Rapid Responses to:

LETTERS: Karen M Silvers, Michael J Epton, and Chris M Frampton Study was not designed to test the hypothesis BMJ 2007; 335: 899 [Full text]

Rapid Responses: Submit a response to this article

Rapid Responses published:

Testing hypotheses about the health effects of breast feeding

Michael S. Kramer   (4 November 2007)

Testing hypotheses about the health effects of breast feeding 4 November 2007
Michael S. Kramer, Professor, Scientific Director McGill University Faculty of Medicine, Montreal, Quebec, Canada H3H 1P3 Send response to journal: Re: Testing hypotheses about the health effects of breast feeding	Silvers et al seem confused about several aspects of PROBIT, our cluster-randomised trial of a breast feeding promotion intervention. It is true that the primary outcome investigated in the trial was the risk of gastrointestinal infection in the first year of life. Every RCT has a primary outcome that serves as the basis for estimating sample size requirements and for obtaining funding. That did not prevent my colleaugues and me from examining other outcomes (respiratory infection and atopic eczema) in the first year of life, nor from seeking support for subsequent follow-up of the PROBIT cohort to assess important health outcomes at older ages that observational studies have associated with infant feeding. Randomised allocation of the breast feeding promotion intervention yielded 2 groups that differed substantially in the duration and exclusivity of breast feeding. The fact that those group differences were created by randomisation, rather than by the mothers' choices or the recommendations of the infants' physicians, created a unique scientific opportunity to study longer term child health outcomes. This opportunity was cited in the editorial [1] that accompanied the publication of the trial's results [2] and was one of the main reasons why the Canadian Institutes of Health Research funded the 6.5-year follow-up, for which the major hypotheses concerned effects on growth, adiposity, neurocognitive development, behaviour, and dental caries, as well as asthma and allergy. In our recent BMJ paper [3] reporting on the latter outcomes, my colleagues and I did not infer that "breast feeding has no effect" on those outcomes. We did not compare any breast feedng with no breast feeding. All PROBIT infants were breast fed at birth; the difference between the 2 randomised groups was limited to the duration and exclusivity of breast feeding. Our inference was thus that prolonged and exclusive breast feeding did not protect against asthma and allergy, an inference justified by the randomised design, intention to treat analysis, and observed results. Finally, the wide confidence intervals noted by Silver et al around the cluster adjusted odds ratios for the skin prick test results have absolutely nothing to do with "important confounding and predictor variables." As shown in Table 1 of our BMJ paper [3] and as expected from the randomised allocation, the 2 groups were very similar in baseline characteristics. As we point out in the Results section of our paper and as shown by the high intraclass correlation coefficients in Table 5, the wide confidence inervals are a function of the high degree of clustering for the skin prick test results. Clustering was far less evident for the allergic symptoms and diagnoses (Table 4) and for the sensitivity analysis for the skin prick tests (Table 6), with considerably narrower confidence intervals. References: 1. Lawrence RA. Breastfeeding in Belarus. JAMA 2001;285:463-4. 2. Kramer MS, Chalmers B, Hodnett ED, Sevkovskaya Z, Dzikovich I, Shapiro S, et al. Promotion of breastfeeding intervention trial (PROBIT): a randomized trial in the Republic of Belarus. JAMA 2001;328:702-8. 3. Kramer MS, Matush L, Vanilovich I, Sevkovskaya Z, et al. Effect of prolonged and exclusive breast feeding on the risk of allergy and asthma: cluster randomised trial. BMJ 2007;335:815-8. Competing interests: None declared