Establishment And Validation Of The Model Of Estimating Population Cause-Specific Mortality Fractions From In-Hospital Mortality

Background: At present, the coverage of all-cause mortality surveillance in China is very limited, only about 20% out of the whole country. The rest has no other data but hospital reports. In China, about 80% of the people died at home, of which a considerable portion without doctor's diagnosis of causes of death. If not querying according to standardized verbal autopsy, it is very difficult to ensure the correctness of the causes of death. Most hospital reports are highly reliable with credible diagnosis of the causes of death. As a developing country, China is not able to build a quality and population-based cause-of-death registration system in a short period of time.Objective: To build a multi-factor model based on analysis of factors affecting places of death, particularly in hospital or not. On the basis of existing hospital data, estimate the nationwide and regional cause-specific mortality fractions, and evaluate the effects of the estimates.Methodology: A multi-factor logistic regression model on factors affecting places of death will be built, based on data from the third retrospective survey on the cause-of-death in 2004, with whether died in hospital as a dependent variable, and diseases, individual factors and regional factors as independent variables. Verify the model using cause-of-death data of 2005 in the above-mentioned survey as verification data. Comparing the true cause-of-death distribution of the whole population, urban and rural, and eastern, central and western regions in China, with the inferred cause-of-death distribution from hospital data, determine the accuracy of the model, and explore the reasons for estimation errors. Use the established model and hospital mortality data to evaluate the 2008 population-based cause-of-death data from the national disease surveillance system. Assess the estimation effect of the model by comparing the estimated cause-specific mortality fractions of the model with the actual cause-specific mortality fractions, and also evaluate the accuracy of the routine surveillance of the cause-specific mortality fractions. Results: A multi-factor logistic regression model on factors affecting places of death is built based on data from the third retrospective survey on the cause-of-death in 2004, with whether died in hospital, and diseases, individual factors and regional factors as indicators. The average systematic errors of the model in terms of estimation of the whole country, urban and rural, and eastern, central and western regions are 10.46%, 7.98%, 17.14%, 19.46%, 11.70% and 13.46% respectively.The results of the verification of model over 2005 cause-of-death data show that the model has average relative errors over the whole country, urban and rural, and eastern, central and western regions at 11.59%, 7.34%, 17.23%, 18.92%, 11.39% and 14.16% respectively, which are close to the corresponding systematic errors of the model and the verification effect is good.According to the estimation of the model, in 2008, the average relative errors of the cause-specific mortality fractions of the whole country, urban and rural, and eastern, central and western regions are 13.99%, 12.17%, 19.72%, 18.69%, 10.69% and 18.93% respectively. With exceptions of the whole country, urban area and western area which are 3.65, 4.19 and 5.47 percent higher than the model's systematic errors respectively, the model's estimation errors are close its systematic errors. In general, the estimated cause-specific mortality fractions of the model is very close to the actual cause-specific mortality fractions, and the estimation effect is promising.Conclusion: A regression model on population-based cause-of-death data based on the data of in-hospital mortality can estimate the cause-specific mortality fractions of the whole country, urban and rural areas and the eastern, central and western regions with lower relative estimation errors; The study provides a population-based cause-specific mortality fractions estimation method to areas without all-cause mortality surveillance. Meanwhile, the methodology may, to some extent, evaluate the accuracy of population-based cause-specific mortality fractions, suggest missed reporting or incorrect classification of diseases, and find clues for on-site problems.