Global report on preterm birth and stillbirth (1 of 7): definitions, description of the burden and opportunities to improve data

Introduction This is the first of seven articles from a preterm birth and stillbirth report. Presented here is an overview of the burden, an assessment of the quality of current estimates, review of trends, and recommendations to improve data. Preterm birth Few countries have reliable national preterm birth prevalence data. Globally, an estimated 13 million babies are born before 37 completed weeks of gestation annually. Rates are generally highest in low- and middle-income countries, and increasing in some middle- and high-income countries, particularly the Americas. Preterm birth is the leading direct cause of neonatal death (27%); more than one million preterm newborns die annually. Preterm birth is also the dominant risk factor for neonatal mortality, particularly for deaths due to infections. Long-term impairment is an increasing issue. Stillbirth Stillbirths are currently not included in Millennium Development Goal tracking and remain invisible in global policies. For international comparisons, stillbirths include late fetal deaths weighing more than 1000g or occurring after 28 weeks gestation. Only about 2% of all stillbirths are counted through vital registration and global estimates are based on household surveys or modelling. Two global estimation exercises reached a similar estimate of around three million annually; 99% occur in low- and middle-income countries. One million stillbirths occur during birth. Global stillbirth cause-of-death estimates are impeded by multiple, complex classification systems. Recommendations to improve data (1) increase the capture and quality of pregnancy outcome data through household surveys, the main data source for countries with 75% of the global burden; (2) increase compliance with standard definitions of gestational age and stillbirth in routine data collection systems; (3) strengthen existing data collection mechanisms—especially vital registration and facility data—by instituting a standard death certificate for stillbirth and neonatal death linked to revised International Classification of Diseases coding; (4) validate a simple, standardized classification system for stillbirth cause-of-death; and (5) improve systems and tools to capture acute morbidity and long-term impairment outcomes following preterm birth. Conclusion Lack of adequate data hampers visibility, effective policies, and research. Immediate opportunities exist to improve data tracking and reduce the burden of preterm birth and stillbirth.

on achieving high coverage of evidence-based interventions to prevent preterm delivery and to improve survival for preterm newborns [5]. In some high-income countries (HICs), preterm birth has been high on the maternal, newborn and child health (MNCH) agenda for two decades, but is now starting to receive wider public health attention because of increasing preterm birth rates, particularly in the United States [6]. However, only recently has this issue started to reach the attention of higher-level policy makers in low-and middle-income countries (LMICs). Many countries, particularly in Latin America, have recognized the importance of preterm birth and are looking for solutions in prevention as well as improved care. Understanding and improving the current data are critical to setting priorities for action and for tracking progress.
Another adverse pregnancy outcome that is closely linked to preterm birth is stillbirth, which remains invisible on global policy agendas, as stillbirths are not included in MDG targets or tracking [7]. Each year there are an estimated 3.2 million stillbirths-almost as many as neonatal deaths. Attention to stillbirths has increased notably in the last few years. Important signs of change include the fact that estimation of disability-adjusted life years for stillbirth were calculated and included in the most recent edition of Disease Control Priorities for Developing Countries [8].
It is widely recognized that MDG 5 to improve maternal health has shown the least progress among all MDGs [9]. Maternal mortality is strongly correlated with stillbirth [10]. Increasing attention for preterm birth and stillbirth interventions, alongside increasing investment for mothers, will accelerate progress for these inextricable maternal, fetal, newborn and child health outcomes. Improved data on these pregnancy outcomes are crucial to guiding investment and tracking progress.
Th is is the fi rst of seven articles in a global report on preterm birth and stillbirth. In this article we present estimates of the current burden, assess the quality of these estimates, review trends, and make recommendations to improve data. Th e second article discusses the process of pregnancy and childbirth, etiologies of preterm birth and stillbirth and opportunities through discovery science to identify pathways, and potential interventions [11]. Other articles discuss eff ectiveness of existing interventions [12], barriers and opportunities for scaling up interventions [13], advocacy [14], and ethical considerations [15]. Th e fi nal article presents a Global Action Agenda created by about 200 MNCH stakeholders [16].

Preterm birth and stillbirth: assessing the status and quality of global estimates
Less than 5% of the world's births occur in countries with complete vital registration or networks of representative, facility-based data. One-third of the world's births occur at home. Th erefore, global-level data rely heavily on household surveys and modelled estimates. Global epidemiological estimation is a new science and builds on principles established for reviewing evidence for public health interventions-particularly with its focus on systematic literature reviews. However, approaches to standardizing the steps and assessing the quality of estimates are yet to be well-defi ned [17]. GRADE is a system designed to review the quality of evidence supporting health interventions [18]. Here, we have adapted the GRADE system to provide a summary assessment of the quality of major epidemiological parameters related to preterm birth and stillbirth, including rates, causes and impairment outcomes. Th e following two sub-sections, Preterm Birth and Stillbirth, start with a summary "Epidemiological GRADE" table (Table 1 and  Table 5, respectively). Th ey assess the input data and methods used to generate current global esti mates for these parameters, current gaps and new work in progress.

Defi ning preterm birth
Th e preterm birth rate is defi ned as the percentage of babies born before 37 completed weeks of gestation (Table 1). In addition, more granularity would be helpful for programs, such as dividing moderately preterm (33 to 36 completed weeks of gestation), very preterm (<32 weeks) and extremely preterm (<28 weeks). Particularly in countries where caesarean section is common, diff erentiating spontaneous and medically induced preterm birth is of programmatic relevance. Trend analysis in Canada suggests that a signifi cant contribution to increasing preterm birth prevalence is related to more aggressive policies for caesarean section for poor fetal growth-which The risk of ROP is increased for preterm babies exposed to hyper-oxygenation. CLD is defi ned as persistent need for oxygen therapy in order to maintain oxygen saturation above 88% to 36  http://www.biomedcentral.com/1471-2393/10/S1/S1 may reduce stillbirth but increase preterm birth [19,20]. Although there is consensus on the broad defi nition, it is clear that preterm birth is a manifestation of a complex network of causal pathways. Consensus around the phenotypes and comparable case defi nitions are an impor tant next step in better understanding this syndrome of preterm birth [11].

Preterm birth prevalence rates
A recent publication estimates about 13 million preterm babies are born each year worldwide [21]. However, country-level data are unavailable for most LMICs. Globally, around one-third of babies are born at home with little or no information on birthweight, gestational age or even survival. For those born in health care facilities, data on birthweight are often lacking or not recorded and compiled. Gestational age is rarely recorded and where recorded, tends to be based on self-reported last menstrual period (LMP), which is fairly imprecise.
Diff ering methods such as LMP, clinical assessment, and ultrasound assessment have varying levels of accuracy.
Despite the data gaps, estimates of the prevalence of low birth weight (LBW) are published each year in UNICEF's State of the World's Children report for most nations [22]. Th ese estimates rely on available data in national household surveys, especially the Demographic and Health Survey (DHS) and UNICEF's Multiple Indicator Cluster Survey, applying adjustments for maternal reporting of the child's size and for heaping of birth weights on multiples of 500 grams [23]. Birth weight is only an indirect surrogate for gestational age, and many neonates-those either small or large for gestational age-will be incorrectly misclassifi ed as preterm or term, respectively. Preterm birth rates in the published literature range from 5% in HICs to 25% in LMICs [24,25]. Populationbased data for most LMICs are scarce, especially from Africa. Th e current status of the global data is sum marized in Table 1. Th e lack of systematic country esti mates for the prevalence of preterm birth is an important gap in the visibility of preterm birth. Th e WHO Special Programme of Research, Development and Research Training in Human Reproduction has recently published estimates of preterm prevalence at global and regional levels ( Table 2) [21]. Rates are highest in least developed regions, especially Africa, but are also high in North America. A new exercise in partnership with the neonatal team at the Child Health Epidemiology Reference Group (CHERG) is a systematic review and modelling of pre term prevalence for WHO country-level estimates and that will also be used in the Global Burden of Disease (Table 1).

Preterm birth rate disparities within countries
Preterm birth rates vary greatly within countries and by sociodemographic characteristics. For example, in the United States, great disparities exist between racial and ethnic groups-in both preterm birth rates and outcomes. Th e most striking diff erences are between African American women and non-Hispanic white, Asian and Pacifi c Islander women. In 2005, the preterm birth rates among these groups varied from 18.4% among African American to 11.7% among non-Hispanic white women and 10.8% among Asian and Pacifi c Islander women [26]. Th e overall preterm birth rate has increased since 1990, due primarily to a 38% increase in non-Hispanic white preterm births and a 10% increase in Hispanic preterm births [26].
American Indians and Alaska Natives also have high preterm birth rates, reported to be 13.5% in 2005 [27]. Among US Indigenous populations, Native Hawaiians experi ence the highest infant and neonatal mortality rates [27,28]. Th e Pregnancy Risk Assessment Monitoring System (PRAMS) conducted by the US CDC estimates that one-half of infant deaths among Indigenous populations in the United States are attributable to low birth weight or preterm birth.
Th e recording of births and deaths, as well as the likelihood of medical intervention have been shown to be aff ected by medical caregivers' perceptions of viability of the baby. Babies that are very preterm may be less likely to be recorded or even to receive care despite reasonable chances of survival [29,30]. In countries without neonatal intensive care, few babies below the gestational age of 32 weeks survive and even at 30 weeks may be called "abortions" and not recorded [31]. Th is is very diff erent than countries with intensive care, where although few babies born alive at 22 weeks may survive intact, by 25 weeks the majority survive [32,33]. Hence even extremely preterm babies may be aggressively resusci tated Lawn et al. BMC Pregnancy and Childbirth 2010, 10(Suppl 1):S1 http://www.biomedcentral.com/1471-2393/10/S1/S1 and data fully recorded, although practices still vary between countries. Th e Nuffi eld Council on Bioethics recom mends that below 22 weeks of gestation resuscitation should not be attempted, even if a baby is born with signs of life [34]. Table 3 provides trends in preterm birth for a number of selected HICs and LMICs, including preterm prevalence in non-representative populations published by WHO in 1995 [35]. Reported preterm birth rates among European and other HICs range from 5% to 9%, and similar to the United States, have been on the rise over the past three decades [36]. A signifi cant contribution to the rise in preterm birth rates refl ects an increase in preterm delivery due to medical indication of either the mother or the fetus. In absolute terms, however, medically-indicated preterm births made up less than half of all preterm births in the year 2000 in the United States [36,37]. In LMICs, data on trends in preterm birth are very limited and results are mixed. In general, LMIC rates tend to be higher than in HICs. In Latin America, rates are increasing in Brazil, possibly related to elective cesarean sections and labor inductions [38]. In an analysis of more than 1.7 million births that took place in 51 maternity hospitals in Latin America, for which Uruguay and Argentina contributed half the births, the rates of preterm birth were essentially the same between 1985 Lawn et al. BMC Pregnancy and Childbirth 2010, 10(Suppl 1):S1 http://www.biomedcentral.com/1471-2393/10/S1/S1

Preterm birth prevalence trends
and 2003 (around 9%). However, there was a marked increase in the proportion of preterm births associated with induction/elective caesarean sections during this period [39]. For countries outside of Latin America, such as China, Indonesia, and Bangladesh, the available studies use sub-national samples and should be interpreted with care.

Preterm birth as a cause-of-death, acute morbidity, and disability
Systematic estimates for the causes of neonatal deaths in 192 countries were undertaken by the CHERG based on vital registration data for 45 countries (N=96,797 deaths) and modelled estimates for 146 countries (input database of N=13,685 deaths). Th ese were published in Th e Lancet Neonatal Survival Series [2], incorporated in the World Health Report 2005 [40], and in Disease Control Priorities in Developing Countries [8,41] (Figure 2). Th e methods are described in detail elsewhere and also summarized in Table 1. At the global level, these estimates place preterm birth as the single largest direct cause of the world's four million neonatal deaths [2]. In addition to being the leading direct cause of neonatal deaths (Figure 2), preterm birth also increases the risk of dying due to other causes, especially from neonatal infections [2]. An example is a moderately preterm baby who dies of infection after a few days of life. Hence, as well as being the leading direct cause of neonatal deaths, preterm birth is a crucial risk factor for neonatal deaths due to infection. A systematic risk factor analysis is planned ( Table 1).
As shown in Figure 3, the proportion of neonatal deaths attributed to preterm births is inversely related to the rates of neonatal mortality, because in countries with very high neonatal mortality, more deaths occur due to infections such as syphilis or tetanus, as well as to intrapartum-related "birth asphyxia" [2]. However, although the proportion of deaths due to preterm birth is lower in LMICs than in HICs, the cause-specifi c rates are much higher in LMICs than in HICs. For example, in Nigeria the estimated cause-specifi c rate for neonatal deaths directly due to preterm birth is 13.5 per 1000 compared to the UK where it is under 2 per 1000. Th is is due to the lack of even simple care for preterm babies. Neonatal mortality rates are higher in LMICs than in HICs, partly because of poorer access to health services and quality of maternal and newborn interventions [5].
Mortality rates increase proportionally with decreasing gestational age (and hence decreasing birth weight). Mortality and morbidity are highest among infants born at less than 32 weeks gestation. Infants born from 32 to 36 weeks represent about 75% of all preterm births and the group of infants who make up the fastest-growing proportion of the preterm births in HICs, with a 25% increase during 1990-2005 [6]. While improvements in medical care have led to improved survival and longterm outcomes among moderately and extremely preterm babies in HICs, these babies still account for the majority of deaths, especially in LICs where even simple care is lacking.
In Southern Brazil, preterm babies experience high mortality rates due to respiratory infections, diarrhea, and other infections that were eight, fi ve, and six times higher, respectively, than rates of term babies [42]. In the United States during 1995-2002, the mortality rate for  Lawn et al. BMC Pregnancy and Childbirth 2010, 10(Suppl 1):S1 http://www.biomedcentral.com/1471-2393/10/S1/S1 term newborns was 2.4-3.0 per 1,000 live births. Among babies who were born between 34 and 36 weeks gestation, the mortality rate was 7.9-9.5 per 1,000 live births [43]. Few studies in the literature evaluate gestational age-specifi c neonatal mortality rates. Th e comparison of three such studies in Table 4 illustrates the diff erences in survival among low-, middle-, and highincome countries.
Th e major focus in HICs is now on the extremes of gestational age and survival. In a comparative analysis of data from France and England in 1997, 19% and 27% of babies born at less than 26 weeks survived to discharge; 57% and 68% of those born at 26-28 weeks gestation survived to discharge; and 86% and 92% of those born at 28-32 weeks survived to discharge, respectively [44]. In a cohort of extremely preterm infants from the United Kingdom from 1995, 26% of babies born at 24 weeks survived to discharge, and among those born at 25 weeks, 44% survived to discharge [45]. Similarly, in a Canadian cohort of babies born between 1996 and 1997, 57% of babies born at 24 weeks and 76% of babies born at 25 weeks survived to discharge [46].

Preterm morbidity and long-term sequelae
Th e complications of preterm birth arise from immature organ systems that are not yet prepared to support life in the extrauterine environment. Th e response of the infant's organ systems to the demands of the extrauterine environment and the life support provided have an important impact on the infant's short-and long-term health and neurodevelopmental outcomes. Th ese outcomes are also infl uenced by the etiology of the preterm birth; maternal and family risk factors; and the extrauterine environment, including the neonatal intensive care unit; and the home and community.
Babies born preterm have an increased risk of morbidity due to diff erent mechanisms. Some are directly related to their immaturity, as with hyaline membrane disease due to the lack of pulmonary surfactant, and retino pathy of prematurity due to the excessive use of oxygen to treat hyaline membrane disease. Preterm birth may also be a marker for other problems that produce disease, such as fetal infection and systemic infl ammation, which are themselves associated with intracranial haemorrhage, cerebral white matter damage, cerebral palsy, and chronic lung disease (bronchopulmonary dysplasia) [47].

Defi ning stillbirth
Th e International Classifi cation of Diseases, 10 th revision (ICD-10) [48] defi nes a fetal death as "death prior to the complete expulsion or extraction from its mother of a product of conception, irrespective of the duration of pregnancy; the death is indicated by the fact that after such separation the fetus does not breathe or show any other evidence of life, such as beating of the heart, pulsation of the umbilical cord, or defi nite movement of voluntary muscles" without specifi cation of the duration of pregnancy. Although birth weight has been the preferred criterion in the ICD to identify a late fetal death, gestational age is an additional requirement for reporting for international comparative purposes. ICD classifi es late fetal deaths (greater than 1000 gms or after 28 weeks) and early fetal deaths (500 to 1000 gms or 22-28 weeks) ( Table 5).
It should be noted that "stillbirth" is not a technical term. In this article "stillbirth" refers to late fetal deaths to conform to the WHO recommendation that late fetal deaths be reported for purposes of international comparison. Th e rationale for restricting international reporting to stillbirths of greater than 1000 gms or after 28 weeks is to assure comparability, as the countries where most stillbirths occur mostly still do not capture even these larger more mature deaths reliably and data remain uncertain [49]. In countries lacking neonatal intensive care, few babies below the gestational age of 30 weeks survive [31]. However, in many countries where neonatal intensive care units are available, the gestational age for viability has decreased, and the gestational age criterion to defi ne stillbirth has been adapted accordingly. Current gestational age thresholds for stillbirth vary from 16 to 28 weeks of gestation across countries.

Stillbirth rates estimates
Prior to 2006, no organization had published global, regional or country-specifi c stillbirth rates. Two global series of stillbirth estimates for the year 2000 were published in 2006 (hereafter referred to as the SNL/immpact and WHO estimates) [7,50], with both exercises generating estimates of just over three million stillbirths (3.2 million, with wide uncertainty: 2.5-4.1 million; and 3.3 million, respectively). SNL/immpact represents a collaboration between Saving Newborn Lives/Save the Children USA and the Initiative for Maternal Mortality Programme Assessment, at the University of Aberdeen, Scotland. Lawn et al. BMC Pregnancy and Childbirth 2010, 10(Suppl 1):S1 http://www.biomedcentral.com/1471-2393/10/S1/S1 Defi nition For international comparison stillbirth rates refer only to late fetal deaths (>1000g or >28 weeks gest).
ICD-10 defi nes a fetal death as «death prior to the complete expulsion or extraction from its mother of a product of conception, irrespective of the duration of pregnancy; the death is indicated by the fact that after such separation the fetus does not breathe or show any other evidence of life, such as beating of the heart, pulsation of the umbilical cord, or defi nite movement of voluntary muscles» without specifi cation of the duration of pregnancy. The denominator is all live births plus late fetal deaths.

Continued overleaf
Lawn et al. BMC Pregnancy and Childbirth 2010, 10(Suppl 1):S1 http://www.biomedcentral.com/1471-2393/10/S1/S1 http://www.biomedcentral.com/1471-2393/10/S1/S1 Figure 4 presents the SNL/immpact numbers of stillbirth by region. Given the very diff erent methods used in these two estimation exercises and the dearth of stillbirth data available from developing countries, the results are remar kably similar. Table 6 summarizes regional stillbirth rates from the two series of estimates. Stillbirth rates are very similar for sub-Saharan Africa and South Asia (32 per 1000 births). However, there is little agreement between the remaining regional estimates and even less agreement at the country level, where the data are most needed for planning purposes; for example, two-to three-fold diff erences in both directions exist between the two series of estimates for some countries (data not shown). Figure 4 presents the estimated number of stillbirths by world region.
Th e methods for both series of stillbirth estimates have been summarized in Table 5. Any global estimation exercise is by defi nition an attempt to make the best of sub-optimal data. Both series of estimates suff er from a lack of quantity and of quality input data. Th is leads to decisions in the modelling process that are easy targets for criticism. Our summary assessment of these two exer cises is "moderate" at best when judged according to the criteria outlined in Table 5. An updated series of stillbirth rates and numbers for 2005 will be undertaken jointly by WHO and CHERG with GAPPS and undergo external review prior to the next global burden of disease exercise.

Availability of stillbirth rate data
In HICs, national vital registration systems usually have high coverage and reasonably reliable cause-of-death data for live births, but the stillbirth data are often more questionable [7]. Globally, only about 2% of late stillbirths are accounted for via vital registration. In countries lacking complete vital registration on stillbirths, but with high institutional birth rates, health facility-based data are also an important source of representative data on pregnancy outcomes. In LMICs, by far the largest source of data on stillbirths comes from population-based household surveys. Other sources include demographic surveillance sites, or special studies. In LMICs lacking high institutional birth rates, health facility data can still be a valuable resource if compiled regionally or nationally, especially if selection bias is taken into account. Notable examples are the Latin American Center for Perinatology (CLAP) database [51] and the South African Perinatal Identifi cation Programme [52,53].

Stillbirth causes of death
Currently there are no global, systematic estimates for stillbirth causes of death. Where data do exist, the lack of comparability across studies greatly inhibits interpre tation. More than 30 diff erent stillbirth classifi cation systems have been identifi ed in the literature [54], with some encompassing up to 37 causes [55]. Most focus on stillbirths in HICs where determination of the most prevalent causes requires fetal surveillance and sophisticated diagnostics [49]. Later in the article, options for improving stillbirth cause-of-death comparability will be discussed.
Comparable data regarding the timing of stillbirths relative to delivery are more widely available. Intrapartum stillbirths are generally defi ned as stillbirths occurring after the onset of labour, or as "fresh stillbirths" (with skin still intact, implying death occurred less than  Lawn et al. BMC Pregnancy and Childbirth 2010, 10(Suppl 1):S1 http://www.biomedcentral.com/1471-2393/10/S1/S1 12 hours before delivery) weighing more than 1,000 grams and more than 28 weeks of gestation, but exclude severe lethal congenital abnormalities [56]. Th is increased availability of data permitted publication in 2005 of intrapartum stillbirth rates for 192 countries. Details regarding the input data for this series of estimates are included in Figure 5 and methods are summarized in Table 5. Based on these estimates, one million intrapartum stillbirths occur annually (uncertainty bounds: 0.66-1.48), representing one-third of stillbirths globally [56]. Despite the caveats inherent in the interpretation of the intrapartum stillbirth estimates, these estimates clearly highlight the magnitude of loss of life just minutes and hours prior to birth. Hospital-based studies suggest that from 25-62% of intrapartum stillbirths are avoidable with improved obstetric care and more rapid responses to intrapartum complications, including reducing delays in seeking care from home [57][58][59][60].
At the time of writing of this supplement, a systematic review of the literature on stillbirth cause-of-death is under way. Data permitting, the distribution of causes of stillbirth will be estimated, using methods similar to those used to estimate neonatal cause-of-death [61]. Approxi mately 65 studies from 36 countries have been identifi ed that provide at least minimal stillbirth causeof-death data. Data have been extracted into the following categories: congenital abnormality (physically visible); maternal conditions (including pregnancyinduced hypertension, eclampsia/preeclampsia, diabetes, other antenatal); antepartum hemorrhage (abruption); infections (including syphilis, other maternal and other fetal infection); intrapartum stillbirth (including obstruction, and breech); preterm labor of undetermined cause; other and unclassifi able.

Preterm birth data improvement Improving measurement of preterm birth prevalence
Th ere are many opportunities to improve data now in both low-and high-income countries (Table 7). Th e defini tion for preterm birth (less than 37 weeks of completed gestation) is well-known. Th e challenge is the current low priority given to collecting gestational age data, and the complexity of measurement (apart from the use of last menstrual period). Further eff ort is needed to infl uence the content of midwifery and medical pre-and in-service education and to establish gestational age assessment as an integral component of routine care.
In HICs, gestational age assessment has surpassed birth weight as the measurement of choice, with a much closer correlation to short-and long-term outcomes. A number of methods exist for the assessment of gestational age. In Lawn et al. BMC Pregnancy and Childbirth 2010, 10(Suppl 1):S1 http://www.biomedcentral.com/1471-2393/10/S1/S1 middle-income countries, gestational age is increasingly available, even with ultrasound dating (the gold standard). In most countries, a very small proportion of births have reliable gestational age assessment. Even estimates of gestational age based on last menstrual period are often not recorded or known, particularly in African settings. In most survey-based data, women are asked to state their gestational age in completed months. Th is is the practice in the DHS contraceptive calendar, for example.

Option 1: Birth weight as a surrogate measure
In LMICs, low birth weight is often used as the criterion for identifying preterm births given the paucity and quality of self-reported data on gestational age. Reliance on LBW is problematic, however, as 58% of babies in LMICs are not weighed at birth (Table 8), and homebased births, those most likely to be of low birth weight, are not represented [23]. In middle-income countries, notably in Latin America, many countries have a record of birth weight for the majority of babies (83%), but in South Asia and sub-Saharan Africa, where the majority of neonatal deaths occur, only a fourth to a third of babies have a record of birth weight. Th ese fi gures parallel the coverage of skilled attendance at birth, though even with a facility-based birth by a skilled attendant, the birth weight may not be recorded due to a lack of scales, skilled staff , and standard protocols [13]. Table 9 shows the proportion of preterm babies in diff erent birth weight groups. Only about half of the newborns at 2000-2499-grams were born preterm. Th ese data suggest that using a cut-off of 2000 grams may be more appropriate than the traditional LBW defi nition in  Lawn et al. BMC Pregnancy and Childbirth 2010, 10(Suppl 1):S1 http://www.biomedcentral.com/1471-2393/10/S1/S1 identifying preterm births. Th e two studies described in Table 9 are from Latin America, and these proportions may diff er in other regions, such as South Asia where intrauterine growth restriction is highly prevalent. When data are available for birth weight and age-at-death of stillbirths and neonatal deaths, a simple cross tabulation of birth weight by age-at-death can be a useful guide for programmatic priority setting [62]. For example, full-size babies dying during birth have very diff erent solutions to very small babies dying after birth.

Option 2: Clinical assessment of gestational age
Given the need for a paradigm shift to use gestational age instead of birth weight for the identifi cation of preterm births, the possibility of simplifi ed gestational age assessment by lower cadres of workers is of interest. A recent systematic review of methods for gestational age assessment identifi ed 17 diff erent methods using a combination of neurological and physical criteria or physical criteria alone [63]. Methods requiring complex technology or neurological assessment alone were excluded. Of these 17 methods, fi ve were considered "complex, " nine were "intermediate, " and three were "simple, " based on the number of characteristics examined. As compared against varying standards (only some of which were ultrasound) all methods were accurate within plus or minus three weeks. Th e number of methods to choose from and the varying levels of complexity allow for recommendations to be made appropriate to two settings: tertiary care hospitals and district-level health facilities. Further uptake of these methods are needed by international and medical professional associations to infl uence the content of midwifery and medical pre-and in-service education as a means of establishing gesta tional age assessment as an integral component of routine care. Evaluation of use at large-scale settings and data validity could further refi ne recommendations by setting. However, none of the methods which were compared against an acceptable standard were applied by community health workers. Hence, further research is required to identify the most feasible and acceptably accurate methods for community-based gestational age assessment.

Improving measurement of other parameters related to the burden of preterm birth
A new analysis would be required to better delineate the eff ect of preterm compared to term gestational age, to defi ne the risk of varying gestational ages for death, and to separate direct from indirect risks. Individual-level data on birth weight, gestational age, mortality outcome and ideally comparable cause-specifi c mortality would be required for such an analysis (Table 1).
To improve the assessment of long-term outcomes of preterm birth, particularly impairment outcomes, an international consensus group is required to agree to standard defi nitions for these parameters. Protocols and tools are required to ensure standard measurement, especially for disability and cognitive function at various ages. Table 10 summarizes a number of opportunities that are immediately available to improve stillbirth data through existing data collection mechanisms.

Option 1 -Vital registration
Improved measurement of stillbirths in HICs requires a focus on highly standardized reporting of stillbirths via vital registration or other comprehensive national registries. Th e most important data intervention is the establish ment of a stillbirth death certifi cate. Given the plethora of data available from HIC health facilities, standard ized reporting is entirely feasible. At issue is the political will to demand such an intervention. Establishment of a stillbirth death certifi cate could address both improved counting of events, as well as improved standardization of the causes of stillbirth.
In LMICs, one should capitalize on the current increased interest in improving vital registration by also introducing a standard perinatal death certifi cate. Complete registration may be a distant goal, but as the foundation for improved data is being established, stillbirths should be included or countries will miss the opportunity to show mortality change concurrent with the implementation of maternal and neonatal programs.
At the international level, seizing the opportunity of the upcoming revision of ICD codes to refl ect recent advances in diagnosing stillbirth cause-of-death is essential to future data improvement. Regarding the identifi cation of avoidable factors for the prevention of stillbirth, expanding the use of national audits or other forms of confi dential inquiry is recommended. In addition to Lawn et al. BMC Pregnancy and Childbirth 2010, 10(Suppl 1):S1 http://www.biomedcentral.com/1471-2393/10/S1/S1 investigating traditional defi ciencies in quality of care, these audits can be adapted to specifi c contexts to also examine socioeconomic disparities and demographic or behavioral characteristics of the population of interest.

Option 2 -Population-based surveys
Th e Demographic and Health Survey (DHS) website has posted national data on stillbirth rates for 49 surveys from 38 countries [64]. Th ese surveys are by far the largest source of national data from LMICs. Given the lack of vital registration data on stillbirths in LMICs, reliance on survey-based estimates is inevitable for the near future, and given that 98% of stillbirths occur in LMICs-this data source cannot be ignored. Th e majority of DHSs consist of a complete live birth history for each woman of reproductive age in the sample. Many also include a contraceptive calendar in which monthly data on each respondent's contraceptive use, pregnancy status, and pregnancy outcomes are collected for the 60-month period prior to interview. Th ese data permit calculation of stillbirth rates. DHS stillbirth rates range from 3.4 per 1000 (in Ukraine) to 37.0 per 1,000 (in Bangladesh). Excluding the surveys in Bangladesh and Nepal, DHS stillbirth rates do not surpass 20 per 1,000. However, the Bangladesh DHS estimate is similar to the high-quality estimate from demographic surveillance data in Matlab, Bangladesh [65].

Opportunities Immediately Available Research Priorities (Focus on high Opportunities
High-Income Settings Low-Income Settings mortality, low quality data settings) Comparable case Use 28 week cut-off for international Prioritize improved collection of Development of simple and feasible proxy defi nitions for stillbirth comparisons and 22 week cut-off for representative population-based data indicators for gestational age (e.g., weight) High-Income Country comparisons.
for last trimester and intrapartum Local defi nitions can be used for stillbirths. local purposes.

Mechanisms for
Improve vital registration data by Increase attention to training and fi eld Validation of existing approaches for counting all births, establishing specifi c death certifi cates supervision for DHS-type household pregnancy loss data collection compared to (including stillbirths) for stillbirth and neonatal deaths. surveys which rely on retrospective pregnancy loss data from sentinel surveillance Cross-link data from vital registration reporting of all births. Consider adding sites and health facility surveillance.
stillbirth data collection to MICS surveys. Analyze existing pregnancy loss data from sentinel surveillance sites and increase the number of sentinel surveillance sites which prospectively collect stillbirth data. Improve vital registration systems and register stillbirths. Use specifi c death certifi cates for stillbirths/neonatal deaths.
Classifi cation for Obtain consensus on a single classifi cation system with a limited number of Evaluation of validity and feasibility of a simple stillbirth cause-of-death programmatically relevant, comparable categories, that can be distinguished in standard classifi cation system for stillbirth low income settings through verbal autopsy, but can also be directly cause-of-death incorporated into more detailed sub groups necessary in high income settings attribution for cause-of-death. and costs less than expert assessment of verbal autopsy data Counting avoidable Increase the number of national audit Develop or modify audit systems Evaluation of simple audit tools and a factors, using data systems .Consider confi dential enquiry. linking maternal/fetal and neonatal mechanism to maximize resultant change in in programmes deaths. Compile national data and/or policy and programs. promote sentinel sites in varying health system contexts to ensure that the information is useful for policy prioritization, even if not representative of the population. Consider focusing on few indicators initially (e.g. Intrapartum Case Fatality Rate). Use existing data (e.g., facility birth registers) for local monitoring and programmatic decision-making. http://www.biomedcentral.com/1471-2393/10/S1/S1 Evidence from countries with adequate historical data suggest that an SBR:ENMR ratio of approximately 1.2 can be expected in high mortality countries [50]. Only 5 of the 49 DHS surveys show ratios greater than one. For sub-Saharan African countries, the regional (populationbased) ratio is only 0.55, and ranges from 0.61-0.64 for the remaining regions, suggesting extreme under-reporting in the large majority of these countries. Moldova stands out as an extreme outlier with a ratio of 3.2, suggesting likely misclassifi cation between stillbirths and early neonatal deaths. In a separate analysis of stillbirth rates from multiple data sources, DHS calendar-based stillbirth estimates were found to be approximately 30% lower than other population-based studies, after controlling for other study and population characteristics [7]. As currently implemented, the contraceptive calendar is not a reliable source of stillbirth data.
Over the past 20 years limited research attention has been applied as to how best to collect population-based pregnancy loss data. In 1989, Casterline analyzed the pregnancy loss data in 41 World Fertility Surveys and concluded that these pregnancy histories in their various formats detected from 50-85% of recognizable pregnancy losses, as compared to results from prospective, clinical studies in Western countries [66]. As expected, late fetal losses tended to be better reported than earlier miscarriages. Garenne noted the highly reliable reporting on perinatal mortality in Niakhar, Senegal, when comparing pregnancy history data to DSS data [67]. Goldman et al. [68], Westoff et al. [69], and Becker and Sosa [70] studied the eff ects of using a truncated pregnancy history in Peru, the Dominican Republic, and Costa Rica, with varying results depending on the outcome studied. Stanton found the reliability of reporting stillbirths in two national DHS-type surveys using pregnancy histories from the Philippines to be lower than for early neonatal or infant deaths [71].
To date the most rigorous examination of the validity of self-reported pregnancy outcome data was undertaken by Espeut in Bangladesh: comparing DSS data from Matlab, Bangladesh, matched to respondents in a DHS survey in which respondents were randomly assigned a questionnaire with a live birth or pregnancy history. In summary, a 91% sensitivity rate was found for reporting in the survey on stillbirths. In contrast, the sensitivity rate for early neonatal deaths varied from 79-81% in live birth and pregnancy histories; among stillbirths in the DSS, 3% were misclassifi ed as live births, and 9% were misclassifi ed as abortions (suggesting diffi culty in recalling gestational age) compared to self-report in the surveys [72]. Th e goal of future validation eff orts should not be restricted to identifi cation of the highest quality approaches but should quantify the loss of data quality in choosing, for example, a truncated vs. complete live birth or pregnancy history, or a survey covering wide-ranging issues vs. a highly focused questionnaire. Immediate progress can be made by carefully reviewing the wide variation in current data collection processes-including the formulation of questions that would elicit reporting on pregnancy losses. Likewise, assuring increased attention to the defi nition of stillbirth during interviewer training and improving supervision in the fi eld could also lead to immediate improvements in data quality.

Option 3 -Demographic surveillance sites and special research studies
Demographic surveillance sites (DSS), in which the vital events and background characteristics associated with all residents are recorded prospectively, should be an important data source on early pregnancy loss, stillbirths, and preterm births. INDEPTH, a network of researchers from DSS around the world, promotes the registration of pregnancy and pregnancy outcomes as a means of early registration of births but also for identifying stillbirths and abortions [73]. Although the collection of pregnancy loss data is highly recommended in DSS, it is unclear how many current DSS actually collect pregnancy loss data, and among those that do, how many regularly or ever report such results. Few DSS data could be located in the published or web-based literature. In contrast, the DSS in Matlab, Bangladesh, includes stillbirth data in their routine reporting [65]. Th e evaluation of existing but publicly unavailable data on late pregnancy loss from demographic surveillance sites demands immediate attention and could potentially off er important clues to improved data collection.

Improving stillbirth cause-of-death data
While lack of data on stillbirth cause-of-death is a large hurdle to overcome, another major barrier is the lack of a classifi cation system that is feasible for low-income countries and which is based on categories that can be mapped alongside more complex classifi cations which are useful in high-income settings [49]. Currently, twothirds of the world's stillbirths lack programmatically meaningful cause-of-death categories which could be used to inform prevention strategies.
Stillbirth classifi cation systems have proliferated over the years and a review suggests at least 33 are in use [54]. Most of these are designed for high-income countries and involve laboratory and pathological examination of the baby and the placenta, so are impractical for use when the only information for most stillbirths is through verbal autopsy occurring a year or even longer after the loss. International consensus for standard classifi cation and comparable attribution of cause are essential to improve the comparability and use of stillbirth cause-ofdeath data. Th is can only be achieved if the classifi cation Lawn et al. BMC Pregnancy and Childbirth 2010, 10(Suppl 1):S1 http://www.biomedcentral.com/1471-2393/10/S1/S1 system is practically applicable and serves the needs of high-as well as low-mortality settings.
High-mortality settings require broad causal categories which can be distinguished through simple clinical observations or even through verbal autopsy and which are programmatically relevant in that they identify conditions associated with large numbers of deaths. One useful distinction for stillbirth prevention strategies is between macerated (antepartum) and fresh (intrapartum) stillbirths. Rates of fresh stillbirths are assumed to refl ect the quality of intrapartum care (care in labor), while rates of macerated stillbirths are assumed to refl ect the quality of fetal growth and of care during the antenatal period. Th e antepartum/intrapartum distinction can generally be explored in verbal autopsy studies with questions pertain ing to the appearance of the infant's skin. Such questions have been used and are believed to be well understood by respondents, though they have not yet been systematically validated. Th ere is some potential for misclassifi cation between these categories. For example, in settings with major delays in access to health care, stillbirths may die during labor, but not be delivered for days by which time they are classifi ed as macerated. Conversely, some intrapartum stillbirths may be due to infections or congenital causes. Also, women who have delivered stillbirths may not be shown the infant, and therefore could not adequately respond to questions about the infant's appearance. Th e extent of this mis classi fi cation may vary locally and requires further research [56].
Once these two major categories (antepartum and intrapartum) are defi ned, a more detailed set of programmatically relevant causal groups can be distinguished. Th is intermediate level of detail is possible with clinical data and achievable in most facility deaths in LMICs (e.g., the South African National Saving Babies data) [74,75]. For high-income countries, the existing complex classifi cation systems often require sophisticated investiga tion but can be mapped onto simpler clinical categories ( Figure 6). In verbal autopsy data and even in clinical assessment, some causal groups will be systematically underestimated. For example, congenital abnormalities are underestimated even in high-income countries but are markedly underestimated in verbal autopsy data because only obvious external abnormalities are detected and important internal structural and metabolic disorders are missed. Data from the literature show that around 5-15% of stillbirths are attributed to a congenital cause. Another important cause of stillbirth that is often missed is maternal syphilis. Figure 6 proposes groupings allowing a layered approach with increasing complexity of causal attribution in varying settings. Much could be learned by reclassifying existing data on stillbirth causes of death using the classifi cation system proposed in Figure 6 (or some adaptation thereof ) via collaboration with the original authors responsible for data collection. Such an exercise would quickly and inexpensively test this proposed classifi cation and identify any caveats in the interpretation of the results.
Such a classifi cation system for stillbirth cause-of-death would allow comparability between diff erent data collection systems, such as verbal autopsy and more complex data systems ( Figure 6). Several verbal autopsy tools now exist, thus gaining consensus on a standardized verbal autopsy tool would be an important advance. Such a tool would need to be tested in a wide variety of contexts. Data on avoidable factors contributing to stillbirths could also be addressed within the verbal autopsy questionnaire by adding a social autopsy module covering questions regarding care-seeking and benefi cial or harmful traditional practices, for example. Much more in-depth information would be available through the use of a stillbirth audit, and there are a number of LMICs attempting to increase the coverage and quality of their audit networks. South Africa is an example of a country which has achieved both high coverage and high quality of perinatal audit data that are used for national decisionmaking [76].

Conclusion
Despite more than three million annual stillbirths and approximately one million neonatal deaths directly due to preterm birth, these burdens and the associated loss to families and nations are rarely highlighted in global health policy and research agendas. Th e impact of the com bined numbers of deaths from stillbirth and preterm birth, plus the morbidity and long-term disability associa ted with preterm birth, is considerable. Clinical researchers and epidemiologists face formidable barriers in collecting and analyzing data about prevalence and interventions, particularly in South Asia and sub-Saharan Africa where two-thirds of these events occur. Th e places with the highest risk currently have the least information available. Yet, the quantity and quality of information could be improved, even in the short-term by: (1) seizing opportunities to add or test the measurement of stillbirths and preterm births to ongoing data collection mechanisms; (2) using consistent defi nitions and classifi cation systems across current data collection mechanisms and research studies; and (3) improving global estimates for both outcomes. Research into etiologic mechanisms responsible for stillbirth and preterm birth has been hampered by the lack of standardized defi nitions and measurement protocols for assessing these outcomes. Th e global economic burdens related to these outcomes remain a signifi cant research gap.
From this review, the priority gaps in existing estimates and in the data for eff ective program design include the following: Lawn et al. BMC Pregnancy and Childbirth 2010, 10(Suppl 1):S1 http://www.biomedcentral.com/1471-2393/10/S1/S1 • Systematic estimates for causes of stillbirth are required to increase visibility and prioritize action to reduce these deaths. Agreement around a simplifi ed classification system is a key step to underpin global estimates. • Th e lack of systematic country-level estimates for the prevalence of preterm birth, based on well-defi ned and standard phenotype classifi cation, is an important gap aff ecting the visibility of preterm birth globally. Th e lack of information for preterm prevalence is most marked in Africa and the Eastern Mediterranean. Virtually no consistent data on preterm prevalence trends are available from LMICs. Th e development of methods to permit reliable population-based data on trends in preterm birth in these countries is a key priority. • New analysis is required to better defi ne the risk of death at varying gestational ages, and to separate direct from indirect risks. Input data sets would need to include individual-level data on birth weight, gestational age, mortality outcome, and ideally, comparable causes of death.
• Acute morbidity and long-term sequelae of preterm birth remain virtually unstudied in LMICs, despite the fact that survival is now increasing in some of these settings. Tracking morbidity is crucial. Standard tools and protocols to assess morbidity and long-term sequelae across varying cultures are lacking. Attempts at these global estimates are severely hampered by the lack of data. Opportunities highlighted by this review that could improve the availability and quality of data, even in the short term, include: • Improve the capture and quality of pregnancy outcome data through household surveys, which is the main data source for the countries with 75% of the global burden, and undertake validation studies. Th e expanded number of demographic surveillance sites currently func tioning in various LMICs off er excellent oppor tunities to compare prospective versus retrospective reporting on pregnancy outcomes.

PLACE OF DEATH AND LEVEL OF DATA INPUT AVAILABLE
Lawn et al. BMC Pregnancy and Childbirth 2010, 10(Suppl 1):S1 http://www.biomedcentral.com/1471-2393/10/S1/S1 • Increase awareness of, and compliance with, standard defi nitions for stillbirth and preterm birth, and more frequently include stillbirth and gestational age data in existing data collection systems (vital registration, facility-based data and research studies). Current ICD 10 codes for both stillbirth and preterm birth need to be updated to refl ect defi nitions currently in use and advances in understanding made in the last decade. A simplifi ed classifi cation system for stillbirth cause-ofdeath could also be incorporated into the ICD 11. Th is would allow data from a standardized verbal autopsy tool and other data collection systems in LMICs to improve input data for future global estimates. • Expand and strengthen the coverage and quality of existing data collection mechanisms, especially vital registration, and facility data by instituting a standard death certifi cate for stillbirth and neonatal death linked to revised International Classifi cation of Diseases coding. • Validate a simple, standardized classifi cation system for stillbirth cause-of-death that is feasible though verbal autopsy. • Improve systems and tools to capture acute morbidity and long-term impairment outcomes following pre term birth and other adverse pregnancy or neonatal events. In addition to these priority actions to improve preterm birth and stillbirth data in the immediate future, there is an extensive research agenda around the epidemiology of preterm births and stillbirths and many possible research questions too detailed to list here. Th e fi nal article in this report presents a Global Action Agenda developed by global stakeholders at the GAPPS International Conference on Prematurity and Stillbirth held in May 2009, and includes short-and long-term objectives related to the epidemiology of preterm birth and stillbirth [16].
Th e numbers discussed in this report are large-on par with the issues considered the greatest priorities in global health today, and indeed larger than some that receive major attention, such as two million annual HIV/AIDS deaths. Yet, preterm birth and particularly stillbirth are not included amongst global priorities. Th is invisibility is partly an issue of data, but remains a reality despite increasing quality and progress for global estimates. Another critical issue is the value put on a baby's life-a newborn baby remains the most vulnerable human and a preterm newborn is even more vulnerable.
Each of these losses is a bereavement for families and may leave a deeper scar than a death which is openly acknowledged and mourned. Long-term follow-up studies show that 20 years after a stillbirth, a woman may remain in a delayed grief response [77,78]. Th e societies where stillbirth and preterm birth have become priorities are those where such babies are expected to live, and women and families can express their loss. Indeed, the power of these families to use data for change may be likened to the power of individuals who lost loved ones from HIV/AIDS and advocated successfully for change. Data alone will not result in change until society and leaders recognize that these deaths are a loss that can and must count and be prevented.