In this cohort study, we have analysed how length of breastfeeding is conditioned by maternal characteristics (age, education, occupation activity, smoking habit), pregnancy features (duration, twin/single, type of delivery) and newborn attributes (gender, weight at birth). In a multivariate analysis that provides length of breastfeeding ratios, we found that breastfeeding was shorter in children born from mothers who were younger, without university studies or smoked in pregnancy, and those born from twin pregnancy or weighting less than 2500 g at birth.
The lack of quality information in previous studies prevented Cohen et al. from including maternal age in their meta-analysis on factors associated with breastfeeding initiation and continuation [13]. According to our results, each additional year in maternal age increases breastfeeding duration by 2% (i.e. time ratio = 1.02); in this regard, if women aged 25 are expected to breastfed 2.90 months on median, women 10 years older are expected to breastfed about 20% more time, until 3.50 months on median (Table 4). Scott et al. [15] found that 10 years more doubled the odds of breastfed continuation, although they did not study breastfeeding duration but its continuation until 4 months.
To study maternal education as factor associated with breastfeeding continuation is not straight forward as education attainment could be measured in different ways as educative systems are not always equivalent to each other. The easiest ways of making international results comparable are (i) measuring it in number of years of schooling [15], which assumes a linear effect (i.e. each additional educative year has the same effect on breastfeeding length) and (ii) restricting the analysis to a comparison between the highest and the lowest educational levels [13]. Both strategies mislay part of the information whether assuming a linear effect or omitting intermediate educational levels. We have found that, far from linear, the effect of maternal education in enlengthening breastfeeding only appears in mothers who have university studies and their children were breastfed for 53% more time than those born from mothers with only primary studies. Of note, in the last decades, progressively higher proportions of Spanish women have attained university degrees, making it possible for 37% women in our cohort and even a higher percentage among those breastfeeding at hospital discharge (Table 1). Mechanisms for higher educated women to breastfeed longer are not clear. It has been suggested that they are more aware of the health implications of breastfeeding [16] and that they -being economically more independent than less educated women, and are empowered to make the decision on whether breastfeeding or not by themselves [17].
Smoking in pregnancy has been largely and consistently identified as factor associated with both breastfeeding non-initiation and early discontinuation [10, 11, 13]. What our study adds is that breastfeeding lasted about 40% less in children born from mothers who smoked in pregnancy when compared with those from mothers who did not smoke. Among the identified factors influencing breastfeeding initiation and duration, smoking is probably the more modifiable and the more consistently associated with other deleterious effects on both mothers and children [18,19,20]. In spite of that, about 1 in 8 women in our cohort smoked in pregnancy, a similar percentage as reported in the US [13].
Shorter duration of pregnancy and lower weight at birth have been frequently found related to early breastfeeding discontinuation [8, 9, 21]. The fact that these two factors are strongly associated with each other makes it difficult to separate its effects. According to our results, the association between length of pregnancy and breastfeeding duration disappeared in the multivariate setting when adjusting for weight at birth, suggesting that lower weight at birth is the dominant factor of this finding.
The main point our study adds to literature is to present results as time ratios instead of the usual hazard ratios. Time ratios allows an easier interpretation, especially in the multivariate model where we can take advantage of its multiplicative nature. In this regard, we have shown an example on how a deleterious factor (smoking) could cancel out some positive factors (education level and maternal age), which reinforces the importance of acting on any preventable factor, whatever the exposure to the non-modifiable factors is.
Our study has some limitations. Firstly, length of breastfeeding was reported by mothers, so there is some room for information bias as some women could have informed according to social desires more than according to their actual practice. Secondly, although our sample size is close to 1000 mother-child dyads, some categories in the analysis have few participants, which makes some confidence intervals excessively wide; this could be the case of newborn weight lower than 2500 g or some categories in duration of pregnancy in the multivariate setting. Third, we have limited our research to variables standardly recorded in order to make our results more robust, but this strategy has left aside some important variables related to breastfeeding initiation and continuation, such as breastfeeding self-efficacy, return to work activity or previous breastfeeding experience. The main strength of our study is that women and children have been prospectively followed in a homogeneous way in a single centre committed to breastfeeding practices.