Small Area Statistics Potential and Challenge 28th January 2004 at Trinity College

1
Small Area Statistics Potential and
Challenge28th January 2004 at Trinity College

• Individual Health Data and Small Area Studies
  experiences from Sweden and Scandinavia"
• Dr Åke Sivertun
• GIS-lab IDA, Linköpings Universitet
• akesiv_at_ida.liu.se

2
King Cholera dispenses contagionthe London
Cholera Epidemic of 1866George John Pinwell
3
A CLASSICAL EXAMPLE
From John Snow The Cholera epidemic
The pump at Broad Street
One agent - one disease
4
Individual data the key to understand relations
between patients and exposure

• Areas only the second best
• Depending on the research question but often
  impossible to get the answers with aggregated
  data
• Difficulties to track populations at risk in
  cases with long latecies
• Possible to aggregate but impossible to
  disaggregate data

5
Time Sphere
6
Determinants for Health and bad Health
Genetic
Lifestyle
Environment
Impact from Health organisations
7
(No Transcript)
8
GIS Method

• Land use
• Vegetation
• Geology / soils
• Hydrology / River basins
• Terrain models
• Local climate
• Socio economics
• Health indicators

9
The Epidemiological Public Health Observatory -
some tasks

• Discover
• Incidence changes over time
• Variations in spatial distributions
• Extremes (high or low incidences)
• Monitor changes above or below pre-set levels
• - the alarm function
• Make trend projections out from retrospective
• - longitudinal time-series

10
The Epidemiological Public Health Observatory -
some tasks

• make health impact assessments
• using epidemiological tools for evaluation of
• prevention
• guide in allocation of resources
• investigate cancer, environmental and other
  alerts

11
Essential topics when introducing GIS as a tool
in the Public Health Observatory - some practical
experiences

• 1. Availability of data for use in a GIS
• a. Relevance b. Validity
• c. Sources d. Costs
• e. Resolution
• Problems and benefits pros and cons
• What is the real value
• Presentation of data or analytic tool?
• Spread in the organization

12
Database types

• Map data (back-cloth)
• Exposure data (causes)
• Demographic data (population registers)
• Health outcome (diagnoses)
• Property registers (x- and y-coordinates)

13
Relevance and validity
Contents Quality/Validity/Relevance Map data
Various Demography High Property
register Various/Time-dependent Exposures Extreme
ly varying Health outcome Cancer
registry High (Several sources, crossvalidated)
Discharge registry Medium/Low/Time dependent
Mortality register Medium Congenital malform.
Partially Low Quality registers Clinical
registers Rather High (Often crossvalidated)
Dependent of numerous factors
14
Data sources, suppliers and charges of purchase

• Swedish National Archives Riksarkivet (RA)
  (?)
• Statististics Sweden Central Bureau of
  Statistics (SCB)(?)
• Epidemiological Centre at Swedish National Board
  of
• Health and Welfare Socialstyrelsen, (SoS) (?)
• Central Office of the Swedish National Land
  Survey (?)
• Hospital and Clinical Registers (?)
• Tax authorities (?)
• Questionnaires (Subjective health, life-style,
  hidden figures
• not available from registers)

15
Diagnos registers - What impacts quality?
The number of positions in the ICD code is one
determinating factor. There is an inverse
relationship between resolution of the code of
diagnosis and the validity (ROC-analysis) Routine
s and time lag of reporting (National
registers) Feed-back enhances quality! Retrospec
tive longitudinal designs - a special
problem (translational errors between ICD
versions) Use of validation routines Special
inborn errors (transferrals between clinics,
county councils, check-ups/repeated visits etc.)
16
The Principle for Matching on Coordinates
Property register
Adress Property name X and Y coordinates
Other information
YYYYMMDD-123 4 Property name/ Address
Population register
Personal number ICD code Date of diagnosis
Age at diagnosis
Cancer register
17
Matching of personal data requirespermission

• Permission to handle personal records by the
  Data Inspection
• Personal numbers are coded by Statistics Sweden
  (SCB) after matching to preserve privacy
• Medical data are handled by an Etical Committee
  that gives permissions for medical reserach.
• The individual data are not possible to present
  but in aggregate to preserve privacy.

18
ICD-code Date of diagnosis Age at diagnosis
Sex Year of birth Number of years Risk
level location 1 at location 1 at location 1
Year of entry Age at entry in Number of years
Risk level in location 2 location 2
at location 2 at location 2
Year of entry Age at entry in Number of years
Risk level in location 17 location 17
at location 17 at location 17
Number Tot years Tot years in
Tot years in Tot years in Total years of
changes of observation high risk area low
risk area normal risk in
unknown area risk area
19
The WHOs ICD Code System II

• XXI main groups
• e.g. Infectious diseases (I), tumours (II),
  respiratory organs (X),
• cardiovascular diseases (IX), trauma-toxic
  agents-violence (XIX)
• Acute myocardial infarction I21
• in inferior part I 21.1

Acute transmural infarction in diafragmal wall
(of left ventricel) inferiolateral inferioposter
ior
vers 9 410
vers 8 412. 01
20
(No Transcript)
21
Time-Space-Dose interactions
TIME
SPACE
DOSE
22
Dimensions (resolution) in data

• Time
• Space
• Dos (continous, arbitrary, cathegorized)
• Diagnos (Positions of ICD code)

23
Epidemiological visions.
From a macro perspective
24
(No Transcript)
25
.but also a local perspective
26
The regional .
27
(No Transcript)
28
Degree of resolution I
Key-code area (940) Parish (176) Commune (13)
County (1)
Country level Individual level (426 000)
29
Or an even more local perspektive..
Commun
1
Parish
2
Key code areas
3
4
Individual level
O Löfman -97
30
(No Transcript)
31
Degree of resolution
Commune
Parish
Key-code area
Individual level
32
Swedish administrative subdivisions
County
District
Kommun
Cluster area
Parish
Key-code area
individual
?
33
BAS areas in Stockholm
34
Ward areas
35
Post codes and Kommun missmatch
36
Ecological fallacy

37
Individual data - exposure address points
38
Exposure to noice on buildings
39
Air and traffic noice exposures to preschools
40
Density mapping population age between 6 34
41
Epidemiological Meassures
The Funnel
Incidence
Prevalens
Out movers
Dead
Recovered
42
Epidemiological meassures are quota f ex
incidence/prevalens/RR/SMR
x
coordinate system
Grid (RT90)
y
High Radon area
exposure
Exposure map
cases
Tumour register
population at risk
Population register
backcloth map
Topologic map data
43
Population dynamics
(Model according to Haegerstrand)
still living here
emmigrants
t1
Geographic space
timedimension
death
t0
New born
Living here from the beginning
immigrants
44
Changing exposures - Individual time-space
trajectories
TIME
X
tn
Exposure levels
1 2 3 4 5
t1
Y
t0
45
NOx concentrations distribution in city of
Helsinborg (example, OPSIS Enviman software).
46
(No Transcript)
47
Integrated time-dose
area 3
area 2
t1
dose level d3
area 1
dose level d2
time
transitors
t0
dose level d1
48
Absorption Distribution Metabolism Exkretion ADME
Health Effect
Bio-effective Dose
Biological sensitivity
Internal Dose
Individual sensitivity
External Dose
Exposure Dose x Time
49
Some GIS Individual data based health studies
in Östergötland Sweden
50
Some Epidemiological Analyses of possible
Environment-Health (causal?) interactions

• Ground Radon and childhood Leukaemia
• Socioeconomic grouping and cardiac disease
• Water hardness and cardiovascular disease
• Acid precipitation och aluminium exposure
• Fluorides in drinking water and fragility
  fractures
• Trihalometanes and congenital malformations
• Accident analysis

51
All malignancies and Ac L. Leukemi
?
Urban case Rural case
?
Norrköping
Motala
Linköping
Lake Vättern
Baltic Sea
52
Radon
53
Research question
What relation can be found between persons living
on ground with a risk for radon and malign
tumours among children? Design - ecologic
correlation study with approx 53 000 children
(population under risk)
Material al children born in E-län 1979 1992
Follow up time 1979 1995 at least 3 years.
Östergötland (E)
54
Material Metod I
Registers
Data rutines
Digitizing (ares with expected
exposures) Matchning of databases Overlay-analysi
s in GIS to identify individual in different
exposure classes
Population register Property register Regional
cancer register Estimated Exposures
(SGU/SIG) Koordin
GIS
Outcome measures
Standard Morbidity Ratio (SMR) related to
Swedish national average age and gender
specific expected numbers
Relative Risk (RR) Related to low-risk areas as
standard (RR1)
55
Linköpings kommun, Östergötland
Total county area 10 000 km2 Total population
406 000 13 communes (4338 - 124 352
inhabitants) of which radon risk areas has been
completely mapped in 8 communes and partly
mapped in 5 communes
56
Swedish geological survey office radon risk
classification
building codes and standards
57
Motala river
58
Airborne measurements of gamma radiation in
SouthEast Sweden
Linköping
Data from SSI 1998
59
Water
Ground
Building material
60
Standardised health risk for individuals living
in areas with estimated Normal- and /ore
High-risk during the studied period
Akut lymfatisk leukemi
SMR
RR
61
Acute lymphatic leukaemia, risk for exposure at
the birth place and distribution of the cases
Riket
62
Population related to radiation calculated in a
GRID with 200 x 200 m resolution (n244 038)
Population year 98, County of Östergötland
63
Exposed population by risk class and commune
Commune
Low
Normal
High
Probably
risk
risk
risk
high risk
Boxholm
531
3210
1995
0

Finspång
-
-
-
-

Kinda
-
690
320
-
Linköping
57350
50378
16557
67
Mjölby
-
3248
3239
-
Motala
-
327
1345
1181
Norrköping
31437
89319
0
0
Söderköping
6916
5154
1502
0
Vadstena
80
6495
980
0
Valdemarsvik
0
6846
337
1687
Ydre
103
3937
298
0

Åtvidaberg
-
-
5340
-

Ödeshög
-
-
2438
174
Total county
96417
169604
34351
3109

Data not valid Not complete data
, Löfman, et al Report 1996
64
Conclusions

• Insidence of Ac Lymph Leukeamia among children
  is related to radon risk class
• There is a stronger relation between incidence
  of leukeamia and persons living in high-risk
  areas then with risk class at the address at
  birth.
• No other tumour types have in this study a
  correspondance with estimated high radon levels
• Normal-risk- class should be regarded as a high
  risk class from the experiences in this
  epidemiological study
• Orig-artikel Kohli, Noorlind Brage, Löfman.
  Childhood leukemia in areas with different radon
  levels a spatial and temporal analysis using GIS
• Journal of Epidemiology and Community Health, Nov
  2000 Vol 5411822-826

65
Socioeconomic grouping and cardiac disease
All cases with heart diagnosis 1997 in hospital
care and/or died the same year
Central district

66
Socio-economic grouping and cardiac disease
socio-economic variables
factor analysis (data/variable
reduction) cluster analysis on key-code
areas selection av cluster according
Wards metod (euklidian distances) coordinates
geocoding
67
Method
Raw data
Factor analysis
Cluster analysis
Register of Death certificates
Geographic definition of cluster population
Discharge register
Matching of cardiac diseases
Comparison of cardiac morbidity between cluster
areas
68
Socioeconomic index for respective
cluster
Method
I
II
III
IV
V
J Byrsjö October 2000
Total
Occupational index
Income index
Education index
Social welfare index
69
III
II
I
IV
70
(No Transcript)
71
Socio-economic differences in Motala
72
Finspang community with 5 km gridnet and the
number of observations in each 25 km2 cell
congenital malformations n 95
73
Raster data calculations
853
1127
Number of persons Area (km2) Population
density
97
653
117
4
54
44
66
23
213.3
25.6
1.8
3.5
28.4
74
6 IDDM cases within a circle of 250 meter radius
75
(No Transcript)
76
(No Transcript)
77
(No Transcript)
78
(No Transcript)
79
Criteria of causality
ty

• Consistency
• Stregth of association
• Specificity
• Temporality
• Biological gradient

• Plausibility
• Coherence
• Experimental evidence

Hill, A. B. The environment and disease
Association or causation. Proc. R. Soc. Med.
1965 58 295-300
80
Causality models
Contributing causes
Necessary cause
E
A
H
A
J
A
B
D
B
G
F
I
C
F
C
Sufficient cause I
Sufficient cause III
Sufficient cause II
K. Rothman. Modern Epidemiology 1998
81
Amyotrophic Lateral Sclerosis (ALS) Finland

• Cases with ALS (Amyotrophic Lateral Sclerosis)
• ALS a rare neurological disease with no cure
• No clear aetiology (causation) theories
• Globally no evidence of spatial clustering
• Data
• Provided by Statistics Finland
• Derived from Finnish death certificate register
• Retrospective study 1985-1995
• 1000 patients
• Longitudinal dataset of residential histories
  from birth to
• death. All residential moves, geocoded to 1m
  accuracy
• Aim does ALS cluster in space or time?

82
Estimating Exposure

• Here we demonstrate the Modelling Problem, with
• 4 Individuals Life-paths or
• trajectories, in Finland
• Exposure to 2 Risk Factors

83
(No Transcript)
84
Mortality Rate
Birth 1900-1950s. Place of Birth Finnish borders
pre 2nd World War Noticeable cluster in
SE Unstable estimates in sparsely populated
Lapland
85
Relative Risk Surface

• Spatial epidemiology of ALS in Finland
• Individual level data
• Kernel Estimates
• All residences 1965-95
• Statistical Significance tested by Monte-Carlo
  Simulation
• Sabel C E, Gatrell A C, Löytönen M, Maasilta P
  and Jokelainen M (2000) Modelling exposure
  opportunities estimating relative risk for Motor
  Neurone Disease in Finland, Social Science
  Medicine 50(78) 1121-1137.

86
Animation

• Migration Residential
• Space-Time Visualisation
• Space Finland
• Time 1965 1990

87

• Thank you for your attention!
• Also thanks to
• Dr Owe Löfman owe.lofman_at_lio.se HealthGIS LiO
• Dr Clive Sable Clive.Sabel_at_klinvet.ki.se ALS
  Finland
• Clive.sable_at_canterbury.ac.nz
• Dr Ulf Samuelsson Ulf.Samuelsson_at_lio.se diabetes
• Anders Schaerström anders.schaerstrom_at_fhi.se
  Time/Space
• Students at the Masters course in Geoinformatics
  LiU
• http//www.ida.liu.se/education/fsgis/portal/