Robust data normalization and analysis are pivotal in biomedical research to ensure that observed differences in populations are directly attributable to the target variable, rather than disparities between control and study groups. ArsHive addresses this challenge using advanced algorithms to normalize populations (e.g., control and study groups) and perform statistical evaluations between demographic, clinical, and other variables within biomedical datasets, resulting in more balanced and unbiased analyses. The tool’s functionality extends to comprehensive data reporting, which elucidates the effects of data processing, while maintaining dataset integrity. Additionally, ArsHive is complemented by A.D.A. (Autonomous Digital Assistant), which employs OpenAI’s GPT-4 model to assist researchers with inquiries, enhancing the decision-making process. In this proof-of-concept study, we tested ArsHive on three different datasets derived from proprietary data, demonstrating its effectiveness in managing complex clinical and therapeutic information and highlighting its versatility for diverse research fields.

Delirium presents a significant clinical challenge, primarily due to its profound impact on patient outcomes and the limitations of the current diagnostic methods, which are largely subjective. During the COVID-19 pandemic, this challenge was intensified as the frequency of delirium assessments decreased in Intensive Care Units (ICUs), even as the prevalence of delirium among critically ill patients increased. The present study evaluated how the serum molecular fingerprint, as acquired by Fourier-Transform InfraRed (FTIR) spectroscopy, can enable the development of predictive models for delirium. A preliminary univariate analysis of serum FTIR spectra indicated significantly different bands between 26 ICU patients with delirium and 26 patients without, all of whom were admitted with COVID-19. However, these bands resulted in a poorly performing Naïve-Bayes predictive model. Considering the use of a Fast-Correlation-Based Filter for feature selection, it was possible to define a new set of spectral bands with a wider coverage of molecular functional groups. These bands ensured an excellent Naïve-Bayes predictive model, with an AUC, a sensitivity, and a specificity all exceeding 0.92. These spectral bands, acquired through a minimally invasive analysis and obtained rapidly, economically, and in a high-throughput mode, therefore offer significant potential for managing delirium in critically ill patients.

Delirium is a severe and common complication among critically ill patients, particularly those with SARS-CoV-2 infection, contributing to increased morbidity and mortality. Early identification of at-risk patients is crucial for timely intervention and improved outcomes. This prospective observational cohort study explores the potential of electroencephalography (EEG) combined with machine learning (ML) models for predicting delirium in critically ill patients with SARS-CoV-2 infection. A stepwise modeling approach was applied, starting with the independent analysis of specific EEG variables to assess their predictive value. Subsequently, three ML models were developed using data from 70 patients (31 with delirium, 39 without): two relied solely on EEG data, while the third integrated demographic, clinical, laboratory, and EEG data. An additional model analyzed EEG data before and after delirium diagnosis in 11 patients. Several EEG features were identified as predictors of delirium, with increased theta activity emerging as the most consistent. The best EEG-only model achieved an area under the curve (AUC) of 0.733 (sensitivity = 0.645, specificity = 0.692), indicating moderate predictive performance. Including demographic, clinical, and laboratory variables improved performance (AUC = 0.825, sensitivity = 0.613, specificity = 0.795). The model analyzing EEG features before and after delirium diagnosis achieved the highest accuracy (AUC = 0.950, sensitivity and specificity = 0.818), reinforcing the value of EEG-based monitoring. EEG-based ML models show promise for predicting delirium in critically ill patients, with increased theta activity identified as a key predictor. However, their moderate AUC, sensitivity, and specificity highlight the need for further refinement. Graphical Abstract

Delirium is a common and underrecognized complication among critically ill patients, associated with prolonged ICU stays, cognitive dysfunction, and increased mortality. Its multifactorial causes and fluctuating course hinder early prediction, limiting timely management. Predictive models based on data available at ICU admission may help to identify high-risk patients and guide early interventions. This study evaluated machine learning models used to predict delirium in critically ill patients with SARS-CoV-2 infections using a prospective cohort of 426 patients. The dataset included demographic characteristics, clinical data (e.g., comorbidities, medication, reason for ICU admission, interventions), and routine lab test results. Five models—Logistic Regression, Support Vector Machine, Decision Tree, Random Forest, and Naïve Bayes—were developed using 112 features. Feature selection relied on Information Gain, and model performance was assessed via 10-fold cross-validation. The Naïve Bayes model showed moderate predictive performance and high interpretability, achieving an AUC of 0.717, accuracy of 65.3%, sensitivity of 62.4%, specificity of 68.1%, and precision of 66.2%. Key predictors included invasive mechanical ventilation, deep sedation with benzodiazepines, SARS-CoV-2 as the reason for ICU admission, ECMO use, constipation, and male sex. These findings support the use of interpretable models for early delirium risk stratification using routinely available ICU data.