Abstract

BackgroundHypovolemic shock is characterized by a critically reduced central blood volume (CBV). The heart rate and blood pressure responses to a reduced CBV do not predict onset of hypovolemic shock. We hypothesized that a machine learning model detects subtle changes and patterns within hemodynamic signals that predict impending pre‐syncope as an expression of a critically reduced CBV.MethodsIn 33 healthy subjects we provoked pre‐syncope by reducing CBV using continuous lower body negative pressure. We extracted information from the arterial pressure wave and used it as model input. Next to blood pressure and heart rate, we introduced a new set of blood pressure wave features. We trained a support vector machine to predict time remaining till pre‐syncope using leave‐one‐out methods. We expressed the model performance as absolute error and a squared correlation coefficient (r2).ResultsIn 72% of the predictions the model followed the decreasing time remaining towards pre‐syncope. Mean r2 was 0.43 [0.13 – 0.64]. In 13 subjects the model predicted pre‐syncope with an median error of −49 seconds [−312 −27].ConclusionThe model detects a trend towards pre‐syncope conforming to a worsening condition of subjects in a controlled progressive hypovolemia setting, but cannot pinpoint the onset of pre‐syncope exactly. Future research may reveal the most valuable input feature and allow for testing current models during real hemorrhage.Support or Funding InformationEducational grant supplied by Edwards Lifesciences.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call