Ian Blunt: PARR-30: a predictive model for readmission within 30 days


Published on

Published in: Health & Medicine
  • Be the first to comment

  • Be the first to like this

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

Ian Blunt: PARR-30: a predictive model for readmission within 30 days

  1. 1. PARR-30: a predictive model for readmission within 30 days Presenter: Ian Blunt19 June 2012 © Nuffield Trust
  2. 2. Development of a predictive model for readmission within 30days of discharge (PARR-30)Model developed by Billings, Blunt, Steventon, Georghiou, Lewis andBardsley• Motivation• Development• Model performance• Testing in hospitals• Conclusions © Nuffield Trust
  3. 3. Why predict readmissions within 30 days?• Readmissions are costly, suboptimal health care - costs to the NHS estimated at £1.6 billion each year• DH guidance for the NHS proposes commissioners do not pay provider hospitals for emergency readmission within 30 days of a selected index elective admission• Rate of readmissions will also play an important part in monitoring health system performance, as one of the new English Public Health “outcome indicators” © Nuffield Trust
  4. 4. Not first to try this, but…Number of international 30 day models Model From C statisticPredictive tools built in one setting Halfon et al 2008 0.67may not necessarily be accurate whenused in other health care settings Silverstein et al 0.65 2008Used hospital episode statistics (HES)data to develop model for NHS in Van Walraven et al 0.68England 2010 Howell et al 2009 0.65Make PARR-30 freely available foruse across the NHS in England And others…(possibly tablet/smartphone app) See Kansagara et al JAMA 2011 © Nuffield Trust
  5. 5. How is PARR30 different from PARR++?• Readmission in next 30 days vs next 365 days• Tools operate in different ways, trigger different responses• Next year – longer for clinicians and care managers/coordinators to contact and engage with high-risk patients, effect behavioural change• 30 days – highest likelihood of an unplanned admission, focussing their discharge planning efforts and post-discharge support for high-risk patientModel Timescale Run by Input Data Data lag variables sourcePARR++ 12 months PCT ~250 SUS ~ 3 monthsPARR30 30 days Acute 17 PAS/notes None• Aim for speed of low-variable models with accuracy of PARR © Nuffield Trust
  6. 6. How is PARR30 different from PARR++? PARR30 Hospital provides SUS Patient nears dischargePCT runs PARR++ Risk score calculated on ward Patients selected for intervention (via GP) Any extra intervention put in discharge plan Predicts readmission in next year – Predicts readmission in 30 days© Nuffield Trust PPV 65% – PPV ???%
  7. 7. Model developmentDeveloped using 10% sample HESfrom April 2006 to May 2009Index discharges in FY 2008/09Readmissions within 30 daysreflected 2011-12 operatingframeworkLogistic regressions identify variablesthat contributed most to predictionsValidated with split sample © Nuffield Trust
  8. 8. Model development Hospital of current admission Patient age Deprivation (via post code) History of emergency admissions: Current? Last 30 days? Past year?History in the prior two years of eleven major health conditions drawn from the Charlson co-morbidity index © Nuffield Trust
  9. 9. ResultsThe performance of the modelwas respectable, with a positivepredictive value (PPV) of59.2% and area under the ROCcurve (“c-statistic”) of 0.70.For the higher-risk patients (riskscore > 50%), readmission ratesranged from 47.7% up to 88.7%.However, these patients onlyrepresented a small share(1.1%) of all patients analysed. Receiver Operating Characteristic Curve (ROC) for the bootstrapped central estimate (red line) and 95% confidence Intervals (shaded area) © Nuffield Trust
  10. 10. Results £3,000Predictive modelling only as effectiveas the intervention it is used to £2,500trigger. Providers need to know Mean cost of readmissionpotential costs of readmission to £2,000build business case for intervention £1,500For patients risk score > 50%, mean £1,000readmission cost per patient was £500£1,088. Assuming that anintervention can reduce the number £0of readmissions by 10% for this Risk scoregroup, £109 per patient could be Mean of cost readmission (readmitted patients only)spent at breakeven Mean of cost readmission (all patients) © Nuffield Trust
  11. 11. Testing PARR-30 in hospitalsTesting: Royal Berkshire Hospital using spreadsheet version of tool on wards:• Is the tool easy to use? • Completed by junior doctors• Bedside info vs admin systems? • Test tool stored its output• Does ward PPV reflect national? • Later reconciled with admin systems for analysisChelsea & Westminster Hospitalrunning tool direct from their datawarehouse: Applied on four care of the elderly wards in Feb/March 2012• Proved it can be done easily• Looking into PPV and clinical engagement © Nuffield Trust
  12. 12. Testing PARR-30 in hospitalsResults from using spreadsheet on the wards:• Tool was used 88 times • Low risk scores – max 39%• Median time to complete 1m 41s • Em admit in last 30 days diff 10%• Median patient age was 86, • Em admits last year diff 20%, ±1,2 mostly emergency admissions • Even split whether tool or system• Average 1.3 co-morbidities, max 4 has more• 10 patients had emergency • 14% where system has diagnosis readmission within 30 days not ticked as co-morbidity © Nuffield Trust
  13. 13. ConclusionsBuilt a predictive model using a limited set of variables that weregenerated from hospital episode statisticsVariables easily available from patients’ notes or from the patientsthemselves – can calculate from spreadsheet or in PASThe performance of the model was respectable - highest risk patientshad a 88.7% chance of hospital readmission within 30 days – but highrisk patients relatively rareCost data suggests interventions need to be lower-cost to break evenEasily used on wards in trials - less than 2 minutes per applicationSome differences in data on ward, but not huge © Nuffield Trust
  14. 14. www.nuffieldtrust.org.uk Sign-up for our newsletter www.nuffieldtrust.org.uk/newsletter/login.aspx Follow us on Twitter (http://twitter.com/NuffieldTrust)19 June 2012 © Nuffield Trust