Dear RG members. Epidemiological research sometimes deals with time series count data as the dependent variable. But, what is the best regression model for these data? ARIMA has been the classical approach taking into account the autoregressive nature of the data. Recently, other approaches have been proposed, such as the Poisson autoregressive regression models for count data. Any suggestion? Pros and cons of each method? Thank you.
For best way for modeling epidemiological time series data and find regression most appropriate model, you can select there variables, 1. Infection rate 2. Incubation time 3. Morbidity occurred during the study period, Now on another side you can choose 1. most affected section of the population 2. Immune defense natural and artificial 3. mortality occurred. Here you can use Poisson autoregressive regression models. You have to find clinical versus natural immune defense and significant infection rate in immunized and non immunized population if you are doing group study in a population. For maintaining precision you should at least select six samples at a time.
Hi,
For modeling epidemiological time series data:
Refer the following link:
http://csm.lshtm.ac.uk/themes/time-series-regression-analysis/
http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0088075
http://markstat.net/en/images/stories/technical_guide_chapter_3.pdf
ftp://ftp.cordis.europa.eu/pub/tmr/docs/lifmetaepid980435.pdf
https://www.researchgate.net/post/Time_series_of_epidemiological_data
Take a look at Age-Period-Cohort models. See for example: http://www.annualreviews.org/doi/abs/10.1146/annurev.pu.12.050191.002233
Thank you for the answers. And what is the right regression model if I am interested in an intervention effect over the time series of the dependent variable? Thank you.
Hi,
If you are interested to evaluate an intervention effect over time, you may use multivariate interrupted time-series intervention analysis (ITSIA).
ITSIA is considered as a strongest quasi-experimental approach which use segmented and dynamic regression analysis developped by alan Pankratz.
See : Mahamat A et al. J Hosp Infect. 2011,
Mahamat A et al. Int J Antimicrob Agents. 2007
and Mahamat A et al.Clin Microbiol Infect. 2005
Pankratz A. Forecasting with dynamic regression models. New York,
NY: Wiley; 1991
All the best
Dear Andres,
I like the ACP model of Heinen and Rengifo, and their model can incorporate any exogenous shock (an intervention) as an exogenous factor affecting the poisson autoregressive process. I have tried this and it works fine (examining the effect of preventive measures in human-ignited wildfire). Here's a citation of a paper you can examine:
Heinen A., RengifoE (2003) Multivariate modelling of time series count data: an autoregressive conditional poisson model. Center for Operations Research and Econometrics, Catholic University of Louvain, Louvainla-Neuve, Belgium. http://alfresco.uclouvain.be/alfresco/download/attach/workspace/SpacesStore/e8b81e6f-0b93-4fc1-9d1d-643579c8f578/coredp_2003_25.pdf
See
ULRICH HELFENSTEIN-1991- The Use of Transfer Function Models, Intervention Analysis and Related Time Series Methods in Epidemiology, International J. of Epidemiology
http://www.uvm.edu/~dhowell/StatPages/Mixed-Models-Repeated/Mixed-Models-for-Repeated-Measures1.html
- Badges
- Science topic
- Similar topics
- Statistics
- Time Series