Novel Claims-Based Outcome Phenotypes in Survivors of Pediatric Traumatic Brain Injury

Modeling healthcare demands and long-term costs following pediatric traumatic brain injury

Introduction

Traumatic brain injury (TBI) is a leading cause of death and disability in children, but data on the longitudinal healthcare and financial needs of pediatric patients is limited in scope and duration. We sought to describe and predict these metrics following acute inpatient treatment for TBI.

Methods

Children surviving their initial inpatient treatment for TBI were identified from Optum's deidentified Clinformatics® Data Mart Database (2007-2018). Treatment cost, healthcare utilization, and future inpatient readmission were stratified by follow-up intervals, type of claim, and injury severity. Both TBI-related and non-TBI related future cost and healthcare utilization were explored using linear mixed models. Acute inpatient healthcare utilization metrics were analyzed and used to predict future treatment cost and healthcare demands using linear regression models.

Results

Among 7,400 patients, the majority suffered a mild TBI (50.2%). For patients with at least one-year follow-up (67.7%), patients accrued an average of 28.7 claims and $27,199 in costs, with 693 (13.8%) readmitted for TBI or non-TBI related causes. Severe TBI patients had a greater likelihood of readmission. Initial hospitalization length of stay and discharge disposition other than home were significant positive predictors of healthcare and financial utilization at one-and five-years follow-up. Linear mixed models demonstrated that pediatric TBI patients would accrue 21.1 claims and $25,203 in cost in the first year, and 9.4 claims and $4,147 in costs every additional year, with no significant differences based on initial injury severity.

Discussion

Pediatric TBI patients require long-term healthcare and financial resources regardless of injury severity. Our cumulative findings provide essential information to clinicians, caretakers, researchers, advocates, and policymakers to better shape standards, expectations, and management of care following TBI.

Unsupervised Clustering in Neurocritical Care: A Systematic Review

Managing patients with acute brain injury in the neurocritical care (NCC) unit has become increasingly complex because of technological advances and increasing information derived from multiple data sources. Diverse data streams necessitate innovative approaches for clinicians to understand interactions between recorded variables. Unsupervised clustering integrates different data streams and could be supportive. Here, we provide a systematic review on the use of unsupervised clustering using NCC data. The primary objective was to provide an overview of clustering applications in NCC studies. As a secondary objective, we discuss considerations for future NCC studies. Databases (Medline, Scopus, Web of Science) were searched for unsupervised clustering in acute brain injury studies including traumatic brain injury (TBI), subarachnoid hemorrhage, intracerebral hemorrhage, acute ischemic stroke, and hypoxic-ischemic brain injury published until  March 13th 2024. We performed the systematic review in accordance with the Preferred Reporting Items for Systematic reviews and Meta-Analyses guidelines. We identified 18 studies that used unsupervised clustering in NCC. Predominantly, studies focused on patients with TBI (12 of 18 studies). Multiple research questions used a variety of resource data, including demographics,  clinical- and monitoring data, of which intracranial pressure was most often included (8 of 18 studies). Studies also covered various clustering methods, both traditional methods (e.g., k-means) and advanced methods, which are able to retain the temporal aspect. Finally, unsupervised clustering identified novel phenotypes for clinical outcomes in 9 of 12 studies. Unsupervised clustering can be used to phenotype NCC patients, especially patients with TBI, in diverse disease stages and identify clusters that may be used for prognostication. Despite the need for validation studies, this methodology could help to improve outcome prediction models, diagnostics, and understanding of pathophysiology.Registration number: PROSPERO: CRD4202347097676.

Applications of machine learning in pediatric traumatic brain injury (pTBI): a systematic review of the literature

OBJECTIVE

Pediatric traumatic brain injury (pTBI) is a heterogeneous condition requiring the development of clinical decision rules (CDRs) for the optimal management of these patients. Machine learning (ML) is a novel artificial intelligence (AI) predictive tool with various applications in modern neurosurgery, including the creation of CDRs for patients with pTBI. In the present study, we summarized the current literature on the applications of ML in pTBI.

METHODS

A systematic review was conducted following the PRISMA guidelines. The literature search included PubMed/MEDLINE, SCOPUS, and ScienceDirect databases. We included observational or experimental studies focusing on the applications of ML in patients with pTBI under 18 years of age.

RESULTS

A total of 18 articles were included in our systematic review. Of these articles, 16 were retrospective cohorts, 1 was a prospective cohort, and 1 was a case-control study. Of these articles, ten concerned ML applications in predicting the outcome of pTBI patients, while 8 reported applications of ML in predicting the need for CT scans. Artificial Neuronal Network (ANN) and Random Forest (RF) were the most commonly utilized models for the creation of predictive algorithms. The accuracy of the ML algorithms to predict the need for CT scan in pTBI cases ranged from 0.790 to 0.999, and the Area Under Curve (AUC) ranged from 0.411 (95%CI: 0.354-0.468) to 0.980 (95%CI: 0.950-1.00). The model with the maximum accuracy to predict the need for CT scan was a Deep ANN model, while the model with the maximum AUC was Ensemble Learning. The model with the maximum accuracy to predict the outcome (favorable vs. unfavorable) of patients with TBI was a support vector machine (SVM) model with 94.0% accuracy, whereas the model with the highest AUC was an ANN model with an AUC of 0.991.

CONCLUSION

In the present systematic review, conventional and novel ML models were utilized to either predict the presence of intracranial trauma or the prognosis of children with pTBI. However, most of the reported ML algorithms have not been externally validated and are pending further research.

Predicting and improving outcome in severe pediatric traumatic brain injury

INTRODUCTION

Severe pediatric traumatic brain injury (spTBI), including abusive head trauma (AHT) in young children, is a major public health problem. Long-term consequences of spTBI include a large variety of physical, neurological, biological, cognitive, behavioral and social deficits and impairments.

AREAS COVERED

The present narrative review summarizes studies and reviews published from January 2019 to February 2024 on spTBI. Significant papers published before 2019 were also included. The article gives coverage to the causes of spTBI, its epidemiology and fatality rates; disparities, inequalities, and socioeconomic factors; critical care; outcomes; and interventions.

EXPERT OPINION

There are disparities between countries and according to socio-economic factors regarding causes, treatments and outcomes of spTBI. AHT has an overall poor outcome. Adherence to critical care guidelines is imperfect and the evidence-base of guidelines needs further investigations. Neuroimaging and biomarker predictors of outcomes is a rapidly evolving domain. Long-term cognitive, behavioral and psychosocial difficulties are the most prevalent and disabling. Their investigation should make a clear distinction between objective (clinical examination, cognitive tests, facts) and subjective measures (estimations using patient- and proxy-reported questionnaires), considering possible common source bias in reported difficulties. Family/caregiver-focused interventions, ecological approaches, and use of technology in delivery of interventions are recommended to improve long-term difficulties after spTBI.

Discovery of generalizable TBI phenotypes using multivariate time-series clustering

Traumatic Brain Injury (TBI) presents a broad spectrum of clinical presentations and outcomes due to its inherent heterogeneity, leading to diverse recovery trajectories and varied therapeutic responses. While many studies have delved into TBI phenotyping for distinct patient populations, identifying TBI phenotypes that consistently generalize across various settings and populations remains a critical research gap. Our research addresses this by employing multivariate time-series clustering to unveil TBI's dynamic intricates. Utilizing a self-supervised learning-based approach to clustering multivariate time-Series data with missing values (SLAC-Time), we analyzed both the research-centric TRACK-TBI and the real-world MIMIC-IV datasets. Remarkably, the optimal hyperparameters of SLAC-Time and the ideal number of clusters remained consistent across these datasets, underscoring SLAC-Time's stability across heterogeneous datasets. Our analysis revealed three generalizable TBI phenotypes (α, β, and γ), each exhibiting distinct non-temporal features during emergency department visits, and temporal feature profiles throughout ICU stays. Specifically, phenotype α represents mild TBI with a remarkably consistent clinical presentation. In contrast, phenotype β signifies severe TBI with diverse clinical manifestations, and phenotype γ represents a moderate TBI profile in terms of severity and clinical diversity. Age is a significant determinant of TBI outcomes, with older cohorts recording higher mortality rates. Importantly, while certain features varied by age, the core characteristics of TBI manifestations tied to each phenotype remain consistent across diverse populations.

Discovery of Generalizable TBI Phenotypes Using Multivariate Time-Series Clustering

Traumatic Brain Injury (TBI) presents a broad spectrum of clinical presentations and outcomes due to its inherent heterogeneity, leading to diverse recovery trajectories and varied therapeutic responses. While many studies have delved into TBI phenotyping for distinct patient populations, identifying TBI phenotypes that consistently generalize across various settings and populations remains a critical research gap. Our research addresses this by employing multivariate time-series clustering to unveil TBI's dynamic intricates. Utilizing a self-supervised learning-based approach to clustering multivariate time-Series data with missing values (SLAC-Time), we analyzed both the research-centric TRACK-TBI and the real-world MIMIC-IV datasets. Remarkably, the optimal hyperparameters of SLAC-Time and the ideal number of clusters remained consistent across these datasets, underscoring SLAC-Time's stability across heterogeneous datasets. Our analysis revealed three generalizable TBI phenotypes (α, β, and γ), each exhibiting distinct non-temporal features during emergency department visits, and temporal feature profiles throughout ICU stays. Specifically, phenotype α represents mild TBI with a remarkably consistent clinical presentation. In contrast, phenotype β signifies severe TBI with diverse clinical manifestations, and phenotype γ represents a moderate TBI profile in terms of severity and clinical diversity. Age is a significant determinant of TBI outcomes, with older cohorts recording higher mortality rates. Importantly, while certain features varied by age, the core characteristics of TBI manifestations tied to each phenotype remain consistent across diverse populations.

Post-discharge rehabilitation and functional recovery after pediatric injury

INTRODUCTION

Variability in rehabilitation disposition has been proposed as a trauma center quality metric. Benchmarking rehabilitation disposition is limited by a lack of objective measures of functional impairment at discharge. The primary aim of this study was to determine the relative contribution of patient characteristics and hospitalization factors associated with inpatient and outpatient rehabilitation after discharge. The secondary aims were to evaluate the sensitivity of the Functional Status Scale (FSS) score for identifying functional impairments at hospital discharge and track post-discharge recovery.

PATIENTS AND METHODS

We report a planned secondary analysis of a prospective observational study of seriously injured children (<15 years old) enrolled at seven pediatric trauma centers. Functional Status Scale (FSS) score was measured for pre-injury, hospital discharge, and 6-month follow-up timepoints. Multinomial logistic regression identified factors associated with three dispositions: home without rehabilitation services, home with outpatient rehabilitation, and inpatient rehabilitation. Relative weight analysis was used to identify the impact of individual factors associated with inpatient or outpatient rehabilitation disposition.

RESULTS

We analyzed 427 children with serious injuries. Functional impairment at discharge was present in 103 (24.1%) children, including 43/337 (12.8%) discharged without services, 12/38 (31.6%) discharged with outpatient rehabilitation, and 44/47 (93.6%) discharged to inpatient rehabilitation. In multivariable modeling, variables most contributing to prediction of inpatient rehabilitation were severe initial Glasgow coma scale (GCS), injured body region, and functional impairment at discharge. Severe initial GCS, private insurance, and extremity injury were independently associated with disposition with outpatient rehabilitation. Patients discharged without services or with outpatient rehabilitation most frequently had motor impairments that improved during the next 6 months. Patients discharged to inpatient rehabilitation had impairments in all domains, with many improving within 6 months. A higher proportion of patients discharged to inpatient rehabilitation had residual impairments at follow-up.

CONCLUSION

Injury characteristics and discharge impairment were associated with discharge to inpatient rehabilitation. The FSS score identified impairments needing inpatient rehabilitation and characterized improvements after discharge. Less severe impairments needing outpatient rehabilitation were not identified by the FSS score.

Postdischarge health resource use in pediatric survivors of prolonged mechanical ventilation for acute respiratory illness

We aimed to identify characteristics associated with postdischarge health resource use in children without medical complexity who survived an episode of prolonged mechanical ventilation for respiratory illness. We hypothesized that longer durations of mechanical ventilation, noncomplex chronic conditions, and severe acute respiratory distress syndrome (ARDS) would be associated with readmission or an Emergency Department (ED) visit. In this retrospective cohort, we evaluated children without a complex chronic condition who survived a respiratory illness requiring ≥3 days of mechanical ventilation and who had insurance eligibility within the Colorado All Payers Claims Database. We used insurance claims to characterize health resource use and multivariable logistic regression to identify characteristics associated with readmission or an ED visit during the postdischarge year. We evaluated 82 children, median age 12.8 months (interquartile range [IQR]: 4.0-24.1), 20 (24%) with a noncomplex chronic condition and 62 (76%) without any chronic conditions. Bronchiolitis (60%) and pneumonia/aspiration pneumonitis (17%) were the most common etiologies of respiratory failure and 47 (57%) patients had severe ARDS. Forty-six (56%) patients had an ED visit or readmission. Among the 18 readmitted patients, 16/18 (89%) readmissions were for respiratory illness. Forty (49%) patients had ≥2 outpatient pulmonary visits and 45 (55%) filled a pulmonary medication prescription. In analyses controlling for age, illness severity and mechanical ventilation duration, severe ARDS was predictive of ED visit or readmission (odds ratio [OR]: 5.53 [95% confidence interval [CI]: 1.79, 19.09]). Children who survive prolonged mechanical ventilation for respiratory disease experience high rates of postdischarge health resource use, particularly those surviving severe ARDS.

Identifying Long-Term Morbidities and Health Trajectories After Prolonged Mechanical Ventilation in Children Using State All Payer Claims Data

OBJECTIVES

To identify postdischarge outcome phenotypes and risk factors for poor outcomes using insurance claims data.

DESIGN

Retrospective cohort study.

SETTING

Single quaternary center.

PATIENTS

Children without preexisting tracheostomy who required greater than or equal to 3 days of invasive mechanical ventilation, survived the hospitalization, and had postdischarge insurance eligibility in Colorado's All Payer Claims Database.

INTERVENTIONS

None.

MEASUREMENTS AND MAIN RESULTS

We used unsupervised machine learning to identify functional outcome phenotypes based on claims data representative of postdischarge morbidities. We assessed health trajectory by comparing change in the number of insurance claims between quarters 1 and 4 of the postdischarge year. Regression analyses identified variables associated with unfavorable outcomes. The 381 subjects had median age 3.3 years (interquartile range, 0.9-12 yr), and 147 (39%) had a complex chronic condition. Primary diagnoses were respiratory (41%), injury (23%), and neurologic (11%). We identified three phenotypes: lower morbidity (n = 300), higher morbidity (n = 62), and 1-year nonsurvivors (n = 19). Complex chronic conditions most strongly predicted the nonsurvivor phenotype. Longer PICU stays and tracheostomy placement most strongly predicted the higher morbidity phenotype. Patients with high but improving postdischarge resource use were differentiated by high illness severity and long PICU stays. Patients with persistently high or increasing resource use were differentiated by complex chronic conditions and tracheostomy placement.

CONCLUSIONS

New morbidities are common after prolonged mechanical ventilation. Identifying phenotypes at high risk of postdischarge morbidity may facilitate prognostic enrichment in clinical trials.