Background Recent years have seen considerable interest in the developmental origins of eating disorders (ED) but results have been conflicting, perhaps reflecting low power in many studies. Limited power has also prevented robust comparisons of associations with ED subtypes or the use of within-family designs to address the potential for confounding.

Methods We used total population data to create a cohort of 2,011,908 males and females, born 1975-1998 in Sweden to Swedish-born mothers. Birth characteristics included twin/triplet status; gestational age; birthweight; birth length; premature rupture of the membranes; delivery method; Apgar score at 5 minutes; birth traumas; mother’s smoking during pregnancy; and mother’s weight gain during pregnancy. We adjusted for multiple family and social characteristics, and conducted within-family analyses to test for confounding at the maternal/family level. Our outcomes were hospitalisation for anorexia, bulimia or eating disorder not-otherwise-specified (EDNOS)’ after age 12, with follow-up period until end 2010.

Results Anorexia was independently predicted by multiple birth (AOR 1.33 [95% CI 1.15, 1.53] for twin/triplet vs. singleton) and lower gestational age (HR 0.96 [0.95, 0.98] per extra week of gestation). Gestational age showed a clear dose-response pattern. These associations were largely specific to anorexia, and were only seen in the cohort members affected; within-family analyses revealed that the maternal siblings of twins or preterm individuals showed no increased risk, and provided no evidence of residual maternal-level confounding. Higher birthweight for gestational age showed a strong, positive dose-response association with bulimia (HR 1.15 [1.09, 1.22]) per sex-standardised standard deviation increase). Again, this association was specific to bulimia and within-family analyses provided no evidence of residual confounding. By contrast, although mother’s smoking predicted anorexia, this did seem likely to reflect maternal-level confounding. Other birth characteristics showed little or no association with any ED outcome, except a trend towards increased bulimia and EDNOS among mothers who gained excessive weight during pregnancy.

Conclusion These findings are consistent with a causal role of earlier gestational age upon anorexia and higher birthweight upon bulimia. Further research is needed to elucidate the mechanisms, but the dose-response nature of these associations indicates that they do not simply reflect pathological responses at the extremes of the distribution. The strong association with multiple births is noteworthy as many of the largest population-based studies of ED prevalence have been conducted in twins: our findings suggest the possibility that such studies substantially overestimate ED prevalence.