Regional Misclassification Of Cancer Incidence Registry In Iranian Provinces And Bayesian Correction
Hajizadeh NASTARAN , Shahid Beheshti University of Medical Sciences, Iran
MOHAMAD AMIN P. 2
,
AHMAD REZA B. 1
,
ALIREZA A. 1
1 Department of Biostatistics, Faculty of Paramedical Sciences, Shahid Beheshti University of Medical Sciences, Tehran, Iran.
2 Gastroenterology and Liver Diseases Research Center, Research Institute for Gastroenterology and Liver Diseases, Shahid Beheshti University of Medical Sciences, Tehran, Iran
Purpose: Most cancer patients throughout Iran seek diagnostic and medical treatments in neighboring facilitate provinces due to lack of proper facilities in their own residence and don’t mention their permanent residence address. It makes misclassification error in cancer registry data. The aim of this study is to estimate the misclassified rate between neighboring provinces, using Bayesian method.
Methods: For this study gastric cancer data was extracted from Iranian annual of national cancer registration report in 2008. To correct the misclassification effect between each two neighboring provinces, a Bayesian approach was used with Poisson count regression. An informative beta prior distribution was assumed for the misclassified parameter and expected coverage of each province was used as prior for that parameter. Because the misclassified parameter is unknown, a latent variable approach was employed to simplify the full conditional models and estimate the posterior distribution using a Gibbs sampling algorithm. Analyses were carried out using R software version 3.2.0.
Results: After implementation of the Bayesian method, it was estimated that 43% of gastric cancer patients from North and South Khorasans were registered in Razavi Khorasan, 41% from Kohgilouye&boyerahmad in Fars and 8% in Isfahan, 36% from West Azerbaijan in East Azerbaijan, 43% from Golestan in Mazandaran, 46% from Ilam in Kordestan and 28% in Kermanshah, 63% from Hormozgan in Fars, 47% from Bushehr in Kerman, and finally 58% of cancerous from Sistan&balouchestan were registered in Yazd.
Conclusions: Policy makers that employ cancer registry data for programming should notice to regional misclassification, otherwise decisions for cancer control and prevention and allocating the facilities to different provinces will be erroneous. So it is needed to improve the registration system accuracy with employing more expert stuffs, refining foundations, and enhancing hardware and software resources.