CASE FINDING ALGORITHM FOR PATIENTS AT RISK OF REHOSPITALISATION

1
CASE FINDING ALGORITHM FORPATIENTS AT RISK OF
RE-HOSPITALISATION
27 October, 2005
Health Dialog Data Service
NYU Center for Health and Public Service Research
2
WHAT IM GOING TO TALK ABOUT

• The need for more effective health and social
  care management of high cost patients
• The importance of an effective case finding tool
• How the Patients At Risk of Re-hospitalisation
  (PARR) case finding algorithm was developed
• What the PARR algorithm can do
• What the PARR algorithm cant do
• Next steps you might consider

3
THE NEED FOR MORE EFFECTIVEHEALTH AND SOCIAL
CARE MANAGEMENTOF HIGH COST PATIENTS
4
THE NEED TO IMPROVEHEALTH AND SOCIAL CARE
MANAGEMENTCASE 1
5
THE NEED TO IMPROVEHEALTH AND SOCIAL CARE
MANAGEMENTCASE 2
6
THE NEED TO IMPROVEHEALTH AND SOCIAL CARE
MANAGEMENTCASE 3
7
PREDICTIVE CASE FINDING PROJECT

• Phase 1 Literature review
• Phase 2 PARR case finding algorithm
• Inpatient data
• Phase 3 Combined data case finding algorithm
• Inpatient data
• AE data
• Hospital outpatient data
• GP electronic medical record data
• Community data
• Social services data
• Etc.

8
ABOUT THE IMPORTANCE OFCASE FINDING

• An effective case finding tool is likely to be
  critical
• The key to any management improvement initiative
  is matching patient needs with available
  resources
• One size does not fit all
• May invest more in interventions for very high
  risk patients
• Match intervention design to patient need
• Any new interventions are likely to have to meet
  a business case test
• The cost of the intervention offset by savings
  somewhere else
• The somewhere else is likely to be the hospital

9
ABOUT THE IMPORTANCE OFCASE FINDING
Patient Smith
Patient Jones
Patient Shah
10
OUR APPROACHPATIENT AT RISK OF
RE-HOSPITALISATION (PARR)CASE FINDING ALGORITHM
11
PARR ALGORITHMSGENERAL APPROACH

• Developed using 5 years of HES data
• Designed to be used by PCTs, SHAs, GP Groups with
  ClearNet Admitted Patient Care (APC) data
• Can be used either
• In real time (while patients are hospitalised)
• With archived data only (monthly or annually)

12
PARR ALGORITHMSGENERAL APPROACH

• Look for a recent hospitalisation
• In real time (while the patient is hospitalised)
• Or on a monthly/annual basis (reviewing recent
  discharges)
• Use information in hospital discharge records to
  predict patients at high risk for
  re-hospitalisation
• Logistic regression techniques
• An algorithm produces a Risk Score of 0-100 for
  each patient
• Two methods
• Narrow (PARR1) Focus on emergency admissions
  for reference conditions
• Potentially preventable/avoidable
• Often lead to re-hospitalisation
• Broad (PARR2) Look at all emergency admissions

13
PARR ALGORITHMSGENERAL APPROACH FOR
DEVELOPMENTOF PARR ALGORITHM REAL TIME
Admission
Year 4
Year 5
Year 3
Year 2
Year 1
14
PARR ALGORITHMSGENERAL APPROACH FOR
DEVELOPMENTOF PARR ALGORITHM REAL TIME
Examine utilisation for prior 3 years
Admission
Year 4
Year 5
Year 3
Year 2
Year 1
15
PARR ALGORITHMSGENERAL APPROACH FOR
DEVELOPMENTOF PARR ALGORITHM REAL TIME
Examine utilisation for prior 3 years
Predict adm next 12 months
Admission
Year 4
Year 5
Year 3
Year 2
Year 1
16
PARR ALGORITHMSTYPES OF VARIABLES IN MODEL

• Prior utilisation
• Admissions (emergency and elective examined
  separately)
• Number
• Time intervals (recentness)
• Day cases/attendances
• Type and number of different day case specialty
  types
• Diagnostic information from current/prior
  utilisation
• Chronic disease
• Other conditions with high rates of subsequent
  admission
• Multiple conditions/hierarchical DCG grouping
• Patient characteristics Age, gender, ethnicity
• Reference condition re-hospitalisation rates at
  hospital of current admission
• Contextual information about area of residence
• Age/sex adjusted admission rates for reference
  conditions
• Race/ethnicity
• Deprivation index

17
PARR ALGORITHMREFERENCE CONDITIONS

• Conditions potentially responsive to more
  effective care management with high rates of
  re-hospitalisation
• CHF
• COPD
• Diabetes
• Coronary artery disease
• Sickle cell
• Asthma
• Chronic liver disorders
• Chronic pancreatic disease
• Cystic fibrosis
• URI
• Etc, etc, etc

18
PARR ALGORITHMSRESULTS AND CHARACTERISTICSOF
FLAGGED PATIENTS
19
PARR ALGORITHMSRESULTS FROM PARR ALGORITHM
PARR1 REFERENCE CONDITIONS
PARR2 ALL EMERGENCY PATIENTS
Note Data are from use of real time" methods
20
PARR ALGORITHMSNUMBER OF RE-ADMITTED PATIENTS
IDENTIFIED ANNUALLY FOR A TYPICAL PCT REAL
TIME METHOD
PARR2
PARR1
Number of Re-Admitted Patients Identified
Risk Score Threshold
21
PARR2 ALGORITHMNUMBER OF RE-ADMITTED PATIENTS
IDENTIFIED ANNUALLY FOR A TYPICAL PCT
(PARR2 Algorithm)
REAL TIME
MONTHLY
ANNUAL
Number of Re-Admitted Patients Identified
Risk Score Threshold
22
PARR2 ALGORITHMCHARACTERISTICS OF PATIENTSAT
HIGH RISK FOR RE-ADMISSION
Note Data are from PARR2 Algorithm "Real
time" method
23
PARR2 ALGORITHMCHARACTERISTICS OF PATIENTSAT
HIGH RISK FOR RE-ADMISSION
Note Data are from PARR2 Algorithm "Real
time" method
24
PARR2 ALGORITHMCHARACTERISTICS OF PATIENTSAT
HIGH RISK FOR RE-ADMISSION
Note Data are from PARR2 Algorithm "Real
time" method
25
PARR2 ALGORITHMCHARACTERISTICS OF PATIENTSAT
HIGH RISK FOR RE-ADMISSION
Note Data are from PARR2 Algorithm "Real
time" method
26
PARR2 ALGORITHMTOP 25 EMERGENCY
RE-HOSPITALISATIONSFOR FLAGGED PATIENTS Risk
Score 50
Note Data are from PARR2 Algorithm "Real
time" method
27
PARR ALGORITHMSBREAK EVEN FOR TYPICAL PCT
COSTS/SAVINGS(INTERVENTION COST 500/PATIENT)
Note Data are from use of real time" methods
28
PARR ALGORITHMSBREAK EVEN FOR TYPICAL PCT
COSTS/SAVINGS(15 REDUCTION IN RE-ADMISSIONS --
COST 500/PATIENT)
Net Costs/Savings (000s)
PARR1
PARR2
Risk Score Threshold
Note Data are from use of real time" methods
29
PARR2 ALGORITHMBREAK EVEN FOR TYPICAL PCT
COSTS/SAVINGS(INTERVENTION COST 500/PATIENT)
30
PARR2 ALGORITHMBREAK EVEN FOR TYPICAL PCT
COSTS/SAVINGS(REAL TIME METHOD - COST
500/PATIENT)
Net Costs/Savings (000s)
10 Reduction
15 Reduction
20 Reduction
Risk Score Threshold
Note Data are from PARR2 Algorithm "Real
time" method
31
PARR2 ALGORITHMBREAK EVEN FOR TYPICAL PCT
COSTS/SAVINGS(REAL TIME METHOD 15 REDUCTION
IN RE-ADMISSIONS)
Net Costs/Savings (000s)
500 Per Patient
750 Per Patient
1,000 Per Patient
Risk Score Threshold
Note Data are from PARR2 Algorithm "Real
time" method
32
PARR ALGORITHMSEXPECTED REDUCTION IN ANNUAL
EMERGENCY DAYS FORTYPICAL PCT (REAL TIME
METHOD 15 REDUCTION IN ADMS)
PARR2
PARR1
Reduction in Total Emergency Days for PCT
Risk Score Threshold
33
PARR2 ALGORITHMEXPECTED REDUCTION IN ANNUAL
EMERGENCY DAYS FORTYPICAL PCT (REAL TIME
METHOD)
10 Reduction
15 Reduction
20 Reduction
Reduction in Total Emergency Days for PCT
Risk Score Threshold
Note Data are from PARR2 Algorithm "Real
time" method
34
PARR ALGORITHMSPARR1 or PARR2?
PARR1 ALGORITHM
PARR2 ALGORITHM

• Advantages
• Breaks even at a lower risk score
• Patients identified may be more amenable to
  intervention(?)
• Disadvantages
• Finds fewer patients
• For real time method, requires admission DX

• Advantages
• Finds more patients
• Does not require admission DX for real time
  method
• Disadvantages
• Breaks even at a higher risk score
• Patients identified may be less amenable to
  intervention(?)

35
PARR ALGORITHMSPARR1 or PARR2?
PARR1 ALGORITHM
PARR2 ALGORITHM

• Advantages
• Breaks even at a lower risk score
• Patients identified may be more amenable to
  intervention(?)
• Disadvantages
• Finds fewer patients
• For real time method, requires admission DX

• Advantages
• Finds more patients
• Does not require admission DX for real time
  method
• Disadvantages
• Breaks even at a higher risk score
• Patients identified may be less amenable to
  intervention(?)

36
PARR ALGORITHMSREAL TIME or ARCHIVAL?
REAL TIME METHOD
ARCHIVAL METHODS

• Advantages
• Finds more patients
• Permits discharge planning as component of
  intervention
• Disadvantages
• Must be updated daily
• For PARR1 algorithm, requires admission DX

• Advantages
• Easier (no daily updates)
• Comparable business case (for monthly update
  model)
• Disadvantages
• Finds fewer patients
• Does not permit discharge planning as component
  of intervention

37
PARR ALGORITHMSREAL TIME or ARCHIVAL?
REAL TIME METHOD
ARCHIVAL METHODS

• Advantages
• Finds more patients
• Permits discharge planning as component of
  intervention
• Disadvantages
• Must be updated daily
• For PARR1 algorithm, requires admission DX

• Advantages
• Easier (no daily updates)
• Comparable business case (for monthly update
  model)
• Disadvantages
• Finds fewer patients
• Does not permit discharge planning as component
  of intervention

38
PARR ALGORITHMWHAT IS REQUIRED FOR PCTs, SHAs,
orGP PRACTICES TO RUN THE ALGORITHM

• Three years of archived ClearNet Admitted Patient
  Care (APC) data with patient identifiers for area
  residents
• For Real Time method - Daily updated list of
  admitted patients with identifiers
• Access software and IT staff with moderate level
  of Access knowledge

39
WHAT THE PARR ALGORITHM CANT DOAND POSSIBLE
NEXT STEPS
40
AMONG THE THINGSTHE PARR ALGORITHM/HES
DATACANT DO

• By itself reduce PCT emergency patient days by 5
• Identify emerging risks (patients who have not
  been admitted previously) lt Perhaps Phase 3 of
  the project
• Provide information on the factors that
  contributed to prior admissions
• Provide information on social context/needs of
  patients flagged by the algorithm
• Provide assurance that it is really possible to
  reduce future admissions
• Tell you everything you need to know to design an
  effective intervention
• Tell you what an intervention would cost

41
PARR ALGORITHMSOME RECOMMENDED NEXT STEPSFOR
PARR USERS

• Run the algorithm in real time to identify
  30-40 patients with
• high risk scores
• Interview these patients and their providers to
  learn
• Circumstances that led to admission
• Factors that would help reduce future admissions
• Design/tweak an intervention based on this
  information (letting a few flowers bloom)
• Implement intervention(s) in a manner to learn as
  much as possible
• Randomise patients/practices/hospitals into
  intervention and non-intervention and track
  outcomes
• or
• Track subsequent utilisation and compare to
  historical controls/expected use rates