Study population
This study used data from the Second National Sample Survey on Disability implemented from 1 April to 31 May 2006. This survey covered all provincial administrative areas in mainland China and aimed to describe the prevalence, causes, and severities of disability, as well as the living conditions and health service utilizations of the disabled. Multistage, stratified random-cluster sampling, with probability proportional to size, was used in 734 counties (districts), 2,980 towns (streets) and 5,964 communities (villages) from 31 provinces, autonomous regions, and municipalities under the Central Government in China. A total of 2,526,145 persons was randomly sampled from 771,797 households, representing 1.9 per 1000 inhabitants of China. Details on this survey have been published previously [19]. Because the typical age of onset for schizophrenia is in late adolescence or early twenties [20], we restricted our analysis to adults aged 18 years or older, and finally included 1,909,205 participants in this study. Figure 1 shows the flowchart of this study.
More than 20,000 interviewers, 6,000 physicians of various specialties, and 50,000 coordinating investigators administered this survey. In the pre-survey phase, households, populations, and suspected disabled people in all surveyed communities were investigated. Face-to-face interview was conducted with every family member in the selected households [21].
Study measures
Schizophrenia assessment
The outcome variable was schizophrenia, which was defined as a binary measure. Schizophrenia was identified by experienced psychiatrists using the International Statistical Classification of Diseases and Related Health Problems, 10th Revision (ICD-10) [22]. The ICD-10 diagnostic criteria had been employed in the diagnosis of schizophrenia among Chinese people and presented satisfactory validity in China [23].
This study used the World Health Organization Disability Assessment Schedule, Version II (WHO-DAS II) to assess the physical and social functioning among individuals with schizophrenia. According to the criterion of WHO-DAS II [24], physical and social functioning was consisted of understanding and communicating, physical movement, self-care, getting along with people, life activities and participation in society. The severities of the functioning were evaluated in Likert scales and were classified into five degrees: without difficulty (WHO-DAS scores <52), mild difficulty (WHO-DAS scores <96 and ≥52), moderate difficulty (WHO-DAS scores <106 and ≥96), severe difficulty (WHO-DAS scores <116 and ≥106) and extremely severe difficulty (WHO-DAS scores≥116) [24].
Measures
The independent variable was education, which was classified into 3 categories: primary school and below, junior high school and senior high school and above. According to previous findings [25, 26], we considered demographic characteristics and socioeconomic conditions as potential confounders. Demographic characteristics were consisted of gender, age (continuous variable), marital status and residence. Of these, gender was defined as either of the two sexes (male and female), which considered with reference to social and cultural differences rather than biological ones between males and females. Age was continuous variable, and marital status (married/unmarried) and residence (urban/ rural) were both dummy variables. Socioeconomic conditions were evaluated by household income per capita and employment status. Of these, employment status was defined by 2 categories: employment and unemployment. Household income per capita were divided into 3 groups based on tertiles, with the first tertile being the lowest group (0-1998 yuan), the second tertiles being the moderate group (2000- 3999.8 yuan) and the third tertile being the highest group (4000-9999 yuan).
Statistical analysis
Descriptive statistics were used to describe and compare the characteristics of participants by gender. Logistic regression models were used to evaluate the association between education and schizophrenia, and the odds ratios (ORs) with 95% confidence intervals (CIs) were calculated. Each regression model was controlled for age, gender, residence, marital status, employment status and household income per capita. We presented the results from 2 models: model 1 included ORs adjusted for demographic and socioeconomic characteristics, and model 2 with further adjustment for the interaction between gender and education. A P-value less than 0.05 was considered statistically significant. The software Stata version 13.0 for Windows (Stata Corp, College Station, TX, USA) was used for statistical analysis.