Summary of our research
The clinical outcome modelling research carried out in Portsmouth is led by Professor David Prytherch and Professor Jim Briggs. Our collaborators include Portsmouth Hospitals NHS Trust (PHT), University of Oxford, Oxford University Hospitals NHS Trust, University of Southampton, Bournemouth University and The Learning Clinic. Our approach is extremely inter-disciplinary, but embedded in all we do are the fundamental principles that information must be acquired by reliable means and reasoned about rigorously; all applied in a clinical context.
We collect and use clinical data to model adverse patient outcome. The models enable clinicians to predict which patients are at risk of deterioration, and medically intervene. Our research has built on work done in the late 1990s and up to 2003 to develop models (P-POSSUM) of outcomes in surgery (Prytherch, Whiteley, Higgins, Weaver, et al, 1998). P-POSSUM was a success and has been widely adopted, but is only applicable to surgical cases. This led us to investigate ways to model outcomes in general medicine cases, using pathology data. We have shown that biochemistry and haematology outcome models (BHOM) can be used to identify patients at risk of mortality with very high discrimination (Prytherch, Sirl, Schmidt, Featherstone, et al, 2005; Prytherch, Briggs, Weaver, Schmidt and Smith, 2005).
Other monitoring and surveillance systems (e.g., Dr Foster, CHKS and HES) require coded administrative data only available after discharge. Our techniques add clinical context to these, and have obvious uses in clinical governance and clinical performance management as well as direct patient care. Our approach only uses data routinely collected and available immediately after a patient's admission to hospital.
We know that serious physiological abnormalities frequently precede primary events (defined as in-hospital deaths, cardiac arrests, and unanticipated intensive care unit admissions) (Kause, Smith, Prytherch, Parr, et al, 2004). The P-POSSUM / BHOM work led to our collaboration with The Learning Clinic Ltd (TLC). In return, TLC provided a means to collect vital signs data quickly and accurately in an electronic format. As a result we have access to probably the biggest database of vital signs data anywhere in the world.
Using that and related data we have shown that:
- innovative techniques can be used to join different databases in such a way that clinical significance is not lost or corrupted (unpublished work due to commercial confidentiality)
- vital signs data can be used to devise an early warning score (EWS) system that can both identify patients whose condition is deteriorating and minimise unnecessary false alarms (Smith et al, 2006)
- an EWS devised from vital signs data (ViEWS) performs better than any of the 33 other EWS systems in the literature (Prytherch et al, 2010)
- an EWS can be devised from blood test data (Jarvis et al, 2013)
- decision tree data mining techniques can be used to develop new early warning score systems (DT-EWS) quickly (Badriyah et al,, 2013)
- aggregate National Early Warning Score (NEWS) values are more important than high scores for a single vital signs parameter for discriminating the risk of adverse outcomes (Jarvis et al, 2014)
- introducing an electronic physiological surveillance system (employing NEWS) reduces hospital mortality (Schmidt et al, 2015)
- NEWS performs better than the other EWSs in discriminating risk of death within 24 h of an observation set, irrespective of the method used to select the observations (Jarvis et al, 2015a)
- a simplified version of NEWS (binary NEWS) works almost as effectively at discriminating patients at risk of adverse outcome, but may result in a higher workload for clinical staff (Jarvis et al, 2015b)
- NEWS performs as well in the care of surgical patients as it does in the care of the general medicine patients for which it was first devised (Kovacs et al, 2016)
- NEWS is a better discriminator of a range of adverse outcomes and produces less workload than any of the MET calling criteria (Smith et al, 2016)
Our EWS models can be applied to any patient under clinical care, but are increasingly used to allow nurses to determine which of their patients are deteriorating and when to summon assistance (e.g. a doctor), without causing too many false alarms (which would overburden hospital resources).
Impact of our work
Our work has had considerable impact (and formed a highly-rated case study for the 2014 REF), including:
- Our work heavily influenced the Royal College of Physicians' (RCP) new standard for the assessment of the severity of acute illness (known as the "National Early Warning Score" or NEWS). The specific recommendation is for adoption by NHS bodies, but is already being adopted internationally.
- The chairman of the RCP working party estimated that our work could result in the saving of thousands of lives per year.
- Our work is incorporated in the system developed by Vitalpac Ltd (formerly The Learning Clinic Ltd (TLC)), and (as of 2017) was deployed to more than 50 hospitals.
More details on the impact of this work
We are currently working on two externally funded projects:
- HAVEN (2015-2018) looks at what other data in hospital information systems can be used to predict unanticipated intensive care unit admission (with University of Oxford, Portsmouth Hospitals NHS Trust and Oxford University Hospitals NHS Trust).
- Missed Care (2015-2017) looks at how the provision of nurses in hospitals affects the care and safety of patients (with University of Southampton, Portsmouth Hospitals NHS Trust and University of York). Currently in the writing up phase.
In addition, the following PhD students are doing work on clinical outcome modelling:
- Aya Awad: can we devise ways to make predictions about patients in intensive care earlier than existing models?
- Simarjot Dahella: how frequently do vital signs observations need to be made?
- Caroline Kovacs: NEWS was developed for general medicine patients; does it also work for surgery patients?