Home › Recommendations for Infant Feeding Research

Recommendations for Infant Feeding Research

in

Related Materials

application/pdf icon
Criteria for Research on Infant Feeding

 

Suggested Criteria for Science-Based Research on Infant Feeding and Health Outcomes

 
Executive Summary:
Research on infant feeding and health outcomes is challenging for many reasons. Prospective, randomized, blinded clinical trials, the gold standard for epidemiologic research, cannot be conducted without compromising a mother’s right to make a choice relative to infant feeding. Therefore, other research models such as randomization of breastfeeding interventions, cohort studies, sibling pair studies, and/or surveys must be used. However, due to the limited utility of these types of studies, one must be careful not to over-interpreting the findings. For example, studies designed to show associations cannot be assumed to show cause and effect. Additionally, since there currently are no standard definitions of the terms frequently used in breastfeeding research, comparing different studies via meta-analyses, as in other research fields, the data can be neither meaningful nor interpretable. Following is a description of the pros and cons of various types of research study designs appropriate for research on infant feeding and health outcomes.
 
Background:
Research on infant feeding and health outcomes is challenging for many reasons. First, it is very difficult, if not impossible, to design a comparative nutrition study that can rise to the standards of those considered adequate for other areas of medical research such as in the pharmaceutical industry. For example, there are few nutrients that can be studied in isolation as is common in placebo controlled pharmaceutical studies.
 
Second, there are few distinctive nutritional endpoints that do not have a number of confounding variables. When considering research on infant feeding regimens, the choice of the family to breast feed or to bottle feed, in itself, identifies and introduces the bias that these, by nature of their choice, are different populations. Additionally, with long-term outcome studies, it must be recognized that both human milk and the infant formulas currently available have different composition than what would have been fed to a study population, as several years and sometimes decades are required to evaluate later life outcomes. Important factors such as genetics, environment, cultural, familial variables, and a myriad of other factors are confounding variables that cannot be eliminated by study design. In addition, the “adjustment” for these confounding variables in research designs is often insufficient and incomplete, so, the bias between the feeding groups may persist, in spite of the statistical methods employed.
 
Third, there currently are no standard definitions of the terms frequently used in breastfeeding research. For example, breastfeeding may mean “exclusive” breastfeeding (no formula but water allowed, or no other substance allowed except breast milk), “predominant” breastfeeding, supplemented with formula or other food, or “any” breastfeeding. The development of a standard set of definitions must be created and consistently used by all researchers for infant feeding studies. This would allow for more accurate comparisons of different studies to be made and combined into meta-analyses.
 
Fourth, prospective, randomized, blinded clinical trials, the gold standard for epidemiologic research, cannot be conducted without compromising a mother’s right to make a choice in infant feeding. Other research models can be used. Prospective cohort studies offer benefits, such as allowing relative risk to be determined. Surveys have benefits as well, such as the opportunity for long term follow up. Keeping surveys current and revising with relevant measures helps provide the most useful information. Finally, meta-analyses allow many individual studies to be grouped and trends to be determined, but identical definitions of variables of interest are critical for conclusions to be meaningful. The following are general criteria for consideration in designing research protocols for infant feeding studies:
 
Clinical Trials:
 
Prospective, randomized double-blinded, placebo controlled clinical trials that are the mainstay of the pharmaceutical industry are impossible to conduct in an infant feeding population without compromising the mother’s right to choice of feeding modality. Further, such trials would suffer from the complications of having to bottle feed expressed breast milk to mask the feeding method. Some methodologies have been developed to overcome these difficulties. For example, randomizing breastfeeding promotion, as done by Dr. Michael Kramer in Belarus, is one way to examine the effect of breastfeeding on health outcomes, without interfering with a woman’s choice as how to feed her baby. It is key, however, that the data be interpreted and conclusions be appropriately made, based on the study design used.
 
Cohort Studies:
 
Due to the impossibility of conducting a prospective, randomized, and blinded trial design, the preferred remaining option is a prospective cohort study. Prospective cohort studies are not able to definitively establish cause and effect relationships, but can provide useful associative information if designed to answer specific questions prospectively and the analysis is carried out without identification of the feeding group. One obvious disadvantage of cohort studies is cost, as they require large homogeneous population groups. Prospective cohort studies also allow relative risk to be determined; however in terms of relative risk related to infant feeding, other factors may have a far greater impact on disease. 
 
Specific issues related to cohort studies exist, which should be considered by researchers before undertaking such studies on infant feeding. Cohort studies would require a complete and specific definition of “breast feeding” (e.g., acceptable duration, definition of “exclusive,” etc.). Additionally, such studies require an understanding of the nutrient composition of human milk and of the specific infant formula (different formulas have different compositions) consumed throughout the duration of the study. Even if a comprehensive understanding could be reached, a further complication is the fact that the nutrient content of human milk is extremely variable compared to infant formula. Additionally, in cohort studies of infant feeding, daily nutrient intake must be collected (e.g., calories, macronutrient profile, electrolytes) and any ancillary or complementary feedings should be recorded and included in the data analysis. Further, it is critical that the sample size be adequate to provide statistical power after all confounding variables (some briefly described above) have been evaluated. Lastly, the duration of feeding should be for at least the first 12 months of life with follow up to 6 years for long term health outcome data.
 
One promising cohort study design is the sibling pair study design; by comparing infant feeding method among siblings, many maternal factors, which can often confound the relationship between breastfeeding and health outcome, are removed from the analysis.   Examples of studies using sibling pairs are attached (Nelson et al. 2005; Evenhouse et al. 2005: O’Tierney et al. 2009; Rees et al. 2009).
 
US surveys:
 
US surveys can provide very valuable information regarding infant feeding practices and health outcomes. Such surveys include the National Health and Nutrition Examination Survey conducted by the National Center for Health Statistics (NCHS). These surveys could be strengthened if additional endpoints were incorporated to further evaluate health and development outcomes of different feeding modalities. Recommended endpoints to include are age of achievement of gross motor milestones and performance on different measures of intelligence (achievement, verbal, cognitive ability).   Intelligence tests could include the Bayley’s tests of mental and psychomotor development, Peabody individual achievement test (PIAT) for cognitive ability, Raven’s test of non-verbal cognitive ability, and the Wechsler Abbreviated Scales of Intelligence (WASI) for full scale intelligence. These tests have been used in a variety of studies evaluating outcome differences associated with different feeding practices.  
 
Information obtained from on-going US surveys would be most beneficial if these surveys examined outcomes related to actual feeding practices of US mothers. In order to assess this, we recommend including the following categories of infant feeding practices: 1) moms who never breastfeed and 2) those who mix feed (feed breast milk and formula) 3) those who breastfeed and then switch to formula feeding, 4) moms who practice exclusive breastfeeding. This type of information would benefit scientists and policy makers alike, as it would provide a picture of real-world infant feeding patterns and any associated health outcomes.
 
Meta-Analyses:
 
Meta-analyses can provide useful information, if conducted correctly. However, as noted above, problems with some existing meta-analyses of breastfeeding include variables measured often are not defined or measured the same in each study; thus, when not comparing the same variables, the data are neither meaningful nor interpretable. Further, the range of potential outcome variables is limited to what has already been studied. Thus, IFC recommends that standard definitions of breastfeeding be created and used by all researchers for infant feeding studies.
 
General Considerations for Future Infant Feeding and Health Outcomes Research:
 
Infant feeding research is a very broad topic. Including studies that examine different health outcomes and different patterns of infant feeding would be most beneficial to truly understand how women are feeding their infants and any long term associated health outcomes. Studies that collect data oninterventions aimed to improve breastfeeding rates or duration including WIC and workplace interventions would also be beneficial.
 
 
References Cited:
 
1.      Nelson MC, Gordon-Larsen P, Adair LS. Are adolescents who were breast-fed less likely to be overweight? Analyses of sibling pairs to reduce confounding. Epidemiology 2005;16 (2):247-53.
2.      Evenhouse E, Reilly S. Improved estimates of the benefits of breastfeeding using sibling comparisons to reduce selection bias. Health Serv Res 2005;40 (6 Pt 1):1781-802.
3.      O'Tierney PF, Barker DJ, Osmond C, Kajantie E, Eriksson JG. Duration of breast-feeding and adiposity in adult life. J Nutr 2009;139 (2):422S-5S.
4.      Rees DI, and Sabia JJ. The Effect of Breast Feeding on Educational Attainment: Evidence from Sibling Data. J Human Capital 2009; 3(1): 43-72.