The Pediatric Data Science and Analytics Subgroup of the Pediatric Acute Lung Injury and Sepsis Investigators Network: Use of Supervised Machine Learning Applications in Pediatric Critical Care Medicine Research

Development of a neural network model for early detection of creatinine change in critically Ill children

Introduction

Renal dysfunction is common in critically ill children and increases morbidity and mortality risk. Diagnosis and management of renal dysfunction relies on creatinine, a delayed marker of renal injury. We aimed to develop and validate a machine learning model using routinely collected clinical data to predict 24-hour creatinine change in critically ill children before change is observed clinically.

Methods

Retrospective cohort study of 39,932 pediatric intensive care unit encounters in a national multicenter database from 2007 to 2022. A neural network was trained to predict <50% or ≥50% creatinine change in the next 24 h. Admission demographics, routinely measured vital signs, laboratory tests, and medication use variables were used as predictors for the model. Data set was randomly split at the encounter level into model development (80%) and test (20%) sets. Performance and clinical relevance was assessed in the test set by accuracy of prediction classification and confusion matrix metrics.

Results

The cohort had a male predominance (53.8%), median age of 8.0 years (IQR 1.9-14.6), 21.0% incidence of acute kidney injury, and 2.3% mortality. The overall accuracy of the model for predicting change of <50% or ≥50% was 68.1% (95% CI 67.6%-68.7%). The accuracy of classification improved substantially with higher creatinine values from 29.9% (CI 28.9%-31.0%) in pairs with an admission creatinine <0.3 mg/dl to 90.0-96.3% in pairs with an admission creatinine of ≥0.6 mg/dl. The model had a negative predictive value of 97.2% and a positive predictive value of 7.1%. The number needed to evaluate to detect one true change ≥50% was 14.

Discussion

24-hour creatinine change consistent with acute kidney injury can be predicted using routine clinical data in a machine learning model, indicating risk of significant renal dysfunction before it is measured clinically. Positive predictive performance is limited by clinical reliance on creatinine.

Advancing Pediatric Growth Assessment with Machine Learning: Overcoming Challenges in Early Diagnosis and Monitoring

BACKGROUND

Pediatric growth assessment is crucial for early diagnosis and intervention in growth disorders. Traditional methods often lack accuracy and real-time decision-making capabilities This study explores the application of machine learning (ML), particularly logistic regression, to improve diagnostic precision and timeliness in pediatric growth assessment. Logistic regression is a reliable and easily interpretable model for detecting growth abnormalities in children. Unlike complex machine learning models, it offers parsimony in transparency, efficiency, and reproducibility, making it ideal for clinical settings where explainable, data-driven decisions are essential.

METHODS

A logistic regression model was developed using R to analyze biometric and demographic data from a cross-sectional dataset, including real-world data from public institucions. The study employed a bibliometric analysis to identify key trends and incorporated data preprocessing techniques such as cleaning, imputation, and feature selection to enhance model performance. Performance metrics, including accuracy, sensitivity, and the Receiver Operating Characteristic (ROC) curve, were utilized for evaluation.

RESULTS

The logistic regression model demonstrated an accuracy of 94.65% and a sensitivity of 91.03%, significantly improving the identification of growth anomalies compared to conventional assessment methods. The model's ROC curve showed an area under the curve (AUC) of 0.96, indicating excellent predictive capability. Findings highlight ML's potential in automating pediatric growth monitoring and supporting clinical decision-making, as it can be very simple and highly interpretable in clinical practice.

CONCLUSIONS

ML, particularly logistic regression, offers a promising tool for pediatric healthcare by enhancing diagnostic precision and operational efficiency. Despite these advancements, challenges remain regarding data quality, clinical integration, and privacy concerns. Future research should focus on expanding dataset diversity, improving model interpretability, and conducting external validation to facilitate broader clinical adoption.

External Validation, Re-Calibration, and Extension of a Prediction Model of Early Acute Kidney Injury in Critically Ill Children using Multi-Center Data

Background

Acute kidney injury (AKI) is common among children with critical illness and is associated with high morbidity and mortality. Risk prediction models designed for clinical decision support implementation offer an opportunity to identify and proactively mitigate AKI risks. Existing models have been primarily validated on single-center data, owing partly to the lack of appropriately detailed multicenter datasets.

Objective

To determine the accuracy of a single-center model to predict new AKI at 72 hours of ICU admission across two multicenter datasets and extend this model to improve prediction accuracy while maintaining acceptable alert burden.

Derivation and Validation Cohorts

We separately derived models in two datasets: PEDSNET-VPS, created through the linkage of PEDSnet electronic health record (EHR) extraction with Virtual Pediatric Systems (VPS); and the PICU Data Collaborative dataset, created through EHR extraction and harmonization from eight participating institutions. Derivation datasets comprised temporal and location-specific spit of these datasets (80%), while the holdout test split comprised the remaining (20%).

Prediction Model

We recalibrated an existing single-center model and measured discrimination and accuracy. We then add features guided by precision and recall measures. All features were available at 12 hours of ICU admission. We measure discrimination and accuracy at multiple cut-points and identify the features contributing most to the risk score.

Results

In two datasets comprising 186,540 ICU admissions, we report an incidence of early AKI of 2.2 - 2.7%. Initial recalibration of an existing single-center model demonstrated poor discrimination (AUROC 0.60 - 0.78). Following the addition of new features, we report higher AUROC values of 0.79 - 0.80 and AUPRC values of 0.13 - 0.21 in both datasets. We report accuracy at several cutpoints as well as cross-validate between datasets.

Conclusions

In this first use of two new multicenter datasets, we report improved discrimination and accuracy in a model designed specifically for implementation, balancing sensitivity and precision to predict patients at risk for AKI development.

Incidence of cardiac arrest following implementation of a predictive analytics display in a pediatric intensive care unit

Background

More than 90% of in-hospital cardiac arrests involving children occur in an intensive care unit (ICU) with less than half surviving to discharge. We sought to assess the association of the display of risk scores of cardiovascular and respiratory instability with the incidence of cardiac arrest in a pediatric ICU.

Methods

Employing supervised machine learning, we previously developed predictive models of cardiovascular and respiratory instability, incorporating real-time physiologic and laboratory data, to display risk scores for potentially catastrophic clinical events in the subsequent 12 h. Clinical implementation with risk scores displayed on large screen monitors in multiple areas throughout the ICU was finalized in July 2022. We compared the incidence of cardiac arrest events in the 18-months pre- and post-implementation.

Results

The cardiac arrest incidence rate dropped from 3.0 events (95% CI 2.0-4.4) to 2.4 events (95% CI 1.6-3.5) per 1000 patient days following implementation. We observed a 50% increase in the rate of cardiac arrest events where return of spontaneous circulation (ROSC) was achieved (p = 0.025). The incidence rate of cardiac arrest without ROSC dropped from 1.4 events (95% CI 0.7-2.4) to 0.4 events (95% CI 0.1-0.9) per 1000 patient days (incidence rate difference = 1.0 (95% CI 0.13-1.87), p = 0.01).

Conclusions

We observed a non-significant decrease in the rates of cardiac arrest events and an increase in the rate of cardiac arrests events where ROSC was achieved following the implementation of a predictive analytics display of risk scores.

Predictive analytics in bronchopulmonary dysplasia: past, present, and future

Bronchopulmonary dysplasia (BPD) remains a significant complication of prematurity, impacting approximately 18,000 infants annually in the United States. Advances in neonatal care have not reduced BPD, and its management is challenged by the rising survival of extremely premature infants and the variability in clinical practices. Leveraging statistical and machine learning techniques, predictive analytics can enhance BPD management by utilizing large clinical datasets to predict individual patient outcomes. This review explores the foundations and applications of predictive analytics in the context of BPD, examining commonly used data sources, modeling techniques, and metrics for model evaluation. We also highlight bioinformatics' potential role in understanding BPD's molecular basis and discuss case studies demonstrating the use of machine learning models for risk prediction and prognosis in neonates. Challenges such as data bias, model complexity, and ethical considerations are outlined, along with strategies to address these issues. Future directions for advancing the integration of predictive analytics into clinical practice include improving model interpretability, expanding data sharing and interoperability, and aligning predictive models with precision medicine goals. By overcoming current challenges, predictive analytics holds promise for transforming neonatal care and providing personalized interventions for infants at risk of BPD.

The 2024 Pediatric Sepsis Challenge: Predicting In-Hospital Mortality in Children With Suspected Sepsis in Uganda

The aim of this "Technical Note" is to inform the pediatric critical care data research community about the "2024 Pediatric Sepsis Data Challenge." This competition aims to facilitate the development of open-source algorithms to predict in-hospital mortality in Ugandan children with sepsis. The challenge is to first develop an algorithm using a synthetic training dataset, which will then be scored according to standard diagnostic testing criteria, and then be evaluated against a nonsynthetic test dataset. The datasets originate from admissions to six hospitals in Uganda (2017-2020) and include 3837 children, 6 to 60 months old, who were confirmed or suspected to have a diagnosis of sepsis. The synthetic dataset was created from a random subset of the original data. The test validation dataset closely resembles the synthetic dataset. The challenge should generate an optimal model for predicting in-hospital mortality. Following external validation, this model could be used to improve the outcomes for children with proven or suspected sepsis in low- and middle-income settings.

Digital solutions in paediatric sepsis: current state, challenges, and opportunities to improve care around the world

The digitisation of health care is offering the promise of transforming the management of paediatric sepsis, which is a major source of morbidity and mortality in children worldwide. Digital technology is already making an impact in paediatric sepsis, but is almost exclusively benefiting patients in high-resource health-care settings. However, digital tools can be highly scalable and cost-effective, and-with the right planning-have the potential to reduce global health disparities. Novel digital solutions, from wearable devices and mobile apps, to electronic health record-embedded decision support tools, have an unprecedented opportunity to transform paediatric sepsis research and care. In this Series paper, we describe the current state of digital solutions in paediatric sepsis around the world, the advances in digital technology that are enabling the development of novel applications, and the potential effect of advances in artificial intelligence in paediatric sepsis research and clinical care.