Skip to main content

Archived Comments for: Geovisual analytics to enhance spatial scan statistic interpretation: an analysis of U.S. cervical cancer mortality

Back to article

  1. A valuable advance in the application of the spatial scan statistic

    Francis Boscoe, New York State Department of Health

    26 November 2008

    This is an important paper that engages important interpretive issues regarding the spatial scan statistic that have been generally neglected. Basically, the authors’ method involves distilling complex SaTScan output from multiple program iterations, allowing the maximum circle size to vary. A few years ago, I and three coauthors proposed a method for distilling summary information from within a single program iteration (see reference 15). The two approaches are complementary and could probably be unified.

    When identifying high (or low) rate clusters, the very last geographic unit in a cluster (that is, the one farthest from the center), by definition, must always have a high (or low) rate. This leads to heterogeneous clusters with a dumbbell or ring structure, where the geographic units most contributing to the cluster are found at or near its edge. In this paper, this is seen in the numerous clusters with Los Angeles, California or El Paso, Texas at their very edge. As heterogeneous clusters are difficult for public health officials to interpret and respond to, any method that can help distinguish them from stable core clusters is helpful. The method presented here accomplishes this.

    More work along these lines remains to be done in the temporal dimension. There are many papers that present something like the following scenario: Data is compared over two time periods. A cluster identified in time 1 appears to get smaller, or larger, or moves slightly, or breaks into two separate clusters, in time 2. Authors conclude that the pattern or relationship has changed in some substantial way. In fact, the change was likely trivial. The original cluster is probably still significant in the second time period, it is just no longer the most significant cluster, which is the default SaTScan output. By making use of a broader set of SaTScan output, it should be possible to develop a method or visualization tool that precludes this interpretative error.

    Competing interests

    I have no competing interests. I was not a referee of this paper.