Condition category
Signs and Symptoms
Date applied
Date assigned
Last edited
Retrospectively registered
Overall trial status
Recruitment status

Plain English Summary

Background and study aims
Late recognition of deteriorating patients (i.e their condition getting worse) in hospitals causes delays in treatment for these patients, which then results in an increase in mortality (ill health) and morbidity (death) despite the widespread introduction of vital sign-based “early warning scores”.
Therefore, developing systems that recognise earlier when a patient is at risk of a severe but reversible deterioration in health is a key goal for the NHS. The HAVEN Project aims to produce a hospital-wide IT system that enables a continuous risk assessment in all hospital patients, and predicts those at risk of deterioration.

Who can participate?
Data from the records of all adult patients over the age of 18 who are admitted to the Oxford
University Hospitals and Portsmouth Hospitals NHS Trusts. Staff members from both trusts are also recruited to help design the user interface.

What does the study involve?
An IT system is developed that routinely stores electronic data (including demographics, laboratory results and vital signs recordings) to create a continuous risk assessment. Data is gathered from different local databases; at present they are not integrated or displayed in a way that supports decision making or calculation of patient risk. Risk prediction algorithms are then developed (programs that look at the information and decide whether a patient is likely to be at risk of deteriorating) and tested. These use the records of patients who are admitted to hospital and then admitted to an intensive care unit (ICU) after two or more days in hospital. The information about these patients illustrates the pathway from the first signs of deterioration on the ward to ICU admission.There is no direct involvement for patients who are included in the project. Information from their hospital stay is collected by the NHS IT groups in a way that will allow the university based research team to review the information without knowing who the people are. During the project the automatic system is checked to see whether it has correctly recognised if patients needed intensive care treatment during their hospital stay. Doctors and nurses who are part of the research team need access to individual patient information to check this. The algorithms are used to create an interface allowing clinical staff to identify, rank, review and treat patients who, without acute medical intervention (treatment), will deteriorate and require ICU admission. Members of staff are asked questions about the information they would look for to recognise that a patient was becoming unwell. They are then observed as they carry out their usual daily tasks.

What are the possible benefits and risks of participating?
There are no risks to patients whose data is used in the project and the potential benefit is the
development of a hospital wide IT system that is capable of recognising patients at risk of
deterioration. Clinical participants (healthcare staff) will need to find time to meet with the research team to discuss the project plans in some detail as well as participate in the cognitive task analysis and user interface testing.

Where is the study run from?
The project is run from the University of Oxford. It takes place at the University of
Portsmouth, Portsmouth Hospitals NHS Trust, and Oxford University Hospitals NHS Foundation

When is study starting and how long is it expected to run for?
August 2015 to July 2018

Who is funding the study?
Department of Health and the Wellcome Trust through the Health Innovation Challenge Fund.

Who is the main contact?
Mrs Verity Westgate

Trial website

Contact information



Primary contact

Mrs Verity Westgate


Contact details

Kadoorie Centre for Critical Care Research & Education
Level 3
John Radcliffe Hospital
Headley Way
United Kingdom



Additional contact

Dr Peter Watkinson


Contact details

Department of Clinical Neurosciences
University of Oxford
United Kingdom

Additional identifiers

EudraCT number number

Protocol/serial number


Study information

Scientific title

Hospital Alerting Via Electronic Noticeboard (HAVEN)



Study hypothesis

Production of a hospital-wide IT system that enables a continuous risk assessment in all hospital patients, optimised for end-user functionality.

Ethics approval

1. South Central Oxford C Research Ethics Committee, 20/06/2016, ref: 16/SC/0264
2. Confidentiality Advisory Group, 04/07/2016, ref: 16/CAG/0066

Study design

Observational mixed quantitative/qualitative study

Primary study design


Secondary study design

Mixed quantitative/qualitative methodology

Trial setting


Trial type

Not Specified

Patient information sheet

No participant information sheet available


Patients at risk of deterioration (and requiring intensive care unit admission) in hospital


Patients will be identified by the system as likely (or not) to require intensive care unit admission. The system generated results will be compared against reality and discrepancies reviewed.

Creation of the risk prediction score will be a multi-stage process involving:
1. Candidate Variable Selection
Candidate variables will be selected by a systematic review of variables associated with in-hospital deterioration; analysis of data of 5000 admissions to ICUs at the Royal Berkshire Hospital, the Churchill Hospital, Oxford and the John Radcliffe Hospital, Oxford to identify common factors to patients admitted from general wards; analysis of 127,000 patient episodes in Portsmouth Hospital to obtain information on patents who did not require ICU admission, and compare the frequency of candidate variables between those who deteriorated and those who remained stable; and a modified two-round
Delphi Process to review possible candidate variables identified in the above processes and to obtain further suggestions for variables of interest.
2. Score Creation
To create the risk prediction score, the project will utilise standard statistical approaches(including linear and non-linear regression), machine learning techniques and techniques such as Gaussian processes for trend analysis using trajectories in multi-dimensional input space. Algorithms will be developed in which correlations between the various time-series are learned automatically, allowing early detection of physiological deterioration by identifying unexpected changes in correlation or other changes in dynamics. The algorithms will include probabilistic, Bayesian methods that can cope with missing or artefactual data from one or more input time-series in a principled, robust
manner, as is required for a realistic real-time risk estimation system.
3. Score Validation
The risk prediction score will be validated by comparing observed ICU admissions with predicted ICU admissions. The goodness-of-fit of observed to expected ICU admissions will be described using the Hosmer-Lemeshow (HL) test.
4. Clinical Expert Review
The first risk prediction model will be run on new inpatient episodes. A list of false negatives (patients in whom the algorithm reported a low risk, but the patient was admitted to ICU) and false positives (patients in whom the algorithm reported a high risk, but the patient was not admitted to ICU) will be generated. Clinicials from the project team to examine these patients’ standard clinical records (EPR, paper notes, etc) to determine new factors (or combinations of factors) used by the clinical team to
determine the patients’ need for ICU.
5. Score Re-validation
The score will be re-validated following the modifications made to the risk algorithm after step iv.

Intervention type



Drug names

Primary outcome measures

1. Validated risk prediction score identifying at least 50% of deteriorating patients admitted to the ICU at least 12 hrs ahead of their admission, with a false positive rate of under 60%. This will be validated by comparing observed ICU admissions with predicted ICU admissions. The goodness-of-fit of observed to expected ICU admissions will be described using the Hosmer-Lemeshow (HL) test. This outcome will be measured during the second half of 2017
2. Data aggregation software in place capable of supplying all data fields to the algorithm
3. Working interface complete with functional underlying software and prototype graphics. A minimum success criterion would be usability scale results above average (>68%). This will be designed using Human Factors methods. This will involve process mapping and task analysis of existing workflows used to recognise deterioration of patients on the ward. Knowledge gained will be used to produce a design specification for the interface. This will then be tested formally assessing efficiency, effectiveness and user satisfaction as outlined in ISO 9421

The end of the study will be a key timepoint for overall evaluation of the outcomes from the

Secondary outcome measures

No secondary outcome measures

Overall trial start date


Overall trial end date


Reason abandoned


Participant inclusion criteria

1. Aged 16 years and over
1. Admitted to any hospital from Portsmouth Hospitals NHS Trust since 1st January 2010 or Oxford University Hospitals NHS Foundation Trust since 1st January 2014

Participant type


Age group




Target number of participants

450-500,000 patients

Participant exclusion criteria

1. Patients <16 years of age
2. Patients whose data are not entered into local electronic patient records

Recruitment start date


Recruitment end date



Countries of recruitment

United Kingdom

Trial participating centre

Oxford University Hospitals NHS Foundation Trust

Trial participating centre

Portsmouth Hospitals NHS Trust

Sponsor information


University of Oxford

Sponsor details

University Offices
Wellington Square
United Kingdom

Sponsor type




Funder type


Funder name

Department of Health/Wellcome Trust Health Innovation Challenge Fund

Alternative name(s)

Funding Body Type

Funding Body Subtype


Results and Publications

Publication and dissemination plan

1. A protocol for a systematic review to determine the variables with a statistical association with
unplanned ICU admission will be published in late 2016. This will be followed by publication of the
systematic review itself
2. A protocol for a systematic review examining usability evaluation methods used to assess
information visualisations of patient data will be published in late 2016. This will be followed by
publication of the systematic review itself
3. There will be a final report within 12 months of the end of the project
4. Further interim papers are likely and will be confirmed at a later date

Intention to publish date


Participant level data

To be made available at a later date

Results - basic reporting

Publication summary

Publication citations

Additional files

Editorial Notes