Cancer Incidence Smoothing

1
Geospatial Analysis in Public Health Spatial
Cluster Detection M.J. College, Jalgaon
IndiaSeptember 22-26, 2008 Glen D.
Johnson New York State Department of
Health and The University at Albany School of
Public Health Department of Environmental Health
Sciences
2
Acknowledgement Some of the following graphics
on cluster detection are compliments of Tom
Talbot, MSPH of the New York State Department of
Health - co-teaches GIS in Public Health
with Glen Johnson and Frank Boscoe at the
University at Albany, S.U.N.Y.
3
Cluster

• A number of similar things grouped closely
  together
  
  Websters Dictionary
• Concentrations of health events in space and/or
  time
• 
  Public Health Definition

4

• Clustering of health outcomes may be caused by a
  number of community-level factors
• Occupation mix
• Demographic mix (i.e. Race, Age, Sex)
• Socioeconomic status
• Cultural/Behavioral
• Environmental Exposure (always a big question)
• Time and/or Space (captures unexplained factors
  that co-vary with the outcome)

5
Cluster detection influenced by scaling and
zoning effects
as must be considered for all spatial
statistics and mapping/visualization - the
Modifiable Area Unit Problem (MAUP)
6
(No Transcript)
7
(No Transcript)
8
(No Transcript)
9
Different scale of observational units Coarser
aggregation
10
Different zonation Grid shift
11
(No Transcript)
12
(No Transcript)
13
Cluster Questions

• Does a disease cluster in space?
• Does a disease cluster in both time and space?
• Where is the most likely cluster?

14
More Cluster Questions

• At what geographic or population scale do
  clusters appear?
• Are cases of disease clustered in areas of high
  exposure? - or more generally, Can the cluster
  be explained as being associated with something
  other than chance?

15
Nearest Neighbor AnalysisCuzick Edwards Method

• Count the the number of cases whose nearest
  neighbors are cases and not controls.
• When cases are clustered the nearest neighbor to
  a case will tend to be another case, and the test
  statistic will be large.

16
Nearest Neighbor Analyses
17
Advantages

• Accounts for the geographic variation in
  population density
• Accounts for confounders through judicious
  selection of controls
• Can detect clustering with many small clusters

18
Disadvantages

• Must have spatial locations of cases controls
• Doesnt show location of the clusters

19
Knox Method test for space-time interaction

• When space-time interaction is present cases near
  in space will be near in time, the test statistic
  will be large.
• Test statistic The number of pairs of cases that
  are near in both time and space.
• P value is calculated through random simulations
  of the time value of the cases
• Need to define critical space and time distances.
  i.e. define what is near?

20
Advantages

• Do not have to map controls
• Determines if there is a space-time interaction.
• Can detect space-time clustering even when the
  overall disease rate has remained the same over
  time

21
Disadvantage

• Computationally time consuming with a large
  number of cases.
• Does not determine areas or time periods of where
  clusters occur.

22
Spatial Scan StatisticMartin Kulldorffhttp//ww
w.satscan.org/

• Determines locations with elevated rates that are
  statistically significant.
• Adjust for multiple testing of the many possible
  locations and area sizes of clusters.
• Hypothesis testing based on Monte Carlo
  simulations of the null, completely random,
  spatial distribution

23
Following is an example of how the scan statistic
algorithm delineates all possible circular
clusters, based on census blocks in the city of
Albany
24
(No Transcript)
25
(No Transcript)
26
(No Transcript)
27
(No Transcript)
28
A likelihood ratio is then computed for every
circular window, where each window represents a
potential spatial cluster. For example, assuming
a Poisson distribution of counts, the likelihood
ratio is proportional to
for observed cases c and expected cases Ec
inside the search window, and C total observed
cases throughout the region, including within the
search window.
29
The circle with the maximum likelihood ratio is
then identified as the most likely cluster, and
all others are rank-ordered below the maximum. A
null distribution of maximum likelihood ratios is
obtained by repeating the analysis on a
randomized version of the data, obtaining the
max. likelihood ratio, and repeating this
exercise for, say, 999 times. A p-value is
obtained for each circle by comparing its
likelihood ratio to the simulated null
distribution.So, for a likelihood ratio whose
rank is R within the simulated null values, then
the p-value R/( simulations 1).
30
Example, low birth weight
31
(No Transcript)
32
Note that Ec nC/Nfor population n in the
circle and total number of cases and Population
C and N respectively or
for covariate category i (an indirect
standardization) or Ec may even be predicted
from a regression model.
33
Recent advancements in the spatial scan statistic
aimed at overcoming the restriction of the rather
arbitrary shape of circular clusters

• Patil GP, Taillie C. Upper level set scan
  statistic for detecting arbitrarily shaped
  hotspots. Environ Ecol Stat 2004183-197.
• Duczmal L, Assuncao RM. A simulated annealing
  strategy for the detection of arbitrarily shaped
  spatial clusters. Comp Stat Data Anal 2004
  45269-286.
• Tango T, Takahashi K. A flexibly shaped spatial
  scan statistic for detecting clusters. Int J
  Health Geographics 2005 411.

34
Regression Analysis

• Control for known risk factors before analyzing
  for spatial clustering
• Analyze for unexplained clusters.
• Follow-up in areas with large regression
  residuals with traditional case-control or cohort
  studies
• Obtain additional risk factor data to account for
  the large residuals.

35
Example, Child Lead Poisoning
36
(No Transcript)
37
(No Transcript)