Objectives
The primary objective of our research is providing clinicians with advanced notice of upcoming hypotensive episodes.
In order to achieve this result we’re needing to solve a number of issues:
* Agree a common definition of hypotensive episode – all six of our centres currently work to different definitions
* Build a software interface which collects data from patient monitoring devices – all centres have different equipment
* Find a way of transmitting patient data from the research centres to our data warehouse
* Build software which collects observation, treatment and outcome data at the bedside
* Build a Bayesian Artificial Neural Network and train it to recognise patterns which precede hypotensive episodes
* Build software to enable researchers to analyse actual data, identifying circumstances in which the BANN achieves an accurate prediction and those in which if provides false positives
* Build software which alerts clinicians to potential upcoming episodes.
* Multi stage clinical trial.
In parallel the project includes the need for multi stage ethics approval, support from IT staff in each of the centres, identification and protection of foreground IPR, and multiple routes to market.
We’re also required to promote Framework 7 research and collaborate with other FP7 projects as appropriate.
Results
The starting point for our research has been a database of patient care data amassed as part of the Brain IT project, previously funded under Frameworks 4 and 5. This data is minute by minute readings from monitors, and comprises more than 200 patients with traumatic brain injuries. In some cases the data is incomplete and possibly inaccurate.
In the first stage we analysed the legacy data testing the various definitions of hypotension used by the research centres. Comparison of the definitions and events indicated by them enabled us to agree between all six centres a common definition, on which we could train the BANN.
In the second stage the BANN was used to examine patterns of parameters preceding episodes. Subsequent monitoring of those patterns suggested approximately 35% of future episodes could be identified as outputs from the pattern recognition. In addition some false positive predictions indicated ways the monitored parameters and definitions could be refined to improve accuracy.
In the third stage, working with IT staff at the centres data capture and transmission software has been built and installed. Software for collecting data at the bedside has been built in conjunction with clinical staff and installed. Software enabling researchers to monitor and enhance the accuracy of the prediction engine has been installed and is now in production use.
in the fourth stage we’ll be monitoring and enhancing the BANN prediction capabilities, preparing data which will support our application for ethics approval for stage five – live monitoring and alerting of upcoming episodes.
Throughout the project we’ve been researching questions related to the protection of IPR in the worlds major markets. As of yet we’ve made no application for patent – because we haven’t finished the research. We’re still refining our understanding of monitored parameters and which patterns offer which levels of accurate prediction.
We’ve also collaborated with multiple public bodies, promoting FP7 and offering guidance to potential applicants.
The Avert-IT project has been selected by the Use and Diffuse project as an example of best practice and features in the projects publications.
