Maternal mortality in low-income countries: What interventions have been evaluated and how should the evidence base be developed further?

3.1. Reviews of intervention evaluations 

Thirty reviews were identified. Full papers were retrieved for 25 reviews and 4 were key-worded on abstract alone. It was not possible to retrieve either an abstract or full paper for 1 review (on skilled attendance at delivery).

Eleven were systematic reviews (i.e., they stated their search strategy and inclusion criteria). Four of these reviewed nutrition interventions; the remaining 7 covered a range of topics: complex interventions, private sector involvement, emergency obstetric care (EmOC), maternity referral systems, prenatal care, malaria prevention during pregnancy and Safe Motherhood interventions (please contact the first author for a complete list of identified articles).

Of the 19 non-systematic reviews, 5 were general reviews of interventions to reduce maternal mortality, 5 reviewed traditional birth attendant (TBA) interventions, and 3 focused on prenatal care. Two reviewed nutrition interventions and 1 each reviewed community interventions, interventions involving transport or communication for obstetric emergencies, and skilled attendance at delivery. One used a model-based analysis to estimate the total reduction in maternal mortality in Sub-Saharan Africa, which could be achieved by combining proven effective prenatal interventions.

Less than half of the reviews assessed the quality of evaluations (n=13); a further 3 were unclear. Five of the 6 nutrition reviews were quality assessed.

3.2. Interventions evaluations 

The search identified 109 studies (reported in 124 papers) that met the inclusion criteria. Two additional papers, for which neither an abstract nor full paper could be retrieved, were excluded. Full papers were used to key-word 78 studies and 31 were key-worded using the abstract only.

Studies were identified for only 24 out of a possible 59 countries. The main focus for 25 studies was secondary prevention, 68 focused on tertiary prevention, and 16 covered both (Table 2).

Table 2. Intervention foci of studies

	Focus of intervention	Main focus (No. papers)a	Secondary focus (No. papers)a
	Secondary prevention
	Age	0	0
	Health/nutritional status	19	2
	Fertility status	4	3
	Health service use (non-emergency)	22	1
	Tertiary prevention
	Decision to seek care	35	7
	Access to care	41	7
	Quality of care received	51	4

a Interventions could have more than one focus.

Table 3 shows the topic foci of the studies and the reviews. A greater proportion of the reviews considered secondary prevention (particularly health service use and health/nutritional status) compared with the studies, which focused more on tertiary prevention.

Table 3. Main focus of studies and reviews

	Main focus	% Studies (No.)	% Reviews(No.)
	Secondary prevention
	Age	0 (0)	0 (0)
	Health/nutritional status	17 (19)	37 (11)
	Fertility status	4 (4)	10 (3)
	Health service use preventing immediate causes	21 (22)	40 (12)
	Tertiary prevention
	Decision to seek care	32 (35)	7 (2)
	Access to care	38 (41)	13 (4)
	Quality of care received	47 (51)	30 (9)

3.3. Study design 

Table 4 shows that the majority of studies used a before-and-after design. Relatively few used the more rigorous randomized controlled trial (RCT) design, or even (non-random) trial design.

Table 4. Study designs used

	Study design	No. of studies (n=109)
	Before-and-after evaluation	60
	Randomised controlled trial	17
	Trial	15
	Post-intervention evaluation only	11
	Unclear	6

3.4. Outcomes 

Maternal death was measured in 48 studies. Twenty-two measured the number of maternal deaths or calculated maternal mortality ratios, although for almost all, the samples were either too small to detect significant differences or were not stated. Only 3 reported power calculations (one study stated that it only had power to detect a 90% reduction in maternal mortality).

Fourteen studies used the UN indicators of availability of emergency obstetric services [12], although only 10 measured all indicators. One study reported all indicators except the case fatality rate (CFR) and 2 measured only 2 of the indicators (percentage of births taking place in EmOC facilities and met need for EmOC; number of basic and comprehensive EmOC facilities and met need). One only reported the number of hospitals providing basic and comprehensive EmOC.

In addition to those using the UN indicators, 20 studies measured CFRs. Eighteen studies measured health or nutritional status. Skilled attendance was measured in 14 studies and 5 reported trained TBA (not considered skilled attendants) use. Twelve studies reported the number of women with complications using obstetric services.

Twenty-three studies appeared to measure process indicators for the first “2 delays” (i.e., decision to seek care and access to care), although these were difficult to classify and overlapped considerably with each other. Sixteen of these included measures of emergency transport systems or loan funds for emergency transport or treatment. Ten measured awareness of complications or their danger signs among community members, or knowledge of safe motherhood.

Seventeen studies measured prenatal or postabortion care use. Other outcome measures included various standards of care measures (n=13), family planning use or birth spacing intentions (n=10), knowledge and skills of birth attendants (professional or traditional) (n=10), and referrals to health services (n=9). Outcome measures were not known for 4 studies.

4.1. Interventions evaluated 

Intervention evaluations have been carried out in less than half of the low-income countries with a high MMR. Even countries that have hosted studies do not have evidence on all relevant topics. This highlights the importance of the generalizability of evaluation findings to other settings.

Given the current international policy focus on intrapartum care (i.e., skilled attendance at delivery and emergency obstetric care) it is not surprising that more than twice as many interventions attempted to tackle tertiary prevention than secondary [3], [4]. Within tertiary prevention, more interventions tackled the third delay (i.e., the timeliness and quality of care received) than the first or second delays (i.e., decision to seek care and access to care). This may be due to the medical and health service focus (which have long histories of evaluation) of the third delay rather than infrastructure or social factors. Interventions that address the first or second delays are more likely to be complex, targeting a wider population- or community-level audience, or addressing structural factors. For these reasons, it may be easier to conduct and evaluate interventions targeting medical or health service factors.

A wide range of topics were covered by the reviews and although this is encouraging, it is disappointing that only a few assessed the quality of the studies included. Compared with the studies, reviews were more likely to consider secondary prevention than tertiary. This may be because the focus on skilled attendance and emergency obstetric care has been more recent, with previous attention placed on prenatal care and TBA training [13], [14]. More reviews of primary research should be carried out, particularly focusing on tertiary prevention, as these are lacking in comparison with the number of evaluations carried out. Reviews should consider the quality of studies and weight the evidence accordingly.

4.2. Study designs 

Most of the studies developed adequacy evidence (using before-and-after designs) rather than the more rigorous probability or plausibility evidence (derived from RCTs or trials) [15]. Although not always possible, RCTs can be carried out on complex, non-standard interventions [16]. However, most of the RCTs in this review evaluated relatively simple nutritional supplementation interventions.

If an RCT is not appropriate, the possibility of using a stepped-wedge design should be explored, where the intervention is introduced in a phased manner to sites, which act as comparison sites until then [17]. Failing that, at least one comparison group should be included in the evaluation and if it is not possible to include a comparison group, at the very least outcomes should be measured both before and repeatedly after the intervention [18].

Complex public health interventions ideally require a combination of study designs and data collection methods to develop a more comprehensive assessment of an intervention. For example, process data could highlight who did and did not access the intervention as well as how acceptable and feasible it was. Qualitative data can explore how sociocultural factors interact with intervention use and effects. Dose-response analysis could show if the level of intervention exposure is related to outcomes. Such triangulation of methods is especially important for creating evidence on the generalizability of interventions—an essential yet neglected component of the study designs in this review.

4.3. Outcome measures and indicators 

It is understandable that nearly half of the studies measured maternal death in some way, since this was their primary or secondary focus. However, the lack of information about sample size or the power of the studies highlights the importance of appropriate outcome measures. Even in high mortality areas, deaths are sufficiently rare that studies must have a very large sample size to accurately detect changes in the rate or ratio. In many cases, proxy or process indicators may be more feasible measures to use.

Measuring morbidity, particularly “near miss” (a near miss is defined as “a severe, life-threatening complication, necessitating an urgent medical intervention in order to prevent the likely death of the mother”) has been put forward as an alternative to measuring mortality directly [19]. One advantage is that near misses are more frequent than deaths, allowing greater statistical precision and requiring smaller sample sizes. However, given that current international attentions are focused on treating complications as opposed to preventing them, the number of near misses may rise (or remain the same) while mortality falls [19]. There is also some uncertainty around what is classified as a near miss; the lack of standardization currently inhibits comparisons between settings [20]. Although some studies in the evidence map measured the number of women with complications seen in health facilities, none appeared to measure near miss. Further work is required to ascertain the usefulness of this proxy measure and to determine what should be included in the definition of near miss [21].

In recent years recognition of the impracticalities of measuring maternal mortality (and the remaining uncertainties around proxy measures) has led to an increased focus on process measures. These are particularly useful for complex, programmatic interventions, with long causal pathways. They demonstrate how well different elements of a program are working, or what activities are required, rather than merely providing “good” or “bad” outcome information.

Substantial progress has been made in recent years to develop reliable, feasible, and useful indicators to measure the progress of interventions aiming to improve access and availability of tertiary prevention services. It is encouraging that several studies used the UN indicators to evaluate intervention progress, although some only used one or two rather than the whole set. A major benefit of having a set of commonly-used, clearly-defined indicators is that findings from different studies are easier to compare than if each study calculates its own measures.

Although it is important that, to prevent misinterpretation, the UN indicators are analyzed collectively rather than individually, this has not always been the case [22]. Twenty studies measured CFRs without using any of the other UN indicators, despite the fact that CFRs may change for a variety of interdependent factors, such as quality of care, the distribution of facilities, or the number and severity of complications treated.

It is positive that 23 studies attempted to measure process indicators for reductions in the first and second delay in reaching adequate treatment for obstetric complications. However, many of the studies focusing on these issues did not assess progress. The diversity of measures used makes it difficult to draw comparisons across studies. More work is needed to develop proxy and process indicators for other areas (as has been done, for example, for postabortion care [23]) and to test associations between these indicators and maternal death.