Combined Predictive Model
See also: Predictive Risk Project
As part of our work on managing long-term conditions, the King's Fund, together with New York University and Health Dialog, has developed an algorithm that links inpatient data with other routine data on utilisation of care in order to predict future risk of emergency admission. The 'Combined Predictive Model' final report and technical documentation, which is intended for use at PCT level, is available to download freely from this website.
The Combined Predictive Model integrates accident and emergency, inpatient, outpatient and GP data sources to predict risk of emergency admission to hospital across an entire patient population. It builds on the work undertaken to develop the Patients at Risk of Re-hospitalisation (PARR) models but, because it uses additional primary and secondary care data sources, the combined model is able to identify individuals along the entire continuum of risk as opposed to just those who have already experienced a recent hospital admission.
Additionally, the Combined Predictive Model identifies some patients at very high risk of future admission who are not identified by PARR. It facilitates early identification of people before their conditions deteriorate, which in turn allows differing levels of intervention intensity to be matched to different segments of overall risk. The fact that the Combined Predictive Model is built upon an integrated primary and secondary care data set allows for the development of clinical profiles which provide powerful insights into the types of patients being identified in different risk segments. This facilitates the design of highly targeted interventions proportional to risk.
The model has been developed and tested against datasets from two PCTs. Full results from this process are available in the form of a downloadable report.
Downloading the Combined Predictive Model Documentation
The following downloadable documents are freely available from this website:
-
Combined Predictive Model Final Report (PDF, 462KB)
This document gives more information about the Combined Predictive Model and the results it produced in the development PCTs. -
Combined Predictive Model Final Report and Technical Documentation (PDF, 845KB)
This document includes the Final Report as well as technical guidance on implementing the model in your PCT. Implementing the model locally requires expertise about data sources, a programming application (e.g. SAS or SQL), and expertise in data management and programming. PCTs that need additional technical expertise in implementing the model may consider soliciting consulting help from health care analytics firms, including Health Dialog, or other companies. -
eMedia appendices (available from the King's Fund website )
These contain the look-up files referenced in the technical documentation which are required to build the model. This is a large file and may take time to download. Please read the technical documentation first and download the eMedia appendices only if you are planning to implement the model within your PCT. -
PIAG advice on patient confidentiality (Word, 1.3MB)
Produced by the Patient Information Advisory Group
Questions about the combined model and how it was developed, as well as basic questions about the types of skills and applications needed to implement the model locally, should be submitted to the King's Fund, not to NHS Networks..
Case Study
A number of PCTs across the country are using PARR, and others have been involved in the development of the Combined Predictive Model. One notable example is Croydon PCT, which recently won several HSJ Awards in recognition of its involvement in this innovative work.
Croydon has been using the Combined Predictive Model in its virtual wards project whereby people identified by the model as having a very high risk of future hospitalisation are put on a 'virtual ward'. These people are provided with preventive care in their own homes by a multi-disciplinary team who use the systems, timetable and staffing of a hospital ward but without the physical building.
Admission to the virtual ward is determined solely by the output of the combined predictive model. Patients' risk scores are monitored over time and can be used to prompt the virtual ward staff to discharge patients when appropriate - and offer admission to a patient at higher risk .
-
More information on the Croydon virtual wards (PDF, 79KB)