Participants

According to the inclusion criteria listed below, potential eligible households were identified by the County Maternal and Child Health Office in the selected provinces, which routinely collects information of every case of maternal deaths in the county no matter whether it happens in health facilities or not. After eligibility was confirmed and informed consent was made, the person in charge of the household economy was invited to a face-to-face, standardized interview. The comparison households were identified according to the matching criteria.

Inclusion criteria for the households with maternal death:

• Having maternal death within 3 months of the baseline interview; Maternal death defined as the death of a woman while pregnant (≥28 weeks gestational age) or within 42 days of termination of pregnancy, not for accidental causes unrelated to the pregnancy.

Matching criteria for the households of childbirth without maternal death:

• Having childbirth within 3 months of the baseline interview, living in the same administrative village, with similar economic status evaluated by the administrative village cadres (i.e. rich, moderate, poor), and the household type is the same as that of the affected family before maternal death (i.e. nuclear or extended; with or without older children).

A total of 530 cases of maternal death were identified in Hebei, Henan, and Yunnan provinces between June 2009 and October 2010. Of these cases, 40.9% were determined to ineligible or did not meet the inclusion criteria (217 out of 530) and an additional 22.3% of the cases (118 out of 530) were eligible but refused to be interviewed. The principle reasons for refusing to be interviewed were unresolved medical malpractice legal suits, or extreme grief among the surviving family members. Overall, 36.8% of the total number of maternal deaths were eligible and accepted the baseline interview (195 out of 530). We compared the gestational week and age of maternal deaths between those accepted and refused to be interviewed. No significant difference was found (data not shown). We selected 384 matching households of a childbirth without maternal death according to selection criteria. Six of the affected households were matched with only one comparison due to difficulties in identifying a second eligible comparison family in the remote and under-populated areas. We compared the women's age and gestational week of maternal deaths between those accepted and refused to be interviewed. Results showed that no significant difference was observed (data not shown).

Among households that enrolled in baseline study, 183 affected households (93.8%) and 346 comparison households (90.1%) accepted the follow-up survey at one year after their intake interview. The attrition rate was 6.2% in affected group and 9.9% in the comparison group. The reasons for attrition in affected group included “unavailable for off-farm employment” (6 cases), “unavailable for traffic inconvenience” (4 cases), “refused follow–up survey” (1 case), “moved to another place” (1 case). The reasons in comparison group included “not visiting the comparison households matched for the above 12 affected households lost in follow-up survey” (24 cases), “unavailable for off-farm employment” (11 cases), “moved to another place” (3 cases). We compared the maternal and household characteristics (age, parity, ethnicity, occupation of woman, household size, whether the family was nuclear family, and dependency ratio) of the 183 affected households having both baseline and follow-up data and the 12 affected households only having baseline data. Results showed that no significant difference was observed (data not shown).

Data collection

The baseline survey data were collected between June 2009 and October 2010 and the follow-up surveys were conducted at one year after the baseline interview for each respondent. During each interview, we collected data with quantitative questionnaires designed after a literature review of related studies, expert consultation, ethical review and pilot study.

The questionnaire included questions on the maternal and household demographic status, household economic status, and health status. Maternal and household demographic status included age, parity, ethnicity, and occupation of woman, household size, whether the family was nuclear family, etc. Household economic questions included both subjective and objective indicators. The subjective indicator was self-evaluation of the overall impact of maternal death, i.e. whether their household economy “became worse” in the following year after baseline survey, and the objective indicators were household and individual income, expenditure and debts. In the questionnaire, there were 31 items of income, including salary (1 items), family production income (27 items, produced gain, slaughtered pigs, output of milk, etc.), and income from transfer and estate (3 items, gifts, donations and other forms of transfer from community members). There were 47 items of expenditure, including expenditure of production (9 items, agricultural productive expenditure, livestock expenditure, etc.), living expenses (30 items, food, clothes and commodity, residence, etc.), and expenditure for transfer and estate or tax (8 items). In order to make the annual household income and expenditure comparable in the year before and after the event of maternal death or childbirth, we excluded enormous occasional income or expenditure in calculating annual household income or expenditure, which include cash as gift from relatives or friends for marriage or funeral, or high resettlement subsidies from government, expenditures for wedding or funeral ceremony, etc. In the follow-up survey, we collected data of expenditure for raising the survived newborn after maternal death or childbirth, which was also not included in total annual household expenditure, for this part of expenditure was US$ 0 in the year before maternal death or childbirth.

The baseline survey collected economic data during the year before maternal death or childbirth, while the follow-up survey collected data during the year since the baseline survey interview. Information collected in this study was based on recall of participants. For comparing annual income at individual level derived from our study with national public data, we collected data of rural net income per capita published by National Bureau Statistics [14], [15].

Ethics Statement

The study protocol and data collection instruments were approved by the ethical reviews boards of Peking University, and the World Health Organization. Signed informed consent statements were obtained prior to any interview by the person in charge with household economy, and other family members who also participated in the interview. For one woman whose age was under 18, we obtained written informed consent from herself, her parents and husband. All the interviews were entirely voluntary (no incentives were provided).

Data analyses

The database was developed using the Epidata 3.1 software. Statistical analyses were conducted with SPSS Version 15.0. The demographic characteristics (age, parity, ethnicity, and occupation of woman, household size, whether the family was nuclear family and dependency ratio) were compared between the affected and comparison group. For categorical variables, χ2 or Fisher's exact tests were used for comparison. For continuous variables, difference in means for normally distributed data was tested by t-test, and other continuous variables which had skewed distribution were tested by rank-sum test. p<0.05 was considered to be statistically significant. The average exchange rate in 2011 (US$1 = ¥6.5) was used to convert the money from Chinese RMB (¥) to US$. All income, expenditure, and debt data in the following year after baseline survey were adjusted by the average deflator of 5%.

We conducted logistic regression analysis with backward method to evaluate the subjective estimator of the overall impact of maternal death, analyzing self-reported “household economy became worse” in the following year after baseline survey. The model is logit (P_i) =  β₀+β₁X₁+β₂X₂, in which the dependent variable P_i was a dichotomous variable indicating whether the household self-reported “household economy became worse”, while β₁ was the effect of maternal death, β₂ was the term of vector of covariates which included maternal and household demographic characteristics (age, parity, ethnicity, and occupation of woman, household size and whether the family was nuclear family, etc).

We assessed both “absolute value” and “relative value” of objective estimator of economic status. We used difference-in-difference (DID) approach to evaluate the “absolute value” [16], i.e. the average effects of maternal death on household and individual annual income, expenditure, or debt. Individual data was derived by dividing total household data by household size. We excluded the newborn from household size to eliminate the confounding effect caused by the survival status of newborn in two groups. Our cohort data were analyzed with one form of DID model, ΔY_it = A_i *a+Δe_it, in which Y_it refers to income, expenditure, or debt data of each household during the year before baseline or follow-up survey. ΔY_it equals to the change of annual income, expenditure, or debt of each household between the two years. A_i refers to whether there was maternal death. a is the effect of maternal death we were interested in, and e_it is the residual error. All time-invariant household and individual characteristics which may correlate with the economic indicators were controlled during the process of “differencing” within each group. DID estimators for income, expenditure, and debt were shown in the results.

We then evaluated the “relative value” of the objective estimator of economic status by calculating changes of annual income and expenditure in terms of proportions. Linear regression models with covariance analysis was used, that is, with logarithmically transformed follow-up values as outcomes and respective logarithmically transformed baseline values as explanatory variables. All the income and expenditure data was transformed because of their positively skewed distributions. The antilog of regression coefficient represents the ratio between affected and comparison group [17]. The difference between the ratio and 1 equals the change of annual income and expenditure, and is converted into proportions by multiplying 100.

Demographic characteristics of the affected and comparison groups

Table 1 shows the demographic characteristics of the affected and comparison group before maternal death or childbirth. The two groups did not differ in the proportions of primiparas (p = 0.263), minority (p = 0.255) and peasants or housewives (p = 0.673). Women in the affected group tended to be older (29.5 vs. 26.3, p<0.001) than women in the comparison group. Before maternal death or childbirth, the characteristics of affected households were similar to comparison households, including household size (p = 0.884), proportion of nuclear families (p = 0.809), and dependency ratio (p = 0.329).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Demographic profile of the affected and comparison groups at baseline survey.

https://doi.org/10.1371/journal.pone.0076624.t001

Self-evaluation on the overall impact of maternal death

Sixty-three percent of the affected households and 29.5% of the comparisons self-reported “household economy became worse”, with statistically significant difference (p<0.001). Univariate logistic regression showed that the affected group carried significantly higher risk of worsening household economy (Odd ratio (OR)  = 4.07, 95% CI: 2.78–5.95, p<0.001). We conducted multiple logistic regression by controlling whether the newborn survived, whether the husband remarried, and maternal and household demographic characteristics which were listed in Table 1, including the women's age, whether the woman was primiparas or not, ethnicity (whether she was minority or not), occupation (whether she was peasants or housewives or not), household size, household structure (whether the household was nuclear family or not), and dependency ratio. The results indicated an increased OR (6.04, 95% CI: 3.61–10.08, p<0.001) of maternal death on worsening household economy, and none of the controlled independent variables were statistically significant except the survival of the newborn (OR = 2.26, 95% CI: 1.20–4.25, p = 0.012).

Total annual income and expenditure

We excluded enormous occasional income or expenditure in the calculation of total income or expenditure (as previously described in methods of data collection). As shown in Table 2, the DID estimator of average annual income at household level showed a statistically significant difference between two groups (US$ –869, p<0.001). Average annual income per capita (at individual level) in both groups increased during the following year after baseline survey. Its DID estimator showed that the increase in the affected group was US$ 102 lower than that in the comparison group significantly (p = 0.038). Converted to proportion of change, a 32.0% reduction of average income at household level and 19.9% at individual level was observed between the two groups. Excluding the expenditure for raising the newborn and enormous occasional expenditures, the average annual household expenditure was reduced significantly during the year after baseline survey in affected group (US$ –575, p<0.001), but almost was unchanged in the comparison group (US$ 74, p = 0.429). Similar to annual income, the DID estimator of average annual expenditure at household level showed a statistically significant decrease in affected group (US$ –650, p<0.001). At individual level, average annual expenditure slightly decreased during follow-up period in the affected group (US$ –41, p = 0.242), while it increased significantly in the comparison group (US$ 39, p = 0.047). DID estimator showed that the mean relative difference between two groups was US$–80 (p = 0.047). Converted to proportion of change, a 24.9% reduction of expenditure at household level and 12.4% at individual level was observed between the two groups.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. The results of difference-in-difference analyses (DID) for indicators of economic status between affected and comparison households (Mean, US $).

https://doi.org/10.1371/journal.pone.0076624.t002

Although annual income in both groups probably covered all expenditure in the year (as shown by the means of income and expenditure), the average accumulated debt increased much higher in affected group compared to that in comparison group (DID estimator =  US$ 922, p = 0.024). The self-reported reasons were that households needed money to cover extra expenditure of raising the newborns or unexpected occasional event. In the affected group, the mean expenditure for raising the newborn (102 cases with survived newborns) was as high as US$ 1007 (inter-quartile range US$ 417 to US$ 1436), for 100.0% of them having bottle or formula feeding instead of breastfeeding. Besides, the mean expenditure for remarriage of the husbands in affected group (26 cases) was as high as US$ 4667(inter-quartile range US$ 934 to US$ 6081).

We also analyzed household wealth indices, including index score of family property, live stocks, and ability to maintain basic living including enough food, clothes, medical care, etc. The DID estimators of index score of family property and basic living ability showed significantly decrease in the affected group compared with that in the comparison group (−0.13, p = 0.029; −0.10, p = 0.006).

Since newborn survival associated with increased OR for self-reported “household economy became worse”, we did stratified analyses for quantitative indicators by whether the newborn survived or not. As shown in Table 3, affected households with or without newborn survival was associated with the same trend of reduced DID estimates of total income and expenditure in the year since the baseline, at both household and individual level. They also had increased DID estimates of cumulated debts, and reduced DID estimates of other household wealth indicators. Some estimates were not statistically significant, which may be related with insufficient power of 0.561 for subgroup of newborn survived in affected group, due to the smaller sample size after stratification.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Stratification results of difference-in-difference analyses (DID) for indicators of economic status between affected households with newborn survived or not and comparison households (Mean, US $).

https://doi.org/10.1371/journal.pone.0076624.t003

Categories of annual income and expenditure

Table 4 showed the DID estimator of each category of annual income and expenditure at individual level. Total annual income included salary, family production, and income from transfer and estate. Although the mean differences between two groups were not statistically significant, income from salary, family production, and from transfer and estate had a relative decrease in the affected group compared with that in the comparison group during the following year after baseline survey (from US$ –13 to US$ –55).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Categories of annual income and expenditure at individual level (Mean, US $).

https://doi.org/10.1371/journal.pone.0076624.t004

Total annual expenditure included expenditure of production, living expense, and expenditure for transfer and estate or tax. Among the categories of living expenses, the mean expenditures on food, clothes and commodity, transportation and phone decreased significantly in affected households compared with that in the comparisons (p = 0.037, 0.003, and 0.022, respectively). However, the mean expenditure on cigarette or alcohol in affected households significantly increased compared with that in the comparisons (p = 0.014). During the interview, many male family members in affected households, especially the husbands, said that they tried to relieve their grief and loneliness by frequent drinking or smoking.

We also calculated proportions of household income and expenditure for each category. Results demonstrated lower proportion of clothes and commodity consumption but higher proportion of cigarette or alcohol consumption in affected households with statistical significance (see Table S1).

Comparison with other studies

Until now, there is no quantitative study on the sustained economic impact of maternal mortality at household level in the published or available gray literature that we have been able to retrieve. However, the economic impact of emergency obstetric care (near-miss events) has been explored in a few studies in recent years [7]–[9]. A study conducted in Mali [7] revealed adverse outcome of near-miss events, including loss of income, reduced food consumption and productivity, lack of food and in debt. A study conducted in Burkina Faso [8] reported that more frequent sale of assets, borrowing and slower repayment of debt in the year after near-miss. Both of these studies used small sub-sample analyses and showed the economic impact of near-miss on households, indirectly suggesting the potential impact of maternal death. Our quantitative study on maternal death directly showed its impact on households, including loss of income, reduced food consumption, and accumulated debts.

Most studies related to other illness have not examined economic impact of illness with prospective study design, except one study conducted in South African by Bachmann et al. This cohort study reported economic impact of HIV/AIDS was 13% reduction of annual household income in South Africa [21]. Results from our study showed that annual household income in the affected household reduced by 32.0% during the year after maternal death, and annual household expenditure in the affected group declined by 24.9%, which was similar to a reduction of 26% in the study by Bachmann et al [21]. In addition, we found the significantly reduced expenses of food, clothes and commodity, transportation and telecommunications at the individual level during the year after maternal death. These changes in the pattern of household expenditures were also observed by Over and Dayton, with households that had suffered an adult death exhibiting lower non-food expenditure per capita than non-affected households [22]. Except those reduced expenses, higher consumptions on cigarette or alcohol found by this study may further have adverse impact on the family members' health.

Strength and limitation of the study

To our knowledge, the present study is the first prospective cohort study on the sustained economic impact of maternal death at household level. Although previous studies revealed financial hardship caused by emergency obstetric care, which is a useful complement to investigation of maternal death, this study showed the sustained economic impact of maternal death on household income and expenditure. It provided more direct and reliable evidences that maternal death would make adverse impact on household economy. Sustained worsening economy can impoverish households or push them further into poverty. It is urgent to increase funds for improving maternal health and for off-setting out of pocket expenditures for life-saving medical emergencies.

The sample size of our study (183 affected households and 346 comparison households) was relatively large compared to other, related studies. We enrolled not only the fatalities in hospitals but also those died at home, and the participants were selected from three provinces to represent lower, medium, and higher level of maternal mortality in rural China. In addition, the prospective study design with comparison households adds strength to the results.

In this study, both income and expenditure data was collected directly in each household. However, there is a debate on reliability of self-reported income and expenditure data [17], [23], [24]. Income reflects consumption opportunities; informants usually have better idea of average earnings of each family member, but can potentially conceal the facts for various reasons. Expenditure is an alternative poverty measure, but few informants record daily expenditure in detail and may omit a part. We used both of these measures in the study. Comparing annual income at individual level of comparison households in our study with the indicator of rural net income per capita in China published by National Bureau Statistics [14], [15], we found that data from comparison households in baseline study was not significantly different from the published data in 2009 (US$ 838 VS. 793, p = 0.210), and data from comparison households in follow-up study was also not significantly different from the published data in 2011 (US$ 1044 VS. US$ 1073, p = 0.620). These results increased the confidence in the validity of our findings.

The limitation of this study is that we didn't investigate the mechanism of how maternal death affected household economy. Further studies should focus on the pathway from maternal death to the worsening household economy, providing more implications and recommendations for social assistance interventions.

Implications for policies

The study has several implications for health and social policies. First, catastrophic costs related to maternal death and its adverse lasting impact on household economy justify efforts to improve coverage of preventive measures on maternal mortality. Although the international community has a strong commitment to improve maternal-newborn ill-health (MNIH) in developing countries, direct funding for reproductive health has not increased in recent years, and in many countries it has declined relative to demand for services. Increase of funding on maternal mortality prevention could contribute to poverty reduction [5].

Second, financial supports to cope with high direct costs of maternal death is particularly important, such as expanding the coverage of financing systems, and increasing the reimbursement to protect households experienced maternal death with catastrophic direct costs. One point to be discussed is whether to develop financing mechanisms which eliminate all medical expenses and cover the majority of women treated with emergency obstetric care (EMOC), or to protect and subsidy the minority of households that have suffered from catastrophic costs caused by maternal death. Until now, maternity insurance existed in several provinces of China, providing one-off subsidy for funeral expenses and pension to the family experienced maternal death by using surplus fund. This subsidy is a feasible measure to reduce the economic burden of maternal death for each household. Insurance benefit packages are capped at a maximum amount, which is insufficient for end-of-life medical emergency expenses – for maternal health as well as other conditions. Although extraordinary measures are put into place by local hospitals and government to compensate the families, these measures are ad hoc and, as we saw in this study, insufficient.

Third, the results showed a reduction of both income and expenditure of family production, even lower food consumption per capita in affected households. It may have an adverse impact on physical health status of family members. Ainsworth and Gertler indicated that children who lost their mothers were much more likely to be stunted and maternal death increases the probability of both child death and being malnourished than children with their parents alive [25], [26]. We suggested giving credit to the affected households in the short term to increase labor and land productivity, and consequently avert potential household food insecurity and health problems.

Conclusion

Compared with childbirth, maternal death had adverse impact on household economy, including decreased income and expenditure, increased debts, changed expenditure pattern, and higher risk of self-reported “household economy became worse”. The present study is the first prospective cohort study providing more direct and reliable evidences for adverse economic impact of maternal death at household level. Sustained worsening economy can impoverish households or push them further into poverty. Increase of funding on maternal mortality prevention could contribute to poverty reduction.