Dataset Information

Using a machine learning approach to predict mortality in critically ill influenza patients: a cross-sectional retrospective multicentre study in Taiwan.

ABSTRACT: OBJECTIVES:Current mortality prediction models used in the intensive care unit (ICU) have a limited role for specific diseases such as influenza, and we aimed to establish an explainable machine learning (ML) model for predicting mortality in critically ill influenza patients using a real-world severe influenza data set. STUDY DESIGN:A cross-sectional retrospective multicentre study in Taiwan SETTING: Eight medical centres in Taiwan. PARTICIPANTS:A total of 336 patients requiring ICU-admission for virology-proven influenza at eight hospitals during an influenza epidemic between October 2015 and March 2016. PRIMARY AND SECONDARY OUTCOME MEASURES:We employed extreme gradient boosting (XGBoost) to establish the prediction model, compared the performance with logistic regression (LR) and random forest (RF), demonstrated the feature importance categorised by clinical domains, and used SHapley Additive exPlanations (SHAP) for visualised interpretation. RESULTS:The data set contained 76 features of the 336 patients with severe influenza. The severity was apparently high, as shown by the high Acute Physiology and Chronic Health Evaluation II score (22, 17 to 29) and pneumonia severity index score (118, 88 to 151). XGBoost model (area under the curve (AUC): 0.842; 95% CI 0.749 to 0.928) outperformed RF (AUC: 0.809; 95% CI 0.629 to 0.891) and LR (AUC: 0.701; 95% CI 0.573 to 0.825) for predicting 30-day mortality. To give clinicians an intuitive understanding of feature exploitation, we stratified features by the clinical domain. The cumulative feature importance in the fluid balance domain, ventilation domain, laboratory data domain, demographic and symptom domain, management domain and severity score domain was 0.253, 0.113, 0.177, 0.140, 0.152 and 0.165, respectively. We further used SHAP plots to illustrate associations between features and 30-day mortality in critically ill influenza patients. CONCLUSIONS:We used a real-world data set and applied an ML approach, mainly XGBoost, to establish a practical and explainable mortality prediction model in critically ill influenza patients.

SUBMITTER: Hu CA

PROVIDER: S-EPMC7045134 | biostudies-literature | 2020 Feb

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Using a machine learning approach to predict mortality in critically ill influenza patients: a cross-sectional retrospective multicentre study in Taiwan.

Hu Chien-An CA Chen Chia-Ming CM Fang Yen-Chun YC Liang Shinn-Jye SJ Wang Hao-Chien HC Fang Wen-Feng WF Sheu Chau-Chyun CC Perng Wann-Cherng WC Yang Kuang-Yao KY Kao Kuo-Chin KC Wu Chieh-Liang CL Tsai Chwei-Shyong CS Lin Ming-Yen MY Chao Wen-Cheng WC

BMJ open 20200225 2

<h4>Objectives</h4>Current mortality prediction models used in the intensive care unit (ICU) have a limited role for specific diseases such as influenza, and we aimed to establish an explainable machine learning (ML) model for predicting mortality in critically ill influenza patients using a real-world severe influenza data set.<h4>Study design</h4>A cross-sectional retrospective multicentre study in Taiwan SETTING: Eight medical centres in Taiwan.<h4>Participants</h4>A total of 336 patients req ...[more]

PMID: 32102816

Similar Datasets

Project description:OBJECTIVES:Sepsis is associated with high early and total in-hospital mortality. Despite recent revisions in the diagnostic criteria for sepsis that sought to improve predictive validity for mortality, it remains difficult to identify patients at greatest risk of death. We compared the utility of nine biomarkers to predict mortality in subjects with clinically suspected bacterial sepsis. DESIGN:Cohort study. SETTING:The medical and surgical ICUs at an academic medical center. SUBJECTS:We enrolled 139 subjects who met two or more systemic inflammatory response syndrome (systemic inflammatory response syndrome) criteria and received new broad-spectrum antibacterial therapy. INTERVENTIONS:We assayed nine biomarkers (?-2 macroglobulin, C-reactive protein, ferritin, fibrinogen, haptoglobin, procalcitonin, serum amyloid A, serum amyloid P, and tissue plasminogen activator) at onset of suspected sepsis and 24, 48, and 72 hours thereafter. We compared biomarkers between groups based on both 14-day and total in-hospital mortality and evaluated the predictive validity of single and paired biomarkers via area under the receiver operating characteristic curve. MEASUREMENTS AND MAIN RESULTS:Fourteen-day mortality was 12.9%, and total in-hospital mortality was 29.5%. Serum amyloid P was significantly lower (4/4 timepoints) and tissue plasminogen activator significantly higher (3/4 timepoints) in the 14-day mortality group, and the same pattern held for total in-hospital mortality (Wilcoxon p ? 0.046 for all timepoints). Serum amyloid P and tissue plasminogen activator demonstrated the best individual predictive performance for mortality, and combinations of biomarkers including serum amyloid P and tissue plasminogen activator achieved greater predictive performance (area under the receiver operating characteristic curve > 0.76 for 14-d and 0.74 for total mortality). CONCLUSIONS:Combined biomarkers predict risk for 14-day and total mortality among subjects with suspected sepsis. Serum amyloid P and tissue plasminogen activator demonstrated the best discriminatory ability in this cohort.

Project description:ObjectivesTo identify factors influencing the mortality risk in critically ill patients with COVID-19, and to develop a risk prediction score to be used at admission to intensive care unit (ICU).DesignA multicentre cohort study.Setting and participants1542 patients with COVID-19 admitted to ICUs in public hospitals of Abu Dhabi, United Arab Emirates between 1 March 2020 and 22 July 2020.Main outcomes and measuresThe primary outcome was time from ICU admission until death. We used competing risk regression models and Least Absolute Shrinkage and Selection Operator to identify the factors, and to construct a risk score. Predictive ability of the score was assessed by the area under the receiver operating characteristic curve (AUC), and the Brier score using 500 bootstraps replications.ResultsAmong patients admitted to ICU, 196 (12.7%) died, 1215 (78.8%) were discharged and 131 (8.5%) were right-censored. The cumulative mortality incidence was 14% (95% CI 12.17% to 15.82%). From 36 potential predictors, we identified seven factors associated with mortality, and included in the risk score: age (adjusted HR (AHR) 1.98; 95% CI 1.71 to 2.31), neutrophil percentage (AHR 1.71; 95% CI 1.27 to 2.31), lactate dehydrogenase (AHR 1.31; 95% CI 1.15 to 1.49), respiratory rate (AHR 1.31; 95% CI 1.15 to 1.49), creatinine (AHR 1.19; 95% CI 1.11 to 1.28), Glasgow Coma Scale (AHR 0.70; 95% CI 0.63 to 0.78) and oxygen saturation (SpO2) (AHR 0.82; 95% CI 0.74 to 0.91). The mean AUC was 88.1 (95% CI 85.6 to 91.6), and the Brier score was 8.11 (95% CI 6.74 to 9.60). We developed a freely available web-based risk calculator (https://icumortalityrisk.shinyapps.io/ICUrisk/).ConclusionIn critically ill patients with COVID-19, we identified factors associated with mortality, and developed a risk prediction tool that showed high predictive ability. This tool may have utility in clinical settings to guide decision-making, and may facilitate the identification of supportive therapies to improve outcomes.

Project description:Background and purposeDeep sedation is associated with acute brain dysfunction and increased mortality. We had previously shown that early-assessed brainstem reflexes may predict outcome in deeply sedated patients. The primary objective was to determine whether patterns of brainstem reflexes might predict mortality in deeply sedated patients. The secondary objective was to generate a score predicting mortality in these patients.MethodsObservational prospective multicenter cohort study of 148 non-brain injured deeply sedated patients, defined by a Richmond Assessment sedation Scale (RASS) <-3. Brainstem reflexes and Glasgow Coma Scale were assessed within 24 hours of sedation and categorized using latent class analysis. The Full Outline Of Unresponsiveness score (FOUR) was also assessed. Primary outcome measure was 28-day mortality. A "Brainstem Responses Assessment Sedation Score" (BRASS) was generated.ResultsTwo distinct sub-phenotypes referred as homogeneous and heterogeneous brainstem reactivity were identified (accounting for respectively 54.6% and 45.4% of patients). Homogeneous brainstem reactivity was characterized by preserved reactivity to nociceptive stimuli and a partial and topographically homogenous depression of brainstem reflexes. Heterogeneous brainstem reactivity was characterized by a loss of reactivity to nociceptive stimuli associated with heterogeneous brainstem reflexes depression. Heterogeneous sub-phenotype was a predictor of increased risk of 28-day mortality after adjustment to Simplified Acute Physiology Score-II (SAPS-II) and RASS (Odds Ratio [95% confidence interval] = 6.44 [2.63-15.8]; p<0.0001) or Sequential Organ Failure Assessment (SOFA) and RASS (OR [95%CI] = 5.02 [2.01-12.5]; p = 0.0005). The BRASS (and marginally the FOUR) predicted 28-day mortality (c-index [95%CI] = 0.69 [0.54-0.84] and 0.65 [0.49-0.80] respectively).ConclusionIn this prospective cohort study, around half of all deeply sedated critically ill patients displayed an early particular neurological sub-phenotype predicting 28-day mortality, which may reflect a dysfunction of the brainstem.

Dataset Information

Using a machine learning approach to predict mortality in critically ill influenza patients: a cross-sectional retrospective multicentre study in Taiwan.

Publications

Using a machine learning approach to predict mortality in critically ill influenza patients: a cross-sectional retrospective multicentre study in Taiwan.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure