Computational frameworks for automated detection and quantification of paroxysmal sympathetic hyperactivity among traumatic brain injury patients.

Paroxysmal sympathetic hyperactivity (PSH) is a syndrome that occurs in a large subset of critically ill traumatic brain injury (TBI) patients and is associated with complications and poor recovery. PSH is defined by recurrent episodic vital sign elevations in the appropriate clinical context. However, standard diagnostic criteria rely heavily on subjective judgment, leading to challenges and delays in recognition, monitoring, and management. The objective of this study was to develop automated PSH detection and quantification tools that exclusively utilize objective bedside continuous vital sign data. Using a cohort of 221 critically ill acute TBI patients with at least 14 days of continuous physiologic data (of which 107 were clinically diagnosed with PSH) we developed a high-resolution clinical feature scale based on established PSH-Assessment Measure criteria and two artificial intelligence-based episode detection models including an expert system approach and a machine learning model approach, using a clinician-annotated case example as ground truth. For the episode detection methods, PSH was quantified as the number, duration, and overall temporal burden of detected episodes. To evaluate performance, we compared quantifications across PSH cases and controls and explored precision and recall. All three methods demonstrated initial face validity to delineate PSH cases from non-PSH TBI controls. Future optimization and implementation of the described computational frameworks with real-time patient data could improve the standard monitoring and management of this challenging clinical syndrome.