Gao W L, Lin H, Liu X N, Ren X W, Li J S, Shen X P, Zhu S L
Department of Epidemiology and Health Statistics, School of Public Health, Lanzhou University, Lanzhou 730000, China.
Department of Computer Software, School of Information and Engineering, Lanzhou University, Lanzhou, 730000, China.
Zhonghua Liu Xing Bing Xue Za Zhi. 2017 Mar 10;38(3):400-405. doi: 10.3760/cma.j.issn.0254-6450.2017.03.025.
To evaluate the estimation of prevalence ratio () by using bayesian log-binomial regression model and its application, we estimated the of medical care-seeking prevalence to caregivers' recognition of risk signs of diarrhea in their infants by using bayesian log-binomial regression model in Openbugs software. The results showed that caregivers' recognition of infant' s risk signs of diarrhea was associated significantly with a 13% increase of medical care-seeking. Meanwhile, we compared the differences in 's point estimation and its interval estimation of medical care-seeking prevalence to caregivers' recognition of risk signs of diarrhea and convergence of three models (model 1: not adjusting for the covariates; model 2: adjusting for duration of caregivers' education, model 3: adjusting for distance between village and township and child month-age based on model 2) between bayesian log-binomial regression model and conventional log-binomial regression model. The results showed that all three bayesian log-binomial regression models were convergence and the estimated were 1.130(95: 1.005-1.265), 1.128(95: 1.001-1.264) and 1.132(95: 1.004-1.267), respectively. Conventional log-binomial regression model 1 and model 2 were convergence and their were 1.130(95: 1.055-1.206) and 1.126(95: 1.051-1.203), respectively, but the model 3 was misconvergence, so COPY method was used to estimate , which was 1.125 (95: 1.051-1.200). In addition, the point estimation and interval estimation of from three bayesian log-binomial regression models differed slightly from those of from conventional log-binomial regression model, but they had a good consistency in estimating . Therefore, bayesian log-binomial regression model can effectively estimate with less misconvergence and have more advantages in application compared with conventional log-binomial regression model.
为了评估使用贝叶斯对数二项回归模型估计患病率比（）及其应用，我们在Openbugs软件中使用贝叶斯对数二项回归模型估计了照顾者对其婴儿腹泻风险体征的认知与就医患病率之间的患病率比。结果显示，照顾者对婴儿腹泻风险体征的认知与就医率显著增加13%相关。同时，我们比较了贝叶斯对数二项回归模型与传统对数二项回归模型在照顾者对婴儿腹泻风险体征的认知与就医患病率的患病率比的点估计及其区间估计以及三种模型（模型1：不调整协变量；模型2：调整照顾者教育时长；模型3：在模型2的基础上调整村庄与乡镇之间的距离和儿童月龄）的收敛性方面的差异。结果显示，所有三个贝叶斯对数二项回归模型均收敛，估计的患病率比分别为1.130(95%可信区间: 1.005 - 1.265)、1.128(95%可信区间: 1.001 - 1.264)和1.132(95%可信区间: 1.004 - 1.267)。传统对数二项回归模型1和模型2收敛，其患病率比分别为1.130(95%可信区间: 1.055 - 1.206)和1.126(95%可信区间: 1.051 - 1.203)，但模型3出现误收敛，因此使用COPY方法估计患病率比，其值为1.125 (95%可信区间: 1.051 - 1.200)。此外，三个贝叶斯对数二项回归模型的患病率比的点估计和区间估计与传统对数二项回归模型的略有不同，但在估计患病率比方面具有良好的一致性。因此，贝叶斯对数二项回归模型可以有效地估计患病率比，误收敛较少，与传统对数二项回归模型相比在应用中具有更多优势。
