I often see disease surveillance type articles in which the authors use Moran's I to identify clusters. I also find many articles that use SaTScan for disease surveillance, with usage being done in over 100 countries.
is there any specific advantage to either approach and what is it?
Thank you for your feedback.
Thank you for your reply, Soutrik. I have been using SaTScan for about eight years now. I am used to identifying clusters that can be evaluated by relative risk values in addition to p values. Moran's I would be new to me. Would there be a conflict in philosophy to do both?
Hi Soutnik,
I also run FlexScan after SatScan for some projects so that I check out not just circular shaped (or elliptically shaped) clusters with SaTScan. I like the fact that I can adjust for multiple covariates in SaTScan by first using a Poisson regression, and then using the predicted values in SaTScan as a control. SatScan has become for me a major factor in teaching graduate students to do applied research. I am currently overseeing 11 graduate students with projects based on SaTScan. I am interested in checking out Moran's I simply to learn more and to better understand what other researchers are getting with it.
SaTScan lets you adjust for covariates as follows:
1. Run Poisson regression to predict y from x.
2. Output Y_hat, the predicted y.
3. Have y in the case file and have y_hat in the control file.
4. Run SaTScan.
The resulting clusters are based on y that has been adjusted for x. Martin Kulldorff explained this to me. We are in always touch about SaTScan modeling, and we also do research together with SaTScan. I have learned many things about SaTScan from him that I would never have understood or even found as options in SaTScan.
I agree that the spatial scan test and Moran's I are two different things! Unfortunately this is getting confused in the literature. The scan test is detecting clusters or in other words spatial outliers. On the other hand, Moran's I is testing for spatial dependance. This has also been termed cluster detection and disease clustering and thus things got confused over the similarity of words. However this is not the whole story as strong clustering or dependence can express itself in clusters or local outliers in the sense that too many cases are too close together.
Coming back to the original question: which statistic to use for surveillance? Both can be used. To detect outbreak locations use the scan test, the result is a location on the map. Alternatively use the Getis-Ord G_i. When using Moran's I or Geary's C, the result will be an estimate for the strength of clustering or dependence, which you might want to monitor over time if it is changing, because the virulence of an agent or immunity of a population is changing over time.
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts