Maternal deaths among rural–urban migrants in China: a case–control study

Rural–urban migrants from six provinces and cities (Guangdong, Zhejiang, Shanghai, Jiangsu, Beijing and Fujian) account for 68.5% of the migrating population of China. Our study was conducted in prefecture level hospitals in three of these locations, Guangdong, Zhejiang and Fujian provinces, and in Haidian District of Beijing. China’s permanent residence registry or hukou system has had an especially strong effect on the country’s urban growth [7]. The system registers the residential location of every citizen and divides them into agricultural and non-agricultural groups. People who live in cities hold non-agricultural hukou, whereas farmers and other rural residents hold agricultural hukou. This system denies official residency and access to essential public services for rural–urban migrants, regardless of the duration of their unofficial residence in a city.

For the purposes of this study, we classified women who had arrived from their official place of residence and lived locally for at least half a year and who did not have a local hukou as migrants. Maternal death was defined using the standard WHO definition [16].

A case–control design was employed to investigate associations with maternal death among migrant women. The cases and controls were collected prospective in the above four locations from July 1, 2010 to October 1, 2011. Cases were identified as migrant maternal deaths associated with singleton pregnancy, reported by the national or provincial MCMSS. Twin pregnancies were excluded because of the difficulty identifying a sufficient number of matched controls. Ectopic pregnancy and induced abortion were also excluded because of their lack of attendance at ANC, a perceived key influence on pregnancy outcome. For each case, four matched controls were selected, each a surviving rural–urban migrant woman aged within five years of the case, and who had delivered after a singleton pregnancy during the surveyed period, at the same hospital (or, for a small number of controls aged <25 or >35 years, in nearby hospitals). All cases were reviewed by the local Maternal Death Audit Committee, which includes province, prefecture, and district/county experts in maternal and perinatal health. All relevant documents, including medical charts, experts’ audit documents, and Maternal and Child Health (MCH) Report Forms were reviewed. A final cause of maternal death was assigned by consensus among the committee members in each province, using the same guidelines to assign the cause of death for each case.

Non-attendance at ANC was assumed as a critical risk factor for maternal death [15] and a sample of 120 cases and 480 controls was estimated adequate to determine a 15 percent difference in the proportion of no ANC attendance between the cases and controls, with a power of 80%, alpha = 0.05, odds ratio (OR) =2 (Power Analysis Sample Size software, version 8.0, NCSS, Utah). Five ANC visits are recommended in China.

Ideally, pregnant women in China will register and receive an ANC card, receive follow-up reminders and attend for timely care at the local hospital.

Questionnaires were designed by the survey team to examine key personal, demographic, socio-economic, and reproductive and health status factors for each case and control. Health insurance status and physical and financial health service access were included. For each case, the respondent was the husband/partner, head of the family or another close insider for both case and control group.

The questionnaire was based partly on WHO verbal autopsy standards and took into account standard care for pregnant women in China and related social determinants [17, 18]; it was pre-tested and adjusted by the survey team before finalization. Specialist doctors in the four prefecture’s MCH hospitals were trained by the principal investigator (JZ) in all four locations, and one supervisor in each was assigned to oversee administration of the questionnaires. When a new maternal death of a migrant woman occurred, a trained doctor would attend to conduct the survey after gaining consent, and four matched controls were selected according to the criteria described above. All questionnaires were reviewed carefully by the principal investigator before data entry.

EpiData 3.0 was used to build the database with duplicate data entry used to reduce errors. Analysis was undertaken using SPSS, version 13.0 (SPSS, IBM Corp). Chi-square testing was used to compare the characteristics. Bivariate conditional logistic regression analysis was used to compare the relative frequency of association of the variables assessed among cases and controls. Demographic characteristics such as women’s age group, education level, marital status, gravidity and parity, as well as factors significantly different in bivariate analyses were included in multivariate conditional logistic regression analysis to assess independent associations with maternal death. ORs and corresponding 95% confidence intervals (CI) were used to measure association. Statistical significance was established at the p < 0.05 level (two-tailed).

The research plan, data collection instrument and consent form were approved by the Institutional Review Board at the Peking University Health Science Centre. Interviewers were trained in consent procedures and all interviewees signed a consent form that included a description of the study, noting the possible risks and benefits, and emphasis that participation was voluntary.

The characteristics of cases and controls are summarized in Table 1. The mean age of the cases was not significantly different from the controls, but the percentage of cases aged ≤19 or ≥35 years was higher than among controls. Age was thus included in multivariate analysis. Women who died were poorer educated, had higher parity, were more likely to be unmarried or divorced and their pregnancy was more likely to have been unplanned (according to China’s family planning rules). All these associations were statistically significant (p < 0.001). Given the small sample size, we did not analyze associations with maternal death by province.

Bivariate analysis for risk factors is shown in Table 2. The risk of the family income <2000 renminbi (RMB) per month was 9.6 times higher among cases than the controls. Lack of health insurance and unplanned pregnancy were more common among the cases than the controls, as were several factors related to women’s health awareness and service uptake: not knowing to register the pregnancy and to attend ANC, being unfamiliar with key danger signs during delivery and unaware of the recommendation to deliver in hospital. Regarding health service provision and participation, not receiving initial advice and follow-up reminders to attend ANC were more common amongst cases, as were lacking decision-making power in the household on ANC attendance and not being registered in the healthcare system. Not having any ANC during pregnancy was 27.3 times more common amongst the cases than controls. Attendance at five or more ANC visits was significantly lower among cases than controls. No difference was found between the frequency of pregnancy complications between cases and controls, but assisted vaginal and caesarean section delivery were significantly more common amongst cases of maternal death.

Multivariate conditional logistic regression analysis on associations showed that compared to controls, deceased women were more likely to: have low family income (OR, 4.5; 95% CI, 1.7-11.7); lack health insurance (OR,1.3; 95% CI, 1.1-1.6); not know delivery danger signs (OR, 3.1; 95% CI, 1.3-7.7); have not attended ANC (OR, 22.3; 95% CI, 4.3-116.0) and have ANC only 1–4 times (OR, 5.0; 95% CI, 1.6-15.5) (Table 3). These ORs refer to the model including age as a variable, but do not change significantly when it was omitted from the analysis. The other variables which were significantly associated with death in bivariate analysis (Table 2) were included in an initial model but excluded in the final model.

To our knowledge, this is the first case–control study in China to explore associations with maternal death among urban migrant women. However, limitations should be acknowledged. First, we did not include women who delivered at home or in private clinics in our analysis, because we did not include such locations in our search for controls, potentially causing a selection bias. It is known that women who deliver at home or in private clinics face additional risks [9, 10], but this risk factor could not be assessed. Second, being a known risk factor for maternal mortality, an attempt to match cases and controls for age was undertaken but proved difficult, as the women who died were either younger or older than women who did not die (as expected). We therefore undertook multivariate analyses that both included and excluded age as a variable, but this did not change the identified associations with maternal death. Third, the interviewees for the cases were the deceased women’s husbands or family head, who might not have accurately reflected the knowledge or situation of the women who died. It is possible that the deceased women had more knowledge than their husbands, but were unable to act on it; this emphasizes the importance of the systematic approach recommended below. Fourth, our study used a quantitative methodology, but exploring social determinants would have benefited from qualitative data to better highlight the origins of the associations identified. This may have further explained why some groups of rural–urban migrants are better able to participate in and access services than others. Fifth, the final sample size for cases was lower than the 120 originally calculated to have 80% power to determine a 15% difference in ANC non-attendance. Nonetheless, with the sample of 109 cases and 436 controls, we were still able to show in multivariate analysis that cases were 22 times as likely to have had no ANC and five times as likely to have had fewer than the recommended five visits. Finally, we did not explore deaths due to abortion, a common practice for birth control in China, and possibly an important cause of maternal mortality. However, we believe similar associations would have been identified if this group had been included.