Commentary: Breastfeeding and child health, growth, and survival.

Breastfeeding is associated with reduced risks of gastrointestinal and intestinal infections, and of mortality due to those infections. 1,2 The protective effect is stronger against gastrointestinal infection than against respiratory infection, and in developing than in developed countries for both types of infection, and increases with the duration and exclusivity of breastfeeding. The magnitude of the protective effect wanes with age: highest in the first 3–6 months and diminishing thereafter when complementary foods are introduced in addition to breast milk, but continuing into the second year of life. 1,2 The introduction of complementary foods and complete weaning (the cessation of any breastfeeding) are gradual and complex processes, however, which are affected by many biological, behavioural, and cultural factors. 3 Because of the practical and ethical difficulties inherent in randomizing women and infants to different infant feeding modes, most epidemiological studies have relied on observational designs. Such studies are obliged to consider, and attempt to control for, selection bias, confounding, and reverse causality. Most perinatal and paediatric epidemiologists recognize the importance of control for confounding by such factors as age (young infants are at higher risk of mortality and morbidity than older children), socioeconomic status, and toilet and water facilities, particularly in developing country settings. But confounding by indication is both more insidious and more difficult to control. 4 This source of bias arises when the reason (i.e. the indication) for weaning is strongly associated with the outcome. When that reason is an early manifestation of the outcome itself, the bias is often referred to as reverse causality bias. For example, early signs and symptoms of an infectious illness can lead to supplementation or even complete weaning. Several epidemiological studies of breastfeeding and mortality have attempted to minimize or eliminate this source of bias by assessing the type of feeding that the infant or child received prior to the onset of illness. 2,5–7 The study reported by Jakobsen et al. in this issue of the International Journal of Epidemiology 8 considers confounding by indication not only according to whether the child's illness was the reason for weaning, but also other confounding reasons such as the mother's illness and a subsequent pregnancy. It is not surprising that when maternal illness was the reason for weaning , child mortality risk was higher. An ill mother is less able to feed and care for her child, especially when the child also becomes ill, irrespective …