Predicting deterioration in the PICU

  • Research type

    Research Study

  • Full title

    Predicting adverse outcome in the PICU using a machine learning approach

  • IRAS ID

    269266

  • Contact name

    Nazima Pathan

  • Contact email

    np409@cam.ac.uk

  • Sponsor organisation

    Cambridge University Hospitals NHS Foundation Trust and University of Cambridge.

  • Duration of Study in the UK

    2 years, 0 months, 0 days

  • Research summary

    Early detection of complication or disease is crucial in preventing morbidity and death. In the intensive are unit, there are no validated systems to undertake a proactive approach to preventing deterioration in the critically ill child.
    We have identified a need for a scoring system to evaluate patients in the paediatric intensive care unit (PICU) in a way which will inform healthcare staff if they are in serious danger of deteriorating. This is because the existing paediatric early warning score (PEWS) is designed for patients of lower intensity and does not give useful information about intensive care patients.
    The availability of a rich dataset from the patient electronic health records of children admitted to PICU offers the opportunity to examine whether it is possible to predict organ failure and clinical deterioration. We would like to develop novel clinical decision support systems in order to improve our clinical care and patient outcomes. In our hospital, the use of EPIC as a tool to capture patient data during their admission offers a chance to undertake a machine learning approach to identify children at risk of organ failure. This would allow clinicians to consider escalating care in a more efficient and timely manner.
    We wish to produce a formula which will calculate the risk that a PICU patient will deteriorate at any given time. We will do this by using machine learning algorithms to analyse all available and relevant electronic healthcare data from PICU admissions at Addenbrooke's Hospital. This will involve looking at around 5000 admissions over a 6 year period and using advanced statistical analysis to find appropriate patterns in the data. If we are successful, it would then be possible to incorporate the output into a live electronic healthcare system. The envisaged system would then have the capacity to automatically alert healthcare professionals that a patient was in immediate danger of a deterioration.

  • REC name

    East of England - Cambridgeshire and Hertfordshire Research Ethics Committee

  • REC reference

    19/EE/0245

  • Date of REC Opinion

    6 Aug 2019

  • REC opinion

    Further Information Favourable Opinion