Towards Intelligent Clinical Informatics System for Paroxysmal Sympathetic Hyperactivity among Traumatic Brain Injury Patients
Digital Repository at the University of Maryland (University of Maryland College Park) · 24.07.2026
DOI: 10.13016/m2fwol-jkac ↗Kurz gefasst
Fachartikel in der Zeitschrift Digital Repository at the University of Maryland (University of Maryland College Park), erschienen am 24.07.2026. Verfasst von Sancharee Hom Chowdhury.
Abstract (Originalsprache)
Traumatic brain injury (TBI) is a leading cause of death & long-term disability. A subset of patients develops Paroxysmal Sympathetic Hyperactivity (PSH), a syndrome characterized by episodic tachycardia, hyperthermia, hypertension, tachypnea, motor rigidity, and diaphoresis resulting from autonomic dysfunction. PSH affects approximately 8-33% of TBI patients and is associated with prolonged intensive care, increased morbidity, and poorer physiological outcomes. Although the PSH-Assessment Measure (PSH-AM) standardizes diagnosis using the Clinical Features Scale (CFS) and Diagnosis Likelihood Tool (DLT), its reliance on daily maximum physiological values and a 14-day observation period delays diagnosis. Consequently, important challenges remain in characterizing heterogeneous disease progression, enabling early diagnosis, and evaluating treatment effectiveness. This thesis analyzed a retrospective cohort of 221 adult TBI patients admitted to the R Adams Cowley Shock Trauma Center between January 2016 and July 2018 who underwent retrospective PSH-AM assessment. The first objective characterized longitudinal physiological progression using Group-Based Trajectory Modeling with K-means clustering on 13 consecutive days of CFS scores. Four distinct trajectory groups were identified, representing heterogeneous recovery patterns. Persistently elevated CFS trajectories after four to five days were strongly associated with PSH diagnosis and adverse clinical outcomes, highlighting the importance of early trajectory identification. The second objective developed supervised machine learning models to predict trajectory membership, PSH diagnosis, and day-14 DLT scores using admission characteristics and hourly modified CFS (mCFS) collected during the first 24-96 hours after admission. An odds-based weighting strategy improved prediction of severe clinical presentations. Daily maximum mCFS features from the first 72 hours achieved the best trajectory prediction, while admission characteristics provided the highest performance for PSH diagnosis & DLT prediction. The final objective evaluated the causal effects of commonly used treatments, including β-blockers and bromocriptine, using marginal structural models with inverse probability of treatment weighting. β-blockers consistently reduced next-day CFS across all model specifications, indicating a robust immediate therapeutic effect, whereas bromocriptine alone demonstrated less consistent benefits. Overall, this dissertation presents an integrated framework that combines trajectory discovery, early prediction, and longitudinal causal inference to support earlier diagnosis and data-driven clinical decision-making for patients with PSH following TBI.
Schlagwörter
Zitieren
APA
Chowdhury, S. H. (2026). Towards Intelligent Clinical Informatics System for Paroxysmal Sympathetic Hyperactivity among Traumatic Brain Injury Patients. Digital Repository at the University of Maryland (University of Maryland College Park). https://doi.org/10.13016/m2fwol-jkac
Vancouver
Chowdhury SH. Towards Intelligent Clinical Informatics System for Paroxysmal Sympathetic Hyperactivity among Traumatic Brain Injury Patients. Digital Repository at the University of Maryland (University of Maryland College Park). 2026. doi:10.13016/m2fwol-jkac
BibTeX
@article{Chowdhury2026,
author = {Chowdhury, Sancharee Hom},
title = {Towards Intelligent Clinical Informatics System for Paroxysmal Sympathetic Hyperactivity among Traumatic Brain Injury Patients},
journal = {Digital Repository at the University of Maryland (University of Maryland College Park)},
year = {2026},
doi = {10.13016/m2fwol-jkac}
}
EUVIMED