Predictive Models for Health and Social Care: A feasability study

3,576 views

Published on

Published in: Health & Medicine
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
3,576
On SlideShare
0
From Embeds
0
Number of Embeds
2,386
Actions
Shares
0
Downloads
32
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

Predictive Models for Health and Social Care: A feasability study

  1. 1. Predictive Models for Health andSocial Care: A Feasibility StudyAuthors: Bardsley M, Billings J, Dixon J,Georghiou T, Lewis GH, Steventon A (2011)‘Predicting who will use intensive social care:case finding tools based on linked health andsocial care data’, Age and Ageing 40(2): 265-270February 2011 © Nuffield Trust
  2. 2. BackgroundKey pointsEvidence that admission to a nursing home can be delayed oravoided by means of preventative ‘upstream’ interventionsIncreasingly, public services will have to become more proactive inidentifying and managing risk in older people, in order to mitigatethem as much as possibleIt will be important to identify and support not just those at highestrisk of these costs, but also those currently at lower risk who mightbecome higher risk in futureCurrent methods of assessing risk largely rely on face to faceassessments. In health care this approach has been shown to beless accurate at predicting predicting future events (hospitalisation)compared to statistical models.From: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  3. 3. Predictive modelling• BMJ in paper* in 2002 showed Kaiser Permanente in California seemed to provide higher-quality healthcare than the NHS at a lower cost. Kaiser identify high risk people in their population and manage them intensively to avoid admissions• Modelling aims to identify people at risk of high costs in future• Use pseudonymous, person-level data• In health sector a number of predictive models are available e.g. PARR++ and the combined model.*Getting more for their dollar: a comparison of the NHS with Californias Kaiser Permanente BMJ 2002;324:135-143From: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  4. 4. Predictive Models Identify who will bewhere on next year’s Kaiser PyramidFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  5. 5. Patterns in routine data to identify high-risk people next yearFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  6. 6. Patterns in routine data to identify high-risk peopleFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  7. 7. Patterns in routine data to identify high-risk peopleFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  8. 8. Distribution of Future Utilisation is ExponentialFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  9. 9. Can we predict costly social careevents in the same way?• Do the data exist in local systems?• Can we extract individual, person-level data?• Can we link a wide range of health and social care data at the person level?• Are the data accurate and complete enough to use?• Can we build a valid statistical model?From: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  10. 10. Feasibility Project• Funded by Department of Health Care Services Efficiency Delivery (CSED) programme over 18 months• Worked with 5 sites to extract data sets• Extracted person-level health & social care data sets• Linked data from GP records (2 sites); GP register (all sites); Hospital (all sites); Social Care (all sites)• Undertook exploratory analyses and ran a range of models• Tested the impacts of different data sets on the modelsFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  11. 11. Using health and social care data topredict health and social care useFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  12. 12. Using routine data• Less labour intensive so they can stratify the population systematically and repeatedly• Avoid “non-response bias”• Can identify people with lower, emerging, risk• Important issues of confidentiality and consent to consider• Linking data sources at individual level across health and social care is problematic where there is no NHS number in social care• The tools are never 100% accurate• Data may be missing from routine databases on certain groupsFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  13. 13. Predictive factors – examplesFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  14. 14. Information flowsFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  15. 15. Data collected• From four ‘sites’ (~ PCT areas)• Total seven organisations: 3 PCTs, 1 Care trust, 3 LAs• Total 1.4M pop (range 100,000-700,000)From: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  16. 16. Data linkage - approachFirst instance: NHS number (encrypted) from LA In absence of NHS number: • Central ‘batch tracing’? • Shared PCT/LA databases? Ultimately: •construction of ‘alternative IDs’ 97% of individuals in one site (population ~400,000) were found to have unique ‘alternative ID’.From: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  17. 17. Data linkageGroups of people in social care data – how many are we able tomatch to GP register list (of ages 75+)?Varies, but better for those with > service useFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  18. 18. Data linkageSocial & secondary care interfaceFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  19. 19. Predictive value, sensitivity and specificityof the model incorporating a £5,000 thresholdFrom: Predictive Models for Health and Social Care: A Feasibility StudyPredicting social care costs: feasibility study © Nuffield Trust
  20. 20. Information on Social Care NeedsFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  21. 21. Simplifying and sorting Accommodation Adaptation Adult Placement Advice Given Carer Support Counselling DA1 Day Care Direct Payments Domiciliary Employment/Training Equipment Fostering Group Activity Holiday Payment Housing Advice Given ILS Independent Living Individual Informal Projects Information & Advice Lifeline 400 Lifeline 400 + Smoke Alarm Mobility Training (DAY) Mobility Training (NIGHT) Nursing Nursing (Block Contract) Nursing Block Beds Old Codes OT Advice Given OT Rehab Sessions Other Other Carer Support Services Outreach Permanent Placements with Parents Professional Support R.O. Allowance Referral to Carers Contact Line Referral to Carers Information Service Referral to Crossroads Registration Rehabilitation Residential Residential Block Beds Respite Respite (Carer) Secure Accommodation Sheltered Social Development Special Sheltered Specialist Assessment and/or Treatment Specialist Communication Equipment Specific Approval Support Supported Living Talking Book Machine Time Allocation Vouchers Training Transport WasherFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  22. 22. Individual historiesFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  23. 23. Transitions in care (75+ in one site)From: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  24. 24. Individual health and social care event timelineover a three-year periodFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  25. 25. Predictive ModellingAttempting to predict:• For over 75s a) Admission to care home (or receipt of high intensity home care) or £5,000 increase in social care costs in one year b) Where person had no ‘significant’ costs in prior two yearsFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  26. 26. Original models predicting change at £5k thresholdFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  27. 27. DiagramFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  28. 28. Distillation flaskFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  29. 29. Trade off between PPV (blue line) and sensitivity(red line) according to different risk cut-offsFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  30. 30. Which variables are important in pooled £1k model?From: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  31. 31. Models using lower £1k thresholdsFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  32. 32. Impact of adding new datasetsFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  33. 33. Iteration with Sites - Application• Several sites said they would like to run the Combined Model and the social care model “side-by-side”• Theographs thought to be very useful• Differences between sites: some sites preferred small numbers of clients others wanted large numbers for mail-shots• Regarded concerns over invasion of privacy as “a non-issue if the wording is right”• Keen to use predictive models for existing or planned re-ablement services and multidisciplinary teamsFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  34. 34. Personal observations on social care dataWe could access user level social care data but...• Different systems• Issues of standardisation /coding etc and the absence of standard, structured coding schemes• Some very detailed information collected but can be difficult to use in models• We found it harder to obtain information on descriptions of needs• Local concerns about data quality• Note the history of comparative benchmarkingFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  35. 35. From: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  36. 36. Headlines• We have created linked health/social care data sets in five sites (overall populations about 2 million people aged 75+)• We modelled intensive social care: • Models had satisfactory PPV but low sensitivity at a risk score threshold of 50 (although can be traded-off against each other) • Mainly social care driven • Though accuracy worse than PARR model, accuracy was comparable with SPARRA and the combined model • Discussion with sites – they were less concerned with model performance than we were• We tried dozens of ways to improve the basic model (5K)• We modelled change in social care costs with lower thresholds: 1K models are much better but usefulness may be diminishedFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  37. 37. And the data have so much more to offer.....• Cost modelling• Iso-resource classification (individual budgets)• Trajectories and transitions of care• Evaluation (re-ablement)• Information for - Commissioners and managers - Professionals - Service users and carersFrom: Predictive Models for Health and Social Care: A Feasibility Study © Nuffield Trust
  38. 38. www.nuffieldtrust.org.uk Sign-up for our newsletter www.nuffieldtrust.org.uk/newsletter Follow us on Twitter (http://twitter.com/NuffieldTrust)February 2011for Health and Social Care: A Feasibility StudyFrom: Predictive Models © Nuffield Trust

×