Paroxysmal sympathetic hyperactivity risk modeling based on transients in time series describing the autonomic nervous system and cerebral hemodynamics

Figures

• 

  Fig. 1. Flow chart. Abbreviations: TBI, traumatic brain injury; ICP, intracranial pressure; ICM +, intensive care monitoring software

• 

  Fig. 2. Study design. This retrospective study included 41 traumatic brain injury (TBI) patients hospitalized at Wroclaw University Hospital (Poland), of whom 14 were diagnosed as being at risk for paroxysmal sympathetic hyperactivity (PSH) using the Clinical PSH Assessment Measure (PSH-AM) between days nine and fifteen post-injury. PSH risk modeling was performed using clinical metadata, averaged neuromonitoring parameters, and metrics describing the variability of correlation between autonomic nervous system (ANS) and cerebral hemodynamics parameters over sliding windows of 3, 6, 12, and 24 h. Created at https://BioRender.com

• 

  Fig. 3. The discriminative ability of feature sets to differentiate patients at risk of paroxysmal sympathetic hyperactivity (PSH) from those not at risk is presented as the area under the receiver operating characteristic (ROC) curve (AUC). Feature set III, which includes the variability of the correlation between the autonomic nervous system and cerebral hemodynamics, achieved the highest score

• 

  Fig. 4. The time span of the significance of the autonomic nervous system–cerebral hemodynamics correlation, measured as the zero-crossing rate (ZCR) over a 3-h sliding window, within the paroxysmal sympathetic hyperactivity (PSH) risk model. Feature importance and SHAP absolute values were aggregated across all the models in the stratified five-fold cross-validation. These values were then rescaled using min–max normalization to a range between 0 and 1. The mean aggregated values were 0.55 ± 0.12 for feature importance and 0.54 ± 0.12 for SHAP

• 

  Fig. 5. The most important features for assessing paroxysmal sympathetic hyperactivity (PSH) risk were determined using feature importance and SHAP methods. The results were aggregated for the logistic regression models based on feature set III. The importance analysis of the correlation between the autonomic nervous system (ANS) and cerebral hemodynamics metrics was provided using zero-crossing rate (ZCR) within the first 3 h. Panel A presents clinical metadata, Panel B averaged neuromonitoring parameters, and Panel C variability of rolling correlations between ANS metrics and cerebral hemodynamics. Abbreviations: Hb, hemoglobin; tSAH, traumatic subarachnoid hemorrhage; GCS, Glasgow Coma Scale; SDH, subdural hematoma; tICH, traumatic intracerebral hematoma; ISS, injury severity score; WBC, white blood cell; HR, heart rate; ABP, arterial blood pressure; ICP, intracranial pressure; PRx, pressure reactivity index