Establishment of ICU Mortality Risk Prediction Models with Machine Learning Algorithm Using MIMIC-IV Database

Comparative analysis of dexmedetomidine, midazolam, and propofol impact on epilepsy-related mortality in the ICU: insights from the MIMIC-IV database

BACKGROUND

Dexmedetomidine (Dex), midazolam, and propofol are three distinct sedatives characterized by varying pharmacological properties. Previous literature has indicated the positive impact of each of these sedatives on ICU patients. However, there is a scarcity of clinical evidence comparing the efficacy of Dex, midazolam, and propofol in reducing mortality among people with epilepsy (PWE). This study aimed to assess the impact of Dex, midazolam, and propofol on the survival of PWE.

METHODS

The data were retrospectively retrieved from the Medical Information Mart for Intensive Care (MIMIC)-IV database (version 2.0). PWE were categorized into Dex, midazolam, and propofol groups based on the intravenously administered sedatives. PWE without standard drug therapy were included in the control group. Comparative analyses were performed on the data among the groups.

RESULTS

The Dex group exhibited a significantly lower proportion of in-hospital deaths and a markedly higher in-hospital survival time compared to the midazolam and propofol groups (p < 0.01) after propensity score matching. Kaplan-Meier curves demonstrated a significant improvement in survival rates for the Dex group compared to the control group (p = 0.025). Analysis of Variance (ANOVA) revealed no significant differences in survival rates among the Dex, midazolam, and propofol groups (F = 1.949, p = 0.143). The nomogram indicated that compared to midazolam and propofol groups, Dex was more effective in improving the survival rate of PWE.

CONCLUSION

Dex might improve the survival rate of PWE in the ICU compared to no standard drug intervention. However, Dex did not exhibit superiority in improving survival rates compared to midazolam and propofol.

Exploring Predictive Factors for Heart Failure Progression in Hypertensive Patients Based on Medical Diagnosis Data from the MIMIC-IV Database

Heart failure is associated with a significant mortality rate, and an elevated prevalence of this condition has been noted among hypertensive patients. The identification of predictive factors for heart failure progression in hypertensive individuals is crucial for early intervention and improved patient outcomes. In this study, we aimed to identify these predictive factors by utilizing medical diagnosis records for hypertension patients from the MIMIC-IV database. In particular, we employed only diagnostic history prior to hypertension to enable patients to anticipate the onset of heart failure at the moment of hypertension diagnosis. In the methodology, chi-square tests and XGBoost modeling were applied to examine age-specific predictive factors across four groups: AL (all ages), G1 (0 to 65 years), G2 (65 to 80 years), and G3 (over 80 years). As a result, the chi-square tests identified 34, 28, 20, and 10 predictive factors for the AL, G1, G2, and G3 groups, respectively. Meanwhile, the XGBoost modeling uncovered 19, 21, 27, and 33 predictive factors for these respective groups. Ultimately, our findings reveal 21 overall predictive factors, encompassing conditions such as atrial fibrillation, the use of anticoagulants, kidney failure, obstructive pulmonary disease, and anemia. These factors were assessed through a comprehensive review of the existing literature. We anticipate that the results will offer valuable insights for the risk assessment of heart failure in hypertensive patients.

Explainable artificial intelligence model for mortality risk prediction in the intensive care unit: a derivation and validation study

BACKGROUND

The lack of transparency is a prevalent issue among the current machine-learning (ML) algorithms utilized for predicting mortality risk. Herein, we aimed to improve transparency by utilizing the latest ML explicable technology, SHapley Additive exPlanation (SHAP), to develop a predictive model for critically ill patients.

METHODS

We extracted data from the Medical Information Mart for Intensive Care IV database, encompassing all intensive care unit admissions. We employed nine different methods to develop the models. The most accurate model, with the highest area under the receiver operating characteristic curve, was selected as the optimal model. Additionally, we used SHAP to explain the workings of the ML model.

RESULTS

The study included 21 395 critically ill patients, with a median age of 68 years (interquartile range, 56-79 years), and most patients were male (56.9%). The cohort was randomly split into a training set (N = 16 046) and a validation set (N = 5349). Among the nine models developed, the Random Forest model had the highest accuracy (87.62%) and the best area under the receiver operating characteristic curve value (0.89). The SHAP summary analysis showed that Glasgow Coma Scale, urine output, and blood urea nitrogen were the top three risk factors for outcome prediction. Furthermore, SHAP dependency analysis and SHAP force analysis were used to interpret the Random Forest model at the factor level and individual level, respectively.

CONCLUSION

A transparent ML model for predicting outcomes in critically ill patients using SHAP methodology is feasible and effective. SHAP values significantly improve the explainability of ML models.

Machine Learning for Benchmarking Critical Care Outcomes

OBJECTIVES

Enhancing critical care efficacy involves evaluating and improving system functioning. Benchmarking, a retrospective comparison of results against standards, aids risk-adjusted assessment and helps healthcare providers identify areas for improvement based on observed and predicted outcomes. The last two decades have seen the development of several models using machine learning (ML) for clinical outcome prediction. ML is a field of artificial intelligence focused on creating algorithms that enable computers to learn from and make predictions or decisions based on data. This narrative review centers on key discoveries and outcomes to aid clinicians and researchers in selecting the optimal methodology for critical care benchmarking using ML.

METHODS

We used PubMed to search the literature from 2003 to 2023 regarding predictive models utilizing ML for mortality (592 articles), length of stay (143 articles), or mechanical ventilation (195 articles). We supplemented the PubMed search with Google Scholar, making sure relevant articles were included. Given the narrative style, papers in the cohort were manually curated for a comprehensive reader perspective.

RESULTS

Our report presents comparative results for benchmarked outcomes and emphasizes advancements in feature types, preprocessing, model selection, and validation. It showcases instances where ML effectively tackled critical care outcome-prediction challenges, including nonlinear relationships, class imbalances, missing data, and documentation variability, leading to enhanced results.

CONCLUSIONS

Although ML has provided novel tools to improve the benchmarking of critical care outcomes, areas that require further research include class imbalance, fairness, improved calibration, generalizability, and long-term validation of published models.

Development and verification of a nomogram for predicting short-term mortality in elderly ischemic stroke populations

Stroke is a major healthcare problem worldwide, particularly in the elderly population. Despite limited research on the development of prediction models for mortality in elderly individuals with ischemic stroke, our study aimed to address this knowledge gap. By leveraging data from the Medical Information Mart for Intensive Care IV database, we collected comprehensive raw data pertaining to elderly patients diagnosed with ischemic stroke. Through meticulous screening of clinical variables associated with 28-day mortality, we successfully established a robust nomogram. To assess the performance and clinical utility of our nomogram, various statistical analyses were conducted, including the concordance index, integrated discrimination improvement (IDI), net reclassification index (NRI), calibration curves and decision curve analysis (DCA). Our study comprised a total of 1259 individuals, who were further divided into training (n = 894) and validation (n = 365) cohorts. By identifying several common clinical features, we developed a nomogram that exhibited a concordance index of 0.809 in the training dataset. Notably, our findings demonstrated positive improvements in predictive performance through the IDI and NRI analyses in both cohorts. Furthermore, calibration curves indicated favorable agreement between the predicted and actual incidence of mortality (P > 0.05). DCA curves highlighted the substantial net clinical benefit of our nomogram compared to existing scoring systems used in routine clinical practice. In conclusion, our study successfully constructed and validated a prognostic nomogram, which enables accurate short-term mortality prediction in elderly individuals with ischemic stroke.