500 likes | 663 Views
SPARRA: predicting risk of emergency admission among older people. Steve Kendrick Delivering for Health Information Programme ISD Scotland www.isd.scotland.org/dhip NHS GG&C Public Health Friday Seminar Dalian House, 1 st December 2006.
E N D
SPARRA: predicting risk of emergency admission among older people Steve Kendrick Delivering for Health Information Programme ISD Scotland www.isd.scotland.org/dhip NHS GG&C Public Health Friday Seminar Dalian House, 1st December 2006
Providing information to support ‘Kerr’ and “Delivering for Health” is a key priority for ISD Scotland. The Delivering for Health Information Programme supports a specific focus of “Delivering for Health”.
Long-term conditions + interface with unscheduled care Level 4 Level 3 Individuals with complex needs: case management (3-5%) Emergency admissions Level 2 High risk: disease Management (15-20%) Level 1 Kerr Unscheduled Care Levels Lower risk: supported self-care (70-80%) Interventions Outcomes Public health; health Improvement; health education
Long-term conditions + interface with unscheduled care Level 4 DfHIP Level 3 Individuals with complex needs: case management (3-5%) Emergency admissions Level 2 High risk: disease Management (15-20%) Level 1 Kerr Unscheduled Care Levels Lower risk: supported self-care (70-80%) Interventions Outcomes Public health; health Improvement; health education
DfHIP Priorities around ‘the top of pyramid’ Economics: yield curves SPARRA High risk patients ? VHIUs Very high intensity users Care homes End of life costs End of life care
Economics: yield curves SPARRA High risk patients Primary care SPARRA ? Top 5% bed days Care homes Emergency admissions: comparative trends End of life costs End of life care GP emergency admission rates Information for CHPs LTCs/risk stratification
SPARRA stands for… Scottish Patients At Risk of Readmission and Admission
Purpose of SPARRA • Identify those people at greatest risk of emergency inpatient admission • Current cohort: people aged 65 and over with at least one emergency admission in the previous three years
Steps in implementing model • Develop predictive model (logistic regression) based on patients for whom we do know the outcome – historic data • Identify what determines the likelihood of future emergency admission • Apply model to patients for whom we don’t know the outcome • Calculate individual risks • Feed back results to front line
Developing the predictive model Cohort includes all aged 65+ with an emergency admission in previous three years (around 25% of 65+ pop.) 1st January 2004 Time Period 2002 2004 2003 2001 Outcome year Predictor variables
The shoulders upon which we stand • Substantial American literature see e.g. King’s Fund literature review • King’s Fund: John Billings • NHS Tayside/University of Dundee model – Peter Donnan • Highland; Lanarkshire; Ayrshire and Arran
Our approach • No ‘black boxes’ • Transparent – understand what’s under the bonnet • Collaborative • Evolutionary
Independent variables • Number of previous emergency, elective, day case admissions; total bed days • Time since most recent emergency admission • Age/gender • Deprivation • Most recent admission diagnosis, number of different diagnosis groups. • NHS Board
Results: major factors emerging as predictors • Number of previous emergency admissions • Time since most recent admission • Age • Interaction between age and previous emergency admissions • Deprivation • Number of diagnoses • Most recent diagnosis – especially COPD • NB. NHS Board not significant
Example: individual with very high predicted probability of admission • Predicted probability of admission 86% • Male aged 65 to 69 • Less than one month since most recent admission • 6+ previous emergency admissions • Glasgow – most deprived decile • Most recent admission diagnosis: COPD • Outcome: admitted as emergency
Applying the predictive model Based on previous 3 years of hospital admissions 1st April 2006 Time Period to April 2006- March 2007 March 2006 April2003 Outcome year Predictor variables
How well does the model perform • Reasonable area under the ROC. 0.69 compared with c0.8 when e.g. primary care variables included (c.f 0.685 King’s Fund hospital-based model) • Likely to be identifying the great bulk of the high risk patients out there in the community c 75-90%
Applying the predictive model Now 1st April 2006 Time Period April 2006 to March 2007 March 2006 April 2003 to Outcome year Predictor variables
Usually 6 months until SMR01 data complete enough: how much of an issue? • What might have happened in 6 months • Patient may have • died – must check via local systems • been admitted – increase in future risk • not been admitted – decline in future risk It is an issue, not a showstopper – but not satisfactory
Forms of feedback • Identifiable details of high-risk patients • fed back on CD on receipt of confidentiality form • values of model variables as well as ID and probabilities • Local distributions of risk levels • how many people at all levels of risk • By Board, CHP, practice
The role of SPARRA? Original conception – fairly narrow, mechanical SPARRA identifies a pool of high-risk patients Further local assessment identifies those for whom e.g. case management is appropriate Full stop
Emerging functions: SPARRA as a focus for integration • “international research suggests that integration is most needed and works best when it focuses on a specifiable group of people with complex needs, and where the system is clear and readily understood by service users (and preferably designed with them as full partners)” Integrated Care: A Guide, Integrated Care Network (cited by David Colin-Thome)
Emerging functions: SPARRA as a seed • Local teams often use SPARRA in combination with other sources of local information (e.g. GP registers) • SPARRA may become just one component of a dynamic, multi-source locally owned register of vulnerable people • cf Exeter. Wide range of sources for up-to-date list which ‘keeps tabs on’ vulnerable people. No high tech/IT. Based on commitment and case management
Further development of model • Move to incorporate real-time data: via SystemWatch • Incorporating primary care data. Needs to be led locally • Relation with social care data c.f. Highland – needs to be done locally. • Economic aspects – what could be the pay-off? • Evaluation – SPARRA to help evaluate impact of models of anticipatory care
Current take up of SPARRA • Around 4 Boards motoring • 6-10 Boards/CHPs – very keen – have received data (i.e. around half of CHPs have data either directly or indirectly) • Most of rest – in discussion • A very few – still to start a conversation
The response to SPARRA output • Starting to get feedback: the results seem to be making reasonable sense • Major frustration: based on out-of-date data • This is primary use of healthcare information: helping determine how to deliver the best care to real people • Only the beginning