Predicting Hypotensive Episodes

Our first research project – Avert-IT – targets the prediction of hypotensive episodes occurring during intensive care.

The research breaks down into three main sections:

Analyzing a legacy data warehouse of monitored parameters to find patterns which, when compared with other patterns, can pre-indicate a future episode.
Using a proprietary Bayesian Neural Network, identify those patterns arising in data. collected from current patient monitoring, and pre-indicating actual episodes.
Implementing the technology in Intensive Care Units, in a clinical trial.

As of November 2009, the research is proceeding in accordance with the plan.

Success at predicting upcoming events using the legacy database has been encouraging with clinicians confirming the potential for improving patient care.

We’ve identified areas for deeper analysis. It turns out the prediction can be based on a relatively simple parameter set, but there are much better results likely if we look for a number of conditions (in the data) and then apply some logic to relationships between them.

We’ve built the software environment for the middle phase. and started collecting minute by minute data from multiple patients, in a number of ICUs, and applying the pattern recognition.

During this phase ICU staff will be collecting patient care data, but the results will not be made available to clinicians managing the patient care.

A team of scientists will be receiving the results and evaluating the ability for the technology to offer predictions at an acceptable level of probability.

In the third phase we’ll be feeding back to clinicians in the ICU on the outputs of the pattern recognition.

In the meantime we’ve identified a number of opportunities for commercially exploiting the research results.