Chinese Health and Family Life Survey

With the exclusion of Tibet and Hong Kong, the sample is nationally representative of the adult population of China aged 20 to 64 years. This population sample was drawn probabilistically in 4 steps following standard procedures for complex samples. First, using the 1990 national population census and public health reports of STD infection rates in different provinces and cities, China was divided into 14 strata based on size of urban population and location on the southern and eastern coasts (where STD infection rates have been reported to be high). To capture higher STDs in some regions, coastal regions and large cities were oversampled using known population weights. Second, 2 to 6 administrative units (urban districts, smaller cities, and counties) were selected from each stratum, with the probability of the unit being selected being proportional to the population of that unit. These provided 48 primary sampling units. Third, on arriving at a sampling unit, each survey team arrayed the subunits in the county or city by population size and again picked 1 to 2 subunits (neighborhoods in cities, villages or towns in counties) probabilistically, with more highly populated subunits having a greater probability of being selected. This produced a total of 60 sample communities. Fourth, using the official community registers of households and temporary migrants, the adult population aged 20 to 64 years was arrayed in order. Starting with a randomly chosen person from this list, individuals were picked at fixed intervals to produce approximately 83 individuals per community (5000/60 communities = 83).

The 60 interview sites (villages and urban neighborhoods) were distributed as shown in Figure 1. Except for Tibet, all of the 31 provincial-level municipalities (Beijing, Shanghai, Tianjin, and Chongqing) and provinces had a chance of being included in the sample. Each of the provincial-level units can be ranked by 1996-2000 average prevalence in the national STD reporting system for the 8 STDs mentioned earlier. With the exception of Chongqing, the sample included 2 or more sample sites in each of the 8 provinces with the highest prevalence levels for 8 reportable STDs. Thirteen provinces were not represented in the sample. Tibet was intentionally omitted because of the sparse population and travel difficulties, and 12 happened not to be selected by chance. Of the 13 not included, 10 ranked below the median STD prevalence for all provinces. These include Yunnan (southwest) and Xinjiang (far west) provinces, which had many IDU-related HIV cases but only modest levels of STDs. The 3 unrepresented provinces with above-median STD prevalence were Chongqing Municipality and Jiangxi and Guangxi provinces, with Guangxi having a high rate of IDUs and cases of HIV.

A computerized interview, based in part on the 1992 US National Health and Social Life Survey, was pretested in China in 3 field trials. Also, 50 husband-wife pairs were given shortened versions of the questionnaire. Statistics on their agreements about shared sexual behavior produced values that averaged 0.35, modest but in the same range as agreement about other aspects of family life such as spousal violence and relative social status of parents (average = 0.27). Another 50 respondents had repeat interviews after a gap of 2 months. The 21 items about sexual behavior had average agreement values of 0.75 when the same items were compared across the 2 separate interviews.

Mostly mid- to late-career social workers and researchers in their 40s and 50s (a total of 39 interviewers) had 1 week's training, and then continued with the project for most of a year from August 1999 to August 2000. In the field, interviewers were of the same sex as the respondent.

For the sake of privacy, interviews took place away from the respondent's home. In large cities, these facilities were typically private rooms in a neighborhood hotel. In villages and smaller towns, these facilities were rooms in a larger home or in a village meeting place. Interview team members approached individuals who had been drawn in the sample, explained the purpose of the interviews, and read them an informed consent statement approved by the institutional review boards. Oral and computer-entered consents were obtained prior to the hour-long interview, which began with the interviewer in control of the computerized interview and continued with the computer controlled entirely by the respondent. Only 13% (often older women in the countryside) needed continued interviewer assistance in this last portion of the interview. Respondents were also asked to provide a urine sample. The methods were approved by institutional review boards at the University of Chicago, Chicago, Ill; Renmin University, Beijing, China; and Peking Union Medical College, Beijing, China.

When weighted according to sampling fractions, the data set of successful interviews had too few individuals in their 20s and too few in their 50s and 60s, as judged by national census figures and annual population surveys conducted by the government. Accordingly, analysis weights were further adjusted to make the age distribution approximate what is found in these other data sources. Since those who failed to provide a urine sample were only marginally different from those who did, judged by tests of significance, we made no further adjustments among those who provided urine. After adjusting weights for both sampling fractions and age distributions, percentage distributions for urban residence, age, and education closely paralleled data available in the national census and other national-level statistics, thereby implying that the adjusted sample is representative of the total adult working-age population.

Urine specimens were frozen and taken to a central laboratory where assays were performed within 30 days. Neisseria gonorrhoeae and C trachomatis were measured with ligase chain reaction (Abbott Laboratories, North Chicago, Ill), according to the manufacturer's instructions. The testing was done without further confirmation. Cleaning and maintenance of instruments were performed as instructed by the manufacturer. Notification of results was sent by mail. However, confidentiality in the mail system was a concern. To reduce the number of individuals who would have access to the names of participants in the survey, only participants with positive results were notified if they had requested notification at the time of interview. In the notification, participants were referred to their local public health clinic with a description of their infection and the appropriate treatment.