Descriptive statistics for variables of interest are shown in Table 1. The prevalence of antenatal depression, using a cut-off of EPDS >14, is 7 % at 18 weeks’ pregnancy and 8 % at 32 weeks, similar to rates reported in previous studies . Rates of PPD were between 9 and 12 %, also similar to results from former analyses .
80 % of mothers in this sample initiated breastfeeding and 74 % breastfed for 1 week or more. By 4 weeks only 56 % of mothers were breastfeeding at all and only 43 % were breastfeeding exclusively. The percentages of women feeding for the different durations considered are shown in Table 1; mean durations for breastfeeding and exclusive breastfeeding are also shown.
Table 2 shows the raw relationships between postnatal depressive symptomatology, and (a) prenatal depression, and (b) different measures of breastfeeding duration. A significant degree of correlation is present between postnatal and antenatal EPDS scores; a clear negative relationship also exists between symptoms of maternal depression measured at 8 weeks, and breastfeeding duration. The association between depression and breastfeeding is always negative, but generally statistically insignificant, at 8, 21 and 33 months.
Socio-demographic characteristics for sample members are presented in Table 7 in the Appendix. The mean age of participants was 28.3 years (SD = 4.8). 95 % of the women were white, 86 % were married, 13 % had university degrees, while a further 22 % had high school qualifications at age 18 (“A” levels); and 74 % owned the house in which they lived. In relation to pregnancy and birth, 64 % felt usually well, 55 % percent were working while pregnant, 45 % were primiparous, and only 9 % delivered via Cesarean section. The average gestational age was 39.5 weeks (SD = 1.8). 48 % of mothers and 37 % of fathers had themselves been breastfed as babies. 28 % of the pregnancies were unplanned; 15 % of mothers had lived through their own parents’ divorce before their eighteenth birthday.
Table 3 presents the results of logistic regressions estimating the effect of breastfeeding on PPD.
As explained earlier, three models are estimated: Model A controls only for the child’s sex and parental education; Model B controls in addition for a wide range of socioeconomic and demographic factors, plus information on pregnancy and birth; and Model C also controls for mother’s health (including mental health) in pregnancy, relationship quality and stressful life events.
We consider four different outcomes: EPDS >12 measured at 8 weeks, 8, 21 and 33 months postpartum. For each model/outcome dyad, the model is estimated seven times, for seven different measures of breastfeeding (initiation; any breastfeeding for at least 1, 2 and 4 weeks; and exclusive breastfeeding for at least 1, 2 and 4 weeks). Thus, each coefficient in Table 3 comes from a separate regression.
At 8 weeks postpartum, we observe a pronounced relationship between breastfeeding and PPD, under both Models A and B. The odds ratios for these models indicate that longer durations of breastfeeding are associated with larger reductions in the risk of PPD, and exclusive breastfeeding is associated with a larger reduction than any breastfeeding. However, under Model C, when we control for mothers’ health during pregnancy, these effects largely disappear; the only significant relationship which remains comes from exclusive breastfeeding for 4 weeks or longer (OR 0.81, 95 % CI 0.68, 0.97).
The relationship between breastfeeding and PPD is also weaker, the later the EPDS score is assessed; at 8 months postpartum and thereafter, most of the estimated coefficients are not significantly different from zero (indeed, a few of the results are counter-intuitive, suggesting that breastfeeding may be positively related to an increased risk of depression measured at 33 months postpartum).
Thus, for the sample as a whole, our results demonstrate little evidence for a causal relationship between breastfeeding and the risk of PPD. In the next section, we investigate the possibility that the relationship between breastfeeding and depression varies according to two factors: whether mothers were assessed as at risk of depression during pregnancy, and whether they had been planning to breastfeed their babies. We show that the relationship between breastfeeding and depression is indeed highly heterogeneous, and that this fact explains why little effect is found when considering women as a homogeneous group.
Heterogeneous Effects by Mental Health During Pregnancy and Breastfeeding Intention
We re-estimated Model C separately for mothers who were, and who were not, depressed during pregnancy (in terms of having a score EPDS >14 at least once during pregnancy). As before, we estimated regressions separately for each time at which postnatal depression was assessed (8 weeks, and 8, 21 and 33 months postpartum); for each of these time periods, we estimated seven models, one for each discrete measure of breastfeeding. However, instead of simply controlling for whether or not mothers breastfed for the relevant duration, we identify four groups of women, by whether they had planned to breastfeed, and whether they had actually breastfed for the relevant duration. These four groups are:
Mothers who had not planned to breastfeed, and who did not breastfeed (reference group)
Mothers who had not planned to breastfeed, but who did actually breastfeed
Mothers who had planned to breastfeed, but who did not actually breastfeed
Mothers who had planned to breastfeed, and who did actually breastfeed
Each regression thus generates three coefficients of interest; these coefficients are expressed as odds ratios, relative to the reference group.
Table 4 presents results for mothers without prenatal depression symptoms. Column (2) displays odds ratios and confidence intervals for mothers who did not plan to breastfeed, but who did actually breastfeed; column (3) indicates whether these mothers are significantly different from the mothers in the reference group.
Column (4) presents odds ratios for mothers who planned to breastfeed but who did not breastfeed for the relevant duration; Column (5) present odds ratios for mothers who planned to breastfeed, and who did breastfeed for the relevant duration. Column (6) indicates whether the odds ratios in Column (4) and (5) are significantly different from each other. Thus, the test results in Column (3) indicate whether breastfeeding makes a difference in the case of women who did not originally plan to breastfeed, while the tests in Column (6) indicate whether breastfeeding makes a difference in the case of mothers who had planned to breastfeed.
The strongest result from Table 4 is that breastfeeding is strongly associated with a lower risk of depression at 8 weeks postpartum, for women who had planned to breastfeed. The odds ratios in Column 4 are all well over 1, while the odds ratios in Column 5 are all well below 1; the differences between the two are statistically significant at the 1 % level or better for the first six measures of breastfeeding, and significant at the 5 % level for the remaining measure. The effects are smaller for later assessment periods. At 8 and 21 months, the odds ratios in Column 5 are lower than the odds ratios in Column 4 in almost all cases; however, the differences are not statistically significant. At 33 months, the differences are larger again, and are significant at the 1 % level for three of the seven measures of breastfeeding.
Interestingly, among the group of mothers who had not planned to breastfeed, the risk of depression was higher among women who went on to breastfeed. These differences are statistically significant for depression measured at 21 months, the largest being for any breastfeeding for 2 weeks on EPDS at 21 months (OR 1.62; 95 % CI 1.12, 2.36); at 8 weeks and 33 months the coefficients are all positive, though not generally significant at the 5 % level). To test whether our results were driven by a few mothers with very severe depressive symptoms, we repeated the analysis excluding those mothers with EPDS scores of 20 or more (the cut-off used in general practitioners’ guidelines  ); the results were virtually the same. We also investigated whether the effects depended on whether the mother was primiparous or multiparous, as suggested by ; again, the results were not affected.
Results for mothers who had been assessed as at risk of depression during pregnancy are shown in Table 5. For this group, results are less well defined, at least in part because of the smaller sample size. Our findings suggest that among women who had planned to breastfeed, breastfeeding is associated with a lower risk of PPD (as for mothers not depressed during pregnancy, although with a much smaller effect). However, for previously depressed mothers, there may also be a protective effect from breastfeeding when mothers had not planned to breastfeed. These results should be interpreted with caution: the only significant effect was found on EPDS measured at 8 weeks and for at least 4 weeks’ exclusive breastfeeding (OR 0.42; 95 % CI 0.20, 0.90).