The epidemiology of pregnancy outcomes in rural Burkina Faso

Authors


Corresponding Author J.S. Bell, Immpact, University of Aberdeen, Health Sciences Building, Foresterhill, Aberdeen AB25 2ZD, UK.
Tel.: +44 1224 553672; Fax: +44 1224 555704; E-mail: j.bell@abdn.ac.uk

Summary

Objectives  To describe levels and causes of pregnancy-related mortality and selected outcomes after pregnancy (OAP) in two districts of Burkina Faso.

Methods  A household census was conducted in the two study districts, recording household deaths to women aged 12–49 years from 2001 to 2006. Questions on pregnancy outcomes in the last 5 years for resident women of reproductive age were included, and an additional method – direct sisterhood – was added in part of the area. Adult female deaths were followed-up with verbal autopsies (VA) with household members. A probabilistic model for interpreting VA data (InterVA-M) was used to determine distributions of probable causes of death. An OAP survey was conducted among all women with an experience of pregnancy during the prior 12 months. It aimed to document physical and psychological disabilities, economic and social consequences and discomfort that women may suffer as a result of a pregnancy.

Results  The maternal mortality ratio (MMR) was 441 per 100 000 live births (95% CI: 397, 485), significantly higher in Diapaga [519 per 100 000 (95% CI: 454, 584)] than Ouargaye [353 per 100 000 (95% CI: 295, 411)]. MMRs were associated with wealth quintile, age and distance from a health facility. The causes of death showed higher than expected rates of sepsis (30%) and lower rates of haemorrhage (7%). A substantial proportion of all women had difficulty performing day-to-day tasks as a consequence of pregnancy. Women who had experienced stillbirths or Caesarean sections reported symptom-related indicators of poor physical health more frequently than women reporting uncomplicated deliveries, and were also more likely to be depressed.

Conclusions  Expectations on the levels and causes of pregnancy-related mortality in Burkina Faso may need to be re-examined, and this could have programmatic implications; for example high levels of sepsis could prompt renewed efforts to reach women with skilled attendance at delivery and follow-up during the postpartum period. Further documentation of how complication-induced disabilities affect women and their families is needed. For mortality and morbidity outcomes, demonstrating variation between study districts is important to empower local decision makers with evidence of need at a subnational level.

Introduction

The attention of the international safe motherhood community has recently become more focused on the measurement of maternal mortality. This follows the launch of the fifth Millennium Development Goal (MDG5), which aims to reduce the burden of maternal death by 75% before 2015. National and subnational estimates of this key health outcome are essential if progress towards this target is to be monitored, as well as being valuable for developing policy and planning resource allocation. In many developing countries, however, the data available are inadequate for providing precise estimates, even at a national level. Furthermore, all methodologies for measuring maternal mortality have limitations, in the absence of complete vital registration systems. In recent years, sample surveys have tended to use the sisterhood method, identifying deaths to respondents’ sisters, but these require large sample sizes to yield reliable current or subnational estimates and are very expensive, often prohibitively so (Hill et al. 2001). It has also been demonstrated that sisterhood studies may systematically underestimate the true level of all-cause mortality in some countries (Stanton et al. 1997; Hill et al. 2006).

Restricting the focus of attention to maternal mortality undervalues the burden of disease attributable to pregnancy-related causes, because for every woman who dies many more survive severe obstetric complications and unsafe abortions, with potentially disabling consequences (Filippi et al. 1998; Mantel et al. 1998; UN Millennium Project 2005). These consequences comprise diagnosed and perceived illness or disability following pregnancy (including psychiatric illness) and direct social and economic consequences of events related to the end of pregnancy. Attention to these outcomes, combined with information on maternal death, creates a more holistic and longer-term perspective on maternal health.

In 2001, Family Care International introduced an intervention to improve access to skilled care at delivery in Ouargaye district in south-eastern Burkina Faso: the Skilled Care Initiative. The intervention was implemented so that different components covered various parts of the intervention district at different times. The evaluation of this intervention comprised many elements, which are described in detail elsewhere (Hounton et al. 2008). In this descriptive analysis, we examine maternal mortality and outcomes after pregnancy (OAP) in the two districts of Burkina Faso that comprised the intervention and comparison areas in the evaluation study. The findings presented here are not intended to demonstrate the efficacy of the intervention, but rather to illustrate the nature and magnitude of subnational variation in the outcomes of interest.

The districts were predominantly rural, and the population was poor and largely uneducated (6% female literacy in rural areas, INSD 2004). Healthcare availability was limited, with 31.2% of deliveries taking place in health facilities in rural areas (INSD 2004), and only 5.5% in hospitals in the rural district of Nouna (Jahn et al. 2006). The national Caesarean section rate was one of the lowest in the world at 0.7%, well below the recommended minimum of 5% (UNICEF 1997). The population differed by district in terms of religious practice: over 70% of the population Ouargaye was Muslims, while in Diapaga animism dominated (INSD 2004).

Methods

Maternal mortality

Definitions.

The standard international definition of maternal mortality (World Health Organization 1994) includes deaths due to direct and indirect obstetric causes, but excludes those due to accidental or incidental causes. In practice, it can be difficult to make this distinction, particularly in developing countries where reporting often relies on relatives’ statements, and medical autopsy is extremely rare. In recognition of this, the WHO created a categorisation referred to as ‘pregnancy-related death’, encompassing all deaths during the interval of time from onset of pregnancy until 6 weeks after the end of pregnancy. This is the definition we have employed, but as is still the convention, we refer to certain summary measures using the term ‘maternal’.

Data capture.

A household census was conducted in the two study districts, which had a total population of 515 298 people in 86 378 households. The questionnaire focused primarily on gathering pregnancy-related and perinatal indicators (Hounton et al. 2008). All household deaths were recorded for residents aged 12–49 years over the reference period of the study (January 2001–early 2006). The census questionnaire included questions on all pregnancy outcomes in the last 5 years for resident women of reproductive age (defined as 15–49 years). Given the lack of other estimates of pregnancy-related mortality for this part of Burkina Faso, the decision was made to include an additional method – direct sisterhood1– to the data capture process for a representative portion of both districts; this provided a means of internal validation of the results from the census. The three standard questions2 commonly used to ascertain the timing of death in relation to pregnancy in direct sisterhood methodology were amalgamated, and asked as one question.

In households identified as having one or more adult female deaths, verbal autopsies (VA) were conducted with household members in an attempt to determine the likely cause(s) of the deaths. The VAS were based on previous work in Burkina Faso that identified a list of signs, symptoms and causes of death for women of reproductive age, and was agreed by an experienced international physician panel. Probabilities reflecting the occurrence of each cause and each indicator among female deaths in the 15–49 years age range, and for each indicator given a specific cause, were determined by the panel using a Delphi technique. The InterVA-M model then used a Bayesian approach to process these probabilities, and presented up to three causes of death, each with a percentage likelihood, and a likelihood that each case was related to pregnancy (InterVA-M website 2007; Fottrell et al. 2007). A broader assessment of adult deaths by cause is not part of the current analysis.

Analysis

Information from the census was used to calculate estimates of pregnancy-related mortality: the maternal mortality ratio (MMR), defined here as the total number of pregnancy-related deaths among women aged 15–49 years in the study period, divided by the total number of live births in the same period; and the maternal mortality rate (MMrate), defined as the total number of pregnancy-related deaths among women of reproductive age during the study period, divided by the exposure time of women in the population aged 15–49 years. As deaths to women were reported at a household level, but live births were asked only of women present at the time of the interview, the number of live births was adjusted to account for this. Similar analyses were applied to the data on sisters: using pregnancy-related deaths among sisters as the numerator and the fertility reported in the census overall to generate MMR; and the exposure amongst sisters to generate MMrate.

Maternal outcomes were stratified by health district, wealth quintile, age group and distance to the nearest health facility. The outcome measures obtained using household census deaths were compared with those obtained from the sisterhood method. Wealth quintiles were constructed using the asset-based approach developed by the World Bank3 (Gwatkin et al. 2000). Analysis was conducted using Stata v9.

For all the census data, quality assurance mechanisms for collection and management were in place to ensure standard operating and monitoring procedures across both intervention and comparison districts (Hounton et al. 2008).

Outcomes after pregnancy

An OAP survey was conducted among all women with an experience of pregnancy (uncomplicated or complicated) during the 12 months prior to the census. It aimed to document physical and psychological disabilities; economic as well as social consequences; and discomfort that women may have suffered as a result of a pregnancy and its outcome.

Questions were derived from a concurrent multidisciplinary longitudinal study in Burkina Faso conducted by Immpact, which aimed to document the range of health, economic and social consequences of severe obstetric complications on the health and lives of women (Filippi et al. 2007; Storeng et al. 2008). A subset of these questions were used in the current survey; these included questions on health and disabilities and K10 questions on depression and anxiety (Kessler et al. 2002), applying a score of 14 or above to discriminate women with high risk of depression (Baggaley et al. 2007). Information on stillbirths, early pregnancy loss and Caesarean sections was self-reported by women in the census. In Ouargaye, all eligible women who consented participated in the survey, while in Diapaga only a proportion (44%) of eligible respondents did so because of time and resource constraints. The total sample for the OAP questionnaire was 13 587 women. For this paper two analyses are conducted: comparing women’s reports of their experiences after pregnancy: (i) by district, and (ii) by obstetric complication (using stillbirth and Caesarean section as proxies); and comparing women who had experienced either of these two outcomes to the remainder of the sample, who were inferred to have had an uncomplicated delivery.

The overall Immpact research proposal was approved by the Ministry of Health National Health Research Ethics Committee (Ouagadougou, Burkina Faso). The specific Evaluation and Evidence Research Group protocol was approved by Centre MURAZ (Bobo-Dioulasso, Burkina Faso) Institutional Review Board. Administrative authorisations were obtained at all level of the administrative chain (Ministry of Health, Region Governorates, Regional Directorates of Health, National and Regional Hospital Directorates, Province High Commissioners, District Health Management Teams, Heads of Clinical Services in Hospitals and village community leaders).

Results

Maternal mortality

The census identified a total of 1180 deaths among women aged 15–49 years over the period 2002–2006, of which 385 were subsequently characterised as pregnancy-related deaths (Table 1), giving a mortality rate among women of reproductive age of 2.6 (95% CI: 2.4, 2.8) and proportion maternal among deaths of females (PMDF) of 33%. Data for 2001 were also gathered, but excluded from the analysis because of evident recall bias in the reporting of births, with reduced number of births across all age categories in this year (results not shown). There was no such reporting pattern detected among female deaths or pregnacy related deaths, which were fairly evenly distributed by year. Because data from 2001 were excluded, all the number of births and deaths reported have been adjusted so that a proportion of those events with missing year are included in the total (= 57 for pregnancy-related deaths and = 512 for births). The number of births has also been adjusted to take into account those births to women resident in a household who were not present to be interviewed about their birth histories at the time of the census. In the same period, 112 956 women were interviewed and 82 289 live births documented.

Table 1.   Pregnancy-related mortality by health district (2002–2006), Ouargaye and Diapaga districts, Burkina Faso
 OuargayeDiapagaTotal
  1. MMrate, maternal mortality rate; MMR, maternal mortality ratio.

  2. †Number of deaths adjusted to include the correct proportion of deaths missing date of death (= 57).

  3. ‡Proportion maternal among deaths of females (PMDF) i.e. the proportion of deaths to women of reproductive age that are pregnancy-related.

  4. §Number of births adjusted by area to include the correct proportion missing date of birth and for resident women of reproductive age not interviewed.

  5. ¶General fertility rate = (adjusted live births)/(total exposure).

  6. ††Exposure periods in denominator of MMrate assume that resident women of reproductive age had been resident for a maximum period of the previous 4 years and 4 months, with further adjustments made for age if under 20 years.

Women of reproductive age interviewed51 64061 316112 956
Women of reproductive age deaths†5925881180
Pregnancy-related deaths†143243385
PMDF‡ (%)244133
Live births reported37 20745 08382 289
Live births adjusted§40 45646 77487 230
General fertility rate¶192.6192.7192.7
MMrate/10 000 women-years (95% CI)††6.8 (5.7, 7.9)10.0 (8.7, 11.2)8.5 (7.6, 9.3)
MMR/100 000 live births (95% CI)353 (295, 411)519 (454, 584)441 (397, 485)

The calculated MMR for the study area and period as a whole was 441 per 100 000 live births (95% CI: 397, 485), with a significantly higher level in Diapaga, 519 per 100 000 live births (95% CI: 454, 584) than Ouargaye, 353 per 100 000 live births (95% CI: 295, 411). The MMrates reflected the same pattern, and also showed a statistically significant difference (< 0.001). The PMDF was also significantly higher in Diapaga (41%) than Ouargaye (24%).

Table 2 shows the outcomes obtained using the household and sisterhood methods in the subsamples where both sets of data were collected. For the two districts combined, the MMRs were not significantly different between the methods (= 0.12); although in Diapaga the difference in MMR was significant, with a sisterhood MMR of 591 per 100 000 live births (95% CI: 468, 715) and a household MMR of 394 per 100 000 live births (95% CI: 305, 482).

Table 2.   Pregnancy-related mortality by health district and method of calculation for a subsample of households† in Ouargaye and Diapaga districts, Burkina Faso (2002–2006)
DistrictSourceNo. of respondentsNo. of sistersNo. of WRA deaths‡No. of PR deaths‡ PMDF§ (%)MM rate (95% CI) Live births†† MMR (95% CI)
  1. WRA, women of reproductive age (15–49 years); PR, pregnancy-related; PMDF, proportion maternal among deaths of females; MMrate, maternal mortality rate/1000 women-years; MMR, maternal mortality ratio/100,000 live births.

  2. †Data collected in 234 enumeration areas for the sisterhood analysis, and restricted to these areas for comparison with the household analysis.

  3. ‡Number of deaths adjusted to include the correct proportion of deaths missing date of death (= 57).

  4. §Proportion maternal among deaths of females of reproductive age, i.e. the proportion of pregnancy-related deaths (WHO 2004).

  5. ¶Exposure periods in denominator of MMrate assume that resident women of reproductive age had been resident for a maximum period of the previous 4 years and 4 months, with further adjustments made for age if under 20 years.

  6. ††Number of births adjusted by area to include the correct proportion missing date of birth and for women of reproductive age not interviewed.

OuargayeSisterhood10 70420 28732557176.1 (4.5, 7.7)n/a343 (253, 432)
Household28 398n/a33782247.3 (5.7, 8.8)19 935379 (297, 461)
DiapagaSisterhood916117 319209884211.2 (8.8, 13.5)n/a591 (468, 715)
Household24 540n/a18876408.0 (6.2, 9.8)18 718394 (305, 482)
TotalSisterhood19 86537 606533145278.4 (7.0, 9.8)n/a459 (384, 533)
Household52 938n/a525147287.6 (6.4, 8.8)38 652387 (327, 447)

Looking at the two districts together, the distributions of MMRs and MMrates by wealth quintile show some evidence of an association, with the highest levels among women classified in the middle and poor sections of society, and the lowest levels in the two least poor quintiles (Table 3). With the highest mortality levels in the middle quintile, the relationship is not linear, but still suggestive of a poor–rich gap. This relationship holds true in Ouargaye, but in Diapaga the relationship was reversed, with higher levels of mortality in the richest quintiles, indicating the possibility that the selected assets were less appropriate for this part of Burkina Faso (results not shown).

Table 3.   Pregnancy-related mortality by wealth quintile for Ouargaye and Diapaga districts combined, Burkina Faso (2002–2006)
Wealth quintilePregnancy-related deathsLive births (adjusted) MMR (95% CI) MMrate (95% CI)
  1. MMrate, maternal mortality rate/1000 women-years; MMR, maternal mortality ratio/100,000 live births.

1 Poorest8718 168480 (379, 580)9.4 (7.4, 11.4)
28516 736508 (401, 616)9.9 (7.8, 12.0)
38716 692521 (411, 630)10.2 (8.0, 12.3)
46717 757377 (287, 467)7.3 (5.6, 9.1)
5 Least poor5817 816328 (244, 412)5.9 (4.4, 7.4)

As expected, examining MMR by age group gives a ‘J-shaped’ distribution in both areas (Table 4). However, the differentials were greater in Diapaga, and the high mortality levels among women over 25 years found here were surprising.

Table 4.   Pregnancy-related mortality by age group for Ouargaye and Diapaga districts combined, Burkina Faso (2002–2006)
Age group (years)Pregnancy-related deathsLive births MMR (95% CI) MMrate (95% CI)
  1. MMrate, maternal mortality rate/10,000 women-years; MMR, maternal mortality ratio/100,000 live births.

15–195818 330317 (235, 398)4.80 (3.57, 6.04)
20–247226 393271 (208, 334)7.06 (5.43, 8.70)
25–299619 740487 (390, 584)11.44 (9.16, 13.72)
30–346911 553601 (460, 742)11.93 (9.13, 14.74)
35–39486732714 (513, 915)10.45 (7.50, 13.40)
40–494232851262 (880, 1643)9.60 (6.68, 12.52)

Table 5 shows that in Ouargaye women who live within 5 km of a health facility were less likely to die of pregnancy-related causes than women who live further away [MMR: 292 (95% CI: 222, 363) per 100 000 live births compared with 429 (95% CI: 333, 525)]. There was no such pattern in Diapaga.

Table 5.   Pregnancy-related mortality by distance to nearest health facility in Ouargaye and Diapaga districts, Burkina Faso (2002–2006)
Distance Pregnancy-related deaths† Births†MMratio per 100 000 live births (95% CI)MMrate per 1000 women-years (95% CI)
  1. The totals do not exactly sum within areas as the inflation factor for the increase in births has been calculated for subdivisions within areas.

  2. †All births and deaths have been adjusted for missing year; births have been adjusted for missing women (resident absent); the adjustment factors have been calculated specific to the categorisation under analysis; all analysis has been restricted to the years 2002–2006; and the exposure has been limited to a maximum of 52 months.

Ouargaye
 ≤5 km6722 848292 (222, 363)5.48 (4.16, 6.79)
 >5 km7617 611429 (333, 525)8.56 (6.63, 10.40)
 14340 456  
Diapaga
 ≤5 km8916 669537 (425, 648)9.72 (7.71, 11.74)
 >5 km15230 106506 (453, 584)10.05 (8.46, 11.65)
 24346 774  
Both
 ≤5 km15539 488394 (332, 456)7.26 (6.12, 8.41)
 >5 km22847 696478 (416, 539)9.50 (8.27, 10.74)
 38587 230  

The distributions of probable causes of death for the 864 adult females with completed VAs are shown in Table 6. For the 321 reported pregnancy-related deaths there is a surprisingly low likelihood that the death was due to haemorrhage (7% of the pregnancy-related deaths) and a high likelihood of pregnancy-related sepsis (30% of pregnancy-related deaths). Of the other direct obstetric causes: 2.0% are attributed to abortion; 1.5% to ectopic pregnancy; 1.9% to obstructed labour; 7.0% to pregnancy-induced hypertension and 1.4% to ruptured uterus. The remaining 48.1% of pregnancy-related deaths are classified as indirect causes, with only 0.7% considered ‘indeterminate’.

Table 6.   Percentage distribution of causes of death using InterVA-M for 864 adult female deaths 15–49 years with completed verbal autopsy, Ouargaye and Diapaga districts, Burkina Faso (2002–2006)
Cause of death All deaths (= 864)Reported pregnancy-related† (= 321)Additional pregnancy-related‡ (= 75)Not pregnancy-related§ (= 468)
  1. †Deaths reported in census or verbal autopsy as pregnancy-related.

  2. ‡Deaths not reported in census or verbal autopsy as pregnancy-related, but identified by InterVA-M as pregnancy-related.

  3. §Deaths not reported in census or verbal autopsy as pregnancy-related and not identified as pregnancy-related by InterVA-M.

Abortion-related death22121
Anaemia56242
Cancer1221019
Cardiovascular disease7747
Diabetes2500
Ectopic pregnancy1100
HIV/AIDS-related death54115
Haemorrhage2700
Injury1001
Kidney disease2261
Liver disease3204
Malaria1717618
Non-pregnancy-related infection1001
Obstructed labour1200
Other maternal cause0100
Pregnancy-induced hypertension3730
Pregnancy-related sepsis1330181
Respiratory disease0001
Ruptured uterus0100
Suicide1001
Tuberculosis (pulmonary)172628
Indeterminate5108
Total100100100100

In addition, some deaths were not directly reported by interviewees as pregnancy-related (i.e. not reported as occurring during pregnancy, delivery or up to 6 weeks after the end of pregnancy), but given a high probability of being pregnancy related by the InterVA-M model. These cases included relatively high levels of abortion- and HIV-related deaths (12% and 11%), and of anaemia (24%). If these deaths were included as pregnancy related in the MMR, it would rise to 522 per 100 000 live births (95% CI: 474, 570).

Further results from InterVA-M in Table 7 show how the assigned causes are distributed by age group, parity and timing of the death in relation to pregnancy. Of the 210 pregnancy-related deaths where death occurred within 6 weeks after delivery, 108 (51.4%) delivered at home, 71 (33.8%) delivered in a facility and 31 (14.8%) did not have a place of delivery recorded. Twenty pregnancy-related deaths (2.31%) were reported following Caesarean delivery.

Table 7.   Percentage distribution of causes of death using InterVA-M by age group, parity and timing of death in relation to pregnancy for 321 pregnancy-related deaths aged 15–49 years with completed verbal autopsy, Ouargaye and Diapaga districts, Burkina Faso (2002–2006)
Cause of deathAge group (years)Parity (birth order)Timing
All15–1920–3435+12–4>4PregnantWithin 6 weeks following pregnancy end
Early loss<24 h after delivery>24 h after delivery
n (%)321 (100)49 (15.3)140 (43.6)132 (41.1)52 (16.2)102 (31. 8)167 (52.0)81 (25.2)30 (9.4)55 (17.1)155 (48.3)
Abortion-related death242142202200
Anaemia616937712265
Cancer20140043312
Cardiovascular disease71314141153108
Diabetes524723740123
Ectopic pregnancy11211126000
HIV/AIDS-related death41542543544
Haemorrhage746956744242
Kidney disease21222121602
Liver disease20210312002
Malaria17421874120816181618
Obstructed labour21402314050
Other maternal cause11111213001
Pregnancy-induced hypertension767897716824
Pregnancy-related sepsis302932282431318261946
Ruptured uterus12120125000
Suicide00000101000
Tuberculosis (pulmonary)23232146311
Indeterminate12100200001
Total100100100100100100100100100100100

Differences between the deaths with a completed VA (= 321) and those without (= 64) were checked for evidence of selection bias by examining household attributes. Those with a VA tended to have less assets and amenities recorded, but the differences between the two groups were not large or statistically significant.

Outcomes after pregnancy

Table 8 presents background characteristics of the women participating in the OAP survey, by district and by type of pregnancy outcome. Because of the large sample size, unremarkable differences between districts are likely to be highly significant. However, of particular note, is that this subsample from the census also displayed significant socio-economic differences between the two districts, in particular the wealth quintiles showed very different distributions, with Diapaga being considerably poorer. Women experiencing a stillbirth did not appear to differ markedly from women with an uncomplicated delivery, but women undergoing Caesarean section appeared to have higher socio-economic status: they were more likely to have had some formal education and to be in the least poor wealth quintile. Table 9 presents OAP results by district. Women in Ouargaye were significantly less likely to report symptom-related indicators of poor physical health, such as feeling unwell and having had a serious illness or health problem since the end of pregnancy. However, conversely, they were more likely to report disability-related indicators, such as difficulty fetching wood and water, grinding maize or doing agricultural work.

Table 8.   Outcomes after pregnancy sample background characteristics, by district and by pregnancy outcome, Ouargaye and Diapaga districts, Burkina Faso
 DistrictBirth outcome
OuargayeDiapagaNormal delivery†StillbirthCaesarean section
  1. SD, standard deviation.

  2. †Excludes 55 women with early pregnancy loss.

  3. ‡121 values for age dropped because of suspected errors (age <12 years or >55 years).

Total sample size8121546613 03543166
Mean age in years‡ (SD)< 0.001= 0.232
 27.3 (6.9)26.2 (7.2)26.9 (7.0)26.7 (7.7)25.9 (6.6)
Some formal education< 0.001= 0.001
 4.0%6.9%5.0%6.3%15.2%
Marital status< 0.001< 0.001
 Single2.1%4.1%2.8%5.3%4.6%
 Married, polygamous43.6%40.7%42.8%33.6%37.9%
 Married, monogamous54.3%55.2%54.4%61.0%57.6%
Wealth quintiles< 0.001= 0.258
 Most poor5.0%30.8%15.4%14.4%15.2%
 28.3%27.4%15.9%19.0%12.1%
 319.1%20.0%19.5%18.8%13.6%
 431.4%12.1%23.7%21.1%21.2%
 Least poor36.3%9.7%25.5%26.7%37.9%
Table 9.   Outcomes after pregnancy results for Ouargaye and Diapaga districts, Burkina Faso
 OuargayeDiapagaAdjusted OR (95% CI)‡†
  1. †Taking Diapaga district as baseline.

  2. ‡Adjusted for age, education (some or none), marital status [single, married (monogamous) and married (polygamous)], time since end of pregnancy, wealth quintile and distance to nearest health facility.

  3. §Additional adjustment for death of baby since delivery.

Physical health
 Not feeling well on day of interview 3.3%8.7%0.39 (0.32, 0.47)
 Serious illness since pregnancy10.4%16.2%0.63 (0.55, 0.72)
 Severe anaemia (≤8.0 g/dl)3.5%5.7%0.91 (0.43, 1.93)
 Currently pregnant1.3%3.5%0.36 (0.26, 0.50)
 Difficulty fetching wood or water, grinding maize or doing agricultural work38.3%28.0%1.59 (1.44, 1.76)
Mental health
 Depression: K10 ≥ 14 points19.1%17.7%1.05 (0.94, 1.19)§
Social support
 Negative change in people’s attitude to respondent since pregnancy3.6%3.9%0.93 (0.73, 1.18)
Economic consequences
 Regular income decreased since pregnancy (of those not answering ‘do not know’76.1%47.8%3.22 (2.86, 3.62)
 Not enough food in household since pregnancy10.0%10.7%1.12 (0.96, 1.30)
 Had to borrow to pay for the delivery expenses9.6%4.6%2.55 (2.05, 3.16)

Further interesting results were obtained by examining outcomes by type of delivery (Table 10). Results suggest that there were significant differences in the well-being of women according to the circumstances surrounding their deliveries. Women with stillbirths (= 431) or Caesarean sections (= 66) reported symptom-related indicators of poor physical health more frequently than women reporting an uncomplicated delivery. A substantial proportion of all women, generally around 20% for each task, had difficulty performing day-to-day tasks such as cooking, cleaning, washing and fetching wood or water, grinding maize and performing agricultural work (data for each task not shown). This suggests the extent of the problems that women in general face after childbirth in this setting. There was no significant difference in these disability-related indicators of physical health for women with stillbirths and women with uncomplicated deliveries, but for women with Caesarean sections, the frequency of difficulty with all tasks was significantly greater; for example twice as many of these women reported difficulty in grinding maize compared with the uncomplicated delivery group (< 0.001).

Table 10.   Outcomes after pregnancy results by pregnancy outcome, Ouargaye and Diapaga districts, Burkina Faso
 Normal delivery†StillbirthAdjusted OR‡ (95% CI)Caesarean sectionAdjusted OR‡ (95% CI)
  1. †Excludes 55 women with early pregnancy loss.

  2. ‡Adjusted for age, education (some or none), marital status [single, married (monogamous) and married (polygamous)], time since end of pregnancy, wealth quintile and distance to nearest health facility. Taking normal delivery as baseline.

  3. §Additional adjustment for death of baby since delivery.

Physical health
 Not feeling well on day of interview 5.1%13.2%2.34 (1.63, 3.35)9.1%1.85 (0.72, 4.74)
 Serious illness since pregnancy12.4%19.3%1.46 (1.09, 1.95)28.8%3.24 (1.84, 5.69)
 Severe anaemia (≤8.0 g/dl)4.2%5.7%1.01 (0.13, 7.86)33.3%10.47 (0.93, 118.14)
 Currently pregnant1.7%15.6%13.21 (9.47, 18.41)0.0%-
 Difficulty fetching wood or water, grinding maize or doing agricultural work33.9%37.7%1.18 (0.94, 1.48)50.0%2.36 (1.36, 4.10)
Mental health
 Depression: K10 ≥ 14 points18.1%27.4%1.65 (1.29, 2.11)28.8%1.74 (0.94, 3.21)§
Social support
 Negative change in people’s attitude to respondent since pregnancy3.4%11.4%3.57 (2.53, 5.05)7.6%2.10 (0.82, 5.37)
Economic consequences
 Regular income decreased since pregnancy (excluding those answering ‘do not know’)65.4%65.9%0.86 (0.66, 1.12)68.1%1.69 (0.81, 3.50)
 Not enough food in household since pregnancy10.2%11.9%1.02 (0.72, 1.45)13.6%1.43 (0.67, 3.06)
 Had to borrow to pay for the delivery expenses7.5%7.8%0.89 (0.57, 1.40)21.0%3.50 (1.86, 6.59)

These results extend to mental as well as physical well-being. Using the K10 depression score, and a cut-off of ≥14 to denote women at high risk of depression, a significantly larger proportion of women were at high risk among the stillbirth group (27.4%) and Caesarean section group (28.8%), compared with the uncomplicated delivery group (18.1%). While the significance of these psychological differences was greatly reduced after adjustment for risk factors such as marital status and time since end of pregnancy in the Caesarean section group, this is likely because of the small numbers with this mode of delivery (= 66). In terms of social support, more women in the groups with complications felt that people’s attitudes to them had changed in a negative way since the end of their pregnancy. They also more often reported feeling sidelined by people, or that relations with their husband had worsened since the end of pregnancy (data not shown). However, most of these differences for social support outcomes did not reach statistical significance. There were no significant differences between the groups in terms of economic outcomes, except for borrowing to pay for delivery expenses, where women with Caesarean section were more likely to have borrowed. More than 10% of all women reported not always having enough food in the household since the delivery.

Discussion

It is often difficult to externally validate estimates of pregnancy-related mortality because it is so rarely measured. However, there have been several recent surveys capturing pregnancy-related mortality in Burkina Faso, and the estimate for MMR obtained from the Immpact census (441 deaths per 100 000 live births) compares remarkably well with these. In 1999, a national MMR of 483 was recorded using the direct sisterhood method (INSD 2000), and an Immpact survey, ‘Sampling at service sites: markets’, which was conducted in Ouargaye at the same time as the census (and also used the direct sisterhood method) found an MMR of 397 per 100 000 live births (95% CI: 254, 540). This was very similar to the census estimates for Ouargaye of 353 per 100 000 live births (95% CI: 295, 411) using the deaths in the household, and 343 per 100 000 live births (95% CI: 253, 432) using the sisterhood method. In another rural district, Hounde, a census in 2001 recorded an MMR of 406 per 100 000 live births (95% CI: 281, 566), although the low PMDF (17%) suggests that this may have been an under-estimate (Sombiéet al. 2005). Although the WHO/UNICEF/UNFPA modelled estimate, of 700 maternal deaths per 100 000 live births (uncertainty range: 390, 1000) for the year 2005 (WHO 2007), is higher than all of these empirical survey estimates, they all fall within its range of uncertainty.

The belief that all-cause mortality is under-reported in some sisterhood surveys is the reason why the WHO maternal mortality estimates differ from recent survey estimates (Stanton et al. 1997; Hill et al. 2006). For several countries, including Burkina Faso, MMRs are estimated by taking the age-adjusted PMDFs from DHS household surveys, and applying them to the predicted number of adult female deaths (WHO 2007). However, this is not simply a limitation of the sisterhood method, as evidenced by the lower MMR among household members than among sisters, in the areas where both were measured. The all-cause adult female mortality rate found in this study is lower than that reported in Nouna – 6.0 per 1000 person-years, for the period 1995–1998: calculated using published figures (INDEPTH Network 2002) and by the DHS – 4.5 per 1000 person-years for 1997–2003 (INSD 2004). However, it is not clear whether the differences are due to subnational variation or under-reporting, and this is an area that warrants further investigation. Possible factors that may have led to under-reporting in this census include the data capture process, which used male interviewers to question male heads of household (for household deaths) and female respondents (for the sisterhood method); possible lack of knowledge on the part of the interviewees, together with cultural or religious barriers may limit the information reported; and the low literacy levels in this population may affect the accuracy of reported timings of events such as births and deaths. Different biases may operate for the numerator (deaths) and denominator (births), or change over time. For example: deaths may be more memorable than births, and thus less likely to be subject to recall bias; there may be taboos on reporting recent deaths to adults; and the reporting of deaths in the household depends on the household remaining intact, which is undermined by a recognised tendency for households to dissolve following such a death (United Nations 1998), while the reporting of births only depends on individual women being resident. It is not possible to quantify the effect that such conflicting biases may have on the maternal mortality estimates; of the examples listed one would tend to raise the estimate and the others to lower it.

Other possible limitations include inaccuracies in the estimation of the denominator. For the MMrate the exposure (women-years) is calculated based on the women resident at time of the census, which may result in either over- or under-estimation of the true denominator; it is also possible that women may be counted more than once when a full enumeration is conducted, unless clear rules for residency are applied. For the MMR calculation, data on live births were collected from women who were resident at the time of the survey, and so again the number of births in the household could be under- or over-estimated.

Turning to internal validity, the comparison between the direct household estimation and sisterhood method in a sample of households produced very similar outcomes in Ouargaye, although in Diapaga both rates and ratios were significantly different from one another between methods. Overall, for the combined districts, the MMRs measured by the two methods were not found to be significantly different (= 0.12), and this adds to our confidence in the findings. Also the PMDFs are generally within expected limits. The overall PMDF of 33% is close to the level of 37% used in the WHO, UNICEF and UNFPA model for 2000 (WHO 2004), although somewhat higher than the recently revised estimate for 2005 (PMDF: 25%).

The distribution of pregnancy-related mortality by wealth seen in Ouargaye, with the risk of mortality increasing as poverty increases, has been described previously in other settings (Graham et al. 2004). The reverse pattern observed in Diapaga has not been reported previously, and may result from limitations in the calculation of wealth quintiles in this context where most of the population are rural and there is little variation in asset distribution.

During the VA more deaths were reported as being pregnancy-related than had been reported in response to the census question. It is not possible to ascertain whether this was because of the format of the questions, or some other factor. The three usual time-of-death questions were not used in the Immpact census, being replaced by one combined question, while in the VA three separate questions were asked about timing of death. The use of three separate questions is preferred, as this approach seems likely to improve recall, although no formal studies of data quality have been conducted to test this assumption (Hill et al. 2001).

It is interesting that those deaths not directly reported by interviewees as pregnancy-related, but given a high probability of being pregnancy-related by InterVA-M, showed relatively high levels of abortion and HIV-related deaths. This suggests there may have been a reluctance to report pregnancy status for culturally sensitive causes of death; and that therefore the MMR calculated to include these deaths might reflect true levels of pregnancy-related death more accurately. This pattern of under-reporting is expected, and the ability to detect such misreporting is a potential strength of InterVA-M. Statistical modelling also offers significant advantages in terms of efficiency, consistency and standardisation; although it may not reflect all the subtleties that reviewing physicians might apply to individual cases.

The distribution of causes of death derived using InterVA-M is surprising, and quite different from expectations based on global burden of disease estimates (Graham et al. 2006; Khan et al. 2006). Khan et al.’s 2006 recent systematic review found the following distribution of causes of maternal death for Africa: 68% direct obstetric causes, of which haemorrhage comprised 34%, sepsis 10%, hypertensive disorders 9%, obstructed labour 4% and abortion 4%. The main discrepancies are for haemorrhage, which at 6.7% is much lower than expected, and pregnancy-related sepsis at 30.0%, much higher than expected. Exploring this further, the patterns observed in the haemorrhage deaths were generally consistent with expectation: haemorrhage is seen to increase with age and parity, and levels peak in the 24 h following a delivery. The levels of pregnancy-related sepsis were unexpectedly high, particularly for deaths within 24 h of delivery, which accounted for 19% of all such deaths. However, most of the sepsis deaths occurred after the end of pregnancy, as expected, and deaths from sepsis during pregnancy can possibly be attributed to unsafe abortion. This raises the question of whether sepsis has been under-reported in other series, and deserves further investigation. It is worth noting here that a survey of health facilities in Burkina Faso found inadequate provision for the aseptic care needed to prevent sepsis (Family Care International, 2007).

Assessment of the InterVA-M model in relation to physician coding has already been performed in Burkina Faso, at the Nouna Demographic Surveillance Site. Here the model-generated outputs on cause of death were compared with physician diagnoses, reaching agreement levels of over 80%. Similar, relatively low levels of haemorrhage were recorded in the Nouna analysis (Fottrell et al. 2007).

The high proportion (41.8%) of pregnancy-related deaths that were indirect (non-obstetric) is also unusual, compared with other series (Khan et al. 2006), but may be suggestive of the high overall burden of ill health in African populations (Olsen et al. 2002; Meda et al. 1999). High levels of anaemia and malaria, for example, were observed, and although these causes of death are clearly not peculiar to pregnant and parturient women, it is well known that risk is elevated among women with these conditions. Levels of malaria mortality identified by InterVA-M in the Nouna study were also high, and there was a tendency for physicians to diagnose these deaths as non-pregnancy related infections or HIV (Fottrell et al. 2007). It is possible that the levels of malaria and sepsis reflect the common reporting of fever, to which, in the absence of other specific symptoms, InterVAM assigned a significant probability to these causes of death. Follow-up work is currently being undertaken to assess the validity of cause of death assignment in individual cases. Although the absence of local data on causes of death make it hard to confirm the patterns our data reveal, many of them are consistent with general knowledge on the epidemiology of pregnancy-related death (Ronsmans & Graham 2006): for example, the highest levels of abortion-related mortality are in the youngest age group; the highest levels of cancer, cardiovascular disease and diabetes are in the oldest age group; and malaria is highest among the young, primiparous women. Further analyses comparing reproductive-age sex-specific mortality patterns in similarly high burden of illness populations may also be helpful.

All women with a recent delivery, regardless of whether they had a complication or not, reported high levels of difficulties in their everyday lives, perhaps not surprisingly given their recent pregnancy and the economic context in which they live. Our study found a level of depression which can be considered as high for women in the postpartum period, particularly among women who delivered a stillbirth (27.4%) or women who had a Caesarean section (28.8%). For comparison, a recent review quoted by Lumley et al. (2004) found that the prevalence of postpartum depression ranges between 10% and 18% with a pooled prevalence at 13%. Studies in Africa have reported high prevalence of postpartum depression: including 14.6% in Nigeria (Adewuya et al. 2005), 16% in Zimbabwe (Nhiwatiwa et al. 1998) and 35% in South Africa (Cooper et al. 1999). In our setting, this high level might be partly explained by the rainy season, during which the survey took place, which coincides which a period of stress for the rural population because of difficulties finding food. The relationship between stillbirths and high level of depression has been reported in other populations and can be long lasting (Boyle et al. 1996). But our study challenges a common assumption that the occurrence of depression associated with stillbirth might be lower in populations where infant mortality is high (Boyle et al. 1996).

Our results are useful in showing that women with complications are significantly more at risk of suffering long-term ill health and disabilities. They underline the need for timely interventions – the longer the interval between onset of a complication and intervention, the higher the risk of stillbirth or emergency Caesarean section, with their associated adverse consequences – and also the need for postpartum services. Only a proportion (44%) of eligible respondents completed the OAP questionnaire in Diapaga, which may have led to sampling bias. Non-participants did not differ substantially from participants by wealth quintile (data not shown), but their distance to nearest health facility was significantly greater (7.3 km and 5.4 respectively, = 0.0001), but the analysis controlled for distance to minimise the impact of this bias.

Conclusions

Census and survey data have been used here to describe pregnancy-related mortality and morbidity at a district level. The MMRs measured were similar to those found in previous surveys, but lower than the WHO modelled estimate. Surprisingly high levels of sepsis and low levels of haemorrhage were found as causes of the pregnancy-related deaths. In the light of these findings, expectations on the levels and causes of pregnancy-related mortality in Burkina Faso may need to be re-examined, and if they are confirmed, programmatic implications could follow; for example, the high levels of sepsis detected may prompt a renewal of effort to reach women in the postpartum period and re-emphasise the need for skilled attendance at delivery. In relation to morbidity, women with complicated births were shown to suffer more than those with uncomplicated births in the first year after childbirth, particularly when they delivered by Caesarean section. It is important for further studies to document how these complication-induced disabilities influence the socio-economic status of women, their families and their communities in the long term. For both mortality and morbidity outcomes, the ability to demonstrate variation in estimates between health districts, as shown in this study, is essential to empower local decision makers with evidence of needs at the local level.

Footnotes

  • 1

     Used by the Demographic and Health Surveys (DHS), this is a variant of the sisterhood approach that relies on fewer assumptions than the original method (indirect sisterhood) and collects more information (i.e. the age of all siblings, age at death and year of death of those who have died, in addition to the information obtained by the indirect method).

  • 2

     Was (NAME) pregnant when she died, or did she die during childbirth? Did (NAME) die within 42 days after the end of pregnancy? Did (NAME) die because of complications of pregnancy of childbirth?

  • 3

     Assets included in the construction of wealth quintiles were: household ownership of radio, telephone, television, refrigerator, bicycle, motorbike, lorry, tractor, plough, cart, cows, sheep, goats, donkeys, pigs, chickens; source of drinking water; type of toilet facilities; cooking fuel; materials for construction of roof, walls and floor.

Acknowledgements

Thanks to Cynthia Stanton, who commented on earlier drafts of this paper. This work was undertaken as part of an international research programme – Immpact, funded by the Bill & Melinda Gates Foundation, the UK Department for International Development, the European Commission and USAID. The funders have no responsibility for the information provided or views expressed in this paper. The views expressed herein are solely those of the authors.

Conflict of interest

The authors have not declared any conflicts of interest.

Ancillary