| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Departments of Pediatrics and of Epidemiology and Biostatistics, McGill University Health Centre Research Institute and McGill University Faculty of Medicine, Montreal, PQ, Canada, E-mail: michael.kramer{at}mcgill.ca
Dear Sir:
We thank Rückinger and von Kries for their reflections on statistical power in studies of breastfeeding and subsequent childhood obesity and, more particularly, on the evidence from the Promotion of Breastfeeding Intervention Trial (PROBIT), our cluster-randomized trial of a breastfeeding promotion intervention in the Republic of Belarus. Rückinger and von Kries argue that because of the overlap in breastfeeding behavior between our experimental and control groups, even a study as large as ours (13,889 children followed up at age 6.5 y), statistical power is insufficient to detect protective effects of breastfeeding against subsequent obesity.
Unfortunately, Rückinger and von Kries have ignored 2 key issues and seem unaware of many of our other publications from PROBIT. First, breastfeeding is not an all-or-none behavior. Rückinger's and von Kries's analysis compares breastfed with nonbreastfed children, whereas the PROBIT intervention was designed to increase both the duration and exclusivity of breastfeeding among children who were initially breastfed. Thus, none of the children in PROBIT were formula-fed from birth. If breastfeeding is protective, more exclusive breastfeeding should be more protective. Second, breastfeeding trajectories (ie, entire feeding histories since birth) are important, not just the cross-sectional proportion of children breastfed at a particular age during infancy.
In designing PROBIT, we were fully cognizant that any effect of breastfeeding would be attenuated because of the overlap in breastfeeding behavior in our experimental and control groups. This is precisely why we required a large sample size. That an intention-to-treat analysis with an appropriately randomized intervention can indeed detect causal effects of the experimental intervention is clearly shown by the substantial and statistically significant effects we have reported on gastrointestinal infection, atopic eczema, and growth in infancy (1, 2) and on cognitive development at the age of 6.5 y (3). The absence of any observed effect of the intervention on mean body mass index (BMI; in kg/m2) or risk of obesity cannot, therefore, be attributed to insufficient statistical power. Although we agree that "absence of evidence does not imply evidence of absence," we observed no trend toward reduced mean BMI in the experimental group (15.6 in both groups), and the narrow CI around the cluster-adjusted difference [+0.1 (95% CI: –0.2 to +0.3)] gives no hint of such an effect.
We must therefore conclude that the overall difference in breastfeeding we achieved in our experimental group compared with the control group was sufficient to cause detectable effects on a number of important health outcomes, yet insufficient to cause differences in mean BMI or risk of obesity. In other words, the effect of increased exclusivity and duration of breastfeeding must be smaller for BMI and obesity outcomes than for those for which statistically significant effects were observed.
Rückinger and von Kries mention the systematic review and meta-analysis by Owen et al (4). They cite a reported difference in BMI between subjects who had been breastfed and formula-fed of –0.19; in fact, however, Owen et al's individual-subject data meta-analysis of the 11 observational studies, in which they were able to control for confounding by socioeconomic status, maternal BMI, and maternal smoking during pregnancy, yielded a nonsignificant difference of –0.01 (95% CI: –0.05 to +0.03) (4). Thus, when confounding was adequately controlled for those 11 studies, no effect of breastfeeding was observed.
In summary, despite our detection of statistically significant and clinically important effects on several other outcomes, both during infancy and at 6.5 y of age, we observed no such effect on mean BMI or obesity risk among PROBIT children. This result is consistent with an individual-subject data meta-analysis of observational studies. The problem does not seem to be a lack of statistical power but rather a lack of causal effect.
ACKNOWLEDGMENTS
No conflicts of interest were reported.
REFERENCES
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
