Vinay Gandhi Mukkelli

Background and Aims: Sepsis is a common cause of mortality and morbidity in patients admitted to the intensive care unit (ICU), requiring reliable scoring systems to predict outcomes and guide management. Existing traditional scoring systems inadequately capture sepsis complexity. This study developed the Organ Failure Assessment by Ratio (OFAR) score to improve ICU mortality prediction in sepsis patients. Methods: This retrospective observational study was conducted in the ICUs of a tertiary care centre between January 2021 to December 2023 and included adult patients meeting Sepsis-3 criteria. Demographic data, ICU mortality, and six composite variables were collected: bilirubin-to-albumin ratio, pulse oximetric saturation to fraction of inspired oxygen ratio (SpO ₂ /FiO ₂ ratio), shock index, lactate-to-bicarbonate ratio, urea-to-estimated glomerular filtration rate ratio, and cumulative lung ultrasound score-to-platelet ratio. These variables were evaluated to develop and assess the predictive accuracy of OFAR score for ICU mortality. Results: The OFAR score showed greater discriminatory power with an area under the curve (AUC) of 0.78 than quick Sequential Organ Failure Assessment score (qSOFA) (AUC: 0.70) and Acute Physiology and Chronic Health Evaluation score II (APACHE II) (AUC: 0.73) and was comparable to Sequential Organ Failure Assessment score (SOFA) (AUC: 0.76). With a cut-off score of 51.6, OFAR score demonstrated a sensitivity of 73.7% and a specificity of 72.1%. Conclusion: The OFAR score provides higher predictive power for ICU mortality than qSOFA and APACHE II in our study population, raising its potential as an alternative risk stratification scoring system in sepsis patients. Additional external validation is required to confirm its generalisability across diverse populations.

Purpose Intensive care units (ICUs) handle mechanically ventilated patients with life-threatening conditions, who require intensive monitoring and treatment. In a low physician-patient ratio setting, providing consistent care to all patients is challenging. A survival prediction model using machine-learning can potentially improve prognosis evaluation and resource allocation. This study aims to develop a machine-learning model to predict survival/mortality in mechanically ventilated patients using clinical features recorded at the time of ICU admission and compare its performance with the Sequential Organ Failure Assessment (SOFA) score as a standalone predictor. Methods A dataset consisting of 660 mechanically ventilated patients and 98 clinical parameters (n = 660, Male: Female = 365:295, Age = 44.45 ± 19.36 years) from three ICUs at AIIMS, Delhi, was retrospectively evaluated after institutional ethical approval. Binary classification models were trained using 10-fold cross-validation with 70% data and 30% reserved for testing. The outcome was based on the survival/death of the patient during their ICU stay. Results A total of 39 features were selected using Shapley-Additive-Explanations (SHAP) and Random Forest model. The top three features were SOFA score, International normalized ratio (INR) and respiratory rate with feature importance values of 7.3%, 4.5% and 3.4% respectively. The K-nearest-neighbour (KNN) model using SHAP-selected features achieved the best test performance with an accuracy = 0.80, area-under-receiver-operating-characteristics-curve (AUROC) = 0.84, sensitivity = 0.82, specificity = 0.77, positive-predictive-value (PPV) = 0.78 and negative-predictive-value (NPV) = 0.82, compared to the SOFA-only model showing accuracy = 0.73, AUROC = 0.73, sensitivity = 0.82, specificity = 0.63, PPV = 0.69 and NPV = 0.78. Conclusion The automated machine-learning method for prognosis prediction may assist clinicians in the early triage of patients. These models may offer valuable support to ICU physicians for timely alerts and informed clinical judgment. The study also highlights the continued utility of the SOFA score used by clinicians as the first assessment tool in ICUs, while suggesting that carefully developed machine-learning models may offer complementary support in high-risk ICU settings.

Background: Mechanical ventilation is essential in intensive care units (ICUs) but poses risks such as ventilator-associated complications and high costs. The accuracy of predicting mechanical ventilation duration using clinical information is limited. Predicting ventilation duration accurately can aid clinical decisions like resource-allocation and early tracheostomy-planning. Objective: To develop explainable artificial intelligence (AI) models for predicting mechanical ventilation duration leveraging diverse clinical parameters from ICU patient data. Methodology: This development and testing study analysed 323 mechanically ventilated patients {(n = 323, Male:Female = 160:163, Age = 42.87 ± 19.54 years (mean ± standard deviation)} from three ICUs at AIIMS, Delhi. The dataset included 100-clinical parameters per patient. Two models were developed: (1) A regression model (n = 323) to predict ventilation duration in days, and (2) A classification model (n = 218, non-tracheostomized) to predict short- (≤3 days) vs. long-term (>3 days) ventilation requirements. The misclassification-cost was altered for the classification model. Feature selection was performed using Shapley additive explanations (SHAP) on a random forest model, and training was done with 5-fold cross-validation (80% training, 20% testing). Results: The least-squares boosting regression model achieved root mean squared error (RMSE) of 4.66 days and coefficient of Determination (R²) of 0.65 using 34-SHAP-selected features, with tracheostomy (53.66% importance) being the top predictor. The best classification model, K-nearest neighbours, achieved 79.1% accuracy, Area under the receiver-operating-characteristic-curve (AUROC) of 0.82, sensitivity of 71.4%, and specificity of 86.4% using 47-SHAP-selected features. Key predictors included ICU admission type (8.1%), PO ₂ (5.6%), and pH (5%). Conclusion: AI-driven prediction of ventilation duration can enhance ICU workflows, optimize resource use, and improve personalized care. SHAP-based feature selection promotes AI interpretability, aiding clinical adoption.