Health related quality of life measured by SF-36: a population-based study in Shanghai, China

Shanghai is the biggest city in eastern China. It consists of 18 districts and 1 county that are geographically divided into 3 strata as urban, suburban, and rural regions. Using stratified multiple-stage sampling method, 4 residential areas in the urban region, 2 villages in the rural region, and 3 residential areas in the suburban region were selected following the sequence of district-block-residential area. A total of 1200 subjects older than 18 years of age were randomly sampled from those areas and 1034 subjects actually answered the questionnaires. The sample had 362 respondents from the Huangpu district, 336 from the Pudong district and 336 from the Songjiang district. The overall response rate was 86.17%. In order to analyze the reliability of the results, 10% of the total number of the respondents (i.e., 120 respondents) were randomly selected to take a retest by filling in the questionnaires again 2–7 days after the baseline test. At last 113 subjects took the retest. All respondents signed a written informed consent before participation.

The questionnaire included general information and Mandarin version of SF-36 [See additional file 1: Questionnaire-bilingual]. General information was collected on age, sex, resident region, nationality, marital status, educational level, current job, family monthly income, height, weight, tobacco use, alcohol use, and frequency of activities. Body Mass Index (BMI) was calculated from height and weight. Since the current WHO BMI criterion is suitable for Caucasians rather than Asians, additional BMI categories for Asian populations are recommended by WHO [11]. Many Asian countries have also developed their own criterions, such as Japan [12]. We used the Chinese BMI criterion as follows: underweight was defined as BMI lower than 18.5 kg/m², healthy weight as BMI from 18.5 to 23.9 kg/m², overweight as BMI from 24 to 27.9 kg/m², and obesity as BMI of 28 or more kg/m²[13]. In addition, respondents were asked whether they had been diagnosed by physicians with the following chronic conditions: hypertension, ischemic heart disease, cerebrovascular disorder, diabetes, chronic obstructive pulmonary disease (COPD), asthma, renal disorder, liver disorder, rheumatoid arthritis, osteoarthritis, anxiety, and depression, and at which age the disease had been first diagnosed. The use of medicine at the time of the interview was also recorded.

The Mandarin version of SF-36 was translated from the IQOLA SF-36 Standard UK Version 1.0 by the experts of Zhejiang University, China. Its reliability and validity have been tested in the survey of Hangzhou, the capital of Zhejiang Province, southeast of Mainland China [10]. It was a brief self-administered questionnaire that generated assessment scores across eight dimensions of health: physical function (PF), role limitations due to physical problems (RP), bodily pain (BP), general health (GH), vitality (VT), social function (SF), role limitations due to emotional problems (RE), mental health (MH), and one single item dimension on health transition. The SF-36 dimensions can also be divided into two categories: Physical Component Summary (PCS) and Mental Component Summary (MCS), which represent the physical functioning and wellbeing, and emotional wellbeing, respectively.

The survey was conducted from November 2005 to January 2006, using a self-finished interview method. Respondents filled in the questionnaires by themselves in their household or in local resident committees. The interviewers were social workers on the site who provided explanation without inducement on any unclear questions. Ten percent (10%) of the respondents had a repeatable accuracy check by filling in the questionnaires by themselves again a week later. The interviewers who interviewed the same respondent in the second time were different from the initial ones for the purpose of quality control. The performance of the interviewers was oversaw and coordinated by supervisors who examined questionnaires for any errors and ensured the quality of the survey. A valid questionnaire was the one that had been audited and signed by a supervisor. Both supervisors and interviewers were trained by the experts from Changhai Hospital and the epidemiologists from Second Military Medical University (SMMU).

All valid questionnaires were doubly input into the database by two independent professional data processors in the Department of Health Statistics of SMMU using software EpiData 3.1. Both manual checking and computer checking were conducted to find discrepancies.

In the Pudong District, a total of 112 respondents' questionnaires were withdrawn from the statistical analysis due to one facilitator's failure to adhere to the study protocol. Three questionnaires from the Huangpu District were excluded because more than 80% items were missing. Therefore, after the data checking and validation, 919 effective questionnaires were used for data analyses in this study. Of the 113 respondents who agreed to be re-interviewed, 14 questionnaires were rejected because they were not completed in line with the study protocol, resulting in 99 questionnaires for the retest analysis.

The missing values in the SF-36 dimensions were imputed as follows: if 50% or more items in one dimension were completed, the mean value of the completed items was used to impute the missing values. If more than 50% of the items were missing, the dimension score was excluded from the statistical analysis. In our survey, the item response rates were actually quite high. The average item response rate of the general information was 98.81%; the average response rate of the 36 items in SF-36 was 99.67%, ranging from 98.80% to 99.89%.

The items and dimensions in SF-36 were constructed using the Likert method of summated ratings. The raw score of each of the eight SF-36 dimensions was derived by summing the item scores, and converted to a value for the dimension from 0 (worst possible health state measured by the questionnaire) to 100 (best possible health state). The raw score was then re-calculated across the dimension as follows:

The PCS and MCS scores were calculated using the standard scoring algorithms [14‐17].

The SF-36 questionnaire was evaluated by reliability and validity. Split-half reliability was computed by correlating the scores of the odd half with those of the even half in each dimension of SF-36. Test-retest reliability was assessed by the differences between test and retest scores using a paired-sample t test. It was further assessed by intraclass correlation coefficient (ICC). A questionnaire with ICC value larger than 0.7 was usually considered satisfactory [18]. Internal consistency of the SF-36 items was assessed by Cronbach's α coefficient. A Cronbach's α value of 0.7 or higher was generally considered to be sufficient to demonstrate internal consistency [18]. Construction validity was assessed by correlation analysis and factor analysis using principal component analysis and quartimax rotation. Factor loadings larger than 0.50 within a particular dimension were considered to support its factor construction. The cumulative variance proportion was used to indicate the contributions of the factors [19].

Statistical Analysis System (SAS) 9.1.3 and SPSS 10.0 were used for analyzing the survey data. Student t test, analysis of variance, and multivariate stepwise regression were applied to investigate the impact of various risk factors on quality of life.

A total of 919 subjects were utilized in the statistical analyses, including 509 female and 410 male. The age of all subjects ranged from 18 to 77 years with a mean age of 47 ± 13 years.

Table 2 showed the normative values of the SF-36 dimension scores by age and sex groups. The quality of life was reduced with increasing age. Female had lower scores than male in almost all subgroups, but in some subgroups female did report a better mental health. The SF-36 dimension scores were compared among different Chinese populations (Table 3): (1) every dimension score of the Shanghai population was higher than those of both Hangzhou population and American Chinese [10, 21]; (2) the Shanghai population had higher scores than those of the Sichuan population in the RP, BP, SF, RE, and MH dimensions, but were similar to the Sichuan population in the PF, GH, and VT dimensions [9]; and (3) the SF-36 dimension scores of Hong Kong, Taiwan, American, and Canadian were lower than those of the Shanghai population in six of the eight dimensions, except GH and PF [22‐25].

Analysis of variance was first used to select the risk factors of HRQL. Region, gender, current job, age, current marital status, highest level of education, total income of family per month, frequency of activities, BMI, and chronic diseases were found influencing at least one dimension of SF-36. Multivariate stepwise regression was then applied using the SF-36 dimensions as the dependent variables and the risk factors mentioned above as the independent variables. The statistical significance level was set at 0.15 for both inclusion and exclusion of the independent variables in the stepwise process. The results indicated that the risk factors were different among the eight SF-36 dimensions (Table 4). Chronic diseases were evidently the most common risk factor reducing the scores of all SF-36 dimensions (p-value < 0.05). Its influence was relatively strong because its standardized regression coefficient was the largest one (in absolute value) for most of the SF-36 dimensions. Increasing age reduced quality of life in the PF, BP, GH, and VT dimensions, while manual work led to worse scores in the GH, VT and SF dimensions (p-value < 0.05). In contrast, high family income led to a good quality of life in the RE and MH dimensions. Similarly, frequent activities increased the quality of life in the PF, GH, VT, and MH dimensions (p-value < 0.05). The quality of life was improved with increasing education level in the RP dimension (p-value < 0.05). The unmarried had better quality of life in the VT and MH dimension (p-value < 0.05), but its standardized regression coefficients were quite low. The impact of gender and BMI was not statistically significant (p-value > 0.05). On the other hand, the impact of region was unclear: (1) living in the suburbs led to a better quality of life in the PF, GH, and VT dimensions, while the impact was negative in the SF, RE, and MH dimensions; and (2) living in rural regions resulted in a better quality of life in the RP, BP, and VT dimensions, but the impact was negative in the MH dimension with a high standardized regression coefficient.