GEOGRAPHIC INFORMATION, CLUSTER DETECTION AND SPATIAL SURVEILLANCE

The increased need for cluster detection has coincided with an increasing availability of data, especially data on the location of events.

This is often obtained by geocoding the addresses of individual cases.

This can be done ‘on the fly’ as cases are encountered (Beitel et al., 2004) or with static databases that retain the location of all patients eligible for surveillance (Lazarus et al., 2002).

In its simplest form, geocoding could imply merely obtaining the zip or postal code, but it may also include finding the exact latitude and longitude of an address using geographical information systems (GIS).

In statistical jargon, such data about location is often referred to as ‘spatial’ data.

The value of spatial data for cluster detection is twofold.

First, all attacks are localized at some spatial scale.

That is, an attack could conceivably target a neighborhood, but on a city-wide scale this would be a small area.

Alternatively, an attack could include a whole metropolitan area, but on a national scale this would be a small region.

When surveillance is limited to a single daily count from a neighborhood or city, even sharp increases in relatively small regional counts may be hidden within the natural variation found in the count across a larger area.

Spatial surveillance thus promises to increase the power to detect
events that occur in small regions, relative to surveillance of the total count
only.

Secondly, if an incident cluster is identified, public health officials will need to respond.

If the data are nonspatial, surveillance can only give vague messages of the sort ‘there is an excess of cases in the Boston metropolitan region’; this is unlikely to be of much practical use.

In contrast, spatial surveillance would allow more-specific messages, such as ‘there are excess cases in zip code 02474’.

The job of identifying small regions with extra cases is also referred to as ‘cluster
detection’, where the clustering in this case refers to extra cases in an area on
the map.

The coincidence of suddenly increased need and increasingly available spatial
data has generated new interest in statistical methods for spatial surveillance,
which might be described as the detection of incident clusters in space.

The goal of this book is to provide a snapshot of the state of the nascent art of
incident spatial cluster detection, provided by statisticians involved in traditional
surveillance (of a single statistic), in spatial clustering, and in spatial
surveillance.

(...)

In this context, we are most interested in detecting attacks while they are
ongoing rather than retrospectively.

In statistical terms, we might refer to this as ‘cluster detection’ or ‘incident cluster detection’, where by ‘cluster’ we mean the occurrence of extra cases in a short time span. In the literature on surveillance, this is sometimes referred to as ‘on-line’ surveillance.

(...)

Texto integral em: Spatial and Syndromic Surveillance for Public Health
Andrew B. Lawson and Ken Kleinman