About AVERT-IT

The AVERT-IT project is focused on the creation of a novel mechanism for use in intensive and high-dependency care units to offer predictive capability to clinical personnel of the likelihood, and primary causes of arterial hypotension adverse events. The project envisages the development of a novel bed-side monitoring and alerting system dedicated to the prediction and notification (to clinical staff) of variations in the condition of the patient that are like ly to lead to hypotension without appropriate clinical intervention.

Prediction of detrimental changes to the health of the patient is of paramount importance in all clinical care. Hypotension (sub-normal blood pressure), and its associated sequalia (such as shock) is a condition commonly found within intensive and high-dependency care environments. Emergency treatment of hypotension may involve the administration of an inotropic sympathomimetic such as Dobutamine, but this treatment has associated risks including tachycardia (which may in turn risk myocardial infarction) and peripheral vasoconstriction (which may in turn result in diminished renal function). Clearly, prediction of hypotension, allowing earlier and less aggressive intervention, would dramatically improve patient outcome, leading in turn to a quantifiable reduction in the duration of average patient stay with associated reduction in cost (typically estimated at c. €1,600/patient/day as an average across the EU25).

The project consortium has access to the BrainIT database, a large repository of detailed patient data which holds: one off data (demographic/injury/admission data), periodic time-series (minute by minute physiological monitoring), episodic time-series (intensive care management) and episodic time-series (secondary insult management). The patient cohort to which these data apply were 384 patients drawn from 22 clinical centres in 11 countries, each of which were undergoing intensive care treatment for Traumatic Brain Injury (TBI). This dataset, developed through EC funded projects QLRI-CT-2000-00454 and QLG3-CT_2002-01160 , represents a significant advance in the available information landscape for high-quality, comprehensive patient data. By advancing on the work undertaken through the BrainIT group, we propose a step-change in arterial hypotension prediction within the critical, intensive and high-dependency care environment.

The main scientific objective of our proposed research is the determination (via empirical analysis, univariate analysis and multi-variable linear regression techniques) of the weighted association between multiple patient parameters (drawn from demographic, periodic and episodic datasets) and subsequent arterial hypotension (Milestone 1/Date M6).

This association will then form the basis for the initial definition of a novel Bayesian Neural Network (Milestone 2/Date M12), to be trained against the BrainIT dataset (Milestone 4/Date M18) prior to undertaking a novel clinical trial demonstrating the effectiveness of the AVERT-IT project concept (Milestone 6&7/Date M36). To facilitate robust access to clinical data required by the BANN technology, we will deploy in each of the six clinical test centres an innovative, GRID-based secure access middleware which will interface to existing hospital data sources and EPR systems (Milestone 3/Date M7). On top of the Grid interface layer a secure distributed data access platform (HypoNet) will be built which will allow remote access to data from each centre in a secure role based approached (Milestone 5/Date M12).

The main technological objective of our proposed research is the development of a novel IT-based decision support system (“HypoPredict”), appropriate for deployment within intensive and high dependency care units which is capable of: • Automatically and continually monitoring at least four in-vivo patient parameters (ECG, arterial blood pressure, Oxygen saturation and core temperature), together with open interfaces providing input of key demographic data (age, gender etc.) and periodic data (clinical pathology results etc.) related to the patient. • Outputting a continuous Hypotension Prediction Index (HPi) in the range 0 – 100 (0 = no risk; 100 = patient is currently hypotensive); • Providing primary (P1) and secondary (P2) weighted (0-100; 0 = not considered relevant; 100 = critical importance) causal data (current values of input parameters) in parallel with HPi to facilitate appropriate intervention selection by clinician (for example, elevated core temperature could be indicative of sepsis, a common precursor to hypotension); • Providing updated HPi, P1 and P2 values immediately changes are detected in the patient parameter input set.

Our proposed research meets the objectives of the call (ICT-2007.5.2). We propose the development of a novel decision support system, based on the identification of common patterns in safety relevant events that is capable of predicting adverse events. In order to support this, we will undertake data mining and robust statistical analysis of existing databases and electronic health record systems. We propose the development of a novel adverse event reporting system, validated against large multimedia databases. We propose to validate our developments through clinical trials conducted on a trans-national scale using a novel clinical trial study paradigm supported by an innovative, GRID-based secure access model, and to quantify the benefits in terms of ICU bed/day cost, reduced drug costs and in terms of reduced corrective care, obviated as a consequence of the improved prediction of, and earlier intervention in, hypotension adverse events, and in improved patient outcome with associated reduction in carer-burden.