In order to understand the lack of consistency in the literature, and to make progress, we need not only to draw on knowledge from epidemiology regarding prenatal factors but also on knowledge from psychology regarding post-natal, childhood-related factors. Within alcohol epidemiology there has been a great focus on the prenatal environment, and an almost complete lack of focus on post-natal conditions. However, it is well known from psychology that childhood-related factors and the home environment have a very substantial impact on child development. The point that I will try to make here is that such psychologically-oriented early childhood factors should be taken into account when interpreting the results of the epidemiological studies investigating prenatal exposure to alcohol in relation to child development. That there is too little focus on the post-natal environment has been recognised in the literature (Sayal 2007). Abel noted that “that a child’s postnatal environment affects his/her behaviour is hardly surprising, but it has not received the attention it deserves in the present context (i.e. in the alcohol literature)” (Abel 1998, p. 127). Further, a comprehensive review concluded that “the two most important types of confounding of effects on neurodevelopmental outcome are failure to control for the postnatal environment and failure to control for factors which are strongly genetically influenced and which may be related to both prenatal alcohol exposure and the outcome” (Gray and Henderson 2007a, p. 20).

These childhood-related factors are known as mediators. A mediator (Fig. 2) can be defined as “a variable that occurs in a causal pathway from a causal (independent) variable to an outcome (dependent) variable. It causes variation in the outcome variable and itself is caused to vary by the original causal variable. Such a variable will be associated with both the causal and the outcome variable” (Porta 2008, p. 131). A wide range of mediating factors could be relevant to control for in the statistical analyses including attachment, the child’s IQ (Streissguth 1996), parental mental health in childhood (Sayal 2007), home environment (Abel 1998; Streissguth 1996), quality of parenting (Sayal 2007), changes in living arrangement (Streissguth 1996), violence (Streissguth 1996) and social support (Abel 1998). These factors are especially important in the first 2 years of life when the child’s brain is particularly plastic (Abel 1998). Therefore, a child’s behaviour and cognitive functioning is a product not only of prenatal factors (such as alcohol) but also of postnatal exposures (such as attachment).

Attachment can be defined as “an affectional tie that one person or animal forms between himself and another specific one—a tie that binds them together in space and endures over time” (Ainsworth and Bell 1970). In the 1950s Bowlby was the first to demonstrate the lasting consequences of the mother-child relationship on a wide range of developmental cognitive and mental health outcomes (Bowlby 1950). He concluded that infants who develop a secure attachment style are those with a history of sensitive and responsive maternal care. This style is associated with better emotional regulation, higher self-esteem, and more develop coping skills. In turn this makes the children better able to handle stressful or challenging situations and conversely lowers their risk for poorer mental health outcomes later in life. Secure attachment is also found to be associated with better academic performance in adolescents and better cognitive performance in childhood (Mikulincer and Shaver 2007a, b). On the other hand, children with an insecure attachment are by contrast at greater risk for poor mental health outcomes (Ainsworth and Bell 1970; Sroufe 2005). Therefore, attachment has a “large” effect on mental health development, whereas exposure to low doses of alcohol all other things being equal will have a “small” effect on child development. Therefore, the potential negative effect of alcohol will most likely be masked by the relatively larger effect of attachment and other mediators. Moving back to the model in Fig. 2 it can be said that alcohol is hypothesized to be causally associated with child neurodevelopment, but also causally associated with attachment. Attachment is also hypothesized to be causally associated with child neurodevelopment. The mediator (attachment) thus causes variation in the outcome (neurodevelopment) but is itself caused to vary by the original causal variable (alcohol).

In summary, mediating psychological factors such as attachment are rarely, if ever, controlled for in the statistical analyses of the large-scale epidemiological cohort studies. Because they are strong predictors of mental health they most likely mask the potential, probably small, effect of low dose exposure. Because mediating psychological factors are not included in large scale cohorts we do not know whether large differences exist between exposure groups (as was the case with the confounding factors described above). But it is surely fair to postulate that large differences do exist and that such mediating factors in turn bias the results of the statistical analyses in the studies investigating prenatal exposure to alcohol and child development.

Above I have focused on problems concerning factors that ideally should be included in the statistical analyses, i.e. confounders and mediators. Below I will focus on problems related to the exposures (i.e., the mother’s alcohol intake). What pieces of information do we use to classify the women? Firstly, I will focus on what is known as “dose”, “pattern” and “timing”, after which I will examine the validity of the self-reported measures of alcohol intakes generally used in large-scale cohort studies.

Apart from accepting the importance of dose, the most recent literature also recognised the importance of considering the pattern of the exposure, i.e. the quantity consumed on a typical occasion (O’Leary et al. 2010). Most studies today therefore distinguish between exposure to averaged-over-time (lower) doses of alcohol and to ‘binge’ drinking (Gray and Henderson 2007a; Henderson et al. 2007a, c; Underbjerg et al. 2012; Eriksen et al. 2012; O’Leary et al. 2009; Skogerbo et al. 2012; Kesmodel et al. 2012; Flak et al. 2013). Binge drinking is most often defined as an intake of a minimum of five units of alcohol on a single occasion (Gray and Henderson2007a; Henderson et al. 2007c). The rationale for this distinction is that it has been established that exposure to binge drinking is more devastating for the developing central nervous system because it is the peak blood alcohol concentration that determines the level of the damage (Gray and Henderson 2007a; Henderson et al. 2007c). In other words, binge drinking causes greater harm than exposure to an equivalent amount of alcohol spread over several days, weeks or months (Henderson et al. 2007c; Kesmodel 2001). Generally, there seems to be evidence of a very deleterious effect of being exposed to binge drinking. One review concluded (Gray and Henderson 2007b) that children exposed to binge drinking during their mothers’ pregnancies consistently showed poorer neurodevelopmental outcomes.

Apart from dose and pattern, a third factor, namely timing, seem to be of particular importance to understanding the effects of prenatal exposure to alcohol on child neurodevelopment. Most human studies have only investigated alcohol exposure during early pregnancy, (Sayal et al. 2007, 2009, 2013; Kelly 2010; Underbjerg et al. 2012; Skogerbo et al. 2012; Robinson et al. 2010) despite the fact that there appear to be two critical periods in human fetal development, when the brain is especially vulnerable to insult (Abel 1998). The first period begins during the first trimester, from gestational weeks 12–20. This period is characterised by a rapid rate of nerve-cell proliferation. The second period begins during the third trimester and continues until age 18 months of the child and is characterised by a brain growth spurt. Unfortunately, very few human studies have attempted to investigate possible effects of timing. One study that did so found that exposure to moderate-high levels of alcohol in the third trimester was negatively associated with language delay at age two (O’Leary et al. 2009). Another recent study compared binge drinking in the early and late part of pregnancy (not taking average intake into account) (Niclasen et al. 2013). This study reported that a small but significant association was observed on externalising scores (defined as behaviours that actions direct problematic energy outward) at age seven for the early exposure. However, the effect for the late exposure was much greater.

At least two conclusions can be drawn from the above evidence. One concerns the size of the effect of the exposure. Virtually all human studies focus on early exposure and report small effects of alcohol. But it may be that if the researchers instead focused on late exposure we might generally see stronger effects of alcohol. The second conclusion is related to the combination of the three factors dose, pattern and timing. Most researchers do not investigate all three factors. However, they all have a great impact on the results (small versus large doses; average versus binge drinking; and early versus late pregnancy exposure).

Apart from the questions of dose, pattern and timing, data on exposure can also suffer from reporting bias whereby women with actually a high intake may report only a zero or a very low intake. The former would thus be misclassified into the abstaining group (Abel 1998). It has been demonstrated that mothers who drink often deny their drinking habits (Garcia-Algar et al. 2012), that mothers who admit to drinking do not remember details correctly (Garcia-Algar et al. 2012), and that other women, in order to avoid disapproval, purposely under-report their alcohol intake (Kesmodel 2001; Kesmodel and Frydenberg 2004). Particularly women with a high intake tend to under-report alcohol consumption during pregnancy, especially if the information is not collected in a careful and sensitive manner (Kesmodel 2001). Because the under-reporting probably varies between groups this will also lead to misclassification that varies between intake groups. This, in turn, will inevitably have an impact on the results.