Health insurance and healthcare utilisation for Shenzhen residents: a tale of registrants and migrants?

This cross-sectional telephone survey targeted all households with a residential telephone number in Shenzhen. Based on a hypothesized difference in the proportion of residents and migrants with health insurance coverage of 0.5 (80% power and alpha of 0.05), we estimated a sample size of 1000 respondents was required. The estimatedw?> sample was increased to 1500 respondents to enable subgroup analysis (4 subgroups).

Telephone numbers were computer generated by combining the local residential prefix (+86 755) to a complete randomly generated 8-digits sample. A total of 34,430 residential numbers were randomly generated (Figure 1). Each residential number was dialed a minimum of 3 times, at different times of the day and different days of the week. Upon successful contact, household members aged 18 years or above, who spoke Cantonese, Mandarin or English were recruited for the interview. If more than one qualified household member was available, the individual with the nearest birthday to the date of the telephone call was recruited. The survey was conducted in May 2010 by trained and experience telephone researchers from the Social Sciences Research Centre at the University of Hong Kong.

The questionnaire contained 65 questions that were divided into five sections: the first section collected basic socio-demographics, the second section assessed health condition and risky behaviour, the third section on health seeking behaviours, the fourth section on doctor consultations, and the fifth section on hospital admissions. Questions were formulated and compiled after we reviewed similar reports in the literature [18] on current policy directions, the prevailing characteristics of the Shenzhen health care system as well as drawing specific questions from the Thematic Household Survey in 2002 [19], and the 4th National Health Census [20]. All questions were piloted for comprehensibility, face and content validity on an independent sample (n = 5) of research postgraduate students with personal knowledge of the health care system in the Peoples Republic of China. Minor adjustments were made to the questionnaire after the pilot study.

Data was collected for socio-demographic characteristics (age, gender, marital status [single, married, widowed/separated or divorced], Hukou registration status [registered with a Shenzhen Hukou (temporary or permanent), not registered with a Shenzhen Hukou (migrant)], occupation [executive, professional, managerial, ordinary worker and unemployed], monthly income [sixteen categories between no income and ¥30,000 or above], self-reported chronic health conditions as diagnosed by a Western medicine practitioner [presence of chronic disease or not], age-comparative perceived health status measured on a 5-point Likert scale [1 = much better, 2 = better, 3 = about the same, 4 = worse & 5 = much worse] [21], experience of any health problems (e.g. discomfort, illness or injury) in previous 30 days [yes, no], health service utilisation [out-patient doctor consultations in the past 30 days and in-patient hospital admissions in the past 12 months], point of access for healthcare [tier 1 to tier 3 facilities] [15] and insurance status [uninsured or insured by social insurance schemes or/and additional private insurance schemes].

Verbal informed consent was obtained from all study subjects. Ethics approval was obtained from the Institutional Review Board of the University of Hong Kong/ Hospital Authority Hong Kong West Cluster (UW 10–154).

The primary outcome variables were health insurance status, healthcare utilisation by Hukou registration status and source of care. For analysis, insurance status was grouped into four investigator defined categories (Table 2) and health care utilisation by either doctor consultation in the previous 30 days or hospital admission in the previous and 12 months.

Multiple imputation through bootstrapping from a full Bayesian predictive distribution was used to impute missing values in the demographic and socio-economic variables [22]. Predictive mean matching on binary, categorical, and continuous variables was not required and missing values in the outcome variables were not imputed [23]. Missing variables were imputed ten times and subsequent regressions in analysis were consolidated by the ‘micombine’ function [24].

Marital status was dichotomised (single/separated/widowed and married) for analysis. Income categories were constructed by quintiles, low (<20%), “middle” (20-60%) and “high” (>60%). Age-comparative perceived health status into better (much better and better) and worse (about the same, worse and much worse); smoking status into current smoker (current and past smoker) and never smoker (never smoked) and alcohol drinking patterns into regular drinker (current and past drinker) and never drinker (never drinks) were dichotomised for analysis. Source of healthcare was grouped into three tiers according to the Chinese healthcare system (Table 1).

All sample data were weighted to achieve population representation using standard methodologies against the 4^th National Health Census data for Shenzhen.

Logistic regression models were used to estimate the crude and adjusted odds ratio and the 95% confidence intervals controlling for covariates including age, gender, marital status, local registration, occupation, education, monthly income, age-comparative perceived health status, self-reported chronic health condition, smoking and drinking behaviour, self-reported sickness in the past 30 days and point of healthcare access.

Multiple imputations were performed in R (R Foundation) Version 2.13 and regression analysis was conducted in Stata (StataCorp) Version 10.

Although information on the exact type of social health insurance scheme was obtained from every respondent, we did not further stratify the various social health insurance schemes for analysis. The compression of social insurance schemes for analysis may obscure some of the marginal differences in outpatient and inpatient visits due to different reimbursement compositions. Nevertheless, our aim of the study was to investigate the availability of health protection through social health insurance and additional/ private insurance, but not the impact of different health insurance schemes on health protection, however other studies have already began to inquest such differences [34]. The migrants in our study were relatively ‘healthy’ with fewer self-reported chronic health conditions and better comparative self-rated health status, which is not consistent with health problems amongst migrants as reported in wider literature [30]. We suggest the ‘healthy migrant’ effect may exist in Shenzhen, where the generally younger population are subject to later onset of chronic symptoms, hence delayed diagnosis of chronic illness. The inherent limitation in a cross-sectional study means we can only infer associations rather than causality; and the ‘cold-calling’ nature in telephone surveys has resulted in a less than desirable response rate which may contribute to selection bias. However, the response rate is comparable with other published studies using questionnaire designs. Although our sample was randomised by telephone numbers, migrants who are clustered in factory dormitories or in rental places may lack permanent landline numbers. These unrepresented individuals may differ systematically and may contribute to selection bias.