Ethics Statement

Both the first and the follow-up studies were accepted by the ethics committees in Burkina Faso (Health Research Ethics Committee) and the United Kingdom (Ethics Committee of the London School of Hygiene & Tropical Medicine - LSHTM).

Written informed consent was obtained from all participants before participation in interviews. Where participants had difficulty reading in French, interviewers read out the consent form in participants’ preferred local language to help them understand the study. Once they understood the study, their written consent was then obtained (either they signed or applied their fingerprint on the consent form). All named ethics committee specifically approved this study.

Study site and sampling

The study was carried out in Burkina Faso, a West-African country that has limited resources, poor health outcomes and a weak health system [23], with 46% of its inhabitants living under the poverty line [24]. Maternal mortality is high with a rate of 700 per 100,000 births [25] and only 54% of births are attended by skilled health staff [25]. Large disparities exist in skilled birth attendance ratios between rural and urban populations. Referral systems are weak and in case of emergency, women are obliged to travel long distances to reach hospitals. Since 2006, a national subsidy policy for normal deliveries and emergency obstetric care has been applied in Burkina Faso. Nevertheless, user fees are still charged for delivery services at government hospitals as the policy does not abolish them. Before 2006, the average fee for a normal delivery was CFA27,245 equivalent to US$59; a fee level beyond the monthly budget of a poor household in Burkina Faso. Complicated deliveries requiring surgery or other treatment would incur much higher costs [14].

This study is built on an existing cohort of 1014 women from seven rural and urban referral hospital facilities in Burkina Faso. These women were prospectively recruited in the seven hospital facilities between December 2004 and March 2005. The facilities included the country’s two referral teaching hospitals (located in Bobo-Dioulasso and Ouagadougou), two regional hospitals (located in Dédougou and Tenkodogo) and three district hospitals with surgical capacity (located in Houndé, Nouna and Ouagadougou). The cohort consisted of two groups of comparison: 677 women who had an uncomplicated delivery and 337 women who experienced a severe obstetric complication that would have almost certainly caused death had they not received hospital care (labelled a “near miss” event). The group of women who experienced a severe obstetric complication (n=337) fell into three sub-groups: those who had a live birth (n=199), those who experienced a perinatal death (n=64) and those who had an early pregnancy loss (n=74). The diagnosis of a complicated delivery was based on clinical signs and symptoms of complication as well as case management procedures. Women with uncomplicated deliveries were defined as those who had no documented signs of obstetric complications and who vaginally delivered a live born infant at term.

Data for physical, mental, economic and social indicators were collected from women’s medical records and at-home visits within a week after discharge and at 3 months, 6 months and 12 months. The first interview included questions on demographic, personal and socio-economic factors before the delivery. Data from the first study could be accessed through a request to the Initiative for Maternal Mortality Programme Assessment – Immpact website: http://www.immpact-international.org – University of Aberdeen (UK).

The follow-up study

The follow-up study was done over a period of four months in 2008 and 2009. It managed to track 711 women i.e. 70% of the 2004/2005 cohort. Each woman was located and invited to participate in the study with her husband and/or the head of household. Whenever possible, interviewers who participated in the first study were recruited to work on the follow-up. Following informed consent, women were invited to provide answers to a structured interview on their current socio-demographic characteristics, health, economic and social well-being including their perceived health. Men and other key family members (if the husband was not the head of the household) were also invited to answer questions on income, consumption and assets in separate interviews. Data from the follow-up study could be accessed through a request to the Economic and Social Research Council – ESRC: website http://www.esrc.ac.uk – (UK).

Statistical analysis

To make sensible claims about the impact of a near miss event, we need to find a suitable ‘counterfactual’ to the exposure of such event. In other words, it would have been ideal to know what would have happened, had women with life-threatening delivery complications not been exposed to a near miss event. This is obviously technically impossible as women cannot at the same time be affected and not affected by near miss events. To solve this problem, impact studies compare women exposed to such an event (treatment group) with women not exposed (control group). However, when comparing outcome indicators between women who experienced obstetric complications and women who did not experience such complications, as a measurement of impact, one should be aware that such a comparison may suffer from selection biases. To illustrate this, imagine we are interested in measuring the effect of near miss events on women’s well-being. For this, we compare well-being indicators of women who suffered a near miss event and those who had an uncomplicated delivery. However, differences in well-being indicators may be biased by socio-economic differences that already existed between the two groups, which in this setting is more likely because our sample is hospital-based: women from wealthier households tend to go to hospital irrespective of expecting complications [26-30], while women from poorer households only go to hospital when expecting complications or when unexpected complications suddenly arise. Women with complications who were selected into our sample may, on average, be poorer than women without complications.

It is therefore necessary to adjust samples so that the likelihood of being exposed to a near miss event conditional on important socio-economic characteristics (e.g. wealth) is made similar across the treatment and comparison groups. To achieve this, we resorted to propensity score matching techniques (PSM) [31]. The basic idea of PSM is to match women with uncomplicated deliveries to women with near miss complications who are similar in all observable characteristics that might be correlated with the likelihood of having a near-miss event. Any remaining differences in outcomes of both matched groups can then be attributed to the near miss complication only.

PSM consists of two phases. In the first step, we ran a probit regression, pooling together near miss and uncomplicated delivery women and we estimated the probability that a woman in the sample belonged to the near miss group, controlling for relevant confounding variables. The empirical literature showed us that richer women, women in urban areas, higher educated women, more empowered women or women who had not had many children before were more likely to deliver in a hospital in absence of any complication, whereas poorer women, women in rural areas, lower educated women, women who had had more children, less empowered women only did so if it was really necessary [32-40]. We therefore added variables such as education and residence in the probit regression. We also controlled if women had an economic activity, the number of live births still living with them, gravidity, marital status and a decision-making power dummy equal to 1 if the woman participated in the decision-making on household expenditures. To control for household wealth we added an asset index based on a principal component analysis of wealth variables such as possession of a radio, car, bicycle, fridge etc.

In the second step, women with a near miss complication were matched with women with an uncomplicated delivery and who had similar propensity scores. We used a ‘stratified’ matching algorithm, which matched observations between near miss and uncomplicated delivery women within different strata of propensity scores in order to balance the distribution of the observed predictor variables of near miss between both groups. To ease interpretation of the results, we re-coded all variables so that a negative sign reads as a negative impact. Out of the 711 women we recontacted, complete information was available for only 698 women and all analyses were performed on stata version 11.2 based on these 698 observations.

Outcome indicators

The main outcome variables at the household level used for the analysis were: weekly consumption of a selected relatively expensive food items, estimated income approximated via household expenditures on various items, expenditures on education and index child development measured through their height and weight. The index child is defined as the 2004/5 surviving child of women who had a severe obstetric complication (near miss). The child’s weight was measured with the use of a calibrated balance, with a registered record in a standing position at nearly 100 grams. The child’s height was measured without wearing shoes and standing back against a vertical wall by the use of a tape. The analysis of the weight and height consisted of a comparison of the index child weight and height with, that of the child of women who had had an uncomplicated delivery in 2004/5. We also used food insecurity as measured by the Maxwell food insecurity grid [41]. Each question on the Maxwell grid was weighted from 1 to 4, based on the in-depth knowledge of the research team of consumption habits and food coping strategies in Burkina Faso. We then added up the weights together to produce a food insecurity index with a high score indicating food security at the household level. We also used the WHO quality of life assessment grid that measures the post-partum subjective quality of life [42]. Each question was again given a weight ranging from 1 for total unsatisfaction to 3, meaning entire satisfaction. We added up the scores to produce an index of an overall perceived well-being, a high score meaning more satisfaction for one’s quality of life.

Loss to follow-up and characteristics of the longitudinal sample in 2004/2005 and in 2008/2009

Table 1 and Table 2 respectively present the prevalence of loss to follow-up and the most important characteristics of the sample. Only 70% of the original sample was reached in 2009 (Table 1). Among women who were lost to follow-up, the greatest loss was for the near miss women who had lost their child (42%) followed by near miss women with live birth (34%). Near miss women with live birth were the most represented in our sample of near miss women in 2008/2009 with a percentage of 59% (Table 2). While in 2004/5, only 43% of the women were economically active through formal or informal employment, this percentage was substantially higher (60%) four years later. Most of the women in the sample remained married (84%) in 2009 even if we noticed an overall decrease in the percentage of women married in a monogamous arrangement (71% to 63%). Women with near miss complications were more likely to be separated or divorced from their 2004/5 partners (8% versus 5%), and more likely to be widowed (2% versus 1%) compared to the uncomplicated delivery women in 2009.

Prevalence of disadvantaged households in the sample in 2004/2005

Figure 1 shows that near miss households were poorer than uncomplicated delivery households in 2004/2005. Near miss women with abortion and those with perinatal death appeared to have similar average wealth. Moreover, near miss women with live birth have an average wealth which is higher than women with abortion and perinatal death (p< 0.001).

Download:

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

Figure 1. Household asset index score from principal component analysis in 2004/5.

https://doi.org/10.1371/journal.pone.0080010.g001

Emergency obstetric and normal delivery costs to households in 2004/5

Households of near miss women paid much more for hospital delivery in 2004/2005 than women with uncomplicated deliveries. Each near miss group paid more than twice the amount that the uncomplicated delivery group paid (Figure 2). Within the near miss group, households with live birth and perinatal death incurred the highest costs for the delivery.

Download:

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

Figure 2. 2004/5 hospital event cost by near miss group and for the uncomplicated delivery group.

https://doi.org/10.1371/journal.pone.0080010.g002

Aggregated analysis: comparison of near miss women with the control group

Table 3 presents the results of the probit regression used to estimate individual propensity scores. We found that single women, married monogamous women or either second or more than second polygamous married women were more likely to have a near miss complication. Likewise, women with a larger number of pregnancies were also more likely to have a near miss complication. Conversely, we found that women living in urban areas, women having a low number of children still living with them or those having an economic activity were less likely to have a near miss complication.

The distribution of the predicted probabilities of a near miss event is presented in Figure 3. As expected, we observed that in the group of near miss complications, propensity scores were skewed to the right compared to the uncomplicated delivery group. Table 4 further shows the balancing statistics of women’s characteristics based on standardised differences after matching. After matching, we found a satisfactory balance of the standardised differences between women’s characteristics.

Download:

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

Figure 3. Estimated propensity scores of near miss and control groups.

https://doi.org/10.1371/journal.pone.0080010.g003

Table 5 shows the differences in household economic indicators between near miss and control groups with and without matching. Almost all outcome indicators had a negative sign, suggesting a persistent negative effect of the near miss event on households. However with matching, the effects became weaker and we had higher p-values, which indicates that part of the difference in outcomes is due to baseline differences between comparison groups.

Looking at statistically significant coefficients after matching, we found that child development measured through the height of the index child was significantly affected, 4-5 years after their mothers were hit by the near miss event. We also found that households of near miss women spent significantly less money per school-age child and that they ate significantly less rice a week, up to 4-5 years after the near-miss event. We did not observe any statistically significant difference in total per capita expenditure between the treatment and comparison groups. Additionnally, no significant difference was observed on the proportion of school-age children who actually go to school or on the index child weight. However, we found at the 10% significance level that 4-5 years after the shock, households of near miss women consumed significantly less relatively expensive foods per week.

Besides studying effects of severe obstetric complications on the consumption of relatively expensive food items, we investigated their effects on food security, as measured by the use of strategies to cope with foods shortages, strategies that were described by Maxwell in his paper in 1996 [41].

Table 6 shows a profusion of negative impacts, except for the strategy which consists of picking foods from the bush and for the overall food security index. Again, this may suggest a persistent negative effect of the near miss event on household food security. However, the evidence was not convincing as none of our indicators nor the index of food insecurity were significant after PSM at the conventional level of 5%.

Table 7 shows differences between exposed and control groups on a series of quality of life indicators. Once more, many indicators appeared with a negative sign, even after matching, which may indicate a long term negative effect of the health shock on women’s quality of life. Focusing on statistically significant results, we found that women who were hit by the near miss event were, 4-5 years after the shock, significantly less satisfied with their health and with their perceived quality of life. Surprisingly, we also found at the 10% significance level that near miss women experienced less negative feelings about suicide, anxiety and depression in the month preceding the interview compared to women with uncomplicated deliveries.

Disaggregated analysis: the importance of pregnancy outcome

The near miss group consisted of women with different delivery outcomes including live birth, abortion and perinatal death. The long term outcomes for the near miss group may depend on the delivery outcome. For example, it is likely that giving birth to a live baby may incur extra costs (purchase of care, vaccinations, feeding etc) for women who are already in a distressing situation. To analyze the role of the pregnancy outcome, we disaggregated the near miss group into these three different groups and estimated the long term impact for each category separately.

Table 8 presents results of the propensity score estimations for each type of near miss. Model 1 which estimates the likelihood of a near miss event with live birth showed that economically active women, living in urban areas, were less likely to have a near miss complication with a live birth. On the contrary, single women, first polygamous married women or second or more than second polygamous married women were more likely to have a near miss with live birth. Model 2 which estimates the likelihood of a near miss event with perinatal death showed that single women, first polygamous married women and women having a high number of pregnancies were more likely to have a near miss with perinatal death, while women with a high number of children still living with them were less likely to have a near miss with perinatal death. Finally, Model 3 that estimates the likelihood of a near miss event with early pregnancy loss indicated that women living in urban or rural areas and who were autonomous in decision-making on household expenses were more likely to have a near miss complication with abortion. Model 3 also showed that non-poor primary educated single women, women with a high number of living children with them were less likely to have a near miss with abortion. Table 9 further shows the balancing statistics of women’s characteristics based on standardised differences after matching. After matching, we found a satisfactory balance of the standardised differences of women’s characteristics in each model.

Table 10 presents differences in household economic indicators between each type of near miss and the control group. It is noteworthy that most indicators at household level in the three models appeared with a negative sign (with and without matching). Looking at statistically significant coefficients after matching, we found that the development of the surviving child of near miss women with live birth measured through his height was significantly lower than the one of the uncomplicated delivery child, up to 4-5 years after the health shock Model 1(b). We also found that 4-5 years after the shock, households of near miss women with live birth, in addition to consuming significantly less rice a week, spent significantly less money per school-age child Model 1(b). We were surprised to find that the number of times per week that women received money from their husband/partner for daily cooking seemed higher among near miss women with live birth compared to uncomplicated delivery women. However, this was not significant at the 5% level Model 1(b). Model 2 in the same table which compares household economic indicators between near miss women with perinatal death and their controls indicated that near miss women with perinatal death were significantly less likely to consume relatively expensive foods in a week, and that their households were significantly likely to spend less money on children’s education compared to uncomplicated delivery households Model 2(b). Moreover, we found at the 10% significance level, that near miss women with perinatal death consumed significantly less meat, fish and milk, 4-5 years after the near miss event Model 2(b). However, we were surprised to find that near miss women with perinatal death were more likely to eat the day before the interview as compared to uncomplicated delivery women Model 2(b). Model 3(b) which presents the comparison of household level economic indicators between near miss women with abortion and their controls after matching showed that households of near miss women with abortion consumed significantly less milk a week, up to 4-5 years after the event. We also found at the 10% level, that they spent less money on children’s education.

Table 11 presents the results of the long term effects of each type of near miss complication on household food security. Most of the indicators that measured food security with and without matching at household level appeared with a negative sign for all models. This may be a signal of persisting negative long term effect of the near miss event. Looking again at results after matching, we found a number of statistically significant indicators only in model 3(b). In particular, we found that near miss women with abortion were significantly likely to reduce the number of meals they eat per day. Their households were also more likely to buy food on credit compared to uncomplicated delivery households. We also found, at the 10% significance level, that near miss women with abortion were more likely to limit their meal portions to ensure more food for their children. In addition, they were also more likely to limit household members’ meal portions, to reduce household expenses on other essential needs and to eat non-desired foods, compared to uncomplicated delivery women. We were surprised to find at the 10% level that near miss women with abortion and their household were likely to experience more food security as measured by the food insecurity index compared to their counterparts.

Table S1 presents the results of the long term effects of each type of near miss complication on women’s subjective quality of life. The appearance of the negative sign for most indicators with and without matching in Model 1 which compares women’s quality of life between near miss women with live birth and their controls may suggest a negative impact. Looking at statistically significant results after matching, we found that near miss women with live birth were significantly less likely to be satisfied with their health and with their quality of life. They were aslo significantly less likely to appreciate the meaning of their life, 4/5 years after the near miss event compared to uncomplicated delivery women Model 1(b). In addition, we found at the 10% level that they were less likely to appreciate: the cleanliness of their environment, their access to information in their daily life, their living conditions and finally, less likely to be satisfied with their overall quality of life as measured through the quality of life index, up to 4-5 years after the near miss event Model 1(b).

On the contrary, the occurrence of positive signs with and without matching when comparing women’s quality of life between near miss women with perinatal death and their controls (Model 2) and near miss women with abortion and their controls (Model 3) seems to indicate that the impact of the near miss event is evened out over time. In particular, we found after matching, that near miss women with perinatal death were significantly more likely to appreciate the meaning of their life and to experience less negative feelings about suicide, anxiety and depression in the month preceding the interview compared to women with uncomplicated delivery Model 2(b). Furthermore, we found, at the 10% significance level, that they were likely to be satisfied with their life, with their capacity to find time to relax and with their relations with people Model 2(b). We also found at the 10% significance level, that near miss women with abortion were likely to be satisfied with their relations with people Model 3(b).