To examine the relationship between distance to a health facility, consulting a health professional and maternal mortality.
To examine the relationship between distance to a health facility, consulting a health professional and maternal mortality.
Retrospective cohort study in Matlab, Bangladesh (1987–2005), to collect data on all pregnancies, births and deaths. In Java, Indonesia (2004–2005), an informant-based approach indentified maternal deaths and a population-based survey sampled women who survived birth. Logistic regression was used to examine the influence of distance to a health facility and uptake of a health professional on odds of dying.
Maternal mortality was 320 per 100 000 births (95% CI: 290, 353) in Indonesia and 318 per 100 000 (95% CI: 272, 369) in Bangladesh. Women who lived further from health centres in both countries were less likely to have their births attended by health professionals than those who lived closer. For women who were assisted by a health professional, the odds of dying increased with increasing distance from a health centre [odds ratio per km; Indonesia: 1.07 (95% CI: 1.02–1.11), Bangladesh: 1.47 (95% CI: 1.22–1.78)]. There was no evidence for an association between distance to a health centre and maternal death for women who were not assisted by a health professional.
Even in settings where health services are relatively close to women's homes, distance to a health facility affects maternal mortality for women giving birth with a health professional. Women may only seek professional care in an emergency and may be unable to reach timely care when living far away from a health centre.
Examiner la relation entre la distance à un centre de santé, la consultation d'un professionnel de la santé et la mortalité maternelle.
Etude de cohorte rétrospective à Matlab, au Bangladesh (1987–2005) pour collecter des données sur toutes les grossesses, les naissances et les décès. A Java, en Indonésie (2004–2005), une approche basée sur l'informateur a indentifié les décès maternels et une enquête basée sur la population a permi d’échantillonner des femmes qui ont survécu à l'accouchement. La régression logistique a été utilisée pour examiner l'influence de la distance à un centre de santé et le recours à un professionnel de la santé sur les risques de décès.
La mortalité maternelle était de 320 pour 100.000 naissances (IC 95%: 290–353 en Indonésie) et de 318 pour 100.000 (IC 95%: 272, 369) au Bangladesh. Les femmes qui vivaient plus éloignées des centres de santé dans les deux pays étaient moins susceptibles d'avoir un accouchement assisté par des professionnels de la santé que celles qui vivaient plus proche. Pour les femmes qui ont été assistées par un professionnel de la santé le risque de décès augmentait avec la distance à un centre de santé (odds ratio par km; Indonésie: 1.07 (IC 95%: 1.02 à 1.11), Bangladesh: 1.47 (IC 95%: 1.22 à 1.78)). Il n'y avait aucune preuve d'une association entre la distance à un centre de santé et les décès maternels pour les femmes qui n'ont pas été assistées par un professionnel de la santé.
Même dans les endroits où les services de santé sont relativement proches du domicile des femmes, la distance à un centre de santé affecte la mortalité maternelle pour les femmes qui accouchent avec un professionnel de la santé. Il se peut que les femmes ne recourent à des soins professionnels que dans une situation d'urgence et sont incapables d'obtenir des soins en temps opportun lorsqu'elles vivent éloignées d'un centre de santé.
Examinar la relación entre la distancia al centro sanitario, el consultar a un profesional sanitario y la mortalidad materna.
Estudio retrospectivo de cohortes en Matlab, Bangladesh (1987–2005) para recolectar datos sobre todos los embarazos, partos y muertes. En Java, Indonesia (2004–2005), se identificaron las muertes maternas ocurridas a través de informantes y durante un estudio poblacional, se realizó el muestreo de las mujeres que sobrevivieron al parto. Se utilizó una regresión logística para evaluar la influencia que tenían, sobre la probabilidad de muerte, la distancia al centro sanitario y la atención recibida de un profesional sanitario.
La mortalidad materna en Indonesia era de 320 por 100 000 nacimientos (IC 95%: 290, 353) y 318 por 100 000 (IC 95%: 272, 369) en Bangladesh. Las mujeres que vivían más lejos de los centros sanitarios en ambos países tenían una menor probabilidad de que sus partos fuesen atendidos por profesionales sanitarios que aquellas que vivían más cerca. Para las mujeres que fueron atendidas por un profesional sanitario, la probabilidad de morir aumentaba con la distancia al centro sanitario (razón de posibilidades por km; Indonesia: 1.07 (IC 95%: 1.02–1.11), Bangladesh: 1.47 (IC 95%: 1.22–1.78). No se encontró evidencia de una asociación entre la distancia al centro sanitario y la muerte materna para las mujeres que no fueron asistidas por un profesional sanitario.
Aún en emplazamientos en los que los servicios sanitarios están relativamente cerca a los hogares de las mujeres, la distancia al centro sanitario afecta la mortalidad materna entre aquellas mujeres que dan a luz asistidas por un profesional. Las mujeres podrían buscar ayuda profesional solo en caso de emergencia y podrían no conseguir llegar a tiempo en caso de vivir muy lejos del centro sanitario.
The fifth Millennium Development Goal (MDG-5) calls for reducing the maternal mortality ratio (MMR) by three quarters between 1990 and 2015 (United Nations Development Programme 2008). One of the key strategies to reduce maternal mortality has been to ensure that women give birth with a skilled attendant (Campbell & Graham 2006). The skilled attendant should be a health professional with midwifery skills and have a supporting environment, including access to higher levels of obstetric care in case of complications (Graham et al. 2001). It remains a challenge to provide such environments, especially in remote areas where access to health care is limited.
There are many obstacles to skilled attendant uptake, but physical distance to a health facility is thought to be a key factor (Gabrysch & Campbell 2009). Distance can capture aspects of remoteness such as poverty, low levels of education, strong traditional values, limited access to information and poor road infrastructure (Achadi et al. 2007; Hounton et al. 2008a). Long distances can be an actual barrier to reaching a health facility, but they can also be a deterrent to trying to seek care (Thaddeus & Maine 1994).
The evidence in support of an association between distance to health facilities and uptake of professional care at birth is strong. Studies in Burkina Faso, Mali, Malawi, Bangladesh, Zambia and Cambodia have found a linear relationship between increasing distance to a health centre and decreasing proportions of institutional births (Yanagisawa et al. 2006; Hounton et al. 2008b; Gabrysch et al. 2011) or births with a health professional (van den Broek et al. 2003; Chowdhury et al. 2006; Yanagisawa et al. 2006). Similarly, studies in Indonesia (Achadi et al. 2007), Haiti (Gage & Guirlene Calixte 2006), Cambodia (Yanagisawa et al. 2006) and Burkina Faso (Hounton et al. 2008a) showed that increasing distance to the nearest hospital reduced the level of uptake of a health professional at birth. These associations persisted after adjustment for various socio-economic factors in all sites, except in Cambodia where evidence for an effect was only observed in unadjusted analysis.
The relationship between distance to a health facility and the risk of a woman dying is less clear. Increasing remoteness (Le Bacq & Rietsema 1997; Bartlett et al. 2005), living in rural compared with urban areas (Ronsmans et al. 2009) and time taken to reach a health facility (Urassa et al. 1995; Ganatra et al. 1998; Taguchi et al. 2003; Brentlinger et al. 2005) all are associated with an increasing risk of maternal death. However, the relationship between actual physical distance to a health facility and maternal mortality remains uncertain. In Pakistan, the odds of dying increased 1.5 times only in women living more than 40 miles from a hospital compared with those living within 40 miles. This association was lost once adjusted for maternal age, parity, obstetric history and socio-economic factors (Fikree et al. 1997). In a case–control study in rural west Maharashtra, India, women who died lived significantly further away from an appropriate treatment facility than women who survived childbirth, but no adjustments were made (Ganatra et al. 1998). In Burkina Faso, distance from home to a health centre or nearest hospital was not associated with maternal death (Hounton et al. 2008b).
Understanding the barriers to uptake of obstetric care remains crucial in achieving the MDG-5 goal. The aim of this study is to explore the influence of distance and uptake of professional assistance during birth on maternal mortality. We analysed original studies from Indonesia and Bangladesh, which measured physical distance to health facilities, professional assistance during birth and maternal mortality.
Data collection methods and study populations have been described in detail elsewhere (Chowdhury et al. 2006, 2007; Achadi et al. 2007; Makowiecka et al. 2008; Ronsmans et al. 2009, 2010). The study design and definition of variables for the two study sites are shown in Table 1.
|Population||Study period||Type of study||Definition of a maternal death||Definition of births with a health professional||Definition of distance variable|
|West Java, Indonesia||2004–2005||Case–control study with cases being all maternal deaths and controls being all women identified in the population-based survey||Death during pregnancy or within 42 days post-pregnancy, regardless of cause||Delivery with a midwife or doctor||Distance by road from the centre of the woman's resident village to the nearest hospital or health centrea|
|Matlab, Bangladesh||1987–2005||Population-based cohort of all pregnancies and deaths in a demographic surveillance site||Death during pregnancy or within 90 days of pregnancy termination, regardless of cause||Delivery with a midwife or doctor||Straight line distance between the woman's house and the nearest health centre or hospital using global positioning systema|
The study was performed in Serang and Pandeglang districts in Banten Province, Java, as part of a comprehensive evaluation of the effectiveness of safe motherhood strategies, called IMMPACT (Graham 2002; Achadi et al. 2007; Bell et al. 2008; Makowiecka et al. 2008; Ronsmans et al. 2009). The district of Serang is moderately urbanised, with three hospitals and 36 health centres to serve 1.8 million people. Pandeglang is more remote, with only one hospital and 30 health centres for its 1.1 million population. In 1989, the Indonesian Government launched a safe motherhood programme, which aimed to assign a midwife to every village (The World Bank 1994). During the study period, there were 2.2 midwives per 10 000 population in the two districts, with midwife density per square-kilometre being 30 times higher in urban villages than in remote rural villages (Makowiecka et al. 2008). All maternal deaths between January 2004 and December 2005 were identified through informants (Immpact 2008). Families were visited to collect information about the circumstances and cause of death. A population-based survey was carried out in April–June 2006 on live women who had one or more births between 1 April 2004 and 31 March 2006. The survey was a stratified village cluster random sample of women. 150 villages stratified by urban, rural and rural remote were sampled with probability proportional to size and with replacement. Eight births within each cluster were randomly selected (Ronsmans et al. 2009). Data on maternal age, parity, education, place of delivery, who attended the delivery and household assets were collected for women who died and those who survived. For women who died before delivery, data on uptake of a health professional during pregnancy or at the time of death were collected. Distance by road to the nearest hospital and health centre from the centre of the village where women lived was measured using the network analysis tool (Arcview GIS 3.3, ESRI). Birth with a health professional was defined as a midwife or doctor attending the delivery, regardless of place of birth. A maternal death was defined as any death during pregnancy or within 42 days post-pregnancy, regardless of cause. The proportion of births with a health professional by distance to a health centre or hospital was calculated for the live women from the population-based survey.
A population-based cohort study on all pregnancies and deaths was conducted between 1987 and 2005 in the ICDDR-B service area of Matlab, which has a population of approximately 110 000 (Chowdhury et al. 2006, 2007; Ronsmans et al. 2010). Matlab is a rural area southeast of the capital of Dhaka. In 1987, a safe motherhood programme was introduced in half of the ICDDR-B area. Two midwives were posted at the two health centres with the aim of increasing skilled attendance at delivery at home. A basic obstetric facility was established in Matlab town as a referral site when a midwife considered a home birth to be unsafe. This facility could also be accessed directly by women. In 1990, four more midwives were posted in the two remaining health centres of the ICDDR-B area. Between 1996 and 2001, all four health centres were upgraded to provide basic obstetric care and midwives were asked to stop attending home births. The main referral hospitals performing caesarean sections are located in the town of Chandpur, outside of the ICDDR-B surveillance area. These hospitals can be reached by road or boat.
Data were obtained from the Matlab Health and Demographic Surveillance System (HDSS), periodic censuses and the geographical information systems. The HDSS records all births, deaths, migrations and marriages by carrying out monthly visits to all households. For all pregnancies, data were collected on maternal age, place and year of birth and type of attendant at delivery. Specially designed verbal autopsies were carried for all women who died when aged 10–49 years (Chowdhury et al. 2007). Data on uptake of a health professional during pregnancy, at birth or around the time of death were collected. Household assets were obtained from the 1996 and 2005 censuses. A straight line distance between the women and the nearest health centre or hospital was estimated using global positioning system. A health professional was defined as a doctor, nurse or midwife, regardless of the place of birth. Maternal death was defined as the death of a woman while pregnant or within 90 days of pregnancy termination, regardless of cause. Women who died before delivery were categorised as with a health professional if a midwife or doctor was present near the time of death. The proportion of births with a health professional by distance to a health centre or hospital was calculated for live women only drawn from the women surviving birth and the post-partum from the surveillance data in Bangladesh.
We examined the association between the presence of a health professional during delivery and distance to a health centre or hospital among live women only, by comparing the proportion of women assisted by a health professional during delivery in different distance strata and regression analysis to test for linear trends.
We examined the association between maternal mortality and distance to a health centre or hospital using two approaches. First, we compared the MMR in different distance strata. MMRs were estimated using all maternal deaths and births from the population-based retrospective cohort study in Bangladesh. Mortality was expressed per 100 000 births, and confidence intervals were calculated assuming a Poisson distribution. For Indonesia, maternal deaths were linked to the village where the women lived. Generalised linear models using a Poisson distribution with scale parameter based on the deviance were used to estimate MMRs and their 95% confidence intervals (CI) (McCullagh & Nelder 1984). These models use the total count of maternal deaths per village, which is offset by the expected number of births per village. The expected number of births in each village was calculated by multiplying the population in each village with the crude birth rate estimated from the survey (Ronsmans et al. 2009).
Second, a case–control approach was used to measure the influence of distance to a health facility on the odds of dying. For Indonesia, cases were all maternal deaths identified between January 2004 and December 2005 and controls were all live women with a birth identified from the population survey. For Bangladesh, cases were all pregnancy-related deaths identified between July 1987 and December 2005 and controls were all live women who reported a birth identified from the population-based retrospective cohort study. We used logistic regression to calculate crude and adjusted odd ratios (ORs) with their 95% confidence intervals (CI). Multivariable logistic regression was used to assess potential confounding effects and interactions between distance to a health facility and other variables such as uptake of a health professional at birth, maternal age, education, parity and wealth. Wealth quintiles were computed from the household asset scores using principal component methods (Filmer & Pritchett 2001). For Indonesia, models were adjusted for village clustering and stratification to account for sampling variation (Ronsmans et al. 2009).
The population survey in Indonesia identified 1234 women who survived birth between April 2004 and March 2006. 32.8% births were attended by a health professional with the majority either in the woman's (48%) or the midwife's home (34%). Of the births with a health professional, 53% were with a health-centre midwife, 21% used a village-based midwife, 14% a hospital-based midwife, 4% a private midwife, 1% did not specify the type of midwife, and 7% used a doctor. In Bangladesh, 53 924 women delivered between 1987 and 2005. 23.6% delivered with a health professional, of which 27.1% were in the woman's home and 72.9% in a health facility. All the midwives in this area are ICDDR-B midwives.
Uptake of a health professional during delivery and birth decreased with increasing distance to a health centre (test for linear trend: Indonesia, P = 0.005; Bangladesh, P < 0.001) and increasing distance to a hospital (test for linear trend: Indonesia, P = 0.05; Bangladesh, P < 0.001) (Table 2).
|Births with a health professional, % (n)||Maternal deaths per 100 000 births (95% CI)||Births with a health professional, % (n)||Maternal deaths per 100 000 births (95% CI)|
|Overall||32.8 (443)||320 (290, 353)||23.6 (12 713)||318 (272, 369)|
|Distance to a health centre (km)|
|≤1||57.6 (139)||245 (190, 316)||38.4 (4186)||220 (141, 327)|
|>1–≤2.5||38.9 (55)||324 (241, 436)||21.8 (4324 )||317 (244, 406)|
|>2.5–≤5||36.3 (168)||302 (250, 363)||18.5 (4100)||356 (282, 444)|
|>5–≤7.5||16.4 (31)||323 (258, 406)||10.4 (103)||401 (109, 1026)|
|>7.5–≤10||16.3 (25)||450 (340, 594)||–||–|
|>10||15.4 (25)||389 (297, 508)||–||–|
|Distance to a hospital (km)|
|≤10||53.9 (212)||267 (218, 328)||30.4 (94)||0|
|>10–≤20||28.5 (74)||333 (257, 403)||23.4 (6605)||331 (268, 406)|
|>20 ≤ 30||23.4 (57)||317 (258, 390)||23.8 (6014)||300 (236, 376)|
|>30 ≤ 40||28.7 (36)||282 (201, 396)||–||–|
|>40–≤50||28.0 (28)||379 (254, 564)||–||–|
|>50||14.1 (36)||474 (356, 631)||–||–|
Levels of maternal mortality were similar in Indonesia (320 per 100 000 births (95% CI: 290, 353) and Bangladesh (318 per 100 000 (95% CI: 272, 369) (Table 2). In Indonesia and Bangladesh, there was evidence for an increase in the odds of dying with a linear increase in distance to a health centre in the crude analysis (Indonesia: OR 1.04, 95% CI: 1.00–1.07; Bangladesh: OR 1.15, 95% CI: 0.99–1.33), but this was no longer observed once adjusted for maternal age, education, wealth and parity (Indonesia: OR 1.02, 95% CI: 0.98–1.06, Bangladesh: OR 1.13, 95% CI: 0.97–1.31) (Table 3a,b).
|Distance to a health centre (km)||Overall||Health professional present at time of birth or at time of death||Health professional NOT present at time of birth or at time of death|
|Unadjusted odds ratio (95% CI)||Adjusted odds ratioa (95% CI)||Unadjusted odds ratio (95% CI)||Adjusted odds ratioa (95% CI)||Unadjusted odds ratio (95% CI)||Adjusted odds ratioa (95% CI)|
|>1–≤2.5||1.24 (0.57, 2.73)||1.23 (0.54, 2.81)||1.63 (0.65, 4.09)||2.33 (0.79, 6.87)||1.06 (0.39, 2.89)||0.98 (0.34, 2.79)|
|>2.5–≤5||0.82 (0.46, 1.46)||0.73 (0.39, 1.34)||0.86 (0.40, 1.83)||0.87 (0.37, 2.03)||0.81 (0.40, 1.65)||0.78 (0.37, 1.610|
|>5–≤7.5||1.58 (0.78, 3.20)||1.07 (0.49, 2.32)||4.73 (1.82, 12.31)||2.17 (0.73, 6.49)||0.97 (0.42, 2.21)||0.67 (0.28, 1.60)|
|>7.5–≤10||1.40 (0.67, 2.91)||1.1 (0.50, 2.46)||3.69 (1.36, 10.01)||3.05 (1.03, 9.06)||0.89 (0.37, 2.13)||0.76 (0.31, 1.87)|
|>10||1.62 (0.78, 3.41)||1.1 (0.50, 2.52)||3.18 (1.11, 9.11)||2.11 (0.60, 7.38)||1.43 (0.59, 3.46)||0.93 (0.36, 2.41)|
|Continuous||1.04 (1.00, 1.07)||1.02 (0.98, 1.06)||1.08 (1.05, 1.12)||1.07 (1.02, 1.11)||1.03 (0 .98, 1.08)||1.01 (0.96, 1.07)|
|>1–≤2.5||1.68 (0.99, 2.85)||1.50 (0.88, 2.56)||3.16 (1.50, 6.66)||2.72 (1.29, 5.75)||1.21 (0.57, 2.58)||1.15 (0.54, 2.45)|
|>2.5–≤5||1.84 (1.10, 3.09)||1.69 (1.00, 2.84)||4.27 (2.07, 8.81)||3.90 (1.88, 8.08)||1.20 (0.57, 2.53)||1.17 (0.55, 2.46)|
|>5–≤7.5||2.30 (0.78, 6.81)||1.92 (0.64, 5.74)||12.57 (3.35, 47.14)||9.17 (2.39, 35.14)||0.79 (0.10, 6.25)||0.71 (0.09, 5.65)|
|Continuous||1.15 (0.99, 1.33)||1.13 (0.97, 1.31)||1.05 (1.25, 1.82)||1.47 (1.22, 1.78)||1.00 (0.79, 1.27)||1.00 (0.78, 1.27)|
There was evidence for an interaction between distance to a health centre and assistance by a health professional on maternal mortality in crude analyses (Indonesia, P = 0.038; Bangladesh, P = 0.008), and therefore, the analysis was separated into whether or not a woman was attended by a health professional. For women who were assisted by a health professional, the odds of dying increased with increasing distance from a health centre (adjusted ORs per 1 km increase: Indonesia 1.07, 95% CI: 1.02–1.11; Bangladesh 1.47, 95% CI: 1.22–1.78) (Table 3). In Indonesia, this was more strongly observed among women who lived further than 5 km from a health centre than those who lived within 5 km of a health centre (adjusted OR: 2.2, 95% CI: 1.09–4.5, P = 0.03, Figure 1a). In Bangladesh, the increase in mortality with distance was more linear (Figure 1b). For women who were not assisted by a health professional, there was no evidence for increased odds of dying with increasing distance to a health centre in both Indonesia and Bangladesh (P > 0.05, Table 3). No other interactions were observed between distance to a health centre and either maternal age, wealth, parity or education on maternal mortality.
The above analysis was repeated to explore the relationship between distance to the nearest hospital and uptake of a health professional and maternal mortality. There was no evidence for an association between distance to a hospital and maternal mortality (Indonesia: adjusted OR per 1 km increase 1.00, 95% CI: 1.00–1.01; Bangladesh: adjusted OR per 1 km increase 1.00, 95% CI: 0.96–1.04). There was no evidence for an interaction between distance to a hospital and uptake of professional assistance (data not shown). Excluding deaths before labour and delivery attenuated values of the ORs, but the conclusions remain unchanged (data not shown).
We found that among women assisted by a health professional, there appears to be an increase in maternal mortality with increasing distance to a health centre. In women who did not use a health professional, there is no variation in maternal mortality with distance to a health centre. When uptake of professional birth attendance is generally very low, women will only seek professional care in an emergency and they may do so too late for a midwife or doctor to be able to save their lives. In such circumstances, women who use a health professional have higher levels of maternal mortality than those who do not (Midhet et al. 1998; Brentlinger et al. 2005; Chowdhury et al. 2007; Christian et al. 2008; Hounton et al. 2008b; Ronsmans et al. 2009). The association between maternal mortality and distance to a health centre among women giving birth with a health professional is therefore perhaps not surprising. Reviews by Thaddeus and Maine (1994) and Gabrysch and Campbell (2009) emphasise how distance increases not only the delay in reaching a healthcare facility, but also influences a woman and her family's decision to seek care. Distance can act as a barrier to seek care early and only when the illness becomes serious is the decision made to travel to a health facility. About half of all births with a health professional in Indonesia and a third in Bangladesh were in the woman's home, suggesting that health professionals may face similar barriers to families when trying to overcome long distances. Very little is known about the barriers faced by health professionals in attending home births. Evidence from Bangladesh suggests that midwives found it difficult to travel far away from the health centre, even though transport by boat was provided (Blum et al. 2006). For those women who do not seek care, distance to a health facility has no bearing on their pregnancy-related health outcome. They do not depend on timely access to a health facility to seek care or how long a midwife takes to come to their home.
The relationship between distance to a health facility and birth with a health professional is strong and consistent with several other studies (Hodgkin 1996; Magadi et al. 2000; Tanser 2006). For both countries, previous published work has shown a linear relationship between distance to the nearest hospital (Achadi et al. 2007) or health facility (Chowdhury et al. 2006) and birth with a health professional, regardless of place of delivery. In the Indonesia study, the majority of births occurred at home and only 1.6% of all live births were delivered in a health facility. However, health-centre midwives were the main providers during delivery, and this perhaps also highlights the barriers faced by health professionals in attending home births. In Bangladesh, the safe motherhood programme was introduced in 1987 with the aim of increasing skilled attendance at delivery primarily in the women's home. In 1996, the programme was redesigned to become facility based, and midwives were asked to stop attending home births. The proportion of all births delivered with a midwife in a health facility rose from zero in 1987 to 26.8% in 2001 (Chowdhury et al. 2006). By 2005, 52.5% of all births were delivered in a health facility, and no births at home were attended by a health professional. It is therefore perhaps not unexpected that the associations between distance to a health centre and uptake of professional care during delivery are stronger in Bangladesh than in Indonesia.
The relationship between distance to a health centre and maternal mortality is much less strong than the relationship between distance to a health centre and uptake of professional assistance during delivery. High mortality in only those seeking care is probably due to late seeking care, but low uptake of professional assistance during delivery may be more related to the low quality of care once a woman reaches a health provider. There is increasing evidence that health professionals in poor countries may not have adequate skills or that the health facilities or home environment in which they work do not provide the necessary support (Blum et al. 2006). In 1989, the Government of Indonesia launched a safe motherhood programme, which aimed at assigning a midwife to every village. Within 7 years, midwives were posted in the majority of villages and the proportion of births attended by a health professional rose from 35% in 1989 to 69% in 2000 (Hatt et al. 2006). However, these programmes have focused on the individual providers, neglecting the essential supporting environment. In Bangladesh, basic conditions in rural areas may also prevent midwives from delivering skilled care (Blum et al. 2006). It is thus critical that one of the major roles that rural midwives play in reducing maternal deaths is ensuring timely and appropriate access to adequate obstetric care (Campbell & Graham 2006). However, without the supporting environment, including good transportation, the midwife's capabilities in many emergency situations remains limited. The Indonesian Government has recognised the potential of health centres for increasing deliveries with health professionals. In 2011, Jampersal was introduced, a health programme that provides women with universal free delivery care (Ministry of Health I 2011). However, this programme is only available to women who deliver in a health facility, and thus, barriers to access to a health centre remain the same.
We did not observe any associations between distance to a hospital and maternal mortality. In Bangladesh, the study area is more than 10 km from a hospital, and almost all births occurred between 10 and 30 km of the nearest hospital. Thus, there may not have been enough variation in distance to hospital to explore this association. In Indonesia, only 4% of women delivered in a hospital, such low uptake also limits the ability to explore associations between maternal mortality and distance to a hospital in this region.
There are some strengths to these studies; not only are these prospective studies, but special efforts were made to identify all maternal deaths in the study populations. It is possible though that we still missed some maternal deaths in early pregnancy or due to sensitive causes owing to under-reporting by relatives in both studies, and this may lead to differential misclassification if death reporting is related to distance to a health facility. The two studies have some limitations; we did not examine the relationship between distance to a health facility and specific causes of death. In Bangladesh, haemorrhage was the major cause of direct maternal mortality (Chowdhury et al. 2009) and timely access to emergency care for these women is crucial. However, just over a quarter of maternal deaths were due to indirect causes where the role of the midwife is limited. We do not have data on causes of death for the Indonesian study. In Bangladesh, distance was measured by a straight line between the woman's house and health facility rather than by road distance. In the Indonesian study, distance by road was used, but the data were collected at village level, and we may have underestimated some distances for women who live outside the village centre. This would dilute the effect of distance on risk of dying especially if these women are the most vulnerable and at greatest risk of pregnancy-related complications. The study was not designed to collect information on mode of transport, which may also influence the risk of dying and uptake of professional care. Our analysis also relies on women's usual place of residence, and often women stay with relatives around the time of delivery, which may have distorted our data on distance to a health facility. For the main analysis, we included deaths during pregnancy; these women may be less likely to have seen a health professional than women who died during labour and delivery. However, excluding deaths during pregnancy did not alter our conclusions for either study.
Our findings provide further insight into understanding the determinants of maternal mortality. Distance to a health facility may capture other aspects of remoteness that are difficult to measure quantitatively (Achadi et al. 2007; Hounton et al. 2008a), but may also influence a women's decision to use a health professional during delivery. Clearly, investment in midwives and their professional environment is important. However, improving access to adequate care, in particular for women who live far from health facilities, and encouraging women to seek appropriate and timely care are essential if the MDG-5 goal is to be achieved.
We thank our colleagues in Initiative for Maternal Mortality Programme Assessment (IMMPACT), the District Health Offices in Serang and Pandeglang and those in the Ministry of Health, Jakarta. In Bangladesh, we thank Health and Demographic Surveillance System and Matlab Safe Motherhood Programme for providing their data sources for this study. This work was undertaken as part of international research programme IMMPACT, funded by the Bill & Melinda Gates Foundation, the Department for International Development, the European Commission and USAID. The funders have no responsibility for the information provided or views expressed in this study.