I am running a space-time permutation analysis with occurrence data (case notifications - long., lat. & date) for a viral disease of livestock. SatScan detected several significant clusters some of which look great (i.e. all of the points within such cluster's radius belong to the same period of time and therefore, to the spatio- temporal cluster). However, some of the clusters look really strange because their radii include cases reported from different time periods (both earlier and later than the duration of a cluster). I assume this is supposed to mean that such points do not belong to the cluster, but some of them are actually centroids of these clusters! I have attached an example of such 'good' and 'strange' clusters for illustrative purpose.
My main question is: what may have caused such patterns? Is that an artifact and something is wrong with my data/settings? Or the analysis have captured some real epidemiological aspects, e.g. disease endemicity in the area? I know that SatScan is not designed to detect geographical clusters that are persistent in time however, it's very tempting to assume that such patterns are consistent with a continuous disease circulation and re-appearance in the area.
I would greatly appreciate any thoughts/ideas/tips regarding this problem!
Best regards,
Serhii
Hi Serge,
What is your maximum spatial cluster size set at? Larger clusters (default = 50%) may hide smaller core clusters. You could try running SaTScan for each year and overlaying the clusters to reveal spatiotemporal patterns of disease endemicity.
-Mark
Hi, Mark
Thanks for your reply. The maximum spatial cluster size was set at 150 km. This value has been chosen for several reasons: a) it is a recommended maximal size of the surveillance zone for this infection; b) significant spatial association between consecutive cases has been found within a similar distance range (133-156 km) in previous studies.
I thought about a possibility that 'true' clusters might be obscured by something. For example, there is also a mention that due to a circular nature of the scan window SaTScan might not be able to detect correctly borders of outbreaks which are geographically extensive. Definitely, I will try playing around with different settings and probably different probability models (as far as the limited nature of my data will allow).
What I am really wondering about, is that if it is appropriate at all to interpret such results as shown in attached pictures in terms of disease endemicity? So far, I didn't find many papers discussing these (or similar) topics but I will dig deeper.
Best regards,
Serhii
Hi Serge,
I have in the past used the STPM but I am currently working with the Poisson model and have much more experience with it because I am working with an underlying risk population in my current research. Determining the proper software settings is difficult and no real guidelines exist for users. Below you will find two articles that talk about SaTScan parameters and visualization tips, as well as heterogeneous versus core clusters.
Han, J., Zhu, L., Kulldorff, M., Hostovich, S., Stinchcomb, D. G., Tatalovich, Z., ... & Feuer, E. J. (2016). Using Gini coefficient to determining optimal cluster reporting sizes for spatial scan statistics. International journal of health geographics, 15(1), 27.
Chen, J., Roth, R. E., Naito, A. T., Lengerich, E. J., & MacEachren, A. M. (2008). Geovisual analytics to enhance spatial scan statistic interpretation: an analysis of US cervical cancer mortality. International journal of health geographics, 7(1), 57.
- Badges
- Science topic
- Similar topics
- Medicine
- Epidemiology