Dataset Information

Real-time machine learning model to predict short-term mortality in critically ill patients: development and international validation.

ABSTRACT:

Background

A real-time model for predicting short-term mortality in critically ill patients is needed to identify patients at imminent risk. However, the performance of the model needs to be validated in various clinical settings and ethnicities before its clinical application. In this study, we aim to develop an ensemble machine learning model using routinely measured clinical variables at a single academic institution in South Korea.

Methods

We developed an ensemble model using deep learning and light gradient boosting machine models. Internal validation was performed using the last two years of the internal cohort dataset, collected from a single academic hospital in South Korea between 2007 and 2021. External validation was performed using the full Medical Information Mart for Intensive Care (MIMIC), eICU-Collaborative Research Database (eICU-CRD), and Amsterdam University Medical Center database (AmsterdamUMCdb) data. The area under the receiver operating characteristic curve (AUROC) was calculated and compared to that for the National Early Warning Score (NEWS).

Results

The developed model (iMORS) demonstrated high predictive performance with an internal AUROC of 0.964 (95% confidence interval [CI] 0.963-0.965) and external AUROCs of 0.890 (95% CI 0.889-0.891) for MIMIC, 0.886 (95% CI 0.885-0.887) for eICU-CRD, and 0.870 (95% CI 0.868-0.873) for AmsterdamUMCdb. The model outperformed the NEWS with higher AUROCs in the internal and external validation (0.866 for the internal, 0.746 for MIMIC, 0.798 for eICU-CRD, and 0.819 for AmsterdamUMCdb; p < 0.001).

Conclusions

Our real-time machine learning model to predict short-term mortality in critically ill patients showed excellent performance in both internal and external validations. This model could be a useful decision-support tool in the intensive care units to assist clinicians.

SUBMITTER: Lim L

PROVIDER: S-EPMC10938661 | biostudies-literature | 2024 Mar

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Real-time machine learning model to predict short-term mortality in critically ill patients: development and international validation.

Lim Leerang L Gim Ukdong U Cho Kyungjae K Yoo Dongjoon D Ryu Ho Geol HG Lee Hyung-Chul HC

Critical care (London, England) 20240314 1

<h4>Background</h4>A real-time model for predicting short-term mortality in critically ill patients is needed to identify patients at imminent risk. However, the performance of the model needs to be validated in various clinical settings and ethnicities before its clinical application. In this study, we aim to develop an ensemble machine learning model using routinely measured clinical variables at a single academic institution in South Korea.<h4>Methods</h4>We developed an ensemble model using ...[more]

PMID: 38486247

Similar Datasets

Project description:BackgroundSepsis-associated acute kidney injury (SA-AKI) is common in critically ill patients, which is associated with significantly increased mortality. Existing mortality prediction tools showed insufficient predictive power or failed to reflect patients' dynamic clinical evolution. Therefore, the study aimed to develop and validate machine learning-based models for real-time mortality prediction in critically ill patients with SA-AKI.MethodsThe multi-center retrospective study included patients from two distinct databases. A total of 12,132 SA-AKI patients from the Medical Information Mart for Intensive Care IV (MIMIC-IV) were randomly allocated to the training, validation, and internal test sets. An additional 3,741 patients from the eICU Collaborative Research Database (eICU-CRD) served as an external test set. For every 12 h during the ICU stays, the state-of-the-art eXtreme Gradient Boosting (XGBoost) algorithm was used to predict the risk of in-hospital death in the following 48, 72, and 120 h and in the first 28 days after ICU admission. Area under the receiver operating characteristic curves (AUCs) were calculated to evaluate the models' performance.ResultsThe XGBoost models, based on routine clinical variables updated every 12 h, showed better performance in mortality prediction than the SOFA score and SAPS-II. The AUCs of the XGBoost models for mortality over different time periods ranged from 0.848 to 0.804 in the internal test set and from 0.818 to 0.748 in the external test set. The shapley additive explanation method provided interpretability for the XGBoost models, which improved the understanding of the association between the predictor variables and future mortality.ConclusionsThe interpretable machine learning XGBoost models showed promising performance in real-time mortality prediction in critically ill patients with SA-AKI, which are useful tools for early identification of high-risk patients and timely clinical interventions.

Project description:BackgroundIncreased or decreased blood pressure variability may affect the perfusion of tissues and organs, leading to acute kidney injury and death. This study was conducted to explore the relationship between mean arterial pressure variability and short- and long-term mortality in critically ill patients.MethodsWe used patient data from the MIMIC-III database for cohort study. According to the recorded mean arterial pressure during the first 24 h in the intensive care unit, we calculated each patient's two variability parameters -coefficient of variation and average real variability. The primary outcome was in-hospital mortality and the secondary outcomes were 28-day mortality and 1-year mortality. We conducted smooth spline models to examine the possible nonlinear associations between blood pressure variability and mortality. According to the smoothing curve, we further developed a two-piecewise linear regression model to find out the threshold effect. Multivariable logistic regression or Cox proportional hazards model was used to evaluate the relationship. Kaplan-Meier survival analysis for 28-day and 1-year mortality was performed. Subgroup analysis explored the factors modifying the relationship between them.ResultsA total of 12,867 patients were enrolled in the study, 1,320 in-hospital death, 1,399 28-day death, and 2,734 1-year death occurred. The smooth spline showed death risk was the lowest when average real variability was around 7.2 mmHg. After adjusting for covariates, logistic or Cox regression showed the highest MAP variability level was strongly associated with increased mortality in the hospital (odds ratio: 1.44; 95% CI, 1.21∼1.72), at 28 days (hazard ratio: 1.28; 95% CI, 1.1∼1.5), and at 1 year (hazard ratio: 1.27; 95% CI, 1.14∼1.42) compared with the second level of average real variability group. The survival curve plot showed patients with higher average real variability had a higher risk of 28-day and 1-year mortality. This relationship remained remarkable in patients with low or high Sequential Organ Failure Assessment scores in the sensitivity analysis. The two-piecewise linear regression model showed that lower ARV was a risk factor for 28-day (HR 0.72, 95% CI, 0.57∼0.91) and 1-year mortality (HR 0.81, 95% CI, 0.68∼0.96) when ARV was less than 7.2 mmHg, higher ARV was a risk factor for 28-day mortality (HR 1.1, 95% CI, 1.04∼1.17) and 1-year mortality (HR 1.07, 95% CI, 1.02∼1.12) when ARV was greater than 7.2 mmHg.ConclusionBlood pressure variability predicts mortality in critically ill patients. Individuals with higher or lower mean arterial pressure average real variability during the first day in ICU may have an increased risk of death.

Project description:Background & aimsBody composition is sex dependent and associated with an increased mortality risk in patients with cirrhosis. We evaluated whether it was also associated with short-term mortality in patients critically ill with acute-on-chronic liver failure (ACLF).Patients and methodsWe retrospectively included all patients with cirrhosis and ACLF hospitalised in the intensive care unit (ICU) of Lausanne University Hospital between 2010 and 2019 for whom an abdominal computed tomography (CT) scan performed ±7 days from admission was available. Patients from the ICU of Paul Brousse University Hospital admitted between 2017 and 2020 served as an external cohort. All body composition parameters at the third lumbar vertebral level (L3) were quantified using a deep learning-based method.ResultsIn total, 192 patients from Lausanne were included. Median age was 62 years and 28-day survival rate was 58.2%. In males, variables independently associated with 28-day mortality on days 1 and 3 were Chronic Liver Failure Consortium (CLIF-C) ACLF-lactate and sarcopenia. In females, CLIF-C ACLF-lactate on days 1 and 3 was the only predictor of 28-day survival. We derived two scores combining sarcopenia and the CLIF-C ACLF-lactate score on days 1 and 3, with area under the receiver operating characteristic outperforming the CLIF-C ACLF-lactate score alone in male but not in female patients. Comparable results were found in the external cohort of 58 patients and supported the sex specificity of the performance of the model. Patients with sarcopenia had increased risks of invasive fungal infection and renal replacement therapy.ConclusionSarcopenia was associated with 28-day mortality in male but not in female patients critically ill with ACLF. Although screening for sarcopenia could impact the management of male patients, further studies are needed in female cohorts to investigate whether other body composition parameters are associated with outcomes.Impact and implicationsBody composition, easily assessed by CT, is altered in patients with cirrhosis and associated with outcome; it has never been investigated in patients critically ill with ACLF. The results of the present study, underlining the benefit of sarcopenia evaluation to improve prognosis prediction in males critically ill with ACLF, are of importance for physicians managing such patients to optimise the decision-making process toward continued treatment, liver transplantation, or limitation of care. In a wider sense, besides the number and course of organ failures, the results recall the weight of the general condition of males with ACLF at admission to ICU. In females critically ill with ACLF, in analyses limited by the sample size, none of the body composition parameters was associated with short-term mortality independently of organ failures; this suggests that the number and course of organ failures are the main determinant of mortality in these patients.

Dataset Information

Real-time machine learning model to predict short-term mortality in critically ill patients: development and international validation.

Background

Methods

Results

Conclusions

Publications

Real-time machine learning model to predict short-term mortality in critically ill patients: development and international validation.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets