| BackgroundThree mortality surveys had been conducted since1980s in order to investigate the regional distribution and population distribution of death causes and provide evidence for the disease control programs and healthcare strategies. However, if we could analyze the association between the exposure and the death in the process of describing the death distribution, the information of the mortality survey could be used fully. However, it was difficult to conduct a population-based case-control study in a large-scale mortality survey to explore the association between the exposure and disease. At present, the common solution is the proportional mortality ratio (PMR) method, namely, choosing the dead subjects who die of causes that are probably related with the exposure as cases and the dead subjects who die of causes that being not related with the exposure as control. The strength of PMR is that the bias possibly influencing exposure information of the case and control group is similar. But if the process of control selection is possibly associated with the exposure, the risk between the exposure and the death would be underestimated, which is the main limitation of PMR method. Besides, it seems to be inappropriate if the exposure information of the deceased people was used to estimate the exposure information in the base population.Smoking is a proverbial public health problem which could increase the occurrence and death of several cancers. And China has been the largest tobacco-manufacturing and consuming country. The Chinese tobacco-consuming take up30%of the gross amount of the world. There are0.35billion smokers in China which taking up25%of the whole smokers over the world. So the tobacco will do more harm to Chinese in the future. The peak of the tobacco-consuming in the west countries are40years earlier than that in China. So the studies on the effect of tobacco on the death of several diseases are earlier than that in China. But China was still lack of the data about the harm of tobacco on the disease occurrence and death. In order to make the Chinese to know of the harm caused by tobacco and provide data support for tobacco control strategy, a nationwide large-scale retrospective epidemiological study was conducted for the first time to evaluate the association between the smoke and disease mortality. Professor Liu Boqi put forward a new design innovatively, that is, cases-living spouses crossover design. In this design,67million people, which covered by the nationwide mortality, was defined as base population. All the subjects who died between1986-1988in the base population were defined as case population. And the contemporaneous living spouses of all the deceased subjects were defined as control population. The living spouses provide not only smoking information of the deceased person, but also provide the smoking information of themselves. The design was based on two hypotheses:(1) all the living spouses were regarded as an approximately random sample in the base population.(2) the smoking habits of a couple were independent.The design successfully broke through the problem of control selection for a population-based case-control study in a large retrospective mortality survey and we could evaluate the association between the smoke and several diseases in the novel design. Since the completion of this study the UK and Chinese scientists has separately used PMR method and cases-living spouses crossover method to mine the data and completed a series of reports. However the quantitative evaluation of the differences between the two methods will be the focus of current research.The current study adopted case-living spouses crossover design to evaluate the risk of smoke on the deaths from ten main kinds of cancer among male aged35years or above in the24cities. Meanwhile, the equivalence evaluation methods (parameter interval method and bootstrap percentile method) which were widely used in the bioequivalence and clinical equivalence study were introduced to evaluate the equivalence of two point estimates obtained from PMR method and the case-living spouses crossover method and the consistency of the two control groups could be evaluated. Finally, we conducted a comparative research for the two equivalence evaluation methods to assess the effectiveness and credibility of the result.Objectives:Using living spouses crossover design to estimate the risk of smoking for10major malignant cancer deaths among men aged35years old or above who died between1986-1988in the24large and middle-sized cities in the national mortality survey that was conducted between1989-1991in China. Evaluate the rationality and validity of the living spouses crossover design and provide flexible and reasonable reference for the control selection in the future population-based case-control study. Contents:Calculate the mortality rate as well as the smoking-specific mortality rate of the10primary high incidence cancer(esophagus, stomach, liver, lung, pancreatic, prostate cancer, bladder cancer, small parts cancers(combing lip, oral cavity and laryngeal)), and describe the regional distribution of the primary cancers deaths in the base population. Meanwhile, risks of smoking for primary cancer deaths have been estimated using case-living spouses crossover design.Evaluate the rationality and validity of living spouses crossover design by defining two control groups (defined by the proportional mortality ratio analysis method and case-living spouses cross-over design respectively) and use equivalence test to evaluate the consistence between the two control groups. Evaluate the validity and stability of the statistics obtained using the new design under various sample sizes. Assess the sensitivity and specificity of the two equivalence test method employed in the study under various sample sizes and exposure rates.Methods:24major cities are included in our study. Which are Beijing, Tianjin, Shanghai, Harbin, Hangzhou, Changsha, Yellowstone, Xian, Lanzhou, Chengdu, Chongqing, Zigong, Gui-yang, Guangzhou, Nanjing, Fuzhou, Kunming, Gejiu, Yangquan, Changchun, Jilin, Shen-yang, Dalian and Zibo.The case group in our study is defined as all men aged35years old or above who died of10primary high incidence cancer((esophageal cancer, stomach cancer, liver cancer, lung cancer, pancreatic cancer, prostate cancer, bladder cancer, and small parts of the cancer (combined lip, oral and laryngeal cancer)) between1986and1988; Evaluation control group is defined as living spouses (aged35years old or above)of the women who died of any causes between1986and1988; Reference control group is defined through proportional mortality ratio analysis method (death causes included: infection and parasitic diseases, endocrine, metabolic, immune disorders, blood and blood-forming organs diseases, mental disorders, nervous system diseases, digestive diseases, diseases of the genitourinary system, musculoskeletal and connective tissue disease, injury and poisoning and other diseases).The main indices in our study are the city, age and smoking status-specific mortality as well as age-standardized mortality for lung cancer, liver cancer, stomach cancer, esophageal cancer, bladder cancer, prostate cancer, pancreatic cancer, and small parts of the cancer. The main indices also include the risk of smoking on the deaths from above mentioned cancers.Using logistic regression to get the risk estimates of smoking on the deaths from primary cancers as well as the95%confidence intervals of the risk estimates of reference control group and evaluation controls groups respectively after the adjustment of age and cities. The risk estimates for reference control group and evaluation control group were denoted as OR1and OR1.The PSR (Proportion of Similar Response) method was employed to establish the equivalent limit in our study. The consistence between OR1and OR2was evaluated by using bootstrap percentile method and parametric confidence interval method respectively. Meanwhile, the bootstrap re-sampling method was employed to estimate the correlation between OR1and OR2to overcome the obstacle in the process of equivalence test which caused by the correlation between OR1and OR2.We also used the bootstrap re-sampling method to evaluate the validity and stability of the risk estimation of living spouses crossover control design under various sample sizes. In addition, sensitivity and specificity of the two equivalence test methods under various sample sizes and exposure rates were assessed by using Monte carlo simulation method.Results:(1) The smoking prevalence for the case group, reference control group and evaluation control group was69.2%,56.5%and56.7%respectively. And the smoking prevalence has an obvious regional difference in our study. The top five cities with highest smoking prevalence in the evaluation control group are Kunming/Gejiu, Zigong, Guiyang, Guangzhou, Chengdu/Chongqing, which are important cigarette-producing regions in China. The results of this study also revealed that the mortality of lung cancer in the base population was9.09/ten thousand, which is the highest among the primary caners. The mortality for the other cancers were (from the highest to lowest):gastric cancer (7.74/ten thousand), liver (6.84/ten thousand), esophageal cancer (3.43/ten thousand), cancer of the small parts (1.07/ten thousand), pancreatic cancer (0.83/ten thousand), bladder cancer (0.54/ten thousand) and prostate cancer (0.19/ten thousand). Besides, the regional characteristic of cancer mortalities varies among different cancers. The mortalities for lung cancer, stomach cancer and liver cancer in the eastern regions are higher than that in the central and western regions in China. While the mortality for esophageal cancer in the western regions are higher than that in the central and eastern regions in China.(2) Although the smoking prevalence and cancer mortality vary among different regions the association between smoking and primary cancer deaths were significant statistically in all the cities. However, the regional distribution of the risk of smoking on primary cancer deaths was inconsistent with the regional distribution of the mortalities for primary cancer. That is, for the city with relative high incidence of a primary cancer, the risk of smoking on primary cancer deaths was not necessarily high(such as lung cancer deaths in Harbin). While for the city with high risks of a primary cancer death associated smoking, the mortality for the cancer was not necessarily high(such as gastric cancer deaths in Guangzhou).(3) The results of equivalence test for the two control groups revealed that a relative high consistence can be found between the two control groups when estimating the association between smoking and primary cancer death. And the equivalence conclusions for the equivalence test can be obtained among different cancer. In addition, the subsets were created according to specific smoking information (age begin smoking, types of smoking and number of cigarettes per day). Although statistical significance can be found between the estimates obtained from two control groups under the traditional hypothesis test for some subsets the90%confidence intervals were almost entirely in the equivalent limit.(4) Two control groups were adopted to estimate the relative risk of death from primary cancer for smokers relative to non-smokers under various sample sizes. The results revealed that the risk estimates of the two control groups maintained high consistency. And with the increase of the sample size, the values of odds ratios as well as the gaps between the odds ratios became stable.(5) Monte Carlo simulation results revealed that both of the equivalencet test that employed in the current study showed a relative high sensitivity and specificity, which supported the reliability of the conclusions in our study form a methodological aspect. When the theoretical equivalence was established and the exposure rate remain constant, the probabilities that the equivalence interval was contained by the equivalence limit increased with the increase of sample for the two kinds of equivalence test methods, but the probabilities decrease with the descrease of exposure rate for the two equivalence test methods. Besides, the probability of parameter interval method was higher than that of bootstrap percentile interval method, that is, the power of the parameter interval method was higher than that of bootstrap percentile interval method.ConclusionsIt is valid and appropriate to take the gender-matched living spouses of the deceased people as an alternative control in the population-based case control study. We could get credible result with this new design under the conditions of limited time and funds.InnovationThe equivalent evaluation methods applicable to the clinical independent random sample were applied to epidemiological case-control study to evaluate the consistency of different control groups. We broke through the technical bottlenecks of the bioequivalence evaluation, that is the sample correlation caused by the same set of cases and two different control groups.After the bioequivalence evaluation of the two control selection methods the study evaluated the validity and stability of the risk estimates of the two different control selection methods under various sample size. And no similar study has been reported upon now. |
PDF Full Text Request