层次贝叶斯模型在校正低估以估计慢性病患病率中的应用

目的 探讨层次贝叶斯模型中在慢性病患病率估计研究中的应用,以校正低估的计数数据从而获得潜在的真实患病率。 方法 基于真实的患者人数服从Poisson分布,建档登记的患者人数服从二项分布的假设,考虑与疾病患病水平及建档过程相关的影响因素,建立层次贝叶斯模型对某地17个县(市)2018—2020年建档登记的高血压患者人数进行校正,并对影响患病及建档过程的相关因素的效应进行了估计。 结果 校正之后该地总的高血压患病率为25.62%(95%CI: 22.66%～28.58%),而建档登记的患病率为10.85%。在该地的17个县(市)中,建档率最高为17.70%(95%CI: 14.65%～20.74%),最低为12.96%(95%CI: 8.09%～17.82%)。老龄化率与高血压患病率呈现正相关关系。城镇化率、当地老龄人口的规范化管理率、已建档高血压患者的规范化管理率与高血压患者建档率呈正相关关系。 结论 层次贝叶斯模型在校正低估数据,估计真实的患病率中效果优良,在校正低估从而获得真实的患病人数方面具有潜在的应用价值。

Application of a Bayesian hierarchical model in the estimation of chronic disease prevalence by correcting underestimation

Objective To explore the application of Bayesian hierarchical model in the estimation of chronic disease prevalence, and to correct underestimation and obtain the potential true number of patients. Methods It was assumed that the true number of cases followed a Poisson distribution, and the recorded cases followed a binomial distribution. A Bayesian hierarchical model was established to correct the incomplete recorded cases of hypertension in 17 counties(cities)from 2018 to 2020, combining the relevant influencing factors of hypertension prevalence and the completeness of recording. At the same time, the influence of factors related to the disease and recording completeness was estimated. Results After correction, the overall prevalence of hypertension was 25.62%(95%CI: 22.66%-28.58%), while the prevalence from the recording was 10.85%. In the 17 counties(cities), the registration rate ranged from a high of 17.70%(95%CI: 14.65%-20.74%)to a low of 12.96%(95%CI: 8.09%-17.82%). The rate of aging was positively correlated with the rate of hypertension. The rate of urbanization, the rate of standardized management of the aged and the rate of standardized management of hypertension patients were positively correlated with the hypertension prevalence. Conclusion Bayesian hierarchical model performs well in correcting underestimation data to estimate the true prevalence and has potential application value in correcting underestimation to obtain the true number of patients.