Analyzing a Statistical Method of Estimating Respiratory Deaths based on the Thailand Verbal Autopsy study

Abstract

     The archives record of the causes of death in Thailand using the death registration (DR) system is important source of mortality data. Over the past two decades, Thailand has presented many formats to enhance civil registration and vial statistics systems. More than 30% of death unregistered and about 40% of deaths registered have given the cause of death as “ill-defined”. The aim of this study was to propose a statistical model to estimate percentages of respiratory deaths in Thailand based on a sample of 9,644 deaths from the 2005 Verbal Autopsy (VA) study. Logistic regression was used to predict respiratory deaths classified by three factors, province, gender-age group and cause of each death. The receiver operating characteristic (ROC) curve was used to assess accuracy of the model prediction and the area under the ROC curve measures discrimination model which has ability to correctly predict those with and without respiratory deaths. Province, gender-age group and cause of death were statistically significant associated with respiratory deaths. The area under ROC curve was 0.7 with a false positive rate of 5.52% and sensitivity of 39.2%. Moreover, the results revealed that the under-reporting of respiratory deaths were those registered as tuberculosis, septicemia and other CVD cases. In conclusions, the logistic model in this study can be used for estimating the respiratory deaths from the DR database in Thailand or the DR system in other countries that are under-reporting the death rate.