Non-Spatial Analysis of Relative Risk of Dengue Disease in Bandung Using Poisson-gamma and Log-normal Models: A Case Study of Dengue Data from Santo Borromeus Hospital in 2013

International Seminar on Mathematics, Science, and Computer Science Education 2016

数学;自然科学;计算机科学

Irawan, R.^1 ; Yong, B.^1 ; Kristiani, F.^1

Department of Mathematics, Universitas Katholik Parahiyangan, Jl. Ciumbuleuit 94, Bandung

40141, Indonesia^1

关键词: Bayesian inference;    Dengue haemorrhagic fevers;    Deviance information criterion;    Log-normal model;    Markov chain Monte Carlo method;    Model comparison;    Relative risks;    Spatial analysis;   

Others  :  https://iopscience.iop.org/article/10.1088/1742-6596/812/1/012034/pdf
DOI  :  10.1088/1742-6596/812/1/012034

来源: IOP

【 摘 要 】

Bandung, one of the cities in Indonesia, is vulnerable to dengue disease for both early-stage (Dengue Fever) and severe-stage (Dengue Haemorrhagic Fever and Dengue Shock Syndrome). In 2013, there were 5,749 patients in Bandung and 2,032 of the patients were hospitalized in Santo Borromeus Hospital. In this paper, there are two models, Poisson-gamma and Log-normal models, that use Bayesian inference to estimate the value of the relative risk. The calculation is done by Markov Chain Monte Carlo method which is the simulation using Gibbs Sampling algorithm in WinBUGS 1.4.3 software. The analysis results for dengue disease of 30 sub-districts in Bandung in 2013 based on Santo Borromeus Hospital's data are Coblong and Bandung Wetan sub-districts had the highest relative risk using both models for the early-stage, severe-stage, and all stages. Meanwhile, Cinambo sub-district had the lowest relative risk using both models for the severe-stage and all stages and BojongloaKaler sub-district had the lowest relative risk using both models for the early-stage. For the model comparison using DIC (Deviance Information Criterion) method, the Log-normal model is a better model for the early-stage and severe-stage, but for the all stages, the Poisson-gamma model is a better model which fits the data.