[R-sig-Geo] to forecast a disease risk using a dynamic model

[stevenhuyi]

Hi everyone,

I'm recently struggling with exploring a model to forecast the
schistosomiasis risk. Let me introduce the data I have first.
Parasitological data, including number of infected individuals and
population at risk, are available in a province from 1997 to 2010 at the
county level (i.e., polygon data in county) as well as some risk factors
(e.g., temperature, rainfall, and wetness). I want to forecast the disease
risk, that's the risks in 2011,2012, and etc. 

What comes to me first is the time series model (i.e., ARIMA) or Karman
filtering, but schistosomiasis often occurred in clusters, so the spatial
correlation should be considered. Besides,as the parasitologcal data are
count data, it would be better to fit the data using a Poisson or negative
binomial distribution. Unfortunately, either time series model or karman
filtering can tackle these. I tried the LINA package in R the other day. It
can take into account the issues that I'm concerned about, but it seems that
it can only smooth the data using the historical data, it cannot forecast. 
I have some references, but no example is related to forecast. 

A popular idea to fit spatiotemporal data is to use a hierarchical model,
which is composed of a data level model, a state level model that produce
the data, and finally a parameter level which exists in the former two level
models. Many statistician advocates this model, like Noel Cressie. I think
this would be my target. However, it seems that there is no R package that
can implement the hierarchical model for the polygon data, I know some can
deal with point data like spTimer (but it still cannot deal with count
data).  

What I said above might not be correct and welcome your comments and, most
of all, please give me some suggestions on how to forecast the disease risk
and to implement it, hopefully, in R. Thank you so much, guys! Have a nice
weekend.



--
View this message in context: http://r-sig-geo.2731867.n2.nabble.com/to-forecast-a-disease-risk-using-a-dynamic-model-tp7588373.html
Sent from the R-sig-geo mailing list archive at Nabble.com.