How to cite this article: Patnaik RK, Karan N. Synergizing Survival: Uniting Acute Gastrointestinal Injury Grade and Disease Severity Scores in Critical Care Prognostication. Indian J Crit Care Med 2024;28(6):529-530.

BACKGROUND: There is a growing demand for advanced methods to improve the understanding and prediction of illnesses. This study focuses on Sepsis, a critical response to infection, aiming to enhance early detection and mortality prediction for Sepsis-3 patients to improve hospital resource allocation.
METHODS: In this study, we developed a Machine Learning (ML) framework to predict the 30-day mortality rate of ICU patients with Sepsis-3 using the MIMIC-III database. Advanced big data extraction tools like Snowflake were used to identify eligible patients. Decision tree models and Entropy Analyses helped refine feature selection, resulting in 30 relevant features curated with clinical experts. We employed the Light Gradient Boosting Machine (LightGBM) model for its efficiency and predictive power.
RESULTS: The study comprised a cohort of 9118 Sepsis-3 patients. Our preprocessing techniques significantly improved both the AUC and accuracy metrics. The LightGBM model achieved an impressive AUC of 0.983 (95% CI: [0.980-0.990]), an accuracy of 0.966, and an F1-score of 0.910. Notably, LightGBM showed a substantial 6% improvement over our best baseline model and a 14% enhancement over the best existing literature. These advancements are attributed to (I) the inclusion of the novel and pivotal feature Hospital Length of Stay (HOSP_LOS), absent in previous studies, and (II) LightGBM\'s gradient boosting architecture, enabling robust predictions with high-dimensional data while maintaining computational efficiency, as demonstrated by its learning curve.
CONCLUSIONS: Our preprocessing methodology reduced the number of relevant features and identified a crucial feature overlooked in previous studies. The proposed model demonstrated high predictive power and generalization capability, highlighting the potential of ML in ICU settings. This model can streamline ICU resource allocation and provide tailored interventions for Sepsis-3 patients.

UNASSIGNED: The application of machine learning in health care often necessitates the use of hierarchical codes such as the International Classification of Diseases (ICD) and Anatomical Therapeutic Chemical (ATC) systems. These codes classify diseases and medications, respectively, thereby forming extensive data dimensions. Unsupervised feature selection tackles the \"curse of dimensionality\" and helps to improve the accuracy and performance of supervised learning models by reducing the number of irrelevant or redundant features and avoiding overfitting. Techniques for unsupervised feature selection, such as filter, wrapper, and embedded methods, are implemented to select the most important features with the most intrinsic information. However, they face challenges due to the sheer volume of ICD and ATC codes and the hierarchical structures of these systems.
UNASSIGNED: The objective of this study was to compare several unsupervised feature selection methods for ICD and ATC code databases of patients with coronary artery disease in different aspects of performance and complexity and select the best set of features representing these patients.
UNASSIGNED: We compared several unsupervised feature selection methods for 2 ICD and 1 ATC code databases of 51,506 patients with coronary artery disease in Alberta, Canada. Specifically, we used the Laplacian score, unsupervised feature selection for multicluster data, autoencoder-inspired unsupervised feature selection, principal feature analysis, and concrete autoencoders with and without ICD or ATC tree weight adjustment to select the 100 best features from over 9000 ICD and 2000 ATC codes. We assessed the selected features based on their ability to reconstruct the initial feature space and predict 90-day mortality following discharge. We also compared the complexity of the selected features by mean code level in the ICD or ATC tree and the interpretability of the features in the mortality prediction task using Shapley analysis.
UNASSIGNED: In feature space reconstruction and mortality prediction, the concrete autoencoder-based methods outperformed other techniques. Particularly, a weight-adjusted concrete autoencoder variant demonstrated improved reconstruction accuracy and significant predictive performance enhancement, confirmed by DeLong and McNemar tests (P<.05). Concrete autoencoders preferred more general codes, and they consistently reconstructed all features accurately. Additionally, features selected by weight-adjusted concrete autoencoders yielded higher Shapley values in mortality prediction than most alternatives.
UNASSIGNED: This study scrutinized 5 feature selection methods in ICD and ATC code data sets in an unsupervised context. Our findings underscore the superiority of the concrete autoencoder method in selecting salient features that represent the entire data set, offering a potential asset for subsequent machine learning research. We also present a novel weight adjustment approach for the concrete autoencoders specifically tailored for ICD and ATC code data sets to enhance the generalizability and interpretability of the selected features.

UNASSIGNED: The Charlson Comorbidity Index ≥2, in-Hospital onset, Albumin <2.5 g/dL, altered Mental status, Eastern Cooperative Oncology Group Performance status ≥2, Steroid use (CHAMPS) score is a novel and promising prognostic tool. We present an initial external validation of the CHAMPS score for predicting mortality in acute nonvariceal upper gastrointestinal bleeding (NVUGIB) across multiple clinical outcomes.
UNASSIGNED: A prospective cohort study was conducted on adult patients with NVUGIB admitted to the Department of Gastroenterology between November 2022 and June 2023. The CHAMPS score performance in predicting in-hospital outcomes was evaluated by employing area under the receiver operating characteristic (AUROC) curves, followed by a comparative analysis with five pre-existing scores.
UNASSIGNED: A total of 140 patients were included in the study. The CHAMPS score showed its highest performance in predicting mortality rates (AUROC = 0.89), significantly outperforming the Glasgow-Blatchford Bleeding Score (GBS) as well as the Albumin level <3.0 mg/dL, International normalized ratio >1.5, altered Mental status, Systolic blood pressure ≤90 mmHg, and age >65 years (AIMS65) score (AUROC = 0.72 and 0.71, respectively; all p < 0.05). Subgroup analysis for bleeding-related and non-bleeding-related mortality further confirmed the robust predictive capability of the CHAMPS score (AUROC = 0.88 and 0.87, respectively). The CHAMPS score failed to predict rebleeding and intervention reliably, exhibiting AUROC values of 0.43 and 0.55, respectively. The optimal CHAMPS score cutoff value for predicting mortality was 3 points, achieving 100% sensitivity and 71.2% specificity. In the low-risk category defined by both CHAMPS and GBS scores, mortality and rebleeding rates were 0%. However, within the CHAMPS score-based low-risk group, 58.8% required intervention, contrasting with a 0% intervention rate for the GBS score-based low-risk group (GBS score ≤1).
UNASSIGNED: The CHAMPS score consistently demonstrated a robust predictive performance for mortality (AUROC > 0.8), facilitating the identification of high-risk patients requiring aggressive treatment and low-risk patients in need of localized treatment or safe discharge after successful bleeding control.

UNASSIGNED: Heart failure (HF) is a global health challenge affecting millions, with significant variations in patient characteristics and outcomes based on ejection fraction. This study aimed to differentiate between HF with reduced ejection fraction (HFrEF) and HF with preserved ejection fraction (HFpEF) with respect to patient characteristics, risk factors, comorbidities, and clinical outcomes, incorporating advanced machine learning models for mortality prediction.
UNASSIGNED: The study included 1861 HF patients from 21 centers in Jordan, categorized into HFrEF (EF <40%) and HFpEF (EF ≥ 50%) groups. Data were collected from 2021 to 2023, and machine learning models were employed for mortality prediction.
UNASSIGNED: Among the participants, 29.7% had HFpEF and 70.3% HFrEF. Significant differences were noted in demographics and comorbidities, with a higher prevalence of males, younger age, smoking, and familial history of premature ASCVD in the HFrEF group. HFpEF patients were typically older, with higher rates of diabetes, hypertension, and obesity. Machine learning analysis, mainly using the Random Forest Classifier, demonstrated significant predictive capability for mortality with an accuracy of 0.9002 and an AUC of 0.7556. Other models, including Logistic Regression, SVM, and XGBoost, also showed promising results. Length of hospital stay, need for mechanical ventilation, and number of hospital admissions were the top predictors of mortality in our study.
UNASSIGNED: The study underscores the heterogeneity in patient profiles between HFrEF and HFpEF. Integrating machine learning models offers valuable insights into mortality risk prediction in HF patients, highlighting the potential of advanced analytics in improving patient care and outcomes.

UNASSIGNED: This study investigated potential predictive models associated with natural killer (NK) cell mitochondrial membrane potential (MMP or ΔΨm) in predicting death among critically ill patients with COVID-19.
UNASSIGNED: We included 97 patients with COVID-19 of different severities attending Peking Union Medical College Hospital from December 2022 to January 2023. Patients were divided into three groups according to oxygen and mechanical ventilation use during specimen collection and were followed for survival and death at 3 months. The lymphocyte subpopulation MMP was detected via flow cytometry. We constructed a joint diagnostic model by integrating identified key indicators and generating receiver operating curves (ROCs) and evaluated its predictive performance for mortality risk in critically ill patients.
UNASSIGNED: The NK-cell MMP median fluorescence intensity (MFI) was significantly lower in critically ill patients who died from COVID-19 (p<0.0001) and significantly and positively correlated with D-dimer content in critically ill patients (r=0.56, p=0.0023). The random forest model suggested that fibrinogen levels and NK-cell MMP MFI were the most important indicators. Integrating the above predictive models for the ROC yielded an area under the curve of 0.94.
UNASSIGNED: This study revealed the potential of combining NK-cell MMP with key clinical indicators (D-dimer and fibrinogen levels) to predict death among critically ill patients with COVID-19, which may help in early risk stratification of critically ill patients and improve patient care and clinical outcomes.

UNASSIGNED: Pulmonary hypertension (PH) frequently complicates the course of patients with left heart disease (PH-LHD) and is associated with worse clinical outcomes. Mortality calculators for PH-LHD are lacking, and it is unclear whether any risk prediction tools originally derived from other forms of PH can accurately predict outcomes in patients with PH-LHD.
UNASSIGNED: We retrospectively analyzed data from 161 patients diagnosed with PH-LHD referred to our pulmonary hypertension center from 2016 to 2022. We calculated the Registry to Evaluate Early and Long-Term PAH Disease Management (REVEAL 2.0) risk score and categorized patients as low, intermediate, or high-risk. We assessed survival at 1 and 3 years using Kaplan-Meier and Cox proportional hazards, as well as classification performance using a concordance index.
UNASSIGNED: At the first outpatient visit, 15% of patients were stratified as low-risk, 27% as intermediate, and 57% as high-risk. Cumulative 1-year survival rates were 100%, 94%, and 91% for the low, intermediate, and high-risk strata, respectively. Cumulative 3-year survival rates were 96%, 89%, and 70% for the low, intermediate, and high-risk strata, respectively. We found no difference in outcomes at 1 year between risk groups. High-risk patients had an increased risk of death at 3 years using REVEAL 2.0 (HR 5.32, p < 0.001). However, while REVEAL 2.0 accurately discriminated high-risk patients, the hazard ratio was not statistically different between patients classified as intermediate-risk compared to low-risk.
UNASSIGNED: REVEAL 2.0 accurately predicted 3-year survival in PH-LHD patients with high-risk features. However, the mortality risk between patients classified as intermediate-risk was not different from the low-risk stratum, suggesting inaccurate classification for this group of patients.

BACKGROUND: Norepinephrine (NE) is a cornerstone drug in the management of septic shock, with its dose being used clinically as a marker of disease severity and as mortality predictor. However, variations in NE dose reporting either as salt formulations or base molecule may lead to misinterpretation of mortality risks and hinder the process of care.
METHODS: We conducted a retrospective analysis of the MIMIC-IV database to assess the impact of NE dose reporting heterogeneity on mortality prediction in a cohort of septic shock patients. NE doses were converted from the base molecule to equivalent salt doses, and their ability to predict 28-day mortality at common severity dose cut-offs was compared.
RESULTS: 4086 eligible patients with septic shock were identified, with a median age of 68 [57-78] years, an admission SOFA score of 7 [6-10], and lactate at diagnosis of 3.2 [2.4-5.1] mmol/L. Median peak NE dose at day 1 was 0.24 [0.12-0.42] μg/kg/min, with a 28-day mortality of 39.3%. The NE dose showed significant heterogeneity in mortality prediction depending on which formulation was reported, with doses reported as bitartrate and tartrate presenting 65 (95% CI 79-43)% and 67 (95% CI 80-47)% lower ORs than base molecule, respectively. This divergence in prediction widened at increasing NE doses. When using a 1 μg/kg/min threshold, predicted mortality was 54 (95% CI 52-56)% and 83 (95% CI 80-87)% for tartrate formulation and base molecule, respectively.
CONCLUSIONS: Heterogeneous reporting of NE doses significantly affects mortality prediction in septic shock. Standardizing NE dose reporting as base molecule could enhance risk stratification and improve processes of care. These findings underscore the importance of consistent NE dose reporting practices in critical care settings.

Tissue hypoxia is associated with the development of organ dysfunction and death in critically ill patients commonly captured using blood lactate. The kinetic parameters of serial lactate evaluations are superior at predicting mortality compared with single values. S-adenosylhomocysteine (SAH), which is also associated with hypoxia, was recently established as a useful predictor of septic organ dysfunction and death. We evaluated the performance of kinetic SAH parameters for mortality prediction compared with lactate parameters in a cohort of critically ill patients. For lactate and SAH, maxima and means as well as the normalized area scores were calculated for two periods: the first 24 h and the total study period of up to five days following ICU admission. Their performance in predicting in-hospital mortality were compared in 99 patients. All evaluated parameters of lactate and SAH were significantly higher in non-survivors compared with survivors. In univariate analysis, the predictive power for mortality of SAH was higher compared with lactate in all forms of application. Multivariable models containing SAH parameters demonstrated higher predictive values for mortality than models based on lactate parameters. The optimal models for mortality prediction incorporated both lactate and SAH parameters. Compared with lactate, SAH displayed stronger predictive power for mortality in static and dynamic application in critically ill patients.

OBJECTIVE: Existing approaches to fairness evaluation often overlook systematic differences in the social determinants of health, like demographics and socioeconomics, among comparison groups, potentially leading to inaccurate or even contradictory conclusions. This study aims to evaluate racial disparities in predicting mortality among patients with chronic diseases using a fairness detection method that considers systematic differences.
METHODS: We created five datasets from Mass General Brigham\'s electronic health records (EHR), each focusing on a different chronic condition: congestive heart failure (CHF), chronic kidney disease (CKD), chronic obstructive pulmonary disease (COPD), chronic liver disease (CLD), and dementia. For each dataset, we developed separate machine learning models to predict 1-year mortality and examined racial disparities by comparing prediction performances between Black and White individuals. We compared racial fairness evaluation between the overall Black and White individuals versus their counterparts who were Black and matched White individuals identified by propensity score matching, where the systematic differences were mitigated.
RESULTS: We identified significant differences between Black and White individuals in age, gender, marital status, education level, smoking status, health insurance type, body mass index, and Charlson comorbidity index (p-value < 0.001). When examining matched Black and White subpopulations identified through propensity score matching, significant differences between particular covariates existed. We observed weaker significance levels in the CHF cohort for insurance type (p = 0.043), in the CKD cohort for insurance type (p = 0.005) and education level (p = 0.016), and in the dementia cohort for body mass index (p = 0.041); with no significant differences for other covariates. When examining mortality prediction models across the five study cohorts, we conducted a comparison of fairness evaluations before and after mitigating systematic differences. We revealed significant differences in the CHF cohort with p-values of 0.021 and 0.001 in terms of F1 measure and Sensitivity for the AdaBoost model, and p-values of 0.014 and 0.003 in terms of F1 measure and Sensitivity for the MLP model, respectively.
CONCLUSIONS: This study contributes to research on fairness assessment by focusing on the examination of systematic disparities and underscores the potential for revealing racial bias in machine learning models used in clinical settings.