Epidemiology Survey Of Gastroesophageal Reflux Disease And Analysis Of Missing Data

【Objective】Gastroesophageal reflux disease (GERD) is defined as a condition that develops when symptoms, such as heartburn, are caused by the reflux of stomach contents into esophagus, which can also lead to esophagitis and extra-esophageal tissue impair, such as pharynx, throat and trachea. It is a common disease in western countries, but the prevalence in China is relatively low. But now the prevalence of GERD is increasing worldwidely; in Asia, the trend attracts more and more attention. Many countries have conducted large-scale epidemiology surveys to evaluate the prevalence of GERD, worldwidely, but there is no national population-based survey on the prevalence of GERD in China. Missing data is a common problem in epidemiology survey, but in fact, few studies deal with missing data. This study aimed to conduct a population-based survey, in the meantime, assess different imputation methods used to deal with missing data, and then evaluate the prevalence of GERD in China, estimate the affect of GERD on health related quality of life and sleepiness, and to investigate which demographic characteristics were associated with GERD, based on the the proper imputation methods.【Methods】Firstly, the pilot study was conducted in Shanghai, based on the investigation of 1200 randomly selected residents, to analyze the reliability and validity of the mandarin instruments in order to establish a valid, reliable and comparable epidemiology GERD survey instrument, and to evaluate the prevalence of GERD. Based on the outcomes of pilot study in Shanghai, the investigation was conducted in five cities of China (Beijing, Shanghai, Xian, Wuhan, Guangzhou) from April 2007 to January 2008, lasting for 10 months. A total of 18000 residents aged 18-80 years in both rural region and urban region were sampled using a stratified multiple-stage sampling method, following the sequence of district– block– residential area. A self-finished interview method was used as the subjects filled in the questionnaires by themselves. The trained interviewers provided explanation without inducement on any unclear questions. Instrument included basic information, mandarin version of Reflux Disease Questionnaire (RDQ), Modified ROMEⅡ. In addition, 20% subjects completed additional 36-item Short Form Health Survey (SF-36), Epworth Sleepiness Scale (ESS), and received physical examination. The subjects in Shanghai received blood test, and the volunteers underwent undoscopy. The performance of the interviewers was oversaw and coordinated by supervisors who examined questionnaires for any errors and ensured the quality of the survey. All valid questionnaires were doubly input into the database and cleaned.Based on the cleaned data, several variables were selected to set up complete data sets with different data types, and then mathematics models were established, with the parameters and standard errors estimated. Based on the complete data sets, data sets with missing data were established, with different missing mechanisms and different missing rates, and the simulation were done for 100 times. Then, several methods, including deletion method, dummy variable adjustment, conditional mean imputation, hot deck and multiple imputation, were used to deal with the missing data to get the means of the parameters and standard errors of the 100 simulation missing data sets. The comparison were done to find proper methods for different conditions. Finally, the survey data were analysed after proper imputation of missing data and then the prevalence of GERD was evaluated and the influence of GERD on health related quality of life and sleep were assessed.【Results】It was found that when the missing rate was low and the missing variables were continuous variables, deletion method was better, no matter the missing pattern, but when the missing rate was high, multiple imputation was better. When the missing variables were class variables, deletion method was the best, the outcomes of multiple imputation were not as good as the outcomes when dealing with continuous variables. If the missing data included both continuous variables and class variables, when the missing rate was low, deletion method was better, when the missing rate was high, multiple imputation was better, especially when imputed several times, but imputation times was not the more the better, sometimes the outcomes of imputation 10 times were better than those of imputation 20 times. When missing rate was 50%, the outcomes of several methods were not stable with serious fluctuating。When dealing with missing data in Likert-Type Scales, both item mean substitution method and multiple imputation were good, then was person mean substitution approach, deletion method was not as good as the the others. After analysis, accorrding to RDQ, based on the Montreal definition, the prevalence of GERD was 3.08%, the weekly prevalence of regurgitation was 4.23%, the weekly prevalence of heartburn was 1.83%. Compared to Shanghai, the prevalences of GERD were higher in Wuhan and Xian. Rural region had higher prevalence than urban region, and the prevalence was increasing with increasing age. Correlations between gender, job, marrital status, family income per month, smoking, drinking and GERD were not found. Although BMI was found correlated with GERD in many surveys, our survey didn't find that. In addition, functional dyspepsia, aerophagia, IBS, functional abdominal bloating, functional constipation, unspecified functional bowel disorder and gallbladder dysfunction were found related with GERD, the same as pilot survey. Compared with other surveys, the health related quality of life was higher than the others. The subjects with GERD had lower self-reported health conditions, GERD impaired health related quality of life in all eight SF-36 dimensions, and the significant clinical impairment were seen. The RP, BP, GH dimensions assessing physical health were affected most seriously, and RE dimension assessing mental health was also affected seriously. For the 3214 subjects who completed SF-36 questionnaire, 110 were diagnosed with GERD, and the higher the RDQ scores, the lower the SF-36 scores, which indicated that the seriou/frequent the GERD symptoms, the worse the quality of life. The ESS scores were found significantly different between the subjects with GERD and without, the subjects with GERD had higher ESS scores. We used the ESS criterion as follows: normal was defined as ESS score lower than 10, boederline as ESS score from 10 to 12, excessive sleepiness as ESS score from 13 to 24, and significant difference was found between the subjects with GERD and without, based on the criterion. Sleepiness were more common in subjects with GERD than in those without, 26.36% to 13.34%.【Conclusion】Multiple imputation is a good method used for imputing missing data, but it is not always the best one. The proper method should be selected cautiously according to the missing pattern and the actual condition. It is showed that the prevalence of GERD in China was not high, much lower than that of western countries. GERD affects quality of life seriously, and influences patients'sleep, leading to excessive sleepiness. People should pay more attention to the disease.