Post on 25-May-2015
Risk Prediction in England
Natasha Curry, 14 January 2010
Presentation structure
Why the interest?DH-commissioned project:– Project details– Learning from international evidence– Outputs: PARR & Combined model
Risk prediction uptake & useFuture potential and plansFurther information & plans
Why the interest?
Context:High emergency admissions and rising A&E attendanceRising numbers with LTCsUnsustainable system: need to shift from reactive to proactive care & treat outside hospitalPoor quality care for people with long term conditions: little continuity of care & regular admissionsNeed to strengthen commissioning & tailor services to the needs of population
Why the interest?
Other relevant issues:Government had pledged to put in place 3000 community matrons by 2006 to manage high risk patientsEvercare evaluation showed that the initiative had reduced emergency admissions by 1% at most
Questions arising:
1. Who are the people who will have high numbers of unplanned admissions next year?
2. How do we identify them accurately?3. What can we do to prevent them entering a
spiral of admissions?
DH-Commissioned project
> Research team:> The King’s Fund> New York University> Health Dialog
Timescale:> March 2005 – Dec 2005…
DH-Commissioned project
Three strands:
1.Literature review: what techniques are used to predict risk around the world?
2.Can risk of readmission be predicted using routine inpatient data? PARR
3.Can risk of admission be predicted using linked datasets? Combined Predictive Model
Findings from international literature
International literature revealed 3 main methods for “case-finding”:
1.Clinical knowledge2.Threshold modelling3.Predictive modelling
Non-predictive v. predictive methods
DH-Commissioned project
Three strands:
1.Literature review: what techniques are used to predict risk around the world?
2.Can risk of readmission be predicted using routine inpatient data? PARR
3.Can risk of admission be predicted using linked datasets? Combined Predictive Model
2) Prediction using inpatient data: PARR
Uses just inpatient admissions data
• PARR (2005)• PARR+ (2006)• PARR++ (2007)
Year 1 Year 2 Year 3 Year 4 Year 5
Year of admission
Year of predictionPrior utilisation
Risk score
0
100
DH-Commissioned project
Three strands:
1.Literature review: what techniques are used to predict risk around the world?
2.Can risk of readmission be predicted using routine inpatient data? PARR
3.Can risk of admission be predicted using linked datasets? Combined Predictive Model
3. Combined Predictive Model
High risk
Medium risk
Low risk
PARR
Combined predictive model
Combined Predictive Model: data
Inpatient data
Outpatient data
A&E data
GP data
Social services data Combined
Predictive Model
Risk prediction uptake
Becoming mainstream:– Use of predictive tools is one of WCC skills– Survey suggested:
• 80% of PCTs are using some form of predictive tool
• 67% of PCTs are using PARR• Very few (up to 5%) PCTs are using CPM
due to data challenges & absence of front-end
Use of outputs
Various interventions being tested:– Virtual wards– Telephone health coaching– Integration with social care
Other uses:– Identifying clinical gaps for GPs to address– Informing commissioning decisions/identifying
needIssue: What interventions/approaches are (cost) effective at preventing unplanned admission?
Future plans/potential
Update of PARR & CPM (proposal submitted to DH)Incorporation of more data (e.g. social care) to predict health outcomesPrediction of other outcomes (e.g. nursing home admission; cost)More refined targeting of model (e.g. impactability)Effective interventions
Further information
Details of the models & work on KF website: – www.kingsfund.org.uk