Study population and participants
Data used in the study were from a cross-sectional health behavior survey conducted in Wuhan, China. The target population of the study was rural-to-urban migrants who were 18-45 years old, who had a legal rural residence, stayed in the current city for at least one month, and the main purpose of migration was to earn money.
Non-migrant rural and urban residents were selected as comparison groups. Rural residents were people who aged 18-45, had legal rural residence, and had not migrated to any cities in the past 12 months. Non-migrant urban residents were people who were also within the same age range, had legal urban residence, and had lived in the current city for five years or more. All participants were selected in Wuhan City, the capital of Hubei Province in the Central China. Wuhan has a total population of 10 million and per capita GDP of $17,000 in 2015; and the estimated total number of rural migrants was greater than one million.
Probability sample selected using GIS/GPS-assisted method
Selection of rural migrants
Study participants were selected using the GIS/GPS-assisted sampling method (18) (see Figure 1). To sample the rural-to-urban migrants, four stages were used. First, the targeted urban area of Wuhan was divided into mutually exclusive 100×100m geographic units (named as geounits). The primary sampling frame (PSF) was then constructed using these geounits after exclusion of all non-residential geounits, such as rivers, lakes, mountains, factories, streets and other public places. Second, a total of 60 geounits were randomly selected from the PSF stratified by the seven Districts of the city. Third, a research team consisting of a project coordinator, a GPS expert and a research staff went to the field to physically locate each selected geounit with the assistance of a GPS receiver. Fourth, on a pre-scheduled date and time, a data collection team were dispatched to the located geounit, one at a time, to recruit participants and collect data.
From each selected geounits, only 20 participants (10 male and 10 female) were randomly selected which was pre-determined from a pilot study. This step was completed through random route technique with a natural marker, and then enumerated the selected households to form the secondary sampling frame (SSF). With the SSF, one participant per gender was selected for one household. For household with more than one eligible participants, the Kish Table would be used to randomly select one.
Selection of non-migrant urban and rural residents.
For effective comparison purposes, non-migrant urban residents were selected from the same geounits where the rural migrants were sampled and recruited using the exact the same method described above.
It is practically difficult to select non-migrant rural residents from where the rural migrants in this study come from, because the rural migrants included in this study came from almost all other parts of China. Since most of the migrants in Wuhan come from the surrounding counties, the following protocol was used to sample non-migrant rural residents. First, the target geographic area for rural residents was defined as a band region surrounding Wuhan with the inner radius of 50 kilometers and outer radius of 75 kilometers, covering the place where most rural migrants in Wuhan originated from. Second, the band area was divided into mutual exclusive, 1 km×1km geounits; and the PSF was thus formed after exclusion of the non-residential geounits. Third, the band range was divided into 40 strata using five co-centric bands with 5 kilometers apart and four lines (north-south, west-east, northwest-southeast, and northeast-southwest). Fourth, one geounit was randomly selected from each stratum with a total of 40 geounits. Fifth, the households within the selected geounit formed the SSF, and 30 participants (15 male and 15 female) per geounit were randomly selected from SSF.
<Insert Figure 1 here>
To ensure enough participants enrolled, additional 20% geounits were included. Written informed consent was obtained from all participants. Among the approached 4,215 eligible participants, 6% refused to participate with a final 3951 participants enrolled.
Data collection was completed during 2011 to 2013. The Migrant Health and Behavior Survey, a pilot-tested survey was delivered to the participants through the Audio Computer-Assisted Self Interviewing (ACASI). The survey was anonymous and confidential. Participants were asked to complete the survey independently in a designated room either in the participants’ home or in the local community health center. When completion of the survey, each participant received a $6 reward. The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. All procedures involving human subjects/patients were approved by the Institutional Review Board (IRB) at Wayne State University and Wuhan CDC, and the data analysis was approved by the IRB at University of Florida.
Suicidal ideation was measured based on the response to the question: “In the past 12 months, have you thought about killing yourself?” with answer options of “1=Yes, 0=No”.
Suicidal attempt was measured based on the response to the question: “In the past 12 months, have you tried to kill yourself?” with answer options of “1=Yes, 0=No”.
Socio-economic variables in the study include age (in years, and categorized into 25 or less, 26-35, >35), gender (male/female), marital status (married/unmarried/divorced/widowed, and categorized into married/not married), education attainment (primary or less/middle school/high school/college or more), monthly income (<1000/1000-2000/2000-4000/>4000), number of family members (3 or less/4-5/6 or more), if unemployed in the past 12 months (yes/no), and the intention to move in the next six months (likely/unsure/unlikely).
Descriptive analysis (frequency, proportion, mean and standard deviation) was used to describe the study sample and to assess the prevalence rate of the two suicidal behaviors. Survey method for descriptive statistics were used to obtain point estimates for means and prevalence rate and 95% confidence interval (CI). The survey methods are needed to consider the complex GIS/GPS-assisted probability sampling design, including stratification, clustering, unequal probability, and sample weights. The estimated 95% CI of the estimated prevalence rates was used for comparison with no overlap of two estimated 95% CI as the evidence of significant differences at the level of p <0.05.
Second, to compare the difference in the suicide risk between migrants and non-migrant rural and urban residents while controlling for demographic factors, the survey method of multivariate logistic regression was used. We first compared the rural migrants with rural residents; followed by the comparison of rural migrants with urban residents. Adjusted odds ratio (AOR) with 95% CI was estimated, and the 95% CI not covering one indicates significant differences at the level of p <0.05.
All statistical analyses were conducted using SAS 9.4 (SAS Institute, Cary, NC). PROC SURVEYMEANS were used for descriptive analysis, and PROC SURVEYLOGIUSTIC was used for multivariate logistic regression analysis.