Spatial Analysis of the 2008 Influenza Outbreak of Hong Kong Poh-chin LAI and Kim-hung KWONG

Poh-chin LAI and Kim-hung KWONGDepartment of Geography

The University of Hong Kong

Spatial Analysis of the 2008 Influenza

Outbreak of Hong Kong

23-26 March 2010

Lai & Kwong, HKU Geography, 2010 2

ContensContens

Background Methodology

Data Analytical methods

Results Maps of standard deviational ellipses Nearest neighbor distance statistics Grid-based spatial autocorrelation Kernel density maps

Observations and implications


BackgroundBackground

Hong Kong: flu outbreaks at schools, a hospital and a nursing home for the elderly since 6 March 2008

13 March 2008: suspended classes of all kindergartens, child care centers, and primary schools for two weeks

Measure to shut down all schools did lower the disease incidence but received mixed comments from the public


DataData

Three sets of data were compiled (6-13 March 2008): affected schools, non-affected schools, and background


Analytical Methods Analytical Methods (1)(1)

Statistically testing spatial patterns exhibited by case and control data

Used GIS techniques ArcGIS developed by ESRI GeoDA developed by Luc Anselin

Used a variety of different methods Standard deviational ellipses Nearest neighbor distance statistics Local indicators of spatial autocorrelation Kernel density maps


Results Results (1)(1)

Schools affected About 5.7 percent of the total More primary schools



Standard deviational ellipses locations of mean and

weighted mean (adjusted by student population of each school) centers were indifferent

Infected cases

Control cases



Nearest neighbor distance statistics

Control cases Infected cases Nearest Neighbor Observed Nearest Neighbor Observed Mean Distance = 112.11 Mean Distance = 648.92 Expected Mean Distance = 463.73 Expected Mean Distance

=1456.07 Nearest Neighbor Ratio = 0.24 Nearest Neighbor Ratio = 0.44Z Score = -64.66 Standard Deviations Z Score = -11.52 Standard

Deviations

more compact

Statistically significant clustering



Spatial autocorrelation of quadrat counts

Point data about schools were aggregated into areal units of two different sizes 1 km x 1 km

500 m x 500 m Contains an average of 120 buildings per cell

Ignoring detailed locational information

Data masking to protect individual identity

Keeping the number of cells manageable for desktop computer operations



1km x 1km cells A few patches of ‘high-high’ occurrences or hot spots Hot spots not extensive in their local coverage and

buffered by cells of ‘low-high’ values

1km x 1km Cells1km x 1km Cells



500 m x 500 m cells Similar patterns to 1kmx1km but more

disjoint hot spots Manifests difference of cell sizes on visual

impact

500m x 500m Cells500m x 500m Cells



Kernel density maps

bandwidth yields a smoother surface with low intensity levels

Too small a cell size defeats areal generalization

Cell size: 500m x 500m Cell size: 250m x 250mBandwidth: 1 km Bandwidth: 500 m

Cell size: 1 km x 1 km Cell size: 500m x 500mBandwidth: 1 km Bandwidth: 500 m


Observations Observations (1)(1)

Visual impression of hot spots projected by these maps were quite different even though cell sizes are the same

Kernel density surfaces appeared smoother and the patterns more contoured

Spatial autocorrelation reveals hot spots as a discrete category along with other categories not identifiable on a kernel density surface Pockets of hot spots buffered by spatial outliers

implied that the disease had remained localized


Implications Implications (1)(1)

Graphic, statistical, and spatial analyses work together to provide clues on clustering tendency and cluster areas

The degree of clustering should be evaluated with respect to the usually non-uniform population distribution

Geo-epidemiological models that enable the identification of disease variance in space can help guide interventions in areas with a higher disease burden



A better understanding of spatial distribution of hot and cold spots would help formulate policies to target specific community groups e.g. movements of primary school students are

controlled to school districts thereby reducing cross district interaction

designated isolation of infected primary schools and schools around the hot spots will likely be an effective intervention measure

settings with less movement restrains (such as secondary schools and hospitals) may be modeled in similar fashion but ……

more radical intervention approach may be warranted



Time lags between notification of suspected cases and confirmation of statutory notifiable diseases may distort counts more effective means of communication can

decrease the likelihood of disease transmission and possibly contain a potential flu pandemic

further opportunity to undertake cross-level interactions and how social mixing patterns might affect disease spread

Need for careful assessment of the aggregation level and comparison of different visualization and presentation techniques


More InformationMore Information

Please contact

Dr. P.C. LaiAssociate ProfessorDepartment of GeographyThe University of Hong KongEmail: [email protected]: (852) 2859 2830Fax: (852) 2559 8994

Spatial Analysis of the 2008 Influenza Outbreak of Hong Kong Poh-chin LAI and Kim-hung KWONG

Documents

hot spots hot spots

spatial patterns

expected mean distance

spatial analyses

spatial outliers

nearest neighbor ratio

secondary schools

results1 schools