Low birthweight and preterm birth: trends and inequalities in four population-based birth cohorts in Pelotas, Brazil, 1982–2015

Abstract Background Despite positive changes in most maternal risk factors in Brazil, previous studies did not show reductions in preterm birth and low birthweight. We analysed trends and inequalities in these outcomes over a 33-year period in a Brazilian city. Methods Four population-based birth cohort studies were carried out in the city of Pelotas in 1982, 1993, 2004 and 2015, with samples ranging from 4231 to 5914 liveborn children. Low birthweight (LBW) was defined as <2500 g, and preterm birth as less than 37 weeks of gestation. Information was collected on family income, maternal skin colour and other risk factors for low birthweight. Multivariable linear regression was used to estimate the contribution of risk factors to time trends in birthweight. Results Preterm births increased from 5.8% (1982) to 13.8% (2015), and LBW prevalence increased from 9.0% to 10.1%, being higher for boys and for children born to mothers with low income and brown or black skin colour. Mean birthweight remained stable, around 3200 g, but increased from 3058 to 3146 g in the poorest quintile and decreased from 3307 to 3227 g in the richest quintile. After adjustment for risk factors for LBW, mean birthweight was estimated to have declined by 160 g over 1982–2015 (reductions of 103 g in the poorest and 213 g in the richest quintiles). Conclusions Data from four birth cohorts show that preterm births increased markedly. Mean birthweights remained stable over a 33-year period. Increased prevalence of preterm and early term births, associated with high levels of obstetric interventions, has offset the expected improvements due to reduction in risk factors for low birthweight.


Introduction
Preterm birth and low birthweight are major risks factor for neonatal, infant and under-five mortality. Globally, 16% of all children are born with low birthweight (LBW, <2500 g), 1 and 11.1% are born preterm (less than 37 completed weeks of gestation). 2 In Brazil, data from the National Live Births System (SINASC) which covers 96% of all births in the country, show a prevalence of preterm birth of 11.4% in 2016. 3 Although several countries fail to report on birthweight, it is estimated that 9% of Latin American newborns fall in this category. 1 LBW may result from preterm delivery, intrauterine growth restriction or a combination of both. A systematic review found that each of these conditions contributes to approximately half of all LBW babies. 4 In 2010, preterm births (PTB) were estimated to account for 11% of all live births in the world. 2 PTB complications are the leading cause of death among children under 5 years of age, being responsible for approximately 1 million deaths in 2015. 5 PTB is also associated with long-term consequences including cerebral palsy, sensory deficits, learning disabilities and respiratory illnesses, compared with term birth. 2 Fetal or intrauterine growth restriction (IUGR, defined as a birthweight below the 10th centile of a reference population) was estimated to affect 19.3% of all babies born in low-and middle-income countries in 2012. 6 IUGR is associated with a near 2-fold increase in the risk of neonatal and post neonatal mortality among term infants; the risk is markedly higher for newborns who present both IURG and PTB. 7 IUGR is also a major contributor to child undernutrition and to poor psychomotor development. 7 In 2008, we reported on trends in birthweight and preterm births in three population-based birth cohorts in the city of Pelotas in 1982, 1993 and 2004. 8 We now extend this time-series to incorporate the 2015 birth cohort. Our objectives were to report on time trends in preterm births, LBW and mean birthweight, and to assess inequalities in birthweight according to family income, maternal skin colour and sex of the child.

Methods
Over the course of the years of 1982, 1993, 2004 and 2015, all hospital births in the city of Pelotas were identified through daily visits to all maternity wards, and mothers who lived in the urban area were invited to join the studies. Those who accepted were interviewed by the study team using a structured questionnaire, and anthropometric data were obtained from the women and their newborns. Methodological details of each cohort (1982, 1993, 2004 and 2015) are provided elsewhere. [9][10][11][12] Newborns were weighed within 24 h of birth, using paediatric scales with a precision of 10 g, in each participating hospital. LBW was defined as a birthweight below 2 500 g. 13 The method of assessment of gestational age changed over time. In 1982 and 1993 it was based on the date of the last menstrual period (LMP) provided by the mother, and unknown and unreliable cases were excluded. These represented 21.0% and 10.2% of all newborns, respectively. In 2004 and 2015, we adopted the best obstetric estimate based primarily on first-or second-trimester ultrasound when available. When ultrasound was not available, the estimate was based on the LMP. For the 2004 and 2015 cohorts, we also present estimates based solely on the LMP for comparability with the earlier cohorts.
Mothers were interviewed during the hospital stay and provided information on monthly family income, calculated from the sum of incomes of family members, and divided into quintiles. As noted in a previous publication, 14 income data for 1993 are less reliable than for the other cohorts, due to hyperinflation during that year. Information was also collected on maternal skin colour, categorized as white, black or other by the interviewer, except in 2015 when it was self-reported; in 1982, only two categories (white or other) were recorded.
For family income expressed in quintiles, the slope index of inequality (SII) and concentration index (CIX) were calculated to assess absolute and relative inequalities. The SII can be interpreted as the difference, in percentage points, between prevalence at the top and at the bottom of the income scale; it ranges from -100% to 100% points. The concentration index is a measure similar to the Gini coefficient; a value of zero indicates perfect equality, whereas negative values indicate higher prevalence of the outcome among the poor. 17,18 All analyses were restricted to live, singleton newborns. 13 The chi-square test was used to compare the distribution of maternal characteristics in the four cohorts and trends over time; when applicable the chi-square statistic for linear trend was also calculated. Multivariable linear regression was used to estimate the magnitude of the changes in mean birthweight over time, taking into account changes in the risk factors listed above. All analyses were performed using the Stata 13.1 software. 19 Ethical approval for studies was not required in Brazil until 1996. The 2004 and 2015 studies were each approved by the Ethics Committee of the School of Medicine and School of Physical Education, Federal University of Pelotas, and written consent was obtained from the mothers.

Results
The numbers of liveborn singletons in the four cohorts were 5816, 5168, 4147 and 4164. Birthweight information was missing for 5 (0.09%), 17 (0.33%), 1 (0.02%) and 13 (0.31%) children, respectively. Table 1 shows the distribution of birthweight and gestational age in the four cohorts. Preterm births increased markedly from 5.8% in 1982 to 13.8% in 2015; the prevalence of early term births (37-38 weeks' gestation) also increased from 21.9% in 1982 to 37.5% in 2015, Conversely, full-term births (39-41 weeks) decreased substantially in the period, from 62.1% to 47.9%. For consistency, we reanalysed the data from the four cohorts using only the date of the LMP to assess gestational age, instead of also using ultrasound results in the two more recent cohorts. The resulting prevalences of preterm birth were 5.8%, 11.2%, 16.2% and 17.7% in 1982, 1993, 2004 and 2015, respectively. The corresponding prevalences of early term births were 21.9%, 19.7%, 27.6% and 33.0%.
The distribution of birthweights was remarkably similar in the four cohorts, as was the case for the mean and standard deviation values. The proportion of newborns with low birthweight remained stable at around 8% to 9% throughout the period (P for trend ¼ 0.80). Mean birthweight also remained stable at around 3.2 kg (P for trend ¼ 0.47) As the sources of information on gestational age varied over time, with many missing cases in 1982 and 1993, analyses of socioeconomic and skin colour inequalities are only presented for birthweight. Table 2 shows LBW prevalence according to sex, income and maternal skin colour. As expected due to the use of the same absolute cutoff for both sexes, prevalence was higher among girls than boys in the four cohorts. There was no evidence of a time trend in low birthweight for either sex. In all cohorts, LBW was most frequent in the poorest income quintile and lowest in the richest quintile, except for 2004 when prevalence in the third quintile was slightly lower than in the richest quintile. Regarding maternal skin colour, prevalence was lowest for children born to white mothers, except in 2015 when there was no statistical evidence of a difference.
The magnitude of absolute income-related inequalities was summarized by the slope index, which was equal to -8. Similar analyses for mean birthweight are shown in Table 3. Boys were about 100 g heavier than girls in all cohorts. Mean birthweight increased by 88 g among the poorest quintile (P for trend 0.004) and decreased by 57 g (P ¼ 0.049) in the fourth and by 80 g (P < 0.001) in the richest quintile. The mean difference between the richest and poorest quintile fell from 249 g in 1982 to 81 g in 2015. Regarding maternal skin colour, mean birthweight remained stable for children born to white mothers, but increased by about 60 g (P ¼ 0.01) for those born to black and brown-skinned women.
Because there were important changes in the distribution of risk and protective factors associated with birthweight (Table 4), we used multivariable linear regression to estimate the likely changes in birthweight over time, had the distribution of these factors remained constant (Table 5). In the first analyses, we adjusted for all factors in Table 4 except for diabetes and hypertension. Whereas in the unadjusted analyses mean birthweight in 2015 was only 4 g lower than in 1982, after adjustment for changes in risk factors, mean birthweight became 160g lower in 2015 than in 1982. The effects of adjustment were larger for the richest (-213 g) than for the poorest quintile (-103 g). Further adjustment for reports of diabetes and hypertension changed the overall estimate of the reduction in birthweight over time from 160 g to 153 g [95% confidence interval (CI) 128-177]. The test for linear trend according to maternal skin colour compares white-skinned mothers against black-or brown-skinned mothers, given that in 1982 the information was collected for two categories.

Discussion
Our results show that preterm and early term births increased markedly over a 33-year period in the Brazilian city of Pelotas. For gestational age, changes over time are affected by differences in the methods of assessment of gestational age, and the high proportion of missing cases in 1982 and 1993 when only the date of the last menstrual period was used. Missing data were more common among less educated and poorer mothers, and thus the prevalence of preterm births was likely underestimated in the early cohorts. With increased use of ultrasound during pregnancy, our methods of gestational age assessment changed in the 2004 and 2015 cohorts. In the latter, 84.4% of all newborns had ultrasound results during the first or second trimester recorded in their mother's antenatal cards, and the prevalence of preterm birth for these gestations was 13.8%. During the postpartum interview, 90.7% of the mothers provided information on the date of the last menstrual period, and the prevalence of preterm birth was 18.9%. Even if prevalence for gestations with missing data was twice as high for gestations with existing information as those with data, the corrected prevalence would be 7.0% in 1982 and 12.4% in 1993, which are lower values than for the more recent cohorts. The rise in preterm births was even more marked when we analysed the four cohorts using LMP data. Therefore, in spite of different methods being used in each study, there is strong evidence of an increase in preterm deliveries in Pelotas.
The increased preterm prevalence had been described in our earlier publication 8 and is consistent with other Brazilian studies. A systematic review of peer-reviewed literature showed rising trends from the 1990s onwards, particularly in the southeastern and southern regions where Pelotas is located. 20 In the Ribeirão Preto cohorts, preterm births increased from 7.5% in 1978-79 to 12.8% in 2004. 21 Overmedicalization of childbirth, and in particular the sharp increase in cesarean sections, have been blamed for the current epidemic of preterm deliveries in Brazil. [22][23][24][25]  The test for linear trend according to maternal skin colour compares white-skinned mothers against black-or brown-skinned mothers, given that in 1982 the information was collected for two categories. Due to changes in diagnostic criteria and in data collection methods, and to lower number of antenatal care visits for diagnosis of these conditions, the prevalence of diabetes and hypertension in the earlier cohorts was likely underestimated.
In Pelotas, cesarean sections increased from 27.7% to 65.1% of all deliveries from 1982 to 2015, when it accounted for 86.6% of all births in the richest quintile. 26 Many caesarean sections are scheduled, typically when gestational ages are estimated to reach 38 weeks. Each week of gestational age from 37 to 40 weeks is associated with average gains of 150 g for girls and of 170 g for boys, so that even minor shifts could lead to the effect observed among wealthy women. 27 One important finding of our study is that-for the whole population-mean birthweight and LBW prevalence remained stable. The only other Brazilian study spanning several decades was carried out in the southeastern city of Ribeirão Preto and showed an increase in LBW from 7.2% to 10.7% from 1978-79 to 1994. 21 More recent time-series based on the National Live Birth Registration System showed a slight increase in national low birthweight prevalence from 7.9% in 1995 to 8.4% in 2015. 24 It is paradoxical to find that the highest prevalences of low birthweight are found in the most developed regions of the country (the south and southeast), compared with the poorer regions (the northeast and north). It is postulated that this paradox is explained by excessive cesarean sections in the richest areas. 28,29 Our results are compatible with other Brazilian studies. Prevalence of low birthweight in Latin America is estimated at 9%, with 7% in Argentina and 8% in Uruguay, countries that are closest to Pelotas. 1 Brazilian studies on LBW suggest that the main risk factors include low family income, low education, black or brown skin colour, young maternal age, short stature, low pre-pregnancy weight, primiparity, short birth intervals, lack of prenatal care and maternal smoking during pregnancy. [30][31][32][33] Nearly all of these risk factors evolved favourably in Pelotas from 1982 to 2015 (Table 4). [14][15][16]26 Women became more educated, taller and less likely to smoke or to present with underweight (low body mass index). There are now fewer adolescent mothers, parity is lower and birth intervals longer. The number of antenatal care visits increased substantially. In light of all such changes, one would expect the prevalence of low birthweight to be reduced, and mean birthweight to increase. When we accounted for changes in risk factors over time, mean birthweight in 2015 became 160 g lower than in 1982. The difference was more marked for women in the richest quintile (213g) than in the poorest quintile (103 g). We also ran a model with the above-mentioned covariates plus maternal reports of diabetes and hypertension, for which the time-series (Table 4) was likely affected by changes in diagnostic criteria 34 and by increased use of antenatal care with greater opportunity for diagnosis. As in the previous analyses, the mean birthweight in 2015 was substantially lower than would be expected (152 g on average) given the changes in risk and protective factors over time.
The findings from this regression analysis support the hypothesis that obstetric interventions, which increased over time and are particularly frequent among high-income women, may explain why birthweights failed to increase in accordance with the reduction in the prevalence of risk factors. A simple comparison of birthweights in vaginal and cesarean deliveries is not useful, because the latter include a mixture of procedures with medical indications (associated with lower birthweights due to morbidity) and those without medical indications (which would primarily affect women of high socioeconomic position whose newborns should present higher birthweights). We do not have comparable data on reasons for caesarean sections in the four cohorts, and in addition there are indications that obstetricians may often report medical indications for purely elective procedures; for these reasons, it was not possible to separate these two categories of caesarean sections and assess their specific associations with birthweights. Time trends in low birthweight prevalence according to income groups were not as clear-cut as for mean birthweight, which suggests that the main impact of caesarean sections has been on babies born weighing 2500 g or more. This is consistent with the marked increase in the prevalence of early term deliveries at 37-38 weeks, from 22.3% in 1982 to 37.1% in 2015, which must have had a negative impact on mean birthweight. The hypothesis is also supported by the actual decline in mean birthweight among children born to wealthy women (Table 3), accompanied by an increase among those born to poor women, among whom the prevalence of caesarean sections is much lower.
During the 33-year period covered by our cohorts, there have been substantial improvements in maternal and child health in Brazil as a whole, 25 which are reflected in the data from Pelotas. These positive changes, however, were not reflected in the distribution of birthweights, which remained stable in spite of marked reductions in the prevalence of its main known risk factors. The extremely high rates of caesarean sections may be held accountable for the lack of progress.