Physiological Measurement

BackgroundObstructive sleep apnea (OSA) is a highly prevalent sleep disorder. Major risk factors include obesity, older age, and male sex. Given the well-established association between OSA and cardiovascular disease, markers of cardiovascular health derived from electrocardiography (ECG) may help identify individuals with undiagnosed OSA. This study evaluated whether 12-lead ECG features improve machine learning-based prediction of OSA beyond established clinical factors. MethodsWe conducted a retrospective cohort study of individuals who underwent diagnostic sleep studies at the University of Kansas Health System Sleep Disorders Center (2004-2025) and had a 12-lead ECG acquired within one year. Standard ECG features (intervals, amplitudes, voltage-time integrals, morphology descriptors) were extracted from the representative beat. Latent representations were derived using two variational autoencoder (VAE) models pretrained on >1 million ECGs: a 30-dimensional representative-beat VAE and a 90-dimensional full 10-second ECG VAE. Three LightGBM models were trained: clinical factors only (age, sex, weight, height, body mass index [BMI], Epworth sleepiness scale [ESS]), ECG features only, and a combined model. Feature selection and hyperparameter tuning were performed separately for each model within a 5-fold cross-validation scheme. The outcome was moderate-to-severe OSA (AHI[≥]15 events/h), evaluated by AUROC (bootstrapped 95% CI) in a 10% holdout test set. ResultsThe dataset included 7,699 individuals (52.2% women, mean age 58.0 {+/-} 14.2 years, BMI 33.6{+/-}8.2 kg/m{superscript 2}, ESS 8.8{+/-}5.1; 36.4% with moderate-to-severe OSA) Training AUROC (95% CI) was 0.71 (0.66-0.76) for the combined model, versus 0.65 (0.59-0.70) for the clinical-only model and 0.66 (0.61-0.72) for the ECG-only model. Holdout test AUROC was 0.78 (combined), 0.69 (clinical-only), and 0.71 (ECG-only). Feature importance identified age, BMI, weight, ECG VAE latent variables, height, sex, and P-wave voltage-time integral as relevant contributor to the predictions. ConclusionsLatent ECG features extracted by VAE improve prediction of moderate-to-severe OSA when combined with clinical factors.

Explicit timing and context features may complement short electrocardiogram waveforms, but aggregate gains on imbalanced benchmarks can conceal errors in individual classes. We therefore compared Logistic Regression, Random Forest, XGBoost, and a one-dimensional convolutional neural network (CNN) in four cumulative input stages. Stage S1 used a 255-sample beat waveform alone, and stages S2 to S4 added 6, 13, and 21 engineered features. Using the inter-patient division of the MIT-BIH Arrhythmia Database, with 50,992 development and 49,686 evaluation beats in four classes, we estimated 95% intervals with a paired bootstrap over the 22 evaluation recordings. Adding six RR interval features increased macro F1 by 0.025 to 0.093, with intervals excluding zero for every model family. Further context helped Random Forest and XGBoost, whereas later CNN gains fell within resampling uncertainty. At the final stage, XGBoost and CNN reached macro F1 of 0.580 and 0.574, and the interval for their difference included zero. Despite opposite operating points, their precision for fusion beats was similar at matched recall, suggesting that neither separated fusion beats better. The most accurate pipeline, Random Forest at S2 with 94.45% accuracy, detected only 1.4% of supraventricular beats, and development and evaluation favored the same stage for only one family. The results for the minority classes were based on single recordings. Record 213 supplied 203 of the 207 correct CNN fusion detections, and elsewhere both final models detected at most 4 of 26 fusion beats. Record 232 contains 75% of the evaluation supraventricular beats, and nearly all were missed. Excluding this record nearly doubled XGBoost supraventricular F1, from 0.199 to 0.383. These findings are limited to retrospective analysis with annotated beat positions and a single training seed.

ECG waveform data, which is increasingly required for artificial intelligence analyses, is often unavailable in raw data formats. Digitization can approximate original raw data from ECG images, but digitizers reconstruct ECG waveforms only approximately. Exact waveform reconstruction is needed to guarantee that any errors, even if small, do not unexpectedly propagate into downstream analyses. Portable document format (PDF) files containing vector graphics store waveforms derived directly from raw data. This improves reconstruction accuracy, but existing PDF reconstruction methods remain approximate because raw data values are rounded during PDF encoding/generation. Despite rounding, however, there is a 1:1 mapping between raw data and PDF-encoded values which can be reversed to recover the original raw data with no error. We developed open-source software in MATLAB and Python (http://github.com/BIVectors/pdf2ECG) for bit-exact waveform reconstruction of General Electric MUSE ECG PDFs based on double quantization of ECG waveform data. The software supports multiple PDF layouts and voltage gains, automatically flags waveform clipping under which exact reconstruction is not possible, and certifies if parameter values used for waveform reconstruction are consistent with parameter values used during PDF encoding. Accuracy was assessed by comparing PDF reconstructed waveforms to raw XML waveforms, and reconstruction parameters were perturbed to determine the accuracy of waveform certification. 1,632 PDF-XML pairs were assessed in 5 layouts, each at 3 gains. After software-identified waveform clipping was excluded, the software reconstructed 100% of ECG waveforms with zero error compared to ground-truth XML. Certification tests were extremely sensitive to minor perturbation in encoding parameters, and if passed, confirmed bit-exact waveform reconstruction without need for ground-truth comparisons. Although this method does not replace the need for general ECG image digitization, for MUSE generated PDFs this method is preferable to rasterization followed by image-based digitization.

Abstract Background: Conventional cuff based methods report single systolic and diastolic pressure values and depend either on Korotkoff sounds or on empirical oscillometric algorithms. Methods: We developed a patented cuff based method, termed plethysmometry, that simultaneously records cuff pressure and finger photoplethysmograms from the cuffed and contralateral arms. The contralateral photoplethysmogram serves as an internal physiological reference for pulse amplitude. Following an initial pressure ramp, 16 cuff pressure plateaus are applied. Absolute and normalized pulse amplitudes in the cuffed arm are related to cuff pressure to estimate lower and upper systolic and diastolic pressures. Mean arterial pressure is estimated independently from cuff pressure oscillation amplitude. The estimations are done by automated algorithms developed in MATLAB. Results: The method reports lower and upper systolic, diastolic and mean arterial pressure ranges (during the recording period), together with their average values. Average pulse pressure is also reported. Graphs are provided for reference to check the accuracy of reported pressures. Conclusion: Plethysmometry provides an objective digital record of the cuff pressure and pulse traces and estimates systolic and diastolic pressure ranges from the pulse response of the individual subject to graded cuff pressure. It also reports mean arterial pressure. Comparison of plethysmometric systolic and diastolic pressures with sphygmomanometrically derived pressures and diagnostic agreement are reported in a companion paper.

Background Implantable loop recorders (ILRs) provide continuous rhythm monitoring. Professional society guidelines and expert consensus statements recommend ILR use for patients with recurrent syncope and cryptogenic stroke, but recommendations are weaker for other indications. We assessed ILR use over time and concordance between indication and clinical guidance. Methods We conducted a retrospective cohort study using Epic Cosmos data from 1/1/2020-12/31/2025. Among active patients with at least one of 13 diagnoses potentially eligible for ILR placement, we identified ILR placements and classified them by concordance with professional society guidance: concordant, possibly, or non-addressed. Annual incidence was estimated overall and by indication. We also described the proportion of patients who received longer-term (>7 days) rhythm monitoring in the year prior to placement. Results Of 47,046,706 patients with diagnoses potentially eligible for ILR placement, 201,103 (0.4%) had an ILR placed. ILR placements increased from 3.08 ILRs per 100,000 patient-days in 2020 to 4.13 in 2025 (5.9% annual growth). The most common indications were cryptogenic stroke (28.6%) and atrial fibrillation without prior ablation (21.4%). Overall, 43.9% were guidance concordant, 28.5% possibly concordant, and 24.8% non-addressed. Relative annual growth in ILR placements per 100,000 patient-days was significant across all guidance categories: concordant (1.3%, 95% CI, 0.9%-1.7%), possibly concordant (5.5%, 95% CI, 5.0%-6.0%) and non-addressed (7.4%, 95% CI, 6.8%-8.0%). Among 125,631 outpatient placements, 50,181 (39.9%) had any prior long-term rhythm monitoring. Conclusion ILR placements are uncommon but increasing, with faster annual growth for indications that may not be guidance concordant. More than half of patients lacked long-term rhythm monitoring prior to outpatient ILR placement.

The Muse 2 EEG headset also records forehead photoplethysmography (PPG). We tested whether its pulse intervals can replace Polar H10 RR intervals for ultra-short heart-rate variability in 28 seated recordings (60 s eyes open, then 60 s eyes closed), using fixed processing, no exclusions and eleven sensitivity analyses. Mean heart rate agreed almost perfectly (ICC [≥] 0.999; bias -0.22 and 0.00 beats min-1). SDNN agreement was good to excellent (ICC 0.86 and 0.91; bias +8.6 and +8.2 ms). RMSSD agreement was moderate (ICC 0.60): Muse exceeded Polar in at least 27 of 28 sessions (bias +19.2 and +18.5 ms; about 1.9-fold), and the ratio bias fell as RMSSD rose. Interval differences (SD 23 ms) implied about 16 ms of independent timing error per pulse, which reproduced the RMSSD bias; without ECG, sensor jitter could not be separated from variation in pulse transit time. Tracking of the eyes-closed change in ln RMSSD was not demonstrated ({rho} = 0.35, 95% CI -0.07 to 0.67; detectable {rho} [≥] 0.51). Muse 2 PPG can supply mean heart rate and, with a stated bias, SDNN; its ultra-short RMSSD is not interchangeable with Polar RMSSD and should be reported as pulse-rate variability.

Introduction: Fighting the ventilator (FV) is characterized by a patients active counteraction of respiratory gas flow across multiple breaths during mechanical ventilation. An effective ventilation is prevented during such episodes. The objective of this exploratory study was to evaluate clinicians knowledge, perception and agreement regarding this condition. Methods: In a cross-sectional online survey, medical professionals with experience in intensive care medicine (mostly neonatal and pediatric patients) responded to 11 questions covering various clinical aspects of FV. Results: All respondents were familiar with the phenomenon and assessed it to be clinically relevant. Neonates were perceived to be predominantly affected. Promoting factors of FV included inadequate sedation and ventilation-associated factors. In clinical practice, the participants identified FV by clinical evaluation, analysis of ventilator waveforms and capnography and used both pharmacological and non-pharmacological intervention strategies. There was no consensus concerning the English terminology of the condition. Only few respondents were familiar with local guidelines to manage FV. Conclusion: Although there was consensus on the relevance of FV, diagnostics and management strategies were highly heterogeneous. Many participants agreed that poor patient-ventilator synchrony was an important cause of FV. Future work should focus on characterizing and generating evidence on FV.

Background: Rheumatic mitral stenosis (RMS) disproportionately affects low-resource settings where echocardiography is limited. We evaluated foundation model (FM) feasibility for RMS detection using limited, indirectly labeled ECGs. Methods: An ECG-FM was fine-tuned using 804 RMS-positive ECGs per fold and externally validated across clinical and wearable lead configurations. Results: Internal AUROCs were 0.89, 0.87, and 0.87, for 12-Lead, 6-Lead, and 1-Lead respectively. External AUROCs were 0.82, 0.80, and 0.77. Performance remained comparable across demographic subgroups. Conclusions: FM fine-tuning enables RMS detection using limited ICD-derived labels, supporting further evaluation of reduced-lead ECG screening and echocardiography triage in resource-constrained settings.

Background: Gastrointestinal (GI) motility disorders are prevalent, impose a significant burden, and are associated with disrupted spatial propagation of bioelectrical gastric slow waves. High-resolution endoscopic gastric mapping (EGM) is an emerging minimally invasive technique for imaging gastric electrical activity; however, EGM is performed alongside standard gastroscopy and requires procedural sedation, which may interact with gastric electrical activity, distorting clinical insights. This study explored the influence of procedural sedation on EGM-recorded slow-wave activity. Methods: EGM was performed for {approx}7 min in 32 patients undergoing upper GI endoscopy at two New Zealand hospitals. Endoscopists administered procedural sedation (fentanyl and midazolam) according to best clinical practice. The relationship between slow-wave amplitude and patient characteristics was then investigated using a variety of statistical techniques. Key Results: Gastric slow waves were detected in 30 of 32 patients who underwent EGM. Symptom scores and demographic measures were comparable between sites. Fentanyl (A, 93.4 {+/-} 13.7 g vs B, 137.5 {+/-} 33.7 g, p[≤]0.001) and midazolam (A, 5.0 {+/-} 0.9 mg vs B, 7.4 {+/-} 1.6 mg, p[≤]0.001) doses were higher at Hospital B than Hospital A. Correspondingly, slow-wave amplitudes were lower at Hospital B than Hospital A (A, 2.33 {+/-} 1.61 mV vs B, 0.86 {+/-} 1.01 mV, p=0.004). All other slow-wave metrics were consistent between hospitals. Combined sedative dose was negatively correlated with slow-wave amplitude (p=0.037). Conclusions & Inferences: Higher procedural sedative doses are associated with lower-amplitude gastric slow-waves. Procedural sedation dose and opiate administration should be carefully considered when interpreting GI electrophysiology measurements.

Background. Many Indian special newborn care units (SNCUs) cannot offer amplitude-integrated EEG (aEEG) or early MRI, so bedside judgement decides which newborns receive closer neurological attention. Earlier work found that asphyxiated newborns cry with a higher pitch and shorter phonation than healthy newborns, and a meta-analysis reported higher fundamental frequency, dysphonation and atypical melody in infants with neurological dysfunction, with wide differences between studies. Aim. To determine whether basic acoustic features of a smartphone-recorded pain cry (following a heel prick) can differentiate term newborns with HIE from control newborns. Also, to track changes in cry acoustics over the first 72 hours of life and evaluate their association with the clinical severity and short-term outcomes of encephalopathy. Design. A prospective, single centre diagnostic accuracy and cohort study in a tertiary SNCU. Three groups of term newborns are recorded: newborns with HIE, healthy comparators, and sick comparators who do not have HIE. Every newborn is recorded after the same stimulus, the first clinically indicated heel-prick. Newborns with HIE and sick comparators are recorded again at 24 hours, 72 hours and before discharge. A cry that is late or absent counts as a result and not as a missing value. Three clinicians blinded to group also classify the cries by listening, so the phone-based measures can be compared with the human ear. Analysis. The primary outcome is the area under the ROC curve of a logistic model with five pre-specified acoustic predictors. The target is 180 analyzable newborns (70 with HIE and 110 comparators). Reporting follows STARD and TRIPOD+AI. Relevance. In a South Asian trial, cooling did not reduce death or disability at 18 months and increased death alone. A cry tool in this setting is therefore proposed to help prioritize EEG, seizure surveillance, closer observation and referral. It is not proposed as a trigger for cooling. Keywords: neonatal encephalopathy; infant cry; acoustic analysis; smartphone; diagnostic accuracy; low-resource setting.

Mitochondrial reactive oxygen species (ROS) are implicated in cardiac dysfunction, but complex dynamic interactions between ROS, intracellular calcium, and electrophysiology make it difficult to resolve mechanisms experimentally. We developed the Zukowski excitation-contraction-mitochondrial-ROS (ECM-ROS) model, a computational model that couples human ventricular electrophysiology with mitochondrial calcium handling, energetics, ROS production, and ROS dependent modulation of the ryanodine receptor, SERCA, L-type calcium current, and late sodium current. Model predictions were evaluated against independent experimental measurements of ROS induced changes in action potential duration and intracellular calcium. We then used the model to characterize ROS calcium feedback across physiological pacing rates and applied it to doxorubicin induced oxidative stress. Increasing cytosolic calcium produced nonlinear amplification of mitochondrial activity and cytosolic ROS, with greater amplification at faster pacing rates. Moderate ROS elevation initially enhanced calcium release, whereas greater oxidative stress depleted sarcoplasmic reticulum calcium stores. During simulated doxorubicin exposure, faster pacing lowered the level of oxidative stress required to produce calcium dysregulation and action potential prolongation. Mitochondrial ROS and calcium accumulation preceded detectable electrophysiological remodeling. In populations incorporating variability in ion channel expression, neither doxorubicin induced oxidative stress nor subtle pharmacological IKr block alone produced early afterdepolarizations. However, early afterdepolarizations were predicted to emerge with combined perturbations. These findings establish a computational framework for investigating bidirectional coupling among mitochondrial function, ROS, calcium handling, and human cardiac electrophysiology. The model generates testable predictions regarding the early detection of oxidative cardiac injury and the physiological conditions that increase susceptibility to arrhythmia.

Background: During emergency tracheal intubation, clinicians use waveform capnography or colorimetric carbon dioxide (CO2) detection to confirm exhaled CO2 from the endotracheal tube to determine its location. Both devices reliably yield a negative result when the tube is in the esophagus. However, in most emergency intubations the tube is correctly located in the trachea, and in this circumstance the relative diagnostic accuracy of the two devices is uncertain. If one device more often fails to detect exhaled CO2 despite correct tube placement (a false negative result), this could lead to unnecessary tube removal, repeated intubation attempts, and an increased risk of serious complications. Research Question: Among critically ill adults undergoing emergency tracheal intubation, is waveform capnography or colorimetric CO2 detection more sensitive for identifying an endotracheal tube correctly placed in the trachea? Study Design and Methods: The WAVE study is a pragmatic, prospective, paired diagnostic accuracy study conducted in emergency departments and intensive care units in the United States. The study compares waveform capnography and colorimetric CO2 detection (index tests) against a non-CO2-based reference standard for endotracheal tube location among 2,092 critically ill adults undergoing tracheal intubation. The primary outcome is the sensitivity of each index test for identifying a tube located in the trachea. The secondary outcome is specificity for identifying a tube located outside the trachea. Results: Enrollment began on April 21, 2025 and is ongoing, with completion anticipated in 2026. Interpretation: The WAVE study will provide prospective evidence comparing waveform capnography with colorimetric CO2 detection for determination of endotracheal tube location in critically ill adults. These findings may inform clinical practice and guideline recommendations regarding the optimal test of exhaled CO2 for use in the ED and ICU. Prespecifying the protocol and statistical analysis plan before completion of enrollment enhances the rigor, reproducibility, and transparency of the study.

Optical mapping is a fundamental tool for cardiac electrophysiology research. Processing of these signals can be non-trivial, requiring multiple pre- and post-processing steps to limit interpretation errors. Existing software packages have features siloed across platforms and require users to combine several packages for in-depth analysis. To address this, we developed CADENCE, an open-source MATLAB-based application combining feature breadth (e.g., SVD denoising, conduction velocity, alternans, annotated rotors) and depth (e.g., 15 independent alternans features across both voltage and calcium modalities) in an end-to-end pipeline including data conversion, signal conditioning, feature extraction, and signal analysis. We demonstrate CADENCE's capabilities across datasets spanning four species (rat, rabbit, pig, and human), three tissue preparations (slices, wedge, and whole heart), and three camera configurations (single, dual, and panoramic). CADENCE was validated against three blinded expert observers (15 hearts) and demonstrated agreement comparable to inter-observer variability (SD ratios 1.07/0.95/0.89 for APD80, activation, CaTD80). Dominant frequency was recovered with a median error of 0.03% across 84 of 88 paced recordings with 1:1 capture. CADENCE's code is publicly available and distributed as a compiled standalone application that the end-user can install and run without a MATLAB license to encourage adoption and standardization across laboratories.

Background: Sleep-disordered breathing (SDB) is an established adverse prognostic factor in heart failure (HF). Sleep apnea-specific hypoxic burden (SASHB), which reflects the duration and depth of respiratory event-related oxygen desaturation, has shown stronger prognostic value than the apnea-hypopnea index (AHI) in general populations and patients with coronary artery disease. However, its clinical utility in patients hospitalized due to acute HF remains unclear. Objectives: This study evaluated the prognostic value of SASHB in patients hospitalized for acute HF and compared its performance with AHI for risk stratification and identification of patients who may benefit from positive airway pressure (PAP) therapy. Methods: This post hoc analysis included 230 patients from a prospective single-center observational cohort of adults hospitalized for acute HF. The primary analysis included 178 patients who did not receive PAP treatment. Associations of SASHB and AHI with the composite endpoint of cardiovascular death or rehospitalization for worsening HF were evaluated using Cox proportional hazards models. Model improvement after adding SASHB or AHI to a model including established clinical risk factors was assessed using changes in the C index and Brier score, reflecting changes in model discrimination and prediction error, respectively. In the full cohort, the interaction between PAP treatment and SDB metrics was evaluated using the g-formula. Results: During a median follow-up of 3.7 years, 58 of 178 patients (32.6%) reached the composite endpoint. Both SASHB and AHI were independently associated with the composite outcome after multivariable adjustment. However, only SASHB consistently predicted outcomes across analyses and improved model performance beyond established prognostic factors (bootstrap {Delta} C index, 0.020 [95% CI: 0.008 - 0.031]; bootstrap {Delta} Brier score, - 0.007 [95% CI: - 0.014 to - 0.001]), whereas AHI did not. In the analysis including PAP-treated patients, both SASHB and AHI showed interaction with PAP treatment. Conclusions: In patients hospitalized for acute HF, SASHB showed a consistent independent association with adverse clinical outcomes and provided additional prognostic information beyond AHI. SASHB may improve risk stratification and may help identify patients who are more likely to benefit from PAP therapy.

Purpose Delirium affects around 40% of intensive care unit (ICU) patients and adversely affects survival, length of stay, and cognitive recovery. Despite this, implementation of systematic clinical delirium assessments is variable. We sought to validate a point-of-care electroencephalography (EEG)-based system (Ceribell, Inc.) for the detection of delirium in the ICU. Methods In this multicenter, prospective cohort study, we monitored adult ICU patients with the Ceribell EEG system for 6-8 hours for a minimum of three days. Up to three times per day, expert clinicians assessed participants for delirium and research coordinators completed Confusion Assessment Method for the ICU (CAM-ICU) assessments, which we used to calculate CAM-ICU-7 severity scores. We compared expert clinician diagnoses with Ceribell EEG algorithm output using data from 30-minute periods surrounding each clinician assessment. Results Among 225 participants analyzed (median age: 64 years, 60% male, 21% mechanically ventilated), expert clinicians performed 857 assessments and found 24% of participants were delirious during at least one assessment. The Ceribell EEG system detected delirium with a sensitivity of 81% (95% CI: 76% to 87%) and specificity of 81% (95% CI: 78% to 84%). The continuous Ceribell EEG system score correlated with CAM-ICU-7 delirium severity scores (bootstrapped median rho 0.58, IQR: 0.54-0.62, p <0.001). Conclusion In a mixed adult ICU population, the Ceribell EEG system identified delirium with good sensitivity and specificity. Its continuous output was associated with delirium severity, highlighting its potential ability to characterize underlying physiological patterns across the clinical delirium spectrum.

Rationale: Obstructive sleep apnea (OSA) is highly prevalent yet largely under-diagnosed. Current screening strategies rely on effortful questionnaires with modest accuracy, and prior machine learning models often require resource-intensive inputs and lack external validation. Objectives: To develop and validate machine learning models to predict OSA (apnea-hypopnea index [AHI4%][&ge;]5) and moderate-severe OSA (AHI4%[&ge;]15) using only routinely collected electronic health record (EHR) data, test cross-system transferability and compare performance with clinical screeners and a Bayesian risk-updating framework. Methods: Three tree-ensemble families (XGBoost, random forests, and histogram-based gradient boosting) were trained on an adult cohort with sleep-study results at Kaiser Permanente (KPSC), using full (106-feature) and minimal (age, sex, body-mass index, race/ethnicity) predictor sets and tuned by nested cross-validation. XGBoost models were selected, internally and temporally validated, and externally validated at University of Kansas Medical Center (KUMC). Measurements and Main Results: On our internal hold-out set (n=53,048), the full XGBoost model reached a ROC-AUC of 0.791 for any OSA and 0.768 for moderate-severe OSA. The minimal model retained most performance (0.764 and 0.736). Temporal performance was stable. Externally, frozen models showed lower AUCs (0.718 and 0.690 for full models; 0.725 and 0.706 for minimal models) but similar to KUMC-native models and improved further with transfer learning. The models significantly outperformed STOP-BANG (AUC difference 0.06-0.09 for predicting any OSA; p<0.001). Bayesian updating with select patient-reported items further improved discrimination. Conclusions: Machine learning models based on EHR data can predict OSA probability that outperform existing screeners, transfers across systems, and offers potential for EHR implementation.

Background: Attention-Deficit/Hyperactivity Disorder (ADHD) is a common neurodevelopmental disorder characterized by persistent symptoms of inattention, hyperactivity, and impulsivity that significantly affect cognitive, academic, and social functioning. Although electroencephalography (EEG) has emerged as a promising non-invasive tool for identifying objective neurophysiological biomarkers of ADHD because of its high temporal resolution and cost-effectiveness, many existing machine learning studies rely on single-domain EEG features, which may not adequately capture the complex functional organization of the brain. Methods: This study proposes a subject-level EEG classification framework that integrates three complementary feature domains: spectral characteristics, weighted Phase Lag Index (wPLI)-based functional connectivity, and topological descriptors derived from persistent homology. EEG recordings from 121 subjects were preprocessed and segmented into overlapping windows for feature extraction. Window-level features were aggregated to generate subject-level representations. Recursive Feature Elimination (RFE) was employed to select the 110 most informative features from an initial set of 1,043 descriptors. Five machine learning classifiers, including Support Vector Machine (SVM), Random Forest, XGBoost, CatBoost, and Multi-Layer Perceptron (MLP), were evaluated using stratified five-fold cross-validation. An ablation study was further conducted to assess the contribution of each feature domain. Model interpretability was investigated using SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME), two widely adopted Explainable Artificial Intelligence techniques for interpreting machine learning models. Results: Among the evaluated classifiers, the SVM achieved the highest classification performance, yielding an accuracy of 80.17%, an F1-score of 80.95%, a balanced accuracy of 80.14%, a receiver operating characteristic area under the curve (ROC-AUC) of 86.72%, and a precision recall area under the curve (PR-AUC) of 87.34%. The ablation study demonstrated that combining spectral, functional connectivity, and topological features consistently outperformed individual feature domains, highlighting the complementary nature of these representations. Explainable AI analyses further identified functional connectivity features as the primary contributors to classification performance while confirming the additional value of spectral and topological descriptors. Conclusions: The proposed framework demonstrates that integrating spectral analysis, functional connectivity, and topological data analysis provides an effective and interpretable approach for subject-level ADHD classification from EEG recordings. The combination of recursive feature selection, multi-domain feature fusion, ablation analysis, and Explainable Artificial Intelligence establishes a transparent and reproducible pipeline that may facilitate the development of reliable EEG-based clinical decision-support systems for ADHD.

Abstract Background: In patients with sepsis and pre-existing hypertension, admission findings may not adequately describe the subsequent course of circulatory support and organ dysfunction. We developed and externally validated a model to reassess in-hospital mortality risk after the first 72 hours of intensive care. Methods: This retrospective study used MIMIC-IV for development and eICU-CRD for external validation. Patients had received vasoactive agents and remained in the ICU for at least 72 hours. Static variables were combined with features from six consecutive 12-hour windows. Imputation, feature selection, model fitting, and calibration were performed within training folds during five-fold cross-validation. The final XGBoost model was applied externally without refitting. Evaluation included discrimination, calibration, decision curves, and risk stratification, with an exploratory comparison of static-only and static-plus-dynamic feature sets. Results: The development and external cohorts included 1019 and 1747 patients, with 338 (33.2%) and 516 (29.5%) deaths, respectively. Internal area under the receiver operating characteristic curve (AUC) was 0.752 (95% CI 0.720-0.784). External AUC was 0.713 (95% CI 0.687-0.739), average precision was 0.536, the Brier score was 0.185, and the calibration slope was 0.744. Mortality increased across external risk tertiles from 13.7% to 47.3%. In an exploratory feature-set comparison, the static-plus-dynamic feature set showed a 0.071 higher internal AUC than the static feature set (95% CI 0.042-0.109). Conclusions: Information accumulated during the first 72 hours supported subsequent mortality risk stratification in this selected population. The model retained moderate discrimination externally, supporting further evaluation as an adjunct to clinical reassessment. Local calibration and the clinical consequences of acting on model predictions require evaluation before routine use.

Cardiotocography (CTG) is the standard of care for intrapartum fetal monitoring, yet its poor specificity drives unnecessary intervention without reducing hypoxic-ischemic encephalopathy (HIE). Most artificial intelligence (AI) models are trained exclusively on the final hour before delivery, a window only identifiable retrospectively. We hypothesized that this framework prevents models from learning the early evolving signs of HIE. We analyzed 174,186 deliveries, with gestational age GA[&ge;]35 weeks, from Kaiser Permanente Northern California. Forty CTG features were extracted from 20-minute epochs to train Random Forest classifiers targeting severe acidosis (N=2,636) and clinically validated HIE (N=304). The training window was progressively extended from the final hour to the full duration of labor, and time from labor onset (TLO) was added as a feature. Models were evaluated continuously throughout labor at a fixed 15% false positive rate. Our results showed that the early identification of fetuses at risk of HIE improved as the training window was extended, plateauing at 18 hours. Adding TLO gave the highest early-warning performance: 40.3% of HIE cases identified at least 3 hours and 27.3% at least 6 hours before delivery, absolute gains of 8.4% and 10.7% over the last-hour classifier. Finally, 55.7% of HIE cases were identified at least 40 minutes before delivery. In conclusion, classifiers trained only on the last hour may identify HIE patterns that occur near delivery; time-aware models trained across labor map the temporal evolution of these patterns and provide actionable early warnings. Thus, AI systems for intrapartum monitoring must account for the temporal evolution of labor to shift from end-of-labor diagnosis to early-warning decision support.

Interpersonal physiological synchrony (IPS) between interaction partners has been studied as a marker of social connectedness and coordination. However, little is known about how such interpersonal synchrony relates to the intrapersonal coordination of individuals' own autonomic systems during team interaction. The present study examined whether IPS and intrapersonal physiological synchrony (IAPS), defined as the within-person coordination of heart rate (HR), respiration, and electrodermal activity (EDA), function as competing or complementary regulatory processes. It also examined whether IAPS is state- or trait-like and whether teams high or low in both IPS and IAPS had better outcomes. Fifty five newly formed triads (N = 165; Mage = 23.87, 77% female) participated in a laboratory interaction including passive shared experiences and active collaborative tasks while continuous physiological activity was recorded. Synchrony within individuals and between team members was quantified using multidimensional recurrence quantification analysis. Across interaction segments, IAPS was positively associated with IPS in both HR and EDA, with stronger and more consistent associations during active teamwork versus during passive social contexts. These findings contradict a trade-off hypothesis and instead support a synergistic relationship between IAPS and IPS. IAPS exhibited state-like dynamics and sensitivity to features of the context, i.e., consistency within similar interaction contexts and higher in passive versus active social segments. Further, IAPS had limited associations with individual difference measures. Exploratory analyses indicated that teams characterized by both high IAPS and high IPS demonstrated more favorable outcomes, reflected in either higher objective task performance or greater perceived team efficacy depending on physiological modality. Together, these results suggest that effective team coordination is supported not only by IPS but also by the capacity to maintain coherent internal autonomic organization. They further highlight IAPS as a dynamic component of social functioning that merits exploration alongside interpersonal processes.