The results suggest that in spite of the egalitarian nature of the Swedish society, there were considerable variations in the duration of breastfeeding by maternal education and smoking status during the 1980s. Moreover, we found that the children who were breastfed for shorter than 2 months had higher values for BMI, waist circumference and sum of five skin fold measurements at 15 year compared to their counterparts who were breastfed for 6 months or longer independently of birth weight, age, gender, parental BMI and maternal education. However, adjustment for maternal age and smoking at childbirth substantially reduced these associations. No dose-response relationship between the duration of breastfeeding and any of the studied measures of adiposity were found. We also observed that 15-year old children who were breastfed for shorter than 2 months had 3 times higher odds of being overweight than their counterparts breastfed for more than 6 months in a crude analysis, but this association disappeared after adjustment for other factors. The associations for 9-year olds were in the same direction, although they were less pronounced and did not reach the level of statistical significance.
The representativeness of the sample in relation to the general population in the studied areas [23], prospective registration of the breastfeeding status by health professionals, objective measurements of children's and adolescents' anthropometric characteristics are the main strengths of the study. No differences in socio-demographic factors between mothers with available and missing data on breastfeeding suggest that our estimates of breastfeeding prevalence are likely to be valid for the studied areas. The national breastfeeding statistics for 1989 reported the prevalence of any breastfeeding to be 97.7%, 85.8%, 67.7%, and 49.8% at 2 weeks, 2, 4, and 6 months, respectively, which is slightly above our estimates at 4 and 6 months (98.3%, 89.7%, 64.7%, and 41.6%). This is not surprising given that the studied areas were consistently reported to have breastfeeding rates below the national average [17] in previous national surveys.
Socio-demographic factors are important determinants of breastfeeding [18, 19]. Our findings on variations in breastfeeding duration by maternal education are in line with the results of other studies carried out in Sweden, which also found that low education was associated with shorter duration of breastfeeding [20, 21]. Maternal education in this study is defined as the highest attained education, not education at childbirth. This might be a better indicator reflecting the educational potential of the mother, her ambitions and the possibility of studying after childbirth. The underlying mechanisms of the associations between education and duration of breastfeeding are unclear; though we hypothesize that better educational potential of the mothers might lead to better compliance to the recommendations provided by the health care staff regarding breastfeeding. Although education is considered to be the most informative indicator of women's socioeconomic status in Sweden, there is still a chance for residual confounding by other factors. One study from Sweden reported no association between maternal age and duration of breastfeeding, which is in line with our results [22]. There is convincing evidence that smoking is associated with shorter duration of breastfeeding [19–22, 27, 28]. We obtained similar associations in the present study.
The self-reported nature of the maternal data may threaten the validity of the study, although this is unlikely given that previous studies from Sweden reported acceptable validity of self-reported data on obesity in women [29] and smoking during pregnancy [30]. Moreover, the prevalence of smoking in pregnancy in Sweden was about 30% during the 1980s with some tendency to decrease during the decade [31]. Our data show the prevalence of smoking to be 37% and 34% in 1983 and 1989, respectively, suggesting that underestimation is unlikely.
Similarly to most studies, our findings suggest that infants who are breastfed for shorter than 2 months have higher BMI, larger waistline and thicker skin fold measurements at 15 year than their counterparts breastfed for 6 months or more [3–8, 32, 33], but only before adjustment for maternal age and smoking at the time of childbirth. Several other studies also report attenuation of crude associations between measures of adiposity and breastfeeding after adjustment for confounders [6, 8–11]. No dose-response effect of the duration of breastfeeding on any of the studied measures of adiposity in either 9- or 15-year old children was observed. Interestingly, after adjustment, the lowest values for all measures of adiposity in both 15- and 9-year old children were observed in the group breastfed for 2–3 months reaching the level of statistical significance for sum of skin fold measurements in 15-year-olds.
Similarly to what was observed for the continuous measures of adiposity, crude associations between breastfeeding and overweight were substantially reduced after adjustment for maternal age and smoking status at childbirth. Moreover, adjustment for other studied characteristics resulted in lower odds ratios for overweight for 15-year old children breastfed for 2–3 months (OR = 0.7, 95%CI: 0.2, 2.2) and for 9 year old children breastfed for both 2–3 (OR = 0.6, 95%CI: 0.2, 1.7) and 4–5 months (OR = 0.5, 95%CI: 0.1, 1.5). Although the results were not statistically significant, they are congruent with the results of the analyses performed on BMI, waist circumference and sum of skin fold measurements. Victora et al also found 50% reduction in obesity in children breastfed for 3–5 months making it tempting to relate all these findings to the "critical window" theory of development [9]. However, wide confidence intervals for all associations in our study do not allow any conclusions.
The results of the study should be interpreted with caution, taking into account the limitations of the study. Firstly, the small sample size leads to insufficient statistical power to detect small differences in adiposity measures and odds for overweight by breastfeeding status. Secondly, the use of multiple linear regression while analysing slightly skewed data on BMI, waist circumference and skin fold measurements might be criticized. However, the distributions of the residuals were approximately normal and other assumptions were not violated. Moreover, repeated analyses on logarithmically transformed data yielded similar results.
Although our results do not contradict conclusions of large meta-analyses and systematic reviews on the effect of breastfeeding on obesity in childhood and adolescents, less pronounced differences between the groups of breastfeeding duration may partly be explained by the choice of the reference group. While in often cited reviews [4, 6] and meta-analyses the reference group included formula fed infants, in our study, the reference group was different. Contrary to many other studies, the proportion of never breastfed infants in our setting was very small. Only 2.5% of 15-year old and 1.8% of 9-year old children in our sample were not breastfeed at 2 weeks (Figure 1). Thus, our reference group included children, most of who were breastfed, but for shorter period than two months. Moreover, given that initiation rates and average duration of breastfeeding in Sweden are greater and the prevalence of childhood overweight and obesity is lower than in many other European countries and the USA, the associations between breastfeeding duration and later adiposity may be less visible in Sweden than in countries where breastfeeding is less popular and childhood obesity is more prevalent.