|06. Can Skilled Attendance at Delivery Reduce Maternal Mortality in Developing Countries?|
A key rationale cited for prioritising skilled attendance is the findings from correlational analysis of historical and contemporary data. There are, however, two drawbacks to this approach which should be acknowledged from the outset: firstly, the intrinsic inability to make causal connections using aggregated data, and secondly, the problems of the data - its reliability as well as the choice of independent and dependent variables. Whilst awareness of the former drawback needs to be raised among those using the findings to advocate skilled attendance, there is no way to overcome it completely, although - as will be shown later - multivariate analysis can at least offer some control over known confounders. The drawbacks to the data, on the other hand, are not insoluble, and methodological research could help refine the most commonly-used independent variable - proportion of deliveries with health professionals (Graham and Bell 2000a).
As for the previous section, it is helpful to distinguish between two types of evidence on the link between maternal mortality and skilled attendance at the population level: historical and epidemiological. There is a comparative wealth of information on the historical trends in maternal mortality in modern day industrialised countries, such as Sweden and the United States, and transitional countries, such as Malaysia and China (De Brouwere et al . 1998, Koblinsky et al . 1999, Loudon 1992b). Most of these historical series seek to identify the contributory factors in the downward trend and all conclude that no single factor can be held responsible. Having acknowledged the multi-factorial nature of the decline, the concern has been to establish the relative importance of various factors, and skilled attendance - as reflected in a variety of measures, has emerged as of central importance. Beyond this, there has been an attempt to disentangle the elements of skilled attendance, primarily in terms of place of delivery and type of attendant (doctor or midwife). A justification for this is the perceived relevance of the lessons to contemporary developing countries. Relevance is however hard to assess, since the historical declines occurred when many other demographic, economic, political, cultural and scientific developments were happening in the countries concerned. Some of these factors can be quantified and allowed for in the interpretation, if not the analysis itself, and so, for example, there are authors who argue that maternal mortality was not reduced by broader socio-economic development (Loudon 1992b). The relevance of the historical lessons is also affected by the complexity of the situation today in developing countries, and particularly the limited availability (rather than technological state) of health resources, the new disease challenge of HIV/AIDS, and the declines in fertility - all of which affect maternal mortality.
The historical series all tend to use the maternal mortality ratio as the dependent variable, and to obtain this from vital registration systems. The reliability of these figures cannot be assumed, although they are likely to compare favourably in terms of accuracy with model-based estimates, which is all that is available for many developing countries. As for the independent variable, time is the one most often used, with the occurrence of particular historic events, such as the English Midwives Act of 1902, indicated on the graph. The other independent variables used are place of delivery, often comparing percentage of deliveries at home with those in different types of health institutions, or the type of attendant, usually differentiating between specialists, general physicians, professional midwives, and others (lay persons). The correlation observed, both over time and cross-sectionally, emphasises the crucial importance of quality of care, reflecting both the skills of the provider and the environment in which they practised - including the scientific knowledge and availability of drugs. Thus, for example, (Högberg and Wall 1986) shows for Sweden a correlation between falling maternal mortality and increased deliveries by professional midwives between 1861 and 1894, but for deaths excluding puerperal sepsis since the enabling environment before 1880 did not include knowledge of asepsis. Loudon 1992a provides many examples of higher historical rates of maternal mortality in Europe and the United States among institutional rather than home deliveries and among general practitioner rather than midwife deliveries, in periods before the use of asepsis and the availability of antibiotics and before abuse of anaesthesia and instrumental deliveries was addressed (see Table 4). The correlation between the pace and timing of the fall in maternal mortality with the professionalisation and promotion of midwifery care in different industrialised countries is a further indication of the importance of the enabling environment for these skilled attendants (De Brouwere et al . 1998).
Turning to the analysis of time trends for more recent periods, (Koblinsky and colleagues 1999) identify 4 organisational models for delivery care that they correlate with levels of maternal mortality. In all 4 scenarios, functioning essential obstetric care is assumed to be available. Their findings support the conclusions of other studies, namely that in populations where the majority of deliveries are at home with non-professional attendants, the level of maternal mortality appears not to be reducible to below 100 per 100,000 live births, even with BEOC and/or CEOC available. Of course in the sort of population in which such a model prevails, there may be other factors intervening to keep maternal mortality high, such as poor maternal health status and barrier to access to care, but this type of correlational analysis cannot untangle such influences. In a population in which all women deliver in CEOC facilities with a health professional, Koblinksy and colleagues also found that maternal mortality may remain above 100 per 100,000 live births. Although they say that iatrogenic factors are assumed not to operate in any of the 4 models considered, this may in fact be part of the explanation for the level of mortality where all deliveries are in CEOC institutions. Further research is needed to try to gauge the iatrogenic fraction. The two models which correlated with maternal mortality of less than 50 per 100,000 live births both involved professionals (mostly midwives) attending deliveries, in one case at home (model 2) and the other in BEOC facilities (model 3). Interestingly, (Koblinsky et al 1999) note that all countries in which a model 2 prevailed in the past have now made the transition to model 3, for example Malaysia and Sri Lanka. However they also note that there is insufficient data available to determine which configuration of professional attendance is most cost-effective, and what the constraints are with respect to the enabling environment.
The contemporary epidemiological evidence for the link between maternal mortality and skilled attendance has perhaps been over-interpreted and the constraints of this form of correlational analysis been under-estimated (Graham and Bell 2000a). Figure 4 shows a graphical representation of this link - showing the regression of national estimates of the maternal mortality ratio for developing countries against national coverage statistics, mostly from DHS surveys, on the percentage of deliveries reported by women to have been attended by doctors, nurses or midwives. Depending on the countries included, the coefficients for this regression - and thus the strength of the relationship - vary, but its negative nature holds up - countries with high proportions of deliveries with these professionals tend to have low levels of maternal mortality. Nevertheless, it is important to note that the relationship is considerably weaker if the values for industrialised countries, which all cluster around very low maternal mortality and very high professional attendance, are omitted.
Although this type of exploratory analysis can be helpful in suggesting other relationships to examine, its simplicity also encourages over-interpretation - to infer that increasing the proportion of deliveries with health professionals will itself reduce maternal mortality. There are two inter-related issues here to consider - one to do with the indicators used, and the other the nature of the association between maternal mortality and deliveries with professionals. Whilst it is unlikely that inaccuracies in the dependent and independent variable can totally explain the observed pattern, it is important to acknowledge the crudity of the measures used. Maternal mortality ratios for many of the poorest developing countries are derived from modelling methods, which use the percentage of deliveries with health professionals (PDHP) to predict the level. By removing those countries with modelled estimates from the regression, a relationship remains but the explanatory power is not surprisingly reduced; for mortality data for 1990, the adjusted coefficient of variation falls from 65% to 51%. The PDHP has the benefit of being widely available from national surveys, but the extent to which women can and do report reliably on who attended their births has not been established.
Figure 4. Proportion of deliveries with health professionals1 and the maternal mortality ratio2 for 50 developing countries, ~1990
1 Defined in coverage statistics as doctors, nurses and midwives (WHO 1997)
2 Maternal deaths per 100,000 live births (WHO 1996)
Major uncertainty surrounds the effects of only recording the most qualified person, the definition of attended (e.g. the person who caught the baby or the person attending most of the time), and the confusion over who is a professional in some facility settings. The focus, for example, on recording attendance at delivery only for a womans live births, as in the Demographic and Health Surveys, means that stillbirths are omitted. Moreover, the crude PDHP can disguise many other important differentials between and within countries besides socio-economic status - between regions, urban and rural localities, institutional vs. home settings, private and public facilities, doctors and midwives/nurses, as well as maternal characteristics such as age, parity, education, and birth outcomes.
At the recent five-year review of the programme of action for the International Conference on Population and Development (United Nations 1999), international development targets were set for this indicator, with those for countries with high levels of maternal mortality being 40% of deliveries with skilled attendants by 2005 and 60% by 2015. Figure 5 shows the levels of this indicator, as reflected in the proportion of deliveries with health professionals, for selected world regions in 1996, and highlights the targets. The selection of this so-called benchmark indicator was made owing partly to the acknowledged difficulty of measuring maternal mortality, partly the ready availability of data on PDHP, and partly the inferred causal link. Figure 5 itself challenges the link, since several world regions have already achieved or are close to achieving the targets and yet their estimated levels of maternal mortality remain high.
Figure 5. Deliveries with health professionals in 1996, for selected world regions
The crude relationship can be examined further with bi-variate and multivariate analysis, revealing some intriguing findings with important policy and programme implications. We have disaggregated the data for 50 developing countries according to whether the most qualified person present at delivery was a doctor or midwife, as reported by women. Although some countries with modelled estimates of maternal mortality are included, as mentioned earlier, their removal has little impact on this disaggregated analysis. As the proportion of deliveries attended by doctors increases, the level of maternal mortality appears to exponentially decrease (Figure 6). Looking at countries with less than 15% of deliveries with doctors, there is an enormous range from less than 200 maternal deaths per 100,000 live births to about 1500. Alternatively, looking at countries with maternal mortality ratios of less than 200, the proportion of doctors attending deliveries also varies hugely - from 15-90%. For deliveries with midwives, any pattern is less obvious (Figure 7).
Figure 6. Proportion of deliveries with doctors and the maternal mortality ratio for 50 developing countries, ~1990
Figure 7. Proportion of deliveries with midwives and the maternal mortality ratio for 50 developing countries, ~1990
From a programme perspective, one possible conclusion from this might be that countries should seek to increase access to doctors for deliveries rather than midwives. Another conclusion might be to dismiss the findings as an artefact - reflecting the intrinsic problems of correlational analysis, of confounding, and of the reliability of the data. But our multivariate analysis shows that an association remains between doctors and maternal mortality after controlling for various factors, including GNP, female literacy, antenatal care, and fertility. There is however a further interpretation of Figure 7. The lack of an obvious link between maternal mortality and the proportion of deliveries with midwives hints at the wide variability in the skills of those with this professional label as well as the constraints under which they practice, including inability to refer complicated cases. Merely having a bigger pool of delivery attendants will not work unless they are appropriately skilled, can refer to other professionals as the need arises and have access to an enabling environment. The crucial issue is that a health professional is not necessarily a skilled attendant, and a skilled attendant is not the same as skilled attendance which encompasses both the providers and the environment appropriate to normal as well as complicated cases.
Our analysis also suggests, however, that the mix ratio between medical and midwifery professionals is a powerful correlate of maternal mortality, thus emphasising the importance of partnerships between providers. Figure 8 plots the proportion of deliveries with midwives against those with doctors, and indicates the level of maternal mortality for each country represented. We have called this new indicator the Partnership Ratio (Graham & Bell 2000b). It is expressed as two mutually exclusive figures: the proportion of deliveries with a doctor and those with a midwife, such as PR (10, 55), with the sum indicating the total proportion of deliveries with professional attendance. The PR can also be analysed for key differentials such as region, private/public sector, parity, or maternal education, and so reveal important inequities in access and uptake of delivery care. We acknowledge the need both to avoid overly simple interpretation of this measure and for further developmental work, including case studies of countries with different PRs. The Partnership Ratio does however have the strong advantages of being easily derived from existing data, so giving countries a tool to use now, and of providing insights on effective mix ratios of medical and midwifery staff.
If the goal of 100% of deliveries with health professionals is accepted, along with the need for doctors to attend at least 15% of deliveries (this being the usual estimate of the proportion of cases with life-threatening complications), an optimum point can be plotted - at PR (15, 85). The two countries which come closest to this point on Figure 8 are Sri Lanka and Jamaica - countries frequently cited as success stories in the reduction of maternal mortality. All countries shown in Figure 8 with very high levels of maternal mortality (>1000 deaths per 100,000 live births) have Partnership Ratios lower (for both proportions) than PR (15, 50), and some of these - such as Nepal, have ratios as low as PR (6, 2). It can also be seen that many developing countries with relatively low maternal mortality have more than a third of deliveries with doctors, and in the case of three Central American countries more than two-thirds. This is not however a realistic or affordable option for the majority of poor countries with high maternal mortality where the Partnership Ratio between midwives and doctors needs to be established on a cost minimisation basis. In these countries, some of which already have fairly high proportions of deliveries with midwives, such as Senegal, progress in reducing maternal mortality is more likely to lie in improving the enabling environment, and increasing access to doctors for those women needing emergency obstetric care. The next step in this form of analysis is to look in more detail at those countries which are out-liers in terms of the Partnership Ratio, and to identify the obstacles and facilitators of their higher or lower than predicted levels of maternal mortality.
Although correlational analysis cannot provide the definitive answer to the question can skilled attendance reduce maternal mortality, it does suggest possible mechanisms when combined with our understanding of how skilled attendance may work at the individual level. The earlier diagrams on the Partnership Ratio seems to suggest a threshold effect such that ensuring all deliveries in a population take place with a health professional may not by itself be the most effective nor cost-effective route to lower maternal mortality in the immediate term. This argument does not necessarily extend however to maternal morbidity, since the estimates produced earlier (Table 2 and Table 3) perhaps suggest that the primary prevention role of skilled professionals may be very significant.
Figure 8. Proportion of deliveries with midwives and the proportion with doctors for 50 developing countries, ~1990
There are some populations in which more than a quarter of their deliveries occur without health professionals but the level of maternal mortality is below 250 per 100,000 live births, such as Peru, Tunisia, Egypt and Namibia. Conversely, there are other countries with nearly half of their deliveries with health professionals but maternal mortality remains high - above 500 per 100,000 live births, such as Malawi, Ghana, Bolivia, and Zambia. The key words here are access and quality. Thus those countries with lower than expected maternal mortality may have achieved this not by ensuring that all deliveries occur with skilled attendance but rather that those who need emergency care receive it. Conversely, those countries with higher than expected mortality may have health professionals without a functioning enabling environment and/or professionals who are not in fact skilled.
This is not to suggest that skilled attendance for all deliveries should not be a goal, but it does raise questions about the most effective and efficient intermediate steps to reaching it. Childbirth is undeniably a normal physiological process as well as the cause of some tragedies. Many women in poor developing countries will continue to deliver without skilled attendance for the foreseeable future. An impact on maternal mortality may however be possible with improved mechanisms for referral. Those settings in which skilled attendance is not negatively correlated with maternal mortality raise major questions about the quality of care, and bring us back to the question of the definitions. Skilled attendance implies competent attendants AND an enabling environment. The partnership between these attendants is crucial - and particularly between midwives and doctors, so that their different skills can be used appropriately to meet the different needs of women at the time of delivery. A lack of such partnership was historically an obstacle to progress in developed countries (Loudon 1992a) and elements of this competition indeed remain today. This lesson is highly relevant to those countries that continue to face the challenge of maternal mortality.