data management case studies: enhancing the analysis of e-health data and data on social care
DESCRIPTION
4th ESRC Research Methods Festival St Catherine’s College, Oxford. 5-8 July 2010. Data management case studies: Enhancing the analysis of e-Health data and data on social care. Alison Dawson University of Stirling (DAMES research Node, www.dames.org.uk). The case studies. - PowerPoint PPT PresentationTRANSCRIPT
Data management case studies: Enhancing the analysis of e-Health
data and data on social care
Alison Dawson
University of Stirling
(DAMES research Node, www.dames.org.uk)
4th ESRC Research Methods FestivalSt Catherine’s College, Oxford. 5-8 July 2010
The case studies
1) e-Health – linking eHealth and social science datasets to enhance understanding of risk of suicide in Scotland
2) Social care – ‘fusing’ datasets to enhance analyses of the costs of funding care for older people in Scotland and the UK
Suicides in Scotland In 2008 there were 843 deaths by suicide in
Scotland (defined as deaths from intentional self harm and events of undetermined intent), an age-standardised rate of 16.1 per 100,000 population per year
Suicide rates are around three times as high for men as for women
Rates of suicide in the most deprived areas of Scotland are significantly higher than the Scottish average
(ISD Scotland August 2009)
Previous focus on single factors
Previous studies have tended to focus on single risk / preventative factors associated with suicide:
Health related (e.g. history of mental illness)
Psychological factors (e.g. coping behaviours, religious beliefs)
Social and economic factors (e.g. unemployment, deprivation)
Few attempts to statistically model the interplay between different risk and protective factors in populations completing suicides.
Potential sources of relevant data
SMR and Census data and known risk / preventative factors
Types of data linkage technique
Selecting a data linkage method
CHI No
CHI No. Surname Sex Year of Birth
Postcode
DATASET 1
3456 DAWSON F 1966 FK9 4LA
DATASET 2
3456 DOWSON F 1966 FK9 4LA
DATASET 3
- DAWSON F 2066 FK9 4LA
Concerns when linking SMR and Census datasets
Ethical issues: Informed consentConfidentialityData securityDisclosure
Technical difficulties:Linking datasets from different
domainsProviding infrastructure that addresses
ethical concerns
OPERA - Older PEople’s Resource Allocation model
• Dynamic microsimulation model - takes micro level units (individuals) as the basic unit of analysis and uses simulation techniques to project the sample forward in time in order to investigate the effects of future social and economic policies
• Focuses on events towards end-of-life, particularly the impacts of long-term limiting illness and increasing levels of dependency
More on OPERA
Software – Stata / Mata Main datasets used
Family Resources Survey (FRS) – boosted sample in Scotland
HMRC Survey of Personal Incomes
Outputs Statistics (distributions, panel, survival) Graphics (graphs, plots, choropleth maps)
OPERA – Key elements
Incorporating Home Care Costs into OPERA
Problem 1 - Have a dataset – collected from Welsh Local authorities
Problem 2 - Distribution of costs is highly skewed40% of clients account for 10% of costs10% of clients account for 40% of costs
Model calibration (1)
Estimate determinants of costs of care using Welsh dataset
Estimate determinants of needing care and of being in receipt of local authority care using FRS data
Match FRS disability classification with that used in Welsh survey (IoRN)
Determinants of LA Costs of providing Personal Care
Costs increase with disabilitydecrease with agedecrease with presence of informal carerunaffected by gender and ethnicityvary by local authority
Detailed disability question in FRS
Does this/Do these health problem(s) or disability(ies) mean that you have substantial difficulties with any of these areas of your life? Please read out the numbers from the card next to the ones which apply to you.
PROBE: Which others?
1: Mobility (moving about)
2: Lifting, carrying or moving objects
3: Manual dexterity (using your hands to carry out everyday tasks)
4: Continence (bladder and bowel control)
5: Communication (speech, hearing or eyesight)
6: Memory or ability to concentrate, learn or understand
7: Recognising when you are in physical danger
8: Your physical co-ordination (eg: balance)
9: Other health problem or disability
10: None of these
IoRN Classifications (used in Welsh dataset)Activities of Daily Living (ADLs)
Eating - Transfers (bed to chair) - Toilet
Personal care tasksWash self /hair, bath, dress
Food and drink prepMain meal, snack, hot drink
Mental wellbeing and behaviour
Agitation, disturbance, verbal aggression, resistance,
relationships, risk
Bowel managementAssistance, guiding,
prompting, supervision
Mental wellbeing and
behaviour
MEDIUM HIGHLOW
LOWLOW
LOW HIGH
HIGH
HIGH
HIGH
LOW
MEDIUMMEDIUM
A B D C E G
F H
I
LEVELS OF HELP REQUIRED DETERMINEIoRN CLASSIFICATION
Recipient Sample Before MatchingObs Var X Var Y Var Z
1 x1 z1
2 x2 z2
.. .. ..
nA xNA zNA
Donor Sample Before MatchingObs Var X Var Y Var Z
1 y1 z1
2 y2 z2
.. ..
nB yNB zNB
Recipient Sample After Matching
Obs Var X Var Y Var Z
1 x1 z1
2 x2 z2
.. ..
nA xNA zNA
Data Fusion - General
Fusion proceeds by a number of (mostly numerically intensive) procedures.The objective is to define new variables whose properties differ as little as
possible from those of the (unobserved) underlying data.
Common variable(s)
Values of unobserved
variable(s) are imputed for
recipient sample
Before data fusion:No values for Var Y
After data fusion:Imputed values for Var Y (not true values)
Data Fusion – OPERA and Welsh home care costs data
OPERA Mk 1(Mainly FRS data including: )
AGEINFORMAL CARER?
Derived Disability Classification
Welsh home care costsDataset including:
AGEINFORMAL CARER?
Indicator of Relative Need (IoRN)
Hours of care per weekCost of care to LA per week
OPERA Mk 2AGE
INFORMAL CARER?DISABILITY DATA
+ imputed values for:Hours of care per week
Costs of care to LA per week
RECIPIENT DATASET DONOR DATASET
COMMON VARIABLES
Model calibration (2)
Select most disabled of those receiving LA care in FRS sample to receive personal care – match with proportions receiving LA personal care in Scotland (thus model mimics Scottish policy setting)
Stochastic simulation of model to maintain distribution of costs rather than focus on point estimate
Results weighted using FRS weights to represent UK/Scottish population
LA Costs of providing Personal Care by Age and Gender
LA Costs of Personal Care - Averages over individuals receiving care (£ per week)
0.00
50.00
100.00
150.00
200.00
250.00
15-24 25-34 35-44 45-54 55-64 65-74 75-84 85+
£'s
per
wee
k
Male Female
LA costs of providing Personal Care by Index of Disability (IoRN)
LA Costs of Personal Care - Averages over individuals receiving care (£ per week)
0
50
100
150
200
250
A B C D E F G H I
IoRN
£'s
per
wee
k
Male Female
(Free) Personal Care in Scotland
Scottish Government estimate of cost of providing FPC at home to pensioners in 2003-04 ~ £120m
Model estimate ~ £170mConsistent with LAs spending approx £50m prior to
introduction of policy
What about the personal care costs of those aged under 65 requiring PC?
Model estimate ~ £130m
Fewer clients, higher cost per client
Summary
Two examples of where data management techniques have been used to enhance analysis
The techniques discussed (linkage, fusion) can be technically difficult
The DAMES project is working to create resources that will assist with these kinds of processes