Dataset Information

Explainable Machine Learning to Predict Successful Weaning Among Patients Requiring Prolonged Mechanical Ventilation: A Retrospective Cohort Study in Central Taiwan.

ABSTRACT: Objective: The number of patients requiring prolonged mechanical ventilation (PMV) is increasing worldwide, but the weaning outcome prediction model in these patients is still lacking. We hence aimed to develop an explainable machine learning (ML) model to predict successful weaning in patients requiring PMV using a real-world dataset. Methods: This retrospective study used the electronic medical records of patients admitted to a 12-bed respiratory care center in central Taiwan between 2013 and 2018. We used three ML models, namely, extreme gradient boosting (XGBoost), random forest (RF), and logistic regression (LR), to establish the prediction model. We further illustrated the feature importance categorized by clinical domains and provided visualized interpretation by using SHapley Additive exPlanations (SHAP) as well as local interpretable model-agnostic explanations (LIME). Results: The dataset contained data of 963 patients requiring PMV, and 56.0% (539/963) of them were successfully weaned from mechanical ventilation. The XGBoost model (area under the curve [AUC]: 0.908; 95% confidence interval [CI] 0.864-0.943) and RF model (AUC: 0.888; 95% CI 0.844-0.934) outperformed the LR model (AUC: 0.762; 95% CI 0.687-0.830) in predicting successful weaning in patients requiring PMV. To give the physician an intuitive understanding of the model, we stratified the feature importance by clinical domains. The cumulative feature importance in the ventilation domain, fluid domain, physiology domain, and laboratory data domain was 0.310, 0.201, 0.265, and 0.182, respectively. We further used the SHAP plot and partial dependence plot to illustrate associations between features and the weaning outcome at the feature level. Moreover, we used LIME plots to illustrate the prediction model at the individual level. Additionally, we addressed the weekly performance of the three ML models and found that the accuracy of XGBoost/RF was ~0.7 between weeks 4 and week 7 and slightly declined to 0.6 on weeks 8 and 9. Conclusion: We used an ML approach, mainly XGBoost, SHAP plot, and LIME plot to establish an explainable weaning prediction ML model in patients requiring PMV. We believe these approaches should largely mitigate the concern of the black-box issue of artificial intelligence, and future studies are warranted for the landing of the proposed model.

SUBMITTER: Lin MY

PROVIDER: S-EPMC8104124 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:Background: Comprehensive rehabilitation programs are recommended for patients with prolonged mechanical ventilation (PMV) to facilitate functional recovery and ventilator weaning, but whether the functional status after rehabilitation influences outcome has not been clearly evaluated. This study aimed to investigate the association between post-rehabilitation functional status and weaning and survival outcome in PMV patients. Methods: We retrospectively enrolled PMV patients admitted to the respiratory care center (RCC), a post-ICU weaning facility with protocolized rehabilitation program, from January 2016 through December 2017. Functional status was measured by the de Morton Mobility Index (DEMMI), with a cut-off value set at 20 points. The primary outcomes were the weaning status at RCC discharge and hospital survival. The secondary outcomes were overall survival and survival at 3 months after RCC discharge. We followed patients until 3 months after RCC discharge or death. Logistic and Cox regressions were performed to identify significant parameters associated with weaning success and survival. Results: In total, 320 patients were enrolled. The weaning success rate was 71.6%. The survival rate at RCC discharge, hospital discharge, and 3 months after RCC discharge was 89.1, 77.5, and 66.6%, respectively. Post-rehabilitation DEMMI ≥ 20 (odds ratio [OR], 3.514; 95% confidence interval [CI], 1.436-8.598; P = 0.006) was the most significantly associated with weaning success. The weaning success and higher post-rehabilitation DEMMI were the two most significant independent factors associated with both hospital survival (weaning success, OR, 12.272; 95% CI, 5.281-28.517; P < 0.001; post-rehabilitation DEMMI ≥ 20, OR, 6.298; 95% CI, 1.302-30.477; P = 0.022) and survival at 3 months after RCC discharge (weaning success, OR, 38.788; 95% CI, 11.505-130.762; P < 0.001; post-rehabilitation DEMMI ≥ 20, OR, 4.830; 95% CI, 1.072-21.756; P = 0.040). Post-rehabilitation DEMMI ≥ 20 remained significantly association with overall survival at 3 months after RCC discharge (hazard ratio, 0.237; 95% CI, 0.072-0.785; P = 0.018). Conclusions: Post-rehabilitation functional status of PMV patients was independently associated with weaning success, as well as hospital and 3-month overall survival after RCC discharge. Post-rehabilitation, but not pre-rehabilitation, functional status was a significant parameter associated with weaning success and survival in patients requiring PMV.

Project description:Patients requiring prolonged mechanical ventilation (>21 days) are commonly weaned at long-term acute care hospitals (LTACHs). The most effective method of weaning such patients has not been investigated.To compare weaning duration with pressure support vs unassisted breathing through a tracheostomy collar in patients transferred to an LTACH for weaning from prolonged ventilation.Between 2000 and 2010, a randomized study was conducted in tracheotomized patients transferred to a single LTACH for weaning from prolonged ventilation. Of 500 patients who underwent a 5-day screening procedure, 316 did not tolerate the procedure and were randomly assigned to receive weaning with pressure support (n = 155) or a tracheostomy collar (n = 161). Survival at 6- and 12-month time points was also determined.Primary outcome was weaning duration. Secondary outcome was survival at 6 and 12 months after enrollment.Of 316 patients, 4 were withdrawn and not included in analysis. Of 152 patients in the pressure-support group, 68 (44.7%) were weaned; 22 (14.5%) died. Of 160 patients in the tracheostomy collar group, 85 (53.1%) were weaned; 16 (10.0%) died. Median weaning time was shorter with tracheostomy collar use (15 days; interquartile range [IQR], 8-25) than with pressure support (19 days; IQR, 12-31), P = .004. The hazard ratio (HR) for successful weaning rate was higher with tracheostomy collar use than with pressure support (HR, 1.43; 95% CI, 1.03-1.98; P = .033) after adjusting for baseline clinical covariates. Use of the tracheostomy collar achieved faster weaning than did pressure support among patients who did not tolerate the screening procedure between 12 and 120 hours (HR, 3.33; 95% CI, 1.44-7.70; P = .005), whereas weaning time was equivalent with the 2 methods in patients who did not tolerate the screening procedure within 0 to 12 hours. Mortality was equivalent in the pressure-support and tracheostomy collar groups at 6 months (55.92% vs 51.25%; 4.67% difference, 95% CI, -6.4% to 15.7%) and at 12 months (66.45% vs 60.00%; 6.45% difference, 95% CI, -4.2% to 17.1%).Among patients requiring prolonged mechanical ventilation and treated at a single long-term care facility, unassisted breathing through a tracheostomy, compared with pressure support, resulted in shorter median weaning time, although weaning mode had no effect on survival at 6 and 12 months.clinicaltrials.gov Identifier: NCT01541462.

Project description:BackgroundAn increasing number of patients require prolonged mechanical ventilation (PMV) to survive recovery from critical care. It should be emphasized that PMV is a neglected disease in chest medicine. We investigated 6 years of clinical outcomes and long-term survival rates of patients who required PMV.MethodsWe analyzed retrospectively data from patients in respiratory care center (RCC) to investigate the main causes of respiratory failure leading patients to require PMV. We also studied the factors that influence the ventilator weaned rate, factors that influence the long-term ventilator dependence of patients who require PMV, as well as patients' hospital mortality and long-term survival rates.ResultsA total of 574 patients were admitted to RCC during the 6 years. Of these, 428 patients (74.6%) were older than 65 years. A total of 391 patients (68.1%) were successfully weaned from the ventilator while 83 patients (14.4%) were unsuccessfully weaned. A total of 95 patients (16.6%) died during RCC hospitalization. The most common cause of acute respiratory failure leading to patients requiring PMV was pneumonia. The factor that affected whether patients were successfully weaned from the ventilator was the cause of the respiratory failure that lead patients to require PMV. Our hospital mortality rate was 32.4%; the 1-year survival rate was 24.3%. There was a strong correlation between higher patient age and higher hospital mortality rate and poor 1-year survival rate. Patients with no comorbidity demonstrated good 1-year survival rates. Patients with four comorbidities and patients with end-stage renal disease requiring hemodialysis comorbidity showed poor 1-year survival rates.ConclusionsThe factor that affected whether patients were successfully weaned from the ventilator was the cause of the respiratory failure that lead patients to require PMV. Older patients, patients with renal failure requiring hemodialysis, and those with numerous comorbidities demonstrated poor long-term survival. The reviews of this paper are available via the supplemental material section.

Dataset Information

Explainable Machine Learning to Predict Successful Weaning Among Patients Requiring Prolonged Mechanical Ventilation: A Retrospective Cohort Study in Central Taiwan.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets