed using lab markers from a patient's hospital admission. Feature attribution through local interpretable model-agnostic explanations (LIME) guided development of a weight-of-evidence (WoE) model using only initial laboratory values.

RESULTS: The composite metric of average arterial lactate × average arterial bicarbonate was the most significantly elevated marker in non-survivors and more significant than lactate alone (FDR -adjusted p = 4.49 x 10-4 vs. 2.14 x 10-3). 100x50 MLP architecture achieved 85% accuracy (CI: 80 - 89%), 87% precision (CI: 82 - 92%), F1 of 0.81 (CI: 0.74 - 0.86) and AUC of 0.879 (CI 0.84 - 0.92). LIME identified kidney replacement therapy, surrogate respiratory quotient, lactate gradient, and arterial bicarbonate as metrics incorporated in the WoE model achieving 80% accuracy (CI: 76 - 84%), 70% precision (CI: 63 - 77%), F1 of 0.72 (CI: 0.66 - 0.78), and AUC of 0.886 (CI: 0.853 - 0.920).

CONCLUSIONS: omposite lactate metrics and WoE model improved in-hospital mortality prediction in patients requiring ECMO. Prospective studies and external validation are warranted to confirm these findings.