Careintensive neighbourhoods

1
Care-intensive neighbourhoods

• A comparison of geodemographic systems for
  neighbourhood segmentation of hospital admission
  data
• Jakob Petersen
• Knowledge transfer partnership
• Southwark Primary Care Trust Phil Atkinson
• Geography, UCL Paul Longley Pablo Mateos
• External funding ESRC DTI

2
Outline

• NHS the emerging healthcare market
• Long-term diseases
• Geodemographics
• Segmentation metrics for comparison
• Results
• GP Hospital referral patterns
• What next?

3
NHS the emerging healthcare market

• 1911 Employer contributed health insurance
• 1948 NHS is founded. Universal, comprehensive,
  free at the point of delivery
• Publicly owned and funded by tax
• Integrated no billing internally
• 1980 Transition to a free healthcare market.
• Privatisation of non-clinical tasks like
  cleaning
• Internal market hospital trusts to make own
  income by selling services to health
  authorities. From 1990-1994 245 hospitals had to
  close because they were not profitable.
• 1992 Private Finance Initiative (PFI).
  2000-2005 42 of hospitals built with private
  funds.
• 2000 The New NHS Plan

Talbot-Smith Pollock 2006, Pollock et al. 2007
4
NHS the emerging healthcare market
Talbot-Smith Pollock 2006, Pollock et al. 2007
5
NHS the emerging healthcare market

• The Health Care Market
• Commodity health care services (disease
  accidents?)
• Purchaser PCT, GP
• Seller public or private providers
• Regulator independent MONITOR
• Price fixing the NHS tariff (set by Department
  of Health)
• Market research
• (profitable) local health care needs
• Areas for cost reduction

6
Health care market

• Hospitals
• 19 of contacts
• 58 of costs
• Long-term diseases
• 5-10 of patients use 55 of hospital bed days

Source of figures Talbot-Smith Pollock 2006
7
Long-term diseases

• affects 17 million in UK
• Arthritis 8.5 m
• Asthma 3.4 m 1.5 m children
• Back pain 40 of adults 6 chronically
• Chronic Obstructive Pulmonary Disease1 m
• Coronary Heart Disease (CHD) 2.68 m
• Diabetes Mellitus 1.5 m
• Epilepsy 420,000
• Mental illness 16.4 of adults

DH 2004, 2005 Meldum et al. 2005 Petersen et
al. 2004 Singleton et al. 2001
8
Long-term diseases

• Locating services closer to home
• Avoid hospitalisation
• Improve patients experience
• Community care services
• Community matron to devise individual case
  management plans
• Home visits from specialist nurses or health
  visitors
• Specialist clinics
• Primary care services
• GPs can be paid for taking on patients with
  long-term care needs (QOF)

9
Long-term disease indicators
10
Pearson correlations Log(chronic admission
rates) OAC variables
Correlating variables Age 65 Divorced Single
person hh Single pensioner1 hh Rent public All
flats Longterm ill SIR Provide unpaid
care Unemployed
11
Geodemographics

• Hypotheses
• Deprivation indices are more appropriate than
  geodemographics for explaining variation in
  disease patterns?
• The finer the geographical scale, the better?
  postcode gt output area gt super output area gt ..
• Bespoke classifications are better than general
  classifications?

12
Disease counts and rates

• Ratei diseasedxyz / at-risk populationxyz
• i area
• xyz sex, age group, ethnicity, occupation,

13
Segmentation metrics
14
Segmentation metrics

• Gini
• half the relative mean difference between all
  pairs of observations
• Quartile range ratio (p75/p25)
• GE(2)

15
Contestants

• Postcode
• Mosaic UK Type
• Acorn Type
• Health Acorn Type
• KRON50 rank based on total long-term admissions
  (HES)
• Output area
• OAC subgroup
• LOAC group
• Super output area
• IMD 50 ranked segments
• HESK bespoke classification (HES Long-term
  disease groups)

16
(No Transcript)
17
(No Transcript)
18
(No Transcript)
19
GP hospital referral rates

• Requirements
• Adjusting for sex, age, (ethnicity)
• New adjusting for geodemographics
• Denominators for robust risk estimates

20
Open Geodemographics
21
(No Transcript)
22
LOAC Group Segmentation
23
Risk maps
24
GP referrals

• Age, Sex, and Geodemographic standardisation

25
GP referrals

• Age, Sex
• Age, Sex, Geodemographics

26
Co-morbidity

• Transactional data
• Apriori algorithm

27
MALE individual
ANGINA
0.60
0.72
0.35
0.41
0.16
CHEST PAIN
ACUTE MYO
0.23
28
FEMALE individual
ANGINA
0.46
0.59
0.24
0.45
0.09
CHEST PAIN
ACUTE MYO
0.21
29
Output Area (300 pop.)
ARTHROSES
0.92
0.93
CHEST PAIN
BACK PAIN
ANGINA
0.92
0.92
0.93
0.92
STROKE
30
Hypotheses

• Deprivation indices are more appropriate than
  geodemographics for explaining variation in
  disease patterns?
• No. Uni-directional order of IMD makes it less
  sensitive to neighbourhood differences in
  diseases
• The finer the geographical scale, the better?
  postcode gt output area gt super output area gt ..
  
• No. OAC performs as well as postcode based
  systems
• Yes. Super output area based performs less well
  than OAC
• No. The better support with detailed and timely
  denominators at super output area level will
  allow more rigorous statistical analyses
• Bespoke classifications are better than general
  classifications?
• Not yet.. Transactional data (HES) has too many
  zeros for a classification at fine scale
  geographical level

31
What next?

• Non-parametric clustering?
• Cross-classification of geodemographic systems,
  e.g. in a scenario with targeting 10 worst
  classified areas vs. density estimations
• ML model for long-term diseases sex, age,
  ethnicity, geodemographics, GP

32
(No Transcript)
33
IMD Segmentation
34
(No Transcript)
35
(No Transcript)
36
(No Transcript)
37
(No Transcript)
38
(No Transcript)
39
(No Transcript)
40
(No Transcript)
41
(No Transcript)
42
(No Transcript)
43
(No Transcript)
44
(No Transcript)
45
(No Transcript)
46
(No Transcript)