Live Births and Fertility Amid the Zika Epidemic in Brazil

Statements linking ZIKV to CZS, including microcephaly, were widely announced in Brazilian and international media starting in November 2015, as evidenced through various reports (“On Saturday” 2015; “Ministry of Health confirms” 2015; Sims 2015) and as discussed at length elsewhere (Diniz 2016).

The number of Google searches for the word “microcefalia” went from zero to its highest ever in November 2015.

In work in progress, Rangel et al. (forthcoming) have also found evidence of large changes in birth rates in Brazil during the epidemic, using an index of mosquito infestation to leverage the spread of the ZIKV epidemic across Brazil’s microregions.

Abortion is mostly illegal in Brazil, but it is nonetheless frequent (Diniz and Medeiros 2010; Diniz et al. 2017). A 2016 national household survey using ballot box techniques suggests that one in every five women will have had an abortion by age 40, with abortion more commonly occurring among young women up to age 24 (Diniz et al. 2017). The only data available on abortion in Brazil are hospital admittances owing to abortion complications; to estimate abortion, it is necessary to make assumptions that the characteristics of the women seeking hospitalization after the procedure did not change with the Zika epidemic (Singh and Wulf 1991).

Projections of women at childbearing ages by educational levels are not widely available for Brazil, so we apply the educational structure from the nationally representative PNAD (1995 to 1999, 2001 to 2009, and 2011 to 2016) to projections to estimate the number of women by levels of education (IBGE 2019). We interpolate monthly data from the yearly trend using a nonstochastic exponential interpolation method. Then we calculate the proportions for each interpolated number of women at each five-year age group by month-year and education level and use these proportions to obtain the expected numbers using the Cedeplar totals.

Brazil’s Census Bureau does not make municipal projections available. Thus, we use projections estimated by Cedeplar based on 2010 census data. The overall number of women at childbearing ages is estimated using a combination of small-area estimation procedures, multiregional migration tables, and other standard projection techniques adjusted by expert opinion on the future behavior of each demographic component (Cedeplar 2014).

We eliminate five municipalities out of the 5,570 total number of Brazilian municipalities because they were created after 2010 and therefore do not have corresponding population projections data estimated using the 2010 census. We aggregate localities in the outskirts of Brasília as a single entity; this is usual practice regarding the federal district.

Administrative reports and vital statistics information on live births in developing countries typically face data quality issues, including delay in registration, misreporting, and undercounting. For 2014, undercount at the national level was 4% (Ministério da Saúde 2017).

Live birth data sets were transferred directly from the SINASC website as updated information became available (May 5, 2017; July 29, 2017; March 20, 2018; and April 28, 2018, when 2016 data became final). Finally, on March 21, 2019, we retrieved the whole data set once again.

We also estimate fertility rates using as the denominator projections by the Brazilian Census Bureau (IBGE 2013) to assure that results held regardless of the denominator used (not shown but available upon request). We estimate the observed fertility rates using direct methods and the expected fertility rates by indirect methods, using the census.

We use maternal education as a measure of SES for conceptual and methodological reasons. Conceptually, maternal education is the strongest socioeconomic predictor of live births and fertility in Brazil (Cavenaghi and Berquó 2014; Rios-Neto et al. 2018); it is a more stable measure than income and has been used extensively as the major proxy for SES in Brazil (Lam and Duryea 1999; Marteleto and Dondero 2013; Potter et al. 2010). Methodologically, the data set we use offers two variables that reflect SES: maternal education and maternal occupation. Information on income is not collected. Information on maternal occupation is collected but not made available for public use. Maternal education is thus the only proxy available to us. Fortunately, it is missing in only 1.3% to 2.3% of cases, depending on the year.

The GFR is the ratio between the number of live births and the number of person-years lived by females aged 15–49 years (Preston et al. 2001). Because the GFR depends on the proportion of person-years lived by females aged 15–49 years, the rates are sensible to compositional differences in the age structure of reproductive-aged women. As such, the GFR could potentially change over time simply because of changes in the proportion of women at childbearing age, not because of actual changes in birth patterns. This is an undesirable property of the GFR and the main reason it is seldom used. We estimate GFRs as a counterfactual exercise to predict the expected number of births to highlight the very effect that a changing age structure of women at childbearing ages would have been on the number of live births. We use the GFR’s limitation—its sensitivity to the age structure of women ages 15–49—as a resource to compare actual with expected live births. This is a way to tease out whether the change in the number of live births we document could have been the result of a change in age structure. Because the counterfactuals rely on the number of women at childbearing ages from 2010, prior to the ZIKV epidemic, our exercise shows the contribution of changes in age structure only to the GFR. The difference in estimation and the sign of the difference (positive or negative) provide a sense of the magnitude of live births that would have increased/decreased as a result of the aging structure of women at childbearing ages.

This decline in the observed number of live births we examine is not due to compositional effects in the number of females at childbearing ages, suggesting that we cannot infer that the observed decline and the timing of decline in births are attributable to population composition effects.

Fertility levels are slightly similar across educational groups for older age groups in Brazil (Camarano 2014), but fertility might be qualitatively different for low- versus high-educated according to birth order, as the pattern of first births is highly dependent on mother’s education; high-educated women are having the highest first birth intensity around age 32 while low-educated women have their highest first birth intensity around ages 18–21 (Lima et al. 2018).