A Hypotension Surveillance and Prediction System for Critical Care

Introduction

Blood Pressure (BP) is the force exerted by circulating blood on the walls of blood vessels, and is one of life’s main vital signs. BP is generated by the heart when it pumps blood into the arteries, and is regulated by the response given by arteries to the flow of blood. Hypotension (HT) occurs when there is an abrupt fall of BP leading to below normal values, so low it causes symptoms or signs due to the low flow of blood through the arteries and veins. When the blood flow is too low to deliver enough oxygen and nutrients to vital organs such as the brain, heart and kidney, they do not function normally and can get permanently damaged. Hypotension is classified as a common complication in patients with critical care needs, such as hemodialysis (Daugirdas, 2001).

Hemodialysis is the combination of ultrafiltration of fluid excess and the clearance of solute waste products such as urea by diffusion. This treatment is accompanied by a wide variety of complications. Despite all technical improvements, the most frequent complication is intradialytic hypotension (IDHT), occurring in up to 20% of dialysis sessions (Straver, 2005). IDHT occurs when our normal biological compensatory mechanisms cannot cope with the removal of intravascular fluid in a short period of time in a specific patient. Negative effects of IDHT are the patient’s discomfort and decrease of hemodialysis efficacy due to interventions such as interruption of dialysis and the need for intravenous infusions. IDHT can induce other severe complications such as serious heart, endocrine or neurological and can even lead to the patient’s death. Moreover, a recent study reports an increase of mortality in HT prone hemodialysis patients (Shoji, 2004). Therefore, reducing HT episodes in critical care patients remains a challenge.

Typically, most medical systems store biosignal data for a short time, in order to monitor the patient’s most recent vital signs, and usually do not forecast or issue an alert concerning a patient’s predictable hazardous situation (Bai, 2006). Therefore, most medical systems are typically focused on detecting the current health status of a certain number of vital signs, rather than attempting to predict immediate future trends on any of those features. Thus, standard methods are mainly centered on feature detection rather than feature prediction.

The main issues concerning useful time prediction and detection of HT episodes evolve around two main aspects: 1) How can we efficiently store the biosignal data that is needed in a continuous manner and enable real-time hypotension monitorization and prediction; and 2) Which are the algorithms to be used for efficient hypotension detection and prediction. The Physionet Challenge 2009 (Physionet C, 2009) promoted solutions for predicting HT patient status. The paper (Henriques, 2009), using a neural network approach, was the challenge’s winner, scoring 47 correct predictions out of 50 for the whole event. Our work continues the former research, proposing a full HT monitoring, detecting and predicting medical care system that also promotes medical staff mobility. Regarding efficient real-time biosignal data integration, we present a database for storing all patients BP and heart rate (HR) data. Using this data, we have developed a new HT prediction technique, working with a set of applications capable of monitoring each patient’s status and efficiently predicting if a hypotensive episode will occur during the next 60 minutes.