Ethnicity was subsequently removed as its inclusion inhibitor price did not have an appreciable effect on the result, and its removal appeared to make the model more robust with narrower CIs. For birth weight, the same potential confounders were considered

together with gestational age and smoking during pregnancy. The following variables were significantly associated with the outcome (indicated by Wald, p<0.05) after controlling for other factors in the model: gestational age, education, ethnicity and maternal age. As smoking during pregnancy is likely to be in the causal chain of low birth weight, the model was considered with and without this variable to see to what extent the effect on birth weight is mediated by smoking. The ‘unexposed’ group was used as the reference for these analyses. For univariable analysis, those with missing outcomes of smoking during pregnancy, low birth weight and any breastfeeding following birth were excluded (7 (0.04%), 21 (0.1%) and 4 (0.02%), respectively); all these excluded women were from the 18 201 ‘unexposed’ group. However, 726 women were excluded due to missing data on symptoms of depression (Malaise Inventory score). Women who had been in care were not more likely to have missing

data in this variable than those who had not been in care. In those who had spent some of their childhood in care, 11 of the 291 women had missing data (3.9%). There were no statistically significant differences between those who had missing data and those who did not in terms of age, income, social class and education. Of the women who had not spent any time in care, 715 of the 18 201 women had missing data for symptoms of depression (4.0%). Those who

had missing data were more likely to be in a lower social class, have a lower income and to have lower or no qualifications. For multivariable analysis, a complete case analysis was undertaken. Those excluded due to missing data were less than 10% of the cohort, with resulting sample size ranging from 16 351 to 18 238 (table 4). Table 4 Unadjusted and adjusted ORs (95% CI) for smoking during pregnancy, low birth weight, any breastfeeding and symptoms of depression among mothers according to a history of being in care All analyses took into account the clustered stratified study design by using the survey commands in Stata V.13.0.35 Reported p values and CIs account for clustering, and estimates of proportions and ORs are weighted by sampling weights.36 Results Description of the Entinostat cohort There were 18 552 respondents of the baseline interview of the Millennium Cohort Study. Fifty-seven respondents, who were not the natural mothers of the cohort baby, were excluded, as were three interviewees who did not have data relating to their time in care history. Therefore, our study population included 18 492 natural mothers. In the study population, there were 291 mothers who reported spending time in care as a child, which was 1.6% of the cohort (95% CI 1.3 to 1.8).