Background: Patients suffering from critical congenital heart disease (cCHD) require cardiac intervention within the first year of life. During the postoperative period, patients are at risk of haemodynamic instability resulting in insufficient organ perfusion and subsequent organ failure. To prevent this, patients are placed in the paediatric intensive care unit (PICU) where various vital parameters can be monitored. However, interpreting these continuous data streams can be challenging. Machine learning offers the potential to support clinical decision-making in this setting, but challenges remain, particularly in labelling haemodynamic instability and accounting for varying physiology of this patient demographic. The goal of this study was to improve the retrospective labelling of haemodynamic instability and evaluate the affect of age-stratified subpopulation on model performance.

Methods: This study used a retrospective dataset of continuously measured parameters (heart rate, respiratory rate, mean arterial pressure, central venous pressure, oxygen saturation, and perfusion index) collected from post-operative cCHD patients admitted to the PICU of Erasmus MC Sophia Children's Hospital, the Netherlands, between January 2016 and April 2025. A new scoring system was developed to quantify the haemodynamic support received by patients and to identify intervention times at which support was increased. These interventions were used to label haemodynamic instability in the a period dT prior to intervention. The resulting labelling was applied to train a random forest algorithm to predict haemodynamic instability, and the model was subsequently retrained on age-based subpopulations.

Results: A total of 425 patients were included for this study. The new labelling method resulted in 5.7% of the data being labelled as haemodynamically unstable. The random forest using the new labelling achieved an average (SD) area under precision-recall curve (AUCPR) of 0.233 (0.041) on the test set during cross-validation and final test AUCPR of 0.203. The largest age subpopulations were 0--30 days and 90--180 days. The class prior of instability was 9.2% in the 0--30 days subpopulation and the prediction model achieved an AUCPR of 0.244 (0.064). In the 90--180 day subpopulation the class prior was 4.9% and an AUCPR of 0.221 (0.105) was achieved.

Conclusion: This study proposed a new method of retrospectively labelling haemodynamic instability with the goal of training a predictive model to predict these instabilities. A random forest model trained using the new labelling showed limited improvement, with an AUCPR of 0.204 and an AUCROC of 0.762. Age-based subpopulation analysis indicated potential for reduced data variation, though larger cohorts are needed for better generalisation. Further refinements in the retrospective labelling of haemodynamic instability are required for an effective prediction model to be developed.

Background and objectives: Junctional Ectopic Tachycardia (JET) is a tachyarrhythmia most commonly observed in infants and children in the postoperative setting. An automatic detection algorithm could be valuable for early identification and timely treatment of JET. However, the detection is challenging as the initial changes on the electrocardiogram (ECG) are often subtle and monitor data commonly contains substantial noise and artefacts. The objective of this study was to investigate which features contribute to accurate JET detection and to develop an automated detection model.

Methods: A retrospective study was conducted using monitor ECG data of paediatric patients admitted to the Paediatric Intensive Care Unit. The training set consisted of 17 patients, and the test set of 8 patients. ECG metrics were detected, in order to segment the signal and to derive several features. The two-dimensional vectorcardiogram (VCG) was computed for calculating features representing the beat-to-beat variability of the signal. Automatic feature selection methods were applied to identify which features most effectively differentiate JET from sinus rhythm (SR), based on balanced accuracy. Logistic regression (LR) and random forest (RF) models were finally created and performance was validated.

Results: The LR and RF models achieved balanced accuracy scores of 0,989 and 0,988, respectively, on the training dataset. The selected features included the number of P waves and the variance of the PQ interval. For the RF model, the standard deviation (SD) of the RR interval was also selected. VCG-features did not prove effective in distinguishing JET from SR. A secondary validation on the test set yielded lower scores of 0,899 and 0,892. An analysis of misclassifications revealed that they were all attributed to errors in peak detection, which occurred in cases of deviating ECG morphologies or the presence of artefacts and noise.

Conclusions: This study demonstrates that P wave-related features are most effective for distinguishing JET from SR, with simple machine learning models based on these features showing promising results for automated JET detection. Peak detection is currently the most important limiting factor for the robustness and generalisability of this method. Interpatient variability and the low quality of monitor ECG data remain important challenges. Expanding the dataset, improving the data quality and implementing signal quality assessment (SQA) methods are recommended to improve the robustness of the models.

Blood pressure is a crucial parameter in haemodynamic monitoring, providing an indication of tissue perfusion in critical situations. Commonplace in (paediatric) intensive care units ((P)ICU) are arterial catheters, inserted in the radial artery, which are an invasive means of continuously measuring blood pressure. However, quantifying the degree to which arterial catheter geometries modify the pressure measurement, especially for paediatrics, remains seemingly unexplored in research, while discrepancies between non-invasive and invasive blood pressure monitoring methods have been observed in Erasmus MC (P)ICU departments.

Using FEBio, a computational fluid dynamics (CFD) model was developed to investigate the influence of catheter geometries on the pressure generated at the catheter tip (taken to be the measured pressure) for six age groups ranging from <2 years to 18 years of age. For three age groups (2-5 years, 12-14 years, and 15-18 years), simulations were run with two different catheter gauges (diameters).

Hence, a total of nine different sets of age and catheter gauge were investigated. Apart from comparison between simulations with and without a catheter, the effect of the position of the catheter in the vessel (eccentricity ratio) and changing catheter lengths in the vessel were studied. The results show that arterial catheters (substantially) influence the pressure measured by the catheter. In the worst-case scenarios simulated, namely that in which the catheter is in the centre of the vessel with an intraluminal catheter length equal to 80% of its total length, eight of the nine cases investigated showed systolic pressure values higher than the target values used in the
PICU. Thus, the results suggest that the presence of the catheter may be enough to modify a normal blood pressure into appearing elevated (hypertensive), potentially influencing patient treatment. Shorter catheter lengths were associated with lower pressure overestimations, as was moving the catheter towards the vessel wall. Consequently, these results are clinically significant, and may help physicians in making more informed haemodynamic assessments.

Introduction: In critically ill paediatric patients, sleep is essential for recovery and development, yet sleep disturbances are common in the paediatric intensive care unit (PICU), which highlight the need to integrate sleep monitoring into clinical practice. While automated sleep stage classification using machine learning (ML) on single-channel electroencephalography (EEG) data has shown promise in mostly healthy adult populations, its application to critically ill children is challenged by age-specific sleep architecture, medication effects, pathological conditions, and artefacts. This study evaluates whether deep learning (DL) feature extraction and dynamic models can improve sleep staging performance in this population and explores the use of an unsupervised ML model to gain deeper insight into the complex sleep structures in this population.

Methods: This study utilised EEG recordings from three datasets—healthy adults, non-critically ill children, and critically ill children—to train and evaluate supervised and unsupervised sleep stage classification models. As supervised models, a convolutional neural network (CNN) was used for feature extraction, followed by dynamic models including a long short-term memory (LSTM) network and a hidden Markov model (HMM) to account for temporal dependencies in sleep. Additionally, an unsupervised HMM was applied to explore underlying structures in the sleep EEG data without predefined labels.

Results: Supervised models achieved good performance in healthy adults and non-critically ill children, with maximum accuracies of 90.2\% and 77.4\%, respectively, for three-state classification. The added value of dynamic models over the CNN alone varied per dataset and model type and was not consistent. In critically ill children, classification performance was low, with a maximum accuracy of 61.4\%, and notably low macro-F1 and Cohen’s kappa scores (45.9\% and 26.5\%, respectively). The unsupervised HMM revealed that identifying distinct and stable clusters over time was challenging in all datasets. For critically ill children, the model often failed to identify multiple distinct clusters within individual patients, and substantial variability in cluster assignments was observed across patients.

Discussion: This study demonstrates that DL-based feature extraction and dynamic modelling using single-channel EEG can achieve strong sleep staging performance in healthy adults and non-critically ill children, highlighting the potential for (semi-)automated scoring tools in more stable populations. In contrast, performance in critically ill children was notably lower, likely due to factors such as high variability in sleep architecture, signal artefacts, limited data quality, and the uncertain reliability of manually assigned labels. These results suggest that conventional sleep stages do not generalise well to this population, and a purely data-driven, unsupervised approach does not offer a viable alternative. Overall, the findings emphasise the need for a larger dataset of critically ill children, further evaluation of relevant patient and data characteristics, the inclusion of alternative signals such as electrocardiography, and greater focus on model interpretability.

Background: Neurovascular coupling (NVC) is the physiological mechanism linking neural activity to cerebral blood flow (CBF). When local neural activity increases, NVC ensures a corresponding rise in CBF, which often exceeds the metabolic demand. NVC monitoring could improve neuroprognostication and evaluation of treatment interventions for critically ill children in the pediatric intensive care unit (PICU). This may be particularly relevant for patients following cardiac arrest (CA), where outcome prediction after return of spontaneous circulation (ROSC) remains challenging, but is essential for guiding treatment and informing families.

Objectives: This proof-of-concept study aimed to develop a bedside method for quantifying NVC using simultaneous electroencephalography (EEG) and near-infrared spectroscopy (NIRS) recordings, and to identify biomarkers that best describe NVC.

Methods: Simultaneous EEG and NIRS recordings were obtained from pediatric post-ROSC patients admitted to the PICU at Sophia Children's Hospital between 2017 and 2023. Wavelet coherence analysis was performed between NIRS-derived cerebral tissue oxygen saturation (SctO2) and six EEG features reflecting amplitude, power and continuity aspects of the EEG.

Results: Nine patients were included: five who survived to PICU discharge and four who did not. Survivors exhibited continuous EEG background patterns and showed predominant in-phase coherence between amplitude or power EEG features and SctO2 within the very low-frequency range. Non-survivors, who all exhibited isoelectric EEG patterns, predominantly showed an opposite phase relationship compared to survivors. Significant in-phase coherence between aEEG bandwidth and SctO2 within the 0.2–0.5 and 1–3 mHz frequency ranges yielded the highest percentage of significant in-phase coherence among all EEG features and best distinguished survivors from nonsurvivors. Within this frequency band, survivors showed a median of 6.0% (Q1–Q3: 5.8%–7.2%) significant in-phase coherence, while non-survivors showed a median of 0.8% (Q1–Q3: 0.6–1.0%).

Discussion: Significant in-phase coherence within the 0.2-0.5 and 1-3 mHz frequency ranges between aEEG bandwidth - SctO2 may serve as a biomarker for intact NVC, indicating synchronized increases in cerebral activity and CBF in survivors with a continuous EEG pattern. Validation in larger and more heterogeneous cohorts is required before this biomarker can be used clinically and its prognostic value must be further studied. A key limitation is the need for long, artifact-free recordings, which may limit feasibility in PICU settings.

Background: Continuous monitoring in paediatric intensive care units (PICUs) generates frequent clinical alarms, 87-97% of which are nonactionable, contributing to alarm fatigue and patient safety risks. At the Erasmus MC Sophia Children’s Hospital PICU, improved alarm management is needed. Machine learning offers a promising strategy to distinguish actionable from nonactionable alarms.

Objectives: The primary objective was to develop a machine learning algorithm to classify actionable alarms from multimodal vital signs. Secondary objectives were to characterise the PICU alarm burden, capture stakeholder perspectives and evaluate model feasibility for clinical application.

Methods: Retrospective Dräger monitoring data and alarms from 2,582 PICU patients (Nov 2021 – Oct 2024) were analysed. A machine learning algorithm was developed using ART M, HR, RESP and SpO2 data from 26 patients, with alarms annotated using clinical interventions, small signal deviations and temporal associations with other parameters. Logistic regression, decision tree, random forest and XGBoost were evaluated with nested cross-validation using pre-alarm features. Performance was assessed by sensitivity, specificity, balanced accuracy, with AUROC and F1-score as complementary metrics. Semi-structured interviews with four nurses and one psychologist explored alarm experiences, impacts and reduction strategies.

Results: The best model, a decision tree, showed limited performance (sensitivity 0.36-0.49, specificity 0.51-0.70, balanced accuracy ≈0.50). No discriminative features were identified, and substantial overlap and outliers limited classification. Most alarms involved SpO2 desaturation, with unit variation and temporal patterns linked to ward activity. Interviews highlighted overstimulation and desensitisation, but also that nonactionable alarms can serve as early warnings, with interpretation requiring clinical context.

Discussion: This study provides an evaluation of machine learning-based alarm classification in the PICU and integrates clinician perspectives to guide future interventions. The machine learning model was not suitable for clinical use because of retrospective labelling, the small annotated dataset and the absence of clinical context. Future research should focus on prospective annotation, larger and more diverse datasets and complementary strategies. Alarm dashboards and daily reviews are recommended to reduce alarm burden and mitigate alarm fatigue while maintaining patient safety.

Background: Postoperative junctional ectopic tachycardia (JET) is an arrhythmia associated with increased morbidity and mortality rates in children with congenital heart disease. Developing an automated detection algorithm could aid in early identification and timely treatment of JET.

Methods: A retrospective study was conducted using monitor electrocardiogram (ECG) data of pediatric patients who experienced JET during their admission to the pediatric intensive care unit. A manual decision tree was developed that aimed to differentiate between JET and sinus rhythm based on distinctive characteristics. These features were derived using signal analysis on both two-dimensional vectorcardiograms and ECG data. For the latter, ECG metrics were detected in a fictive lead that was created in the direction with the highest amplitudes. Metrics were identified within adaptive intervals that were dependent on ECG morphology rather than relying on fixed time intervals.

Results: A classification performance was achieved with a sensitivity of 96.3%, specificity of 71.4%, positive predictive value (PPV) of 86.7% and an accuracy of 87.8%. R peaks, Q peaks, S peaks, T peaks and P waves were detected with an accuracy of respectively 99.9%, 95.7%, 89.7%, 98.1% and 54.8%. The computational time of the classification of 41 minutes of data was 4 minutes and 48 seconds.

Conclusion: A manual decision tree algorithm for JET detection was developed, using signal analysis for feature extraction based on JET characteristics. This method with a low computational time and a high sensitivity and PPV holds potential for clinical application as a bedside tool. Implementing this proposed algorithm would allow for treatment in an earlier phase, thereby potentially reducing JET associated morbidity and mortality rates.

Introduction
Approximately 9 in 1000 children are born with congenital heart disease (CHD), of whom a quarter are classified as critical CHD (CCHD) and require an intervention within their first year. Monitoring these patients in the Paediatric Intensive Care Unit (PICU) is crucial, yet with increasing amounts of data, detecting subtle changes that are important for the disease progression interpretation of all vital signs becomes difficult, even for skilled physicians. Machine learning (ML) offers potential solutions, however, challenges such as inter-patient variability and the absence of clear definitions for haemodynamic instability persist. This study aims to develop a ML algorithm for early prediction of haemodynamic instability in CCHD patients with high frequency vital-signs, addressing these challenges through objective labelling methods and stratification approaches.

Methods
Two approaches, on population and patient level, were developed with nested cross-validation (CV). Due to a high inter-patient variability, the patient specific approach was added. A first iteration of objectively labelling haemodynamic instability was proposed, based on medical interventions such as medication administration and fluid therapy. Since it is difficult to retrospectively determine for how long patients were unstable, multiple values for instability duration (dT ) were added to the analysis.
To capture the temporal dependency of time-series data, lag-analysis was performed, adding the relation between the vital signs and their previous values to the model development. Lag-analysis included a sliding window that moved over the data. The width of sliding window (W ) was optimised during the model development. Additionally, a horizon (r ) was implemented, so the data within the sliding window were predicting future timestamps.

Results
This retrospective study included a total of 224 admissions in the analysis. Two random forest classifiers were trained using a nested CV structure to detect haemodynamic instability in CCHD patients. For both approaches the same temporal settings (W : 50 minutes, r : 45 minutes, dT: 120 minutes) were used. This study has shown that the between-patient approach had notable differences between the mean train (85%, AUCPR) and test performance (40%, AUCPR). The in-patient approach, while using 20% and 10% of the test data for training, still yielded a test performance of 96% (AUCPR) and 90% (AUCPR), respectively.

Discussion and conclusion
Generally speaking, the experiments suggest that the first iterations of the models were not robust and generalised poorly. It is most likely caused by a large inter-patient variability and a simple labelling system that is still depending on subjectivity.
This study has shown that the proposed prediction model, which combines high frequency vital signs, labels, and temporal settings (W, r, dT ), requires additional refinement before it can be considered clinically feasible to implement this model as a reliable bedside tool for predicting haemodynamic instability.

Introduction: Congenital diaphragmatic hernia (CDH) is a rare developmental defect of the diaphragm characterised by herniation of abdominal organs into the thoracic cavity during prenatal development. This herniation is usually accompanied by pulmonary hypertension (PH) and cardiac dysfunction (CD). Although several parameters are known to predict the clinical outcomes in CDH, most of these parameters do not monitor the degree of PH and CD and cannot be continuously measured. This retrospective observational trial aimed to monitor the degree of PH and CD in CDH using the oxygen saturation index (OSI), peripheral oxygen saturation (SpO2), heart rate (HR), heart rate variability (HRV), arterial blood pressure (ABP) and derivatives of these parameters.

Methods: the study population consisted of neonates admitted to the paediatric intensive care unit between 2019 and 2022 for treatment of CDH. The degree of PH and CD was determined for each cardiac ultrasound (CUS) performed. A 15-minute window of vital parameters, mechanical ventilator, and electrocardiogram data before each CUS was extracted to calculate the predictors. After preprocessing the data and meeting the statistical assumptions, both univariable and multivariable logistic mixed effects models were fitted and validated.

Results: in total, 136 CUS of 57 patients were included in the study. Of the univariable linear mixed-effects models, the median values of HR, pulse pressure (PP), preductal SpO2, dSpO2, OSI and the interquartile range (IQR) of HR were statistically significant predictors of PH. For the prediction of CD, this was the case for the power of HRV in the very low frequency band (HRV-VLF) and for the median values of HR, mean arterial pressure (MAP) and OSI. The multivariable model for the prediction of PH contained the median values of of dSpO2, HR, PP and OSI and the standard deviation of the normal-to-normal beat intervals (SDNN). The multivariable model for the prediction of CD included the median of dSpO2, HRV-VLF, SDNN and the IQR of the systolic arterial pressure (SAP) as predictors.

Conclusions: the most promising predictors are the median values of preductal SpO2, dSpO2 and OSI for the prediction of PH. For the prediction of CD, HRV-VLF and the median values of HR, OSI and dSpO2 were the most promising predictors. Despite limited predictive performance of the regression models, this study contributes to the improvement of monitoring of patients with CDH, which can lead to more timely interventions and eventually improved outcomes within this patient population.