A novel mortality risk score predicting intensive care mortality in cardiogenic shock patients treated with veno-arterial extracorporeal membrane oxygenation

A Novel, Interpretable Machine Learning Model to Predict Neurological Outcomes Following Venoarterial Extracorporeal Membrane Oxygenation

BACKGROUND

We used machine learning models incorporating rich electronic medical record (EMR) data to predict neurological outcomes after venoarterial extracorporeal membrane oxygenation (VA-ECMO).

METHODS

This was a retrospective review of adult (≥ 18 years) patients undergoing VA-ECMO between 6/2016 and 4/2022 at a single center. The primary outcome was good neurological outcome, defined as a modified Rankin Scale score of 0 to 3, evaluated at hospital discharge. We extracted every measurement of 74 vital and laboratory values, as well as circuit and ventilator settings, from 24 h before cannulation through the entire duration of ECMO. An XGBoost model with Shapley Additive Explanations was developed and evaluated with leave-one-out cross-validation.

RESULTS

Overall, 194 patients undergoing VA-ECMO (median age 58 years, 63% male) were included. We extracted more than 14 million individual data points from the EMR. Of 194 patients, 39 patients (20%) had good neurological outcomes. Three models were generated: model A, which contained only pre-ECMO data; model B, which added data from the first 48 h of ECMO; and model C, which included data from the entire ECMO run. The leave-one-out cross-validation area under the receiver operator characteristics curves for models A, B, and C were 0.72, 0.81, and 0.90, respectively. The inclusion of on-ECMO physiologic, laboratory, and circuit data greatly improved model performance. Both modifiable and nonmodifiable variables, such as lower body mass index, lower age, higher mean arterial pressure, and higher hemoglobin, were associated with good neurological outcome.

CONCLUSIONS

An interpretable machine learning model from EMR-extracted data was able to predict neurological outcomes for patients undergoing VA-ECMO with excellent accuracy.

Mechanical Circulatory Support for Massive Pulmonary Embolism

Up to 50% of patients with pulmonary embolism (PE) experience hemodynamic instability and approximately 70% of patients who die of PE experience an accelerated cascade of symptoms within the first hours of onset of symptoms, thus necessitating rapid evaluation and intervention. Venoarterial extracorporeal membrane oxygenation and other ventricular assist devices, depending on the hemodynamic derangements present, may be used to stabilize patients with massive PE refractory to initial therapies or with contraindications to other interventions. Given the abnormalities in both pulmonary circulation and gas exchange caused by massive PE, venoarterial extracorporeal membrane oxygenation may be considered the preferred form of mechanical circulatory support for most patients. Venoarterial extracorporeal membrane oxygenation unloads the right ventricle and improves oxygenation, which may not only help buy time until definitive treatment but may also reduce myocardial ischemia and myocardial dysfunction. This review summarizes the available clinical data on the use of mechanical circulatory support, especially venoarterial extracorporeal membrane oxygenation, in the treatment of patients with massive PE. Furthermore, this review also provides practical guidance on the implementation of this strategy in clinical practice.

Comparison of Four Intensive Care Scores in Predicting Outcomes After Venoarterial Extracorporeal Membrane Oxygenation: A Single-center Retrospective Study

OBJECTIVE

To assess the capability of the Acute Physiology and Chronic Health Evaluation II (APACHE-II), Sequential Organ Failure Assessment (SOFA) scores, Cardiac Surgery Score (CASUS), and Survival After VA-ECMO (SAVE) in predicting outcomes among a cohort of patients undergoing venoarterial extracorporeal membrane oxygenation (VA-ECMO).

DESIGN

This is an observational retrospective study of 142 patients admitted to the cardiothoracic intensive care unit (CTICU) after undergoing VA-ECMO insertion.

SETTING

CTICU of a tertiary care center.

PARTICIPANTS

All patients admitted to the CTICU for a minimum of 24 hours, post-VA-ECMO insertion, between 2015 and 2022.

INTERVENTIONS

Review of electronic patient records.

MEASUREMENTS AND RESULTS

Scores for APACHE-II, SOFA, and CASUS were calculated 24 hours after intensive care units (ICU) admission. The SAVE score was computed from the last available patient details within 24 hours of ECMO insertion. Relevant demographic, clinical, and laboratory data for the study was retrieved from electronic patient records. Pre-ECMO serum levels of lactates and creatinine were significantly associated with mortality. Lower ECMO flow rates at 4 and 12 hours post-ECMO cannulation were significantly correlated with survival to discharge. The development of arrhythmias, acute kidney injury, and the need for continuous renal replacement therapy while on ECMO were significantly associated with mortality. The APACHE-II, SOFA, and CASUS scores, calculated at 24 hours of ICU admission, were significantly higher amongst nonsurvivors. Following risk score categorization using receiver operating characteristic curve analysis, it was found that APACHE-II, SOFA, and CASUS scores calculated 24 hours post-ICU admission after ECMO insertion demonstrated moderate predictive ability for mortality. In contrast, the SAVE score failed to predict mortality. APACHE-II >27 (area under the curve = 0.66), calculated 24 hours post-ICU admission after ECMO insertion, showed the greatest predictive ability for mortality. Multivariate logistic regression analysis of the four scores showed that APACHE-II >27 and SOFA >14, calculated 24 hours post-ICU admission after ECMO insertion, were independently significantly predictive of mortality.

CONCLUSION

The APACHE-II, SOFA, and CASUS, calculated at 24 hours of ICU admission, were significantly higher among nonsurvivors compared with survivors. The APACHE-II demonstrated the highest mortality predictive ability. APACHE-II scores of 27 or above and SOFA scores of 14 or above at 24 hours of ICU admission after ECMO cannulation can predict mortality and assist physicians in decision-making.

Extracorporal Membrane Oxygenation in Massive Pulmonary Embolism

BACKGROUND

Massive pulmonary embolism (MPE) carries significant 30-day mortality risk, and a change in societal guidelines has promoted the increasing use of extracorporeal membrane oxygenation (ECMO) in the immediate management of MPE-associated cardiovascular shock. This narrative review examines the current status of ECMO in MPE.

METHODS

A literature review was performed from 1982 to 2022 searching for the terms "Pulmonary embolism" and "ECMO," and the search was refined by examining those publications that covered MPE.

RESULTS

In the patient with MPE, veno-arterial ECMO is now recommended as a bridge to interventional therapy. It can reliably decrease right ventricular overload, improve RV function, and allow hemodynamic stability and restoration of tissue oxygenation. The use of ECMO in MPE has been associated with lower mortality in registry reviews, but there has been no significant difference in outcomes between patients treated with and without ECMO in meta-analyses. Applying ECMO is also associated with substantial multisystem morbidity due to systemic inflammatory response, bleeding with coagulopathy, hemorrhagic stroke, renal dysfunction, and acute limb ischemia, which must be factored into the outcomes.

CONCLUSIONS

The application of ECMO in MPE should be combined with an aggressive interventional pulmonary interventional program and should strictly adhere to the current selection criteria.

Acute brain injury risk prediction models in venoarterial extracorporeal membrane oxygenation patients with tree-based machine learning: An Extracorporeal Life Support Organization Registry analysis

Objective

We aimed to determine if machine learning can predict acute brain injury and to identify modifiable risk factors for acute brain injury in patients receiving venoarterial extracorporeal membrane oxygenation.

Methods

We included adults (age ≥18 years) receiving venoarterial extracorporeal membrane oxygenation or extracorporeal cardiopulmonary resuscitation in the Extracorporeal Life Support Organization Registry (2009-2021). Our primary outcome was acute brain injury: central nervous system ischemia, intracranial hemorrhage, brain death, and seizures. We used Random Forest, CatBoost, LightGBM, and XGBoost machine learning algorithms (10-fold leave-1-out cross-validation) to predict and identify features most important for acute brain injury. We extracted 65 total features: demographics, pre-extracorporeal membrane oxygenation/on-extracorporeal membrane oxygenation laboratory values, and pre-extracorporeal membrane oxygenation/on-extracorporeal membrane oxygenation settings.

Results

Of 35,855 patients receiving venoarterial extracorporeal membrane oxygenation (nonextracorporeal cardiopulmonary resuscitation) (median age of 57.8 years, 66% were male), 7.7% (n = 2769) experienced acute brain injury. In venoarterial extracorporeal membrane oxygenation (nonextracorporeal cardiopulmonary resuscitation), the area under the receiver operator characteristic curves to predict acute brain injury, central nervous system ischemia, and intracranial hemorrhage were 0.67, 0.67, and 0.62, respectively. The true-positive, true-negative, false-positive, false-negative, positive, and negative predictive values were 33%, 88%, 12%, 67%, 18%, and 94%, respectively, for acute brain injury. Longer extracorporeal membrane oxygenation duration, higher 24-hour extracorporeal membrane oxygenation pump flow, and higher on-extracorporeal membrane oxygenation partial pressure of oxygen were associated with acute brain injury. Of 10,775 patients receiving extracorporeal cardiopulmonary resuscitation (median age of 57.1 years, 68% were male), 16.5% (n = 1787) experienced acute brain injury. The area under the receiver operator characteristic curves for acute brain injury, central nervous system ischemia, and intracranial hemorrhage were 0.72, 0.73, and 0.69, respectively. Longer extracorporeal membrane oxygenation duration, older age, and higher 24-hour extracorporeal membrane oxygenation pump flow were associated with acute brain injury.

Conclusions

In the largest study predicting neurological complications with machine learning in extracorporeal membrane oxygenation, longer extracorporeal membrane oxygenation duration and higher 24-hour pump flow were associated with acute brain injury in nonextracorporeal cardiopulmonary resuscitation and extracorporeal cardiopulmonary resuscitation venoarterial extracorporeal membrane oxygenation.

Normalized lactate load as an independent prognostic indicator in patients with cardiogenic shock

BACKGROUND

Early prognosis evaluation is crucial for decision-making in cardiogenic shock (CS) patients. Dynamic lactate assessment, for example, normalized lactate load, has been a better prognosis predictor than single lactate value in septic shock. Our objective was to investigate the correlation between normalized lactate load and in-hospital mortality in patients with CS.

METHODS

Data were extracted from the Medical Information Mart for Intensive Care (MIMIC)-IV database. The calculation of lactate load involved the determination of the cumulative area under the lactate curve, while normalized lactate load was computed by dividing the lactate load by the corresponding period. Receiver Operating Characteristic (ROC) curves were constructed, and the evaluation of areas under the curves (AUC) for various parameters was performed using the DeLong test.

RESULTS

Our study involved a cohort of 1932 CS patients, with 687 individuals (36.1%) experiencing mortality during their hospitalization. The AUC for normalized lactate load demonstrated significant superiority compared to the first lactate (0.675 vs. 0.646, P < 0.001), maximum lactate (0.675 vs. 0.651, P < 0.001), and mean lactate (0.675 vs. 0.669, P = 0.003). Notably, the AUC for normalized lactate load showed comparability to that of the Sequential Organ Failure Assessment (SOFA) score (0.675 vs. 0.695, P = 0.175).

CONCLUSION

The normalized lactate load was an independently associated with the in-hospital mortality among CS patients.

Predicting Acute Brain Injury in Venoarterial Extracorporeal Membrane Oxygenation Patients with Tree-Based Machine Learning: Analysis of the Extracorporeal Life Support Organization Registry

Objective

To determine if machine learning (ML) can predict acute brain injury (ABI) and identify modifiable risk factors for ABI in venoarterial extracorporeal membrane oxygenation (VA-ECMO) patients.

Design

Retrospective cohort study of the Extracorporeal Life Support Organization (ELSO) Registry (2009-2021).

Setting

International, multicenter registry study of 676 ECMO centers.

Patients

Adults (≥18 years) supported with VA-ECMO or extracorporeal cardiopulmonary resuscitation (ECPR).

Interventions

None.

Measurements and Main Results

Our primary outcome was ABI: central nervous system (CNS) ischemia, intracranial hemorrhage (ICH), brain death, and seizures. We utilized Random Forest, CatBoost, LightGBM and XGBoost ML algorithms (10-fold leave-one-out cross-validation) to predict and identify features most important for ABI. We extracted 65 total features: demographics, pre-ECMO/on-ECMO laboratory values, and pre-ECMO/on-ECMO settings.Of 35,855 VA-ECMO (non-ECPR) patients (median age=57.8 years, 66% male), 7.7% (n=2,769) experienced ABI. In VA-ECMO (non-ECPR), the area under the receiver-operator characteristics curves (AUC-ROC) to predict ABI, CNS ischemia, and ICH was 0.67, 0.67, and 0.62, respectively. The true positive, true negative, false positive, false negative, positive, and negative predictive values were 33%, 88%, 12%, 67%, 18%, and 94%, respectively for ABI. Longer ECMO duration, higher 24h ECMO pump flow, and higher on-ECMO PaO2 were associated with ABI.Of 10,775 ECPR patients (median age=57.1 years, 68% male), 16.5% (n=1,787) experienced ABI. The AUC-ROC for ABI, CNS ischemia, and ICH was 0.72, 0.73, and 0.69, respectively. The true positive, true negative, false positive, false negative, positive, and negative predictive values were 61%, 70%, 30%, 39%, 29% and 90%, respectively, for ABI. Longer ECMO duration, younger age, and higher 24h ECMO pump flow were associated with ABI.

Conclusions

This is the largest study predicting neurological complications on sufficiently powered international ECMO cohorts. Longer ECMO duration and higher 24h pump flow were associated with ABI in both non-ECPR and ECPR VA-ECMO.

Current status of ECMO for massive pulmonary embolism

Massive pulmonary embolism (MPE) carries significant 30-day mortality and is characterized by acute right ventricular failure, hypotension, and hypoxia, leading to cardiovascular collapse and cardiac arrest. Given the continued high mortality associated with MPE, there has been ongoing interest in utilizing extracorporeal membrane oxygenation (ECMO) to provide oxygenation support to improve hypoxia and offload the right ventricular (RV) pressure in the belief that rapid reduction of hypoxia and RV pressure will improve outcomes. Two modalities can be employed: Veno-arterial-ECMO is a reliable process to decrease RV overload and improve RV function, thus allowing for hemodynamic stability and restoration of tissue oxygenation. Veno-venous ECMO can support oxygenation but is not designed to help circulation. Several societal guidelines now suggest using ECMO in MPE with interventional therapy. There are three strategies for ECMO utilization in MPE: bridge to definitive interventional therapy, sole therapy, and recovery after interventional treatment. The use of ECMO in MPE has been associated with lower mortality in registry reviews, but there has been no significant difference in outcomes between patients treated with and without ECMO in meta-analyses. Considerable heterogeneity in studies is a significant weakness of the available literature. Applying ECMO is also associated with substantial multisystem morbidity due to a systemic inflammatory response, hemorrhagic stroke, renal dysfunction, and bleeding, which must be factored into the outcomes. The application of ECMO in MPE should be combined with an aggressive pulmonary interventional program and should strictly adhere to the current selection criteria.

Prognostic models for mortality risk in patients requiring ECMO

PURPOSE

To provide an overview and evaluate the performance of mortality prediction models for patients requiring extracorporeal membrane oxygenation (ECMO) support for refractory cardiocirculatory or respiratory failure.

METHODS

A systematic literature search was undertaken to identify studies developing and/or validating multivariable prediction models for all-cause mortality in adults requiring or receiving veno-arterial (V-A) or veno-venous (V-V) ECMO. Estimates of model performance (observed versus expected (O:E) ratio and c-statistic) were summarized using random effects models and sources of heterogeneity were explored by means of meta-regression. Risk of bias was assessed using the Prediction model Risk Of BiAS Tool (PROBAST).

RESULTS

Among 4905 articles screened, 96 studies described a total of 58 models and 225 external validations. Out of all 58 models which were specifically developed for ECMO patients, 14 (24%) were ever externally validated. Discriminatory ability of frequently validated models developed for ECMO patients (i.e., SAVE and RESP score) was moderate on average (pooled c-statistics between 0.66 and 0.70), and comparable to general intensive care population-based models (pooled c-statistics varying between 0.66 and 0.69 for the Simplified Acute Physiology Score II (SAPS II), Acute Physiology and Chronic Health Evaluation II (APACHE II) score and Sequential Organ Failure Assessment (SOFA) score). Nearly all models tended to underestimate mortality with a pooled O:E > 1. There was a wide variability in reported performance measures of external validations, reflecting a large between-study heterogeneity. Only 1 of the 58 models met the generally accepted Prediction model Risk Of BiAS Tool criteria of good quality. Importantly, all predicted outcomes were conditional on the fact that ECMO support had already been initiated, thereby reducing their applicability for patient selection in clinical practice.

CONCLUSIONS

A large number of mortality prediction models have been developed for ECMO patients, yet only a minority has been externally validated. Furthermore, we observed only moderate predictive performance, large heterogeneity between-study populations and model performance, and poor methodological quality overall. Most importantly, current models are unsuitable to provide decision support for selecting individuals in whom initiation of ECMO would be most beneficial, as all models were developed in ECMO patients only and the decision to start ECMO had, therefore, already been made.

MELD score is predictive of 90-day mortality after veno-arterial extracorporeal membrane oxygenation support

BACKGROUND

The Model for End-Stage Liver Disease (MELD) score was originally described as a marker of survival in chronic liver disease. More recently, MELD and its derivatives, MELD excluding INR (MELD-XI) and MELD with sodium (MELD-Na), have been applied more broadly as outcome predictors in heart transplant, left ventricular assist device placement, heart failure, and cardiogenic shock, with additional promising data to support the use of these scores for prediction of survival in those undergoing veno-arterial extracorporeal membrane oxygenation (VA ECMO).

METHODS

This study assessed the prognostic impact of MELD in patients with cardiogenic shock undergoing VA ECMO via a single-center retrospective review from January 2014 to March 2020. MELD, MELD-XI, and MELD-Na scores were calculated using laboratory values collected within 48 h of VA ECMO initiation. Multivariate Cox regression analyses determined the association between MELD scores and the primary outcome of 90-day mortality. Receiver operating characteristics (ROC) were used to estimate the discriminatory power for MELD in comparison with previously validated SAVE score.

RESULTS

Of the 194 patients, median MELD was 20.1 (13.7-26.2), and 90-day mortality was 62.1%. There was a significant association between MELD score and mortality up to 90 days (hazard ratio (HR) = 1.945, 95% confidence interval (95% CI) = 1.244-3.041, p = 0.004) after adjustment for age, indication for VA ECMO, and sex. The prognostic significance of MELD score for 90-day mortality revealed an AUC of 0.645 (95% CI = 0.565-0.725, p < 0.001). MELD-Na score and MELD-XI score were not associated with mortality.

CONCLUSION

MELD score accurately predicts long-term mortality and may be utilized as a valuable decision-making tool in patients undergoing VA ECMO.

Biotrauma during ultra-low tidal volume ventilation and venoarterial extracorporeal membrane oxygenation in cardiogenic shock: a randomized crossover clinical trial

Background

Cardiogenic pulmonary oedema (CPE) may contribute to ventilator-associated lung injury (VALI) in patients with cardiogenic shock. The appropriate ventilatory strategy remains unclear. We aimed to evaluate the impact of ultra-low tidal volume ventilation with tidal volume of 3 ml/kg predicted body weight (PBW) in patients with CPE and veno–arterial extracorporeal membrane oxygenation (V–A ECMO) on lung inflammation compared to conventional ventilation.

Methods

A single-centre randomized crossover trial was performed in the Cardiac Intensive Care Unit (ICU) at a tertiary university hospital. Seventeen adults requiring V–A ECMO and mechanical ventilation due to cardiogenic shock were included from February 2017 to December 2018. Patients were ventilated for two consecutive periods of 24 h with tidal volumes of 6 and 3 ml/kg of PBW, respectively, applied in random order. Primary outcome was the change in proinflammatory mediators in bronchoalveolar lavage fluid (BALF) between both ventilatory strategies.

Results

Ventilation with 3 ml/kg PBW yielded lower driving pressures and end-expiratory lung volumes. Overall, there were no differences in BALF cytokines. Post hoc analyses revealed that patients with high baseline levels of IL-6 showed statistically significant lower levels of IL-6 and IL-8 during ultra-low tidal volume ventilation. This reduction was significantly proportional to the decrease in driving pressure. In contrast, those with lower IL-6 baseline levels showed a significant increase in these biomarkers.

Conclusions

Ultra-low tidal volume ventilation in patients with CPE and V–A ECMO may attenuate inflammation in selected cases. VALI may be driven by an interaction between the individual proinflammatory profile and the mechanical load overimposed by the ventilator.

Trial registration The trial was registered in ClinicalTrials.gov (identifier NCT03041428, Registration date: 2nd February 2017).

Supplementary Information

The online version contains supplementary material available at 10.1186/s13613-021-00919-0.

Twelve years of circulatory extracorporeal life support at the University Medical Centre Utrecht

Introduction

Circulatory extracorporeal life support (ECLS) has been performed at the University Medical Centre Utrecht for 12 years. During this time, case mix, indications, ECLS set-ups and outcomes seem to have substantially changed. We set out to describe these characteristics and their evolution over time.

Methods

All patients receiving circulatory ECLS between 2007 and 2018 were retrospectively identified and divided into six groups according to a 2-year period of time corresponding to the date of ECLS initiation. General characteristics plus data pertaining to comorbidities, indications and technical details of ECLS commencement as well as in-hospital, 30-day, 1‑year and overall mortality were collected. Temporal trends in these characteristics were examined.

Results

A total of 347 circulatory ECLS runs were performed in 289 patients. The number of patients and ECLS runs increased from 8 till a maximum of 40 runs a year. The distribution of circulatory ECLS indications shifted from predominantly postcardiotomy to a wider set of indications. The proportion of peripheral insertions with or without application of left ventricular unloading techniques substantially increased, while in-hospital, 30-day, 1‑year and overall mortality decreased over time.

Conclusion

Circulatory ECLS was increasingly applied at the University Medical Centre Utrecht. Over time, indications as well as treatment goals broadened, and cannulation techniques shifted from central to mainly peripheral approaches. Meanwhile, weaning success increased and mortality rates diminished.

Supplementary Information

The online version of this article (10.1007/s12471-021-01552-z) contains supplementary material, which is available to authorized users.

Neural Networks to Predict Radiographic Brain Injury in Pediatric Patients Treated with Extracorporeal Membrane Oxygenation

Brain injury is a significant source of morbidity and mortality for pediatric patients treated with Extracorporeal Membrane Oxygenation (ECMO). Our objective was to utilize neural networks to predict radiographic evidence of brain injury in pediatric ECMO-supported patients and identify specific variables that can be explored for future research. Data from 174 ECMO-supported patients were collected up to 24 h prior to, and for the duration of, the ECMO course. Thirty-five variables were collected, including physiological data, markers of end-organ perfusion, acid-base homeostasis, vasoactive infusions, markers of coagulation, and ECMO-machine factors. The primary outcome was the presence of radiologic evidence of moderate to severe brain injury as established by brain CT or MRI. This information was analyzed by a neural network, and results were compared to a logistic regression model as well as clinician judgement. The neural network model was able to predict brain injury with an Area Under the Curve (AUC) of 0.76, 73% sensitivity, and 80% specificity. Logistic regression had 62% sensitivity and 61% specificity. Clinician judgment had 39% sensitivity and 69% specificity. Sequential feature group masking demonstrated a relatively greater contribution of physiological data and minor contribution of coagulation factors to the model's performance. These findings lay the foundation for further areas of research directions.

Impact of interhospital transfer on clinical outcomes and costs of extracorporeal life support

BACKGROUND

The impact of interhospital transfers for extracorporeal life support have not been studied in large datasets. The present study sought to determine the impact of such patient transfers on survival, complications, and hospitalization costs.

METHODS

The 2010 to 2016 database of the National Inpatient Sample was used to identify all adults who underwent extracorporeal life support. Patients were categorized based on whether or not they were transferred to another facility. Trend analysis and multivariable models were used to characterize the impact of inter hospital transfer on in-hospital mortality, complications, duration of stay, and costs.

RESULTS

Of an estimated 29,298 extracorporeal life support hospitalizations during the study period, 36.8% were transferred from an outside facility. Extracorporeal life support hospitalizations experienced a 7-fold increase with no difference in mortality between transferred and not transferred cohorts in 2016 (4.79% vs 4.79%, P = .97). Mortality rates were less for patients transferred to high volume centers compared to low volume hospitals (48.7% vs 51.6%, P < .001). Transfer to a low volume hospital for cardiogenic shock was associated with greater odds of mortality (adjusted odds Rratio: 2.25, confidence interval 1.01-5.03).

CONCLUSION

Utilization of extracorporeal life support in both transferred and not transferred patients has statistically significantly increased with a decrement in mortality for those transferred. Survival in the transferred cohort is strongly associated with extracorporeal life support procedure volume of the center and this must be taken into account when considering extracorporeal life support transfer.

Predicting Survival After Extracorporeal Membrane Oxygenation by Using Machine Learning

BACKGROUND

Venoarterial (VA) extracorporeal membrane oxygenation (ECMO) undoubtedly saves many lives, but it is associated with a high degree of patient morbidity, mortality, and resource use. This study aimed to develop a machine learning algorithm to augment clinical decision making related to VA-ECMO.

METHODS

Patients supported by VA-ECMO at a single institution from May 2011 to October 2018 were retrospectively reviewed. Laboratory values from only the initial 48 hours of VA-ECMO support were used. Data were split into 70% for training, 15% for validation, and 15% withheld for testing. Feature importance was estimated, and dimensionality reduction techniques were used. A deep neural network was trained to predict survival to discharge, and the final model was assessed using the independent testing cohort. Model performance was compared with that of the SAVE (Survival After Veno-arterial ECMO) score by using a receiver operator characteristic curve.

RESULTS

Of the 282 eligible adult patients who were undergoing VA-ECMO, 117 (41%) survived to discharge. A total of 1.96 million laboratory values were extracted from the electronic medical record, from which 270 different summary variables were derived for each patient. The most important variables in predicting the primary outcome included lactate, age, total bilirubin, and creatinine. For the testing cohort, the final model achieved 82% overall accuracy and a greater area under the curve than the SAVE score (0.92 vs 0.65; P = .01) in predicting survival to discharge.

CONCLUSIONS

This proof of concept study demonstrates the potential for machine learning models to augment clinical decision making for patients undergoing VA-ECMO. Further development with multi-institutional data is warranted.