Menu
Menu
Menu
Menu
Background: Because measles vaccination prevents acute measles disease and morbidities secondary to measles, such as undernutrition, blindness, and brain damage, the vaccination may also lead to higher educational attainment. However, there has been little evidence to support this hypothesis at the population level. In this study, we estimate the causal effect of childhood measles vaccination on educational attainment among children born between 1995 and 2000 in South Africa. Methods and findings: We use longitudinal data on measles vaccination status and school grade attainment among 4783 children. The data were collected by the Wellcome Trust Africa Centre Demographic Information System (ACDIS), which is one of Africa’s largest health and demographic surveillance systems. ACDIS is located in a poor, predominantly rural, Zulu-speaking community in KwaZulu-Natal, South Africa. Using mother fixed-effects regression, we compare the school grade attainment of siblings who are discordant in their measles vaccination status but share the same mother and household. This fixed-effects approach controls for confounding due to both observed and unobserved factors that do not vary between siblings, including sibling-invariant mother and household characteristics such as attitudes toward risk, conscientiousness, and aspirations for children. We further control for a range of potential confounders that vary between siblings, such as sex of the child, year of birth, mother’s age at child’s birth, and birth order. We find that measles vaccination on average increases school grade attainment by 0.188 grades (95% confidence interval, 0.0424-0.334; p=0.011). Conclusions: Measles vaccination increased educational attainment in this poor, largely rural community in South Africa. For every five to seven children vaccinated against measles, one additional school grade was gained. The presence of a measles vaccination effect in this community is plausible because (i) measles vaccination prevents measles complications including blindness, brain damage, and undernutrition; (ii) a large number of number of children were at risk of contracting measles because of the comparatively low measles vaccination coverage; and (iii) significant measles transmission occurred in the community where this study took place during the study observation period. Our results demonstrate for the first time that measles vaccination affects human development not only through its health effects but also through its effects on education.
We used longitudinal data from a health and demographic surveillance system (HDSS) in rural KwaZulu-Natal, South Africa, that was established and is maintained by the Wellcome Trust-funded Africa Centre for Health and Population Studies. The HDSS started in 2000 and covers a demographic surveillance area (DSA) of 438 square kilometers near the market town of Mtubatuba in the predominantly rural Umkhanyakude district of KwaZulu-Natal. The surveillance system covers the entire population of about 85,000 Zulu-speaking people who are members of the 11,000 households in the DSA. Most households are multi-generational, and average household size is 7.9 (SD = 4.7) members. Although this is a predominantly rural area, the principal source of income for most households is waged employment and state pensions rather than agriculture. In 2006, approximately 77% of households in the surveillance area had access to piped water and toilet facilities [34]. Due to the availability of antiretroviral therapy in South Africa’s public-sector health system starting in 2004, adult life expectancy in this community increased from about 49 years in 2003 to 61 years in 2011 [35]. Data on all births in the year of a household’s first interview in the surveillance as well as the previous five years were elicited from all women residing in the DSA. For each child, childhood vaccination data were elicited. We measured the outcome, school grade attainment, up to the year 2007. Our sample for this study consists of all children who were born between 1995 and 2000 and were members of households residing in the DSA in 2007. 1995 was the first year that childhood vaccination data became available in the HDSS; the year 2000 cutoff ensures that every child had the chance to complete at least one year of school by 2007 in longitudinal follow-up. In the sample for complete-case analysis, the total number of children was 4783 and the total number of mothers was 4080. In the sample for multiple-imputation analysis, the total number of children was 7509 and the total number of mothers was 6148. Even though the main effect estimate in our fixed-effects models is based only on the comparison of children who share the same mother but differ in their measles vaccination status (607 in the complete-case analysis and 1031 in the multiple-imputation analysis), we kept all other children in the sample for analysis, because these observations contribute to the estimation of the regression constant and the R2 statistic without affecting the size or significance of the measles effect estimate. The surveillance questionnaires and descriptions of the data sets are available on the website of the Africa Centre for Health and Population Studies (http://www.africacentre.ac.za). Our exposure variable is measles vaccination status at 12 months of age. A child was coded as either vaccinated or unvaccinated for measles by 12 months of age. We coded a child as having received his or her measles dose by 12 months of age if at least one of the following two conditions was met: first, the national vaccination card (the so-called Road-to-Health card) was the data source and the date of vaccination dose was within one year of birth or, second, mother’s report was the data source and indicated that the child had received the vaccination within one year of birth. Mother’s report of her children’s vaccination status has been validated in this community by Ndirangu et al. [36]. If vaccination card information and mother’s report were both available, we used the card information. This approach to coding vaccination data is the same that is used in many other population-based surveys, such as the Demographic and Health Surveys (DHS) [37]. Children with missing card information and missing mother’s report and children with missing covariate information were excluded in the complete-case analyses. To test the robustness of our findings to missing observations, we multiply imputed vaccination status and other missing data and repeated the analyses with the imputed datasets [38]. The sample size for the main, complete-case analysis was 4783; the sample size for the analysis of the multiply imputed data was 7509. To capture educational attainment, we used the highest school grade that a child had attained at the last HDSS household interview up to the year 2007. Not all children were eligible for outcome measurement in 2007 (e.g., because their families had out-migrated). To ensure the comparability of school grade outcomes between children who were born in the same year but had their school grade measured in different years, we controlled for a child’s age at start of the school year in which the household interview was conducted. Also, to ensure the comparability of schooling outcomes between children who were born in different calendar years (and who would therefore be expected to have different levels of school grade attainment in later calendar years), we controlled for year of birth. We excluded a small number of children (49, or 1.0% of the complete-case analysis sample) who had implausible reported grade levels. We defined “implausible reported grade level” as three or more grades ahead of the grade that a child would have attained had she started school (grade 1) at age seven and advanced by one grade per year. (According to the South African Schools Act of 1996 [39], [40], children must start school no later than the calendar year in which they turn seven years old.) At the time of the last measurement of the outcome in this study (school grade attainment), the children were aged six to eleven years. We estimate the effect of childhood measles vaccination on educational attainment using mother fixed-effects analysis. The mother fixed effects control for all observed and unobserved factors that are shared by siblings, including mother and household characteristics that do not vary between siblings, such as risk attitudes, conscientiousness, and aspirations for children’s futures. We also control for a number of factors that can vary between siblings: sex, age at start of the school year in which the household interview was conducted, calendar year of birth, mother’s age at child’s birth, and birth order. Finally, in separate analyses, we additionally control for the number of doses of diphtheria–tetanus–pertussis vaccine (DTP) received. DTP is often used as a proxy for immunization system performance [41]. Here, DTP coverage serves as a powerful control variable to account for any potential sibling-varying confounding that is related to differences between siblings in access to vaccinations in general that are not captured by the other sibling-varying control variables. These sibling-varying confounding factors include differential availability of vaccination between siblings that is not already captured by the calendar-year control variables (e.g., when a family moves their home closer to a vaccination clinic and at the same time moves closer to the nearest school). They also include changes in maternal and paternal knowledge, attitudes, and behaviors that can affect vaccination and school grade attainment. The mother fixed-effects regression has the form: where Yim is the school grade of child i with mother m. Vim is child i’s measles vaccination status at 12 months of age. β is the main parameter of interest in this study: the conditional association between childhood measles vaccination and school grade attainment. Xim is a vector of child i’s characteristics, and Xm is mother m’s age at the time of child i’s birth. μm is the mother fixed effect and ɛim is the error term. We performed four regression analyses: complete-case analyses with and without the DTP covariates and analyses using multiple imputation of missing data, again both with and without DTP covariates. Some data were missing for four of the variables we use in the regression analyses: school grade attainment (missing for 69 of 7509 observations), measles vaccination (missing for 100 observations), DTP vaccination (missing for 2600 observations), and birth order (missing for 117 observations). We carried out 40 imputations in the multiple imputation, which exceeds the commonly recommended minimum numbers of imputations [42] but is unproblematic given today’s computing power. In all models, we clustered heteroskedasticity-robust standard errors at the level of the mother to account for correlation in outcomes among children who share the same mother. Analyses were conducted using Stata version 11 (StataCorp LP, College Station, TX).
N/A
