While various studies document a positive association between duration of breastfeeding and the child’s intelligence, there is an ongoing discussion about the mechanisms behind that association.1,2 The usual approach to distinguish between nutritional and confounding environment-related effects is to control for observable environmental factors in the analysis. Varying availability and measures of control variables make comparisons across studies difficult and lead to different results concerning the partial correlation between duration of breastfeeding and intelligence.1,2 This study provides a new empirical test of the purely nutritional effect of breastfeeding on the child’s intelligence and does not rely on the availability of control variables.
The literature on the nutritional benefits of breastfeeding3,4 suggests that if the main sources of the association between duration of breastfeeding and intelligence are direct nutritional effects, this association should be monotonic. This proposition should especially hold true for breastfeeding in the first weeks of life.3 Accordingly, children who were breastfed for a short duration would be expected to be more intelligent than children who were not breastfed. For some newborns, breastfeeding is not possible for health reasons and thus they cannot receive the nutritional benefits of breastfeeding, but a stimulating social environment (eg, secure attachment, intellectual stimuli) can still be provided. Therefore, this study proposes to test for nonmonotonicity in the association between the duration of breastfeeding and the child’s intelligence at a duration of 0 (ie, when the newborn is not breastfed). As an empirical illustration of this idea, recent data on the Pelotas 1982 Birth Cohort5 are used. Data on IQ at age 30 were collected from June 4, 2012, to February 28, 2013. Data on breastfeeding were collected in 1984 and 1986. The Ethical Review Board of the Faculty of Medicine of the Federal University of Pelotas approved the study. Data analysis was conducted between April 9 and August 28, 2015. The following model was estimated:
where di is a dummy variable of 1 if the child was not breastfed and 0 if the child was breastfed, xi indicates the individual breastfeeding duration, and M indicates the degree of polynomials used to approximate the association. βbreak captures a potential nonmonotonicity in the association at a duration of 0; it indicates the difference between the actual mean IQ of children who were not breastfed and their predicted IQ based on the monotonic approximations. To precisely capture the shape of the association close to a duration of 0, the sample is restricted to durations smaller than 1 month (736 of 3493 observations [21.1%]) and polynomials up to the degree of 3 are used. Results of ordinary least-squares estimations are displayed in the Figure.
Linear, quadratic, and cubic predictions of the association between duration of breastfeeding and IQ based on ordinary least-squares estimates, as well as the actual mean IQ of individuals who were not breastfed.
The actual mean IQ of individuals who were not breastfed is significantly higher than the predictions based on the 3 monotonic approximations. The estimated βbreak coefficients indicate significant upward shifts for children who were not breastfed (2-sided t tests, n = 736; linear model: P = .03; quadratic model: P = .02; cubic model: P = .08). Point estimates and 95% CIs are displayed in the Table.
The identified pattern of a nonmonotonic association contradicts the interpretation of pure nutritional effects as the main source of the raw correlation between duration of breastfeeding and the child’s intelligence. Therefore, there must also be social environment-related factors associated with breastfeeding from which children who are not breastfed can also benefit. Combining the research strategy of this study with the usual approach of adding control variables to isolate nutritional effects bears the potential for new evidence in the debate on the existence of nutritional effects of breastfeeding on intelligence. If βbreak indicates a statistically significant upward shift in a carefully calibrated estimation model, which also includes all plausible and available environmental controls, there is strong evidence that the resulting partial regression coefficient of duration of breastfeeding is still not reflecting pure nutritional effects.
The pattern shown in this study also helps explain why studies comparing children who have ever been breastfed and those who have never been breastfed find no difference in intelligence.6
Corresponding Author: Fabian Kosse, Dr rer pol, Institute for Applied Microeconomics, University of Bonn, Adenauerallee 24-42, 53113 Bonn, Germany (firstname.lastname@example.org).
Published Online: December 21, 2015. doi:10.1001/jamapediatrics.2015.3201.
Conflict of Interest Disclosures: None reported.
Additional Contributions: Data access was provided by Bernardo Lessa Horta, PhD, Universidade Federal de Pelotas. He was not compensated for his contribution.
Kosse F. The Nutritional and Social Environment-Related Effects of Breastfeeding on Intelligence. JAMA Pediatr. 2016;170(2):173-174. doi:10.1001/jamapediatrics.2015.3201