The main finding of this study is that birth weight and ponderal index are related to total cortisol levels at 43 years, which seemed to be explained by increased levels during the latter part of the day, independently of current health behavior and body mass over the life course and unaffected by the exclusion of participants with medical conditions and medications. Interestingly though, while most studies have found birth weight to be negatively related to cortisol levels in adulthood [15–17, 23], we found the opposite direction of association: higher cortisol levels by larger body size at birth, across the normal spectrum of birth weight. This unexpected finding calls for some elaboration as to why our results would differ so substantially from previous findings.
First, most studies on the subject have been performed on small samples and a considerably higher drop-out rate which could introduce selection bias and misestimation of associations, compared to our population-based sample with high participation rate. Second, most studies on adults, and all with larger sample size (e.g., ), have not specifically examined cortisol levels during the latter part of the day, which in our analyses was the period of the circadian rhythm most prominently related to birth weight. Rosmalen and co-workers  have reported no substantial influence of birth weight on either morning or evening cortisol levels, but their large sample consisted of 10-12-year-old children, so their findings might not be directly generalizable to adults. Third, important sample characteristics (e.g. age range at the time of cortisol measurement) differ considerably between studies, rendering direct comparisons of dubious validity. However, yet another issue might be helpful for understanding the discrepancies in results: the management of preterm or low birth weight cases. In an explorative examination, preterm birth or low birth weight (< 2500 g) seemed to be related to higher bedtime cortisol levels of similar magnitude as of the group with the highest birth weight. This suggests that increased basal HPA axis activity might be produced by either harsh pre- or perinatal metabolic circumstances, or in the normal spectrum, also by supposedly favorable circumstances. Discrepant associations for preterm children compared to mature birth children in relation adult cortisol levels have been reported previously by Kajantie et al. , who found a positive relationship in children born at a later gestational age (> 39 weeks), but a negative association in those born early at term (37-39 weeks). A U-shaped association between birth weight and 24-hour glucocorticoid metabolites has also been described in preadolescent children . Although these studies differs from the present one on major methodological points, it corroborates the possibility of a positive relationship between birth size and adult cortisol levels in those maturely born, and a U-shaped association across the entire range of maturity and gestational age at birth.
This finding may warrant a re-evaluation of the hypothesis that only low birth weight might confer health-related physiological dysregulation in adulthood. Elevated evening cortisol levels have been found in men with coronary artery disease, correlated with markers of inflammation  and has also been linked to morbidity of chronic diseases as well as to mortality rates in women, but not in men , indicating that normal but high evening cortisol levels might confer increased health risks and thus emphasizing the potential public health importance of our findings. Although the associations reported in this study were rather weak, the long-term association with measure points 43 years apart would suggest that the observed dysregulations are enduring and stable, which could exert a health-damaging effect possibly over the entire life course. Moreover, even small effects might be of public health importance when generalized to the general population.
Although there is evidence of health-damaging consequences of high evening cortisol levels, the possibility of high cortisol levels representing a healthy pattern and low cortisol levels representing potentially harmful hypocortisolism  has to be considered. Hence, our findings cannot be unambiguously interpreted as representing an increased health risks with increasing birth weight, although we also found numerically high cortisol concentrations in the preterm or low birth < 2500 g subgroup- a group with well-established health risks across the life course [2, 40]. Since the fetal origins hypothesis examining outcomes such as cardiovascular disease has found linear relationships between birth weight and health measures, the pattern of a U-shaped relationship to cortisol seem to be contradictory. Speculatively, it is possible that different mechanisms operate in the normal spectrum of maturity as compared to those of preterm birth or of low birth weight, with different physiological trajectories yielding apparently similar outcomes in adulthood. Nevertheless, the significance of increased cortisol levels for these groups remains to be elucidated.
The analyses on women and men separately entailed lower power compared to the analyses on the total sample, as indicated by the less reliable findings, and should therefore be interpreted with caution. However, the separate analyses suggest that the influence of birth weight on both total (AUC) and bedtime cortisol levels was largely driven by corresponding associations among men, while these associations were considerably weaker and non-significant among women. As pointed out in a recent review , low birth weight is more consistently related to an increased HPA axis reactivity to acute stressors in men than in women, while there seems to be a greater influence on autonomic nervous system response in women. Thus, our results indicate that the circadian HPA axis regulation, in addition to the HPA response to acute challenges, might be more sensitive for early life influences in men than in women.
We did not find any evidence for an effect on the cortisol levels during the earlier part of the circadian rhythm. As described in the introduction, research focusing on parts of the circadian rhythm has almost exclusively studied cortisol levels during the early morning, and perhaps this is a partial reason why the literature on fetal origins of adult cortisol levels is abundant with null findings. There could also be methodological explanations to our null findings concerning CAR (see below).
Major strengths of the study include a large representative sample with low drop-out rate, and the saliva sampling protocol including assessment at different part of the day. Although there was a considerably larger drop-out on the cortisol measures, there was no evidence for selection bias with respect to birth weight.
Saliva was sampled only during one day. This would be expected to contribute to measurement error and attenuation of observed relationships. Sampling over several days would have been a preferred method. Since cortisol levels peak approximately 30 min post-awakening, our lack of findings regarding the cortisol awakening response could possibly be attenuated by the sampling protocol of only 15 min between morning samples. Moreover, since we have no information regarding the accuracy of reported sampling time, non-adherence could introduce random error and attenuation of estimated effects . Although the cortisol levels would not be expected to be influenced greatly by minor variations in sampling time during the later part of the day, the estimation of CAR is sensitive to such non-adherence since cortisol levels change very rapidly during this part of the circadian rhythm. For example, even a short time interval between awakening and the first sample could greatly influence the estimation of CAR [43, 44]. Although time of awakening was not reported by the participants, self-reported sampling time might not be an optimal method . These issues could possibly be methodological explanations of our negative findings regarding cortisol levels in the morning. While objective measures of the time of awakening and sampling would have been an optimal method to reliably assess the CAR, such high-constraint methods might be difficult and costly to implement in large population-based samples.