National and regional estimates of term and preterm babies born small for gestational age in 138 low-income and middle-income countries in 2010

Background: National estimates for the numbers of babies born small for gestational age and the comorbidity with preterm birth are unavailable. We aimed to estimate the prevalence of term and preterm babies born small for gestational age (term-SGA and preterm-SGA), and the relation to low birthweight (<2500 g), in 138 countries of low and middle income in 2010. Methods: Small for gestational age was defined as lower than the 10th centile for fetal growth from the 1991 US national reference population. Data from 22 birth cohort studies (14 low-income and middle-income countries) and from the WHO Global Survey on Maternal and Perinatal Health (23 countries) were used to model the prevalence of term-SGA births. Prevalence of preterm-SGA infants was calculated from meta-analyses. Findings: In 2010, an estimated 32·4 million infants were born small for gestational age in low-income and middle-income countries (27% of livebirths), of whom 10·6 million infants were born at term and low birthweight. The prevalence of term-SGA babies ranged from 5·3% of livebirths in east Asia to 41·5% in south Asia, and the prevalence of preterm-SGA infants ranged from 1·2% in north Africa to 3·0% in southeast Asia. Of 18 million low-birthweight babies, 59% were term-SGA and 41% were preterm-SGA. Two-thirds of small-for-gestational-age infants were born in Asia (17·4 million in south Asia). Preterm-SGA babies totalled 2·8 million births in low-income and middle-income countries. Most small-for-gestational-age infants were born in India, Pakistan, Nigeria, and Bangladesh. Interpretation: The burden of small-for-gestational-age births is very high in countries of low and middle income and is concentrated in south Asia. Implementation of effective interventions for babies born too small or too soon is an urgent priority to increase survival and reduce disability, stunting, and non-communicable diseases. Funding: Bill & Melinda Gates Foundation by a grant to the US Fund for UNICEF to support the activities of the Child Health Epidemiology Reference Group (CHERG). © 2013 Lee et al. We defined small for gestational age as a birthweight lower than the 10th centile for a specific completed gestational age by gender, using the Alexander reference population8 (US National Center for Health Statistics, 1991; n=3 134 879 livebirths). We defined term-SGA as a baby born small for gestational age at 37 or more completed weeks of gestation, and we classified preterm-SGA as infants born small for gestational age at fewer than 37 weeks of gestation. We defined low birthweight as a baby born weighing less than 2500 g. Finally, we defined appropriate for gestational age as a birthweight on or higher than the 10th centile for gestational age, using the Alexander reference. We obtained data from three sources: (1) systematic literature reviews to identify birth cohorts with information on birthweight and gestational age; (2) research networks of birth cohorts; and (3) the WHO Global Survey on Maternal and Perinatal Health.9 We considered datasets for inclusion if they contained information on: birthweight; gestational age measured by last menstrual period, ultrasound, or clinical assessment; and vital status (required for analyses described elsewhere).10 Our exclusion criteria were: datasets missing more than 25% of data for birthweight or gestational age; inaccurate gestational age ascertained by study investigators (ie, poorly linked gestational age–birthweight data, or gestational age reported in months); and gestational age established by symphysis fundal height. We only included weight in analyses if the measurement was made within 72 h of birth. We initially searched Medline and WHO regional databases (LILACS, AIM, and EMRO) in September, 2009, to identify potential birth cohorts with data required for a larger scope of secondary analyses related to preterm birth and small-for-gestational-age-related mortality (appendix p 1).10 We identified additional datasets within maternal-neonatal research networks (ongoing maternal-newborn health studies, demographic surveillance sites, and WHO UNIMAPP11 studies). We contacted investigators to ascertain whether their studies met our inclusion criteria and, if so, we asked them to join the Child Health Epidemiology Reference Group (CHERG) SGA-Preterm Birth working group and contribute data for secondary analyses (appendix pp 2–4).10 We did another literature review in February, 2012, to identify published studies reporting the prevalence of both small-for-gestational-age births, using the Alexander reference,8 and low birthweight to use for statistical modelling, since low birthweight was the primary independent modelling predictor. We implemented two strategies: (1) a Medline search using terms (“small-for-gestational-age” OR “intrauterine growth restriction”) AND “low birthweight” AND (“incidence” OR “prevalence”) AND “developing country”; and (2) a Scopus search identifying all published articles that have cited small for gestational age using the Alexander reference.8 Further details on our search strategy are in the appendix (p 5). We also analysed data from the WHO Global Survey on Maternal and Perinatal Health (appendix p 6),9 which gathered data between 2004 and 2008 from 373 facilities in 24 countries and included 290 610 births. We excluded data from Japan (n=3318) because it is a high-income country. Therefore, a total of 23 countries contributed to the analysis. Details of survey methods are reported elsewhere.9 The WHO Global Survey selected countries randomly from every WHO subregion and then picked facilities at random from the capital city and two other randomly selected provinces. For this facility-based survey, trained data collectors abstracted relevant data from medical records into standardised forms from all births in the facility over a specific period. Several facilities had data with improbable values or unrepresentative data. To exclude these poor data-quality facilities, we omitted those with fewer than 500 births (small sample size), preterm birth rates greater than 40% or less than 3% (outside biological plausibility range), or rates of low birthweight less than 1% (implausible). We aggregated data at the country level. Datasets analysed by the original study investigators were approved by existing site institutional review boards. For datasets shared with the CHERG working group, personal identifiers were not included and, therefore, were deemed exempt by the Johns Hopkins Bloomberg School of Public Health institutional review board. In the first step of the estimation process, we developed a model to estimate the national prevalence of term-SGA, based on the included input data. We then estimated the prevalence of preterm-SGA, using meta-analytical methods, and we applied these proportions to recent national preterm birth estimates developed by members of our working group.2 We used Stata 11.0 for all analyses. We did random-effects regression with logit(term-SGA prevalence) as the dependent variable and study region as the clustered unit of analysis (appendix p 7). Variables tested included: biological factors (prevalence of low birthweight, neonatal mortality rate); health-care access (proportion of deliveries in a facility, proportion of births by caesarean section, proportion of mothers with more than four antenatal care visits); and demographic factors (proportion of cohort in highest risk categories of maternal age, parity, and maternal education). We created categorical dummy variables for: degree of selection bias (population-based or community-level recruitment, facility-based or antenatal care recruitment with some or minimum selection bias, tertiary care or referral facility); study decade; and method of assessment of gestational age (last menstrual period, ultrasound, clinical). To examine candidate models, we included the natural log of low-birthweight prevalence as the main predictor, added individual predictors, and assessed for significance (p<0·05), improvement in adjusted R2, and Akaike information criterion. To select the final model, we did cross-validation12 to compare prediction accuracy between potential models (appendix p 8). We undertook sensitivity analyses using two datasets. In the first (modelling dataset A, n=45), we included all birth cohort data. In the second restricted dataset (modelling dataset B, n=20), we included only population-representative studies, excluding facility-based studies that might have selection bias (WHO studies,9 Pakistan Aga Khan University [ZAB], Uganda 200513). Both datasets A and B resulted in similar estimates of variables and model fit; thus, we present results of the larger dataset A, which enabled cross-validation. We also did multiple imputation14 to incorporate infants with missing birthweight back into the individual cohort studies (appendix p 9). For every cohort in dataset A, we calculated the prevalence of small-for-gestational-age babies within two categories of preterm births: moderate-to-late preterm (between 32 weeks and <37 weeks of gestation) and early preterm (<32 weeks of gestation). We used random-effects meta-analyses to estimate the pooled regional and overall prevalence of infants born small for gestational age between 32 weeks and less than 37 weeks of gestation and those born at less than 32 weeks of gestation. We did sensitivity analyses to look at the effect of region, facility-based versus community-based recruitment, and study quality. We judged studies of a high quality to be those with population-based recruitment, adequate data capture (defined as missing <15% of birthweight data and <30% of birthweight data among neonatal deaths, and the proportion of infants born at <32 weeks comprising at least 5% of preterm births), and biological plausibility (prevalence of small for gestational age >1%). We used the prediction model to estimate term-SGA prevalence in countries of low and middle income (UN Millennium Development Goal [MDG] classification) for the year 2010. We took national neonatal mortality rates from the UN Interagency Group for Child Mortality Estimation15 and low-birthweight rates from several sources (appendix p 10). To obtain the number of small-for-gestational-age liveborn infants, we multiplied the prevalence of term and preterm small for gestational age by the estimated number of livebirths for 2010.16 We used a bootstrap approach to estimate ranges of uncertainty (appendix p 11). In every dataset, we calculated the proportion of term-SGA infants who were low birthweight and did meta-analyses, using random effects to pool the estimate at the major regional level. We multiplied this value by term-SGA estimates for every country and summarised them regionally. The sponsor of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.