Influence of health-related quality of life on health service utilization in Chinese rural-to-urban female migrant workers

A cross-sectional survey was carried out in 2013 in three factories in the Shenzhen-Dongguan economic zone in China. The study was aimed to evaluate HRQOL and health service utilization in rural-to-urban women migrant workers in light-duty large factories (>1000 employees) without strong occupational health risk hazards. With an estimated annual hospitalization rate (a primary outcome) of about 6.0%, allowing a maximal deviation of 1.2%, at an alpha error of 5%, a minimal sample size of 1204 subjects was required. The required sample sizes would have been smaller if the sample size calculations were based on other outcomes (2-week physician visit or HRQOL scores). Three factories (a textile manufacturer, a furniture assembler and a camera assembler) were approached and accepted to allow the implementation of the study protocol. Women in the three factories were all light-duty manual workers. The study subjects were selected by a cluster sampling method. We randomly sampled 20 of 61 workshops (about 1/3) in the three factories. All consented eligible subjects in the sampled workshops were recruited (n = 1438 rural–to-urban female migrant workers). Data were collected via face-to-face interview by trained study personnel. The Research Ethics Board of Guangzhou Medical University approved the study. Written informed consent was obtained from the study participant, or from a parent or legal guardian of the participant if her age was <18 years (a few participants of 16–17 years old).

Well-trained interviewers (medical students from Guangzhou Medical University) collected research data through face-to-face interviews using structured study questionnaires. The interviewers received training and engaged in group discussions and simulated interviews to standardize data collection and recording procedures.

The general study questionnaire included information items on socio-demographic characteristics, health-related factors and health service utilization. Socio-demographic variables included age, marital status, education, medical insurance and duration of employment (years). Health-related variables included body mass index (BMI), any diagnosed chronic disease (yes/no), and recent-month self-reported morbidity (diagnosed disease, or self-perceived disease due to symptoms). BMI was calculated as weight/height² (kg/m2), and categorized as “underweight” (<18.5), “normal weight” (18.5-23.9), “overweight” (24.0-27.9), and “obese” (≥28.0), according to the Chinese BMI reference standards [29]. Recent-month self-reported morbidity was obtained in the question “Have you been ill over the past month?”, an indicator of diagnosed disease plus self-perceived disease (due to uncomfortable symptoms). Health service utilization variables included recent two-week physician visit and annual hospitalization. “Recent two-week physician visit” was obtained in the question “In the past two weeks, have you ever visited a doctor?”. Annual hospitalization” was obtained in the question “In the last one year, have you been hospitalized?”.

The Chinese version of the Short Form (SF-36) Health Survey, which was translated from the standard English version of SF-36, showed satisfactory construct and clinical validity and internal consistency for measuring HRQOL [30-32]. The SF-36 was a generic measure of HRQOL in both physical and mental domains [33-37]. The questionnaire comprises 36 questions covering eight domains: four in the area of physical health including physical functioning (PF), role physical (RP), bodily pain (BP) and general health (GH), and four in the area of mental health including vitality (VT), social functioning (SF), role-emotional (RE) and mental health (MH). Both physical and mental health scores have been empirically validated [38,39]. Total score in each of the eight domains can be converted into 0–100 scale; higher scores indicating better HRQOL. SF-36 has been widely accepted as a valid instrument for measuring HRQOL in the general population and in those with specific health conditions due to its sound psychometric properties, brevity and comprehensiveness [40-42].

Statistical analyses were performed using Statistical Package for Social Sciences (SPSS), version 13.0 (SPSS Inc., Chicago, IL). Mean and standard deviation (SD) are presented for continuous variables. Frequency and percentage are presented for categorical variables. Clustered logistic regression [43] was employed to explore the impacts of socio-demographic, health-related and HRQOL factors (three clusters) on health service utilization. The two dependent (outcome) variables were recent two-week physician visit (“no visit”, “at least one visit”) and annual hospitalization (“no hospitalization”, “at least one hospitalization ”). Significance level was set at P <0.05 for including predictor variables into the regression models.

We first explored the associations of the eight domains of SF-36 with health service utilization without adjustment for other factors. In subsequent clustered regression analyses, independent predictor variables were grouped into three clusters according to nature of the study variables: Cluster 1, socio-demographic variables; Cluster 2, health-related variables; Cluster 3, HRQOL variables. Health-related variables included BMI, self-reported morbidity and chronic illnesses. HRQOL variables included the standard 8 domains as measured by SF-36. Multidirectional associations may exist between the three clusters of independent variables and the dependent variables. Specifically, cluster 1 may affect cluster 2, cluster 3 and the two dependent variables of health service utilization. Similarly, cluster 2 may impact cluster 3 as well as the two dependent variables, while cluster 3 may affect the dependent variables. As a result, simultaneous consideration of variables from the clusters in a free multiple regression model (i.e. a free forward stepwise logistic regression model) might result in confounded inference. Therefore, clustered logistic regression [43] was adopted to analyze whether the addition of HRQOL variables to the models including socio-demographic and health-related variables could significantly increase the explanatory power of the risk-adjustment models. The two health service utilization variables were regressed on the three clusters of independent variables. The final regression model was determined in three steps: (1) A forward stepwise regression of health service utilization for the cluster 1 variables; (2) A forward stepwise regression for the cluster 2 variables with the equation derived from step 1 as a fixed part of the new regression model; (3) A forward stepwise regression for the cluster 3 variables with the equation derived from step 2 as a fixed part of the new regression model.

The independent effect of each cluster on the dependent variable was calculated by the difference in the corresponding R² values between the two regression models (with vs. without the cluster). The independent contribution share of each cluster was calculated as individual R² change/total R² change in the final model × 100%. In logistic regression models, the R² is the Nagelkerke ‘pseudo’ R² which is similar to the classical R² in linear regression models for data interpretation [43].

Results for the pooled data in the whole study cohort are presented, because similar results were observed among subjects from the three factories in exploratory analyses.

To the best of our knowledge, this is the first study demonstrating an association between HRQOL and health service utilization in Chinese rural-to-urban female migrant workers. The results revealed that health service utilization increased modestly and significantly with decreasing scores in certain HRQOL domains. Nevertheless, the independent influence of the HRQOL was smaller than that of socio-demographic variables on annual hospitalization, and than that of health-related variables on recent 2-weeks physician visit. The strong association of hospitalization with socio-demographic variables and the lack of association with health related variables suggest that affordability may be a major determinant in the use of hospitalization care for female migrant workers.

We found that HRQOL was a significant independent predictor of health service utilization in Chinese female rural-to-urban migrant workers. This finding is consistent with the findings in previous reports on the impact of HRQOL on health service utilization in other study populations [13,23-25,44]. In the USA, Nelson and colleagues reported that physical functioning and mental health were two important predictors of both clinic consultation and hospitalization in patients with chronic diseases [23]. Dominick and colleagues reported that HRQOL, especially pain frequency, could be an invaluable tool for evaluation of future health service utilization among patients with osteoarthritis [25]. In northern Europe, Miilunpalo and colleagues found that HRQOL was associated with annual outpatient consultation in a working-age population in Finland [44]. In Hong Kong, Lam CL and colleagues reported that HRQOL was a more important determinant of outpatient consultation than chronic morbidity and socio-demographic factors [24]. In contrast, we found that the independent influence of HRQOL on health service utilization was smaller than that of social-demographic or health related variables. The differential findings may be due to different study context –Hong Kong is quite different from mainland China in socioeconomic environment and health care system. In China, Chen T and colleagues showed that the number of monthly outpatient consultations increased significantly with decreasing HRQOL scores in PF, BP and GH domains, while annual hospitalization rate increased significantly with decreasing scores in PF and GH domains in primary care patients [13]. In contrast, among female migrant workers, our study showed that the likelihood of two-week physician visit increased significantly with decreasing scores in BP and GH domains, while the risk of annual hospitalization increased significantly with decreasing scores in RP and GH domains. Therefore, our findings are consistent with those of Chen’s study with respect to the impact of GH on health service utilization, but somewhat different with respect to the roles of PF and RP domains. The differential findings might be due to different study populations: primary care patients in Chen T’s study vs. mostly healthy rural-to-urban migrant workers in our study. Besides, all the 4 domains - physical functioning (PF), role physical (RP), bodily pain (BP) and general health (GH) are the measures of physical health [38]. Taken together, it seems that physical health domains as measured by SF-36 are important predictors of health service utilization in Chinese population. In both studies, the independent influence of the SF-36 domains on health service utilization was observed to be smaller than that of other clusters of variables (social-demographic or health- related). Similar to the findings in Zhu CY’s study [27], the observed HRQOL scores in most domains (PF, BP, GH, VT, SF) were lower in rural-to-urban migrant female workers in our study population as compared to the norms for Chinese women [32].

Self-reported information is prone to inaccuracies. Nelson and colleagues reported that 5% of subjects over-reported clinic consultations over the last one month, and that self-reported health service utilization could not be taken as actual health service utilization, but only as an indicator of utilization pattern [23]. We expected that such errors in self-reports were random and would not affect the validity of the comparisons. We have data on education, but no data on income which is another important dimension of socioeconomic status associated with HRQOL [45]. Income reporting is a sensitive issue and highly unreliable in China. Nevertheless, it is known that rural-to-urban migrant light-duty manual workers are low-pay workers in urban areas. In addition, we have data on duration of employment which may partly reflect their relative income levels. Our findings are suggestive rather than definitive. We could not draw a firm causal inference from an observational study. HRQOL, health status and health service utilization are inter-correlated, and it is difficult to disentangle causes vs. effects. We do not have detailed information on the specific diseases the study subjects had at the time of investigation, reasons of hospitalization, types of insurance, cost for health service utilization and burdens from out-of-pocket payment. It would be of interest to have these variables in future studies for a more in-depth understanding on the association between HRQOL and health service utilization in Chinese female rural-to-urban migrant workers. The study was based on a sample of rural-to-urban migrant women workers in an affluent economic region. We could not assume that the findings are applicable to other regions without additional validation studies. Nevertheless, it is likely that similar associations may be observed in female rural-to-urban migrant workers in other urban areas in China as they share some similarities in much lower socioeconomic status as compared to local urban residents.