Birth weight in relation to health and disease in later life: an umbrella review of systematic reviews and meta-analyses

Our work constitutes the first comprehensive mapping and appraisal of the association between BW and the risk of subsequent health outcomes, as provided by published systematic reviews and meta-analyses of observational studies. Overall, 78 associations have been examined, including a diverse range of outcomes: cardiovascular, cancer, metabolic, respiratory and mortality outcomes, and disease traits and biomarkers. Despite common belief that the intrauterine environment as assessed by BW is associated with many diseases and disease traits in adult life [1, 6–10], our comprehensive assessment shows that convincing evidence only exists between the associations of low BW and increased risk for all-cause mortality, per 1 kg increase in BW and higher bone mineral concentration in hip and lower risk for mortality from cardiovascular diseases. Furthermore, the association between small-for-gestational-age and childhood stunting in low- and middle-income countries was supported by convincing evidence. There was no convincing evidence supporting associations between high BW and later outcomes; however, the associations with overweight or obesity in later life and all types of leukaemia were highly suggestive.

The associations between BW and cardiovascular disease were amongst the first to be observed in the medical literature [1–5] and our data suggests that the current evidence is highly suggestive. Both meta-analyses looking at low (2500 g) versus high (?4000 g) BW and those examining per 1 SD increase in BW showed highly significant summary effects and small between-study heterogeneity. However, both associations presented evidence for small-study effects and the dose–response association additionally had hints for excess significance bias. The latter may have resulted in inflated effect estimates for an association with cardiovascular disease that needs cautious interpretation [35, 44]. Despite the fact that studies have adjusted for a range of confounders, including socioeconomic status, not all studies were adjusted for gestational age, which is an important confounder and this, as well as other unrecognized confounders, could explain the observed association. In addition, the mechanisms underlying this association remain unclear despite many hypotheses having been suggested, including the one supporting that intrauterine under-nutrition leads to fetal adaptation, which is subsequently related to adverse cardiovascular risk in later life [10]. However, others have provided evidence that at least some of the association between the BW of individuals and their later risk of cardiovascular disease may be genetic and therefore not modifiable via interventions that target the intrauterine environment [45]. The causal pathway linking BW to cardiovascular risk needs further elucidation to allow evidence-based public health interventions.

The observed increased risk of cardiovascular disease associated with lower BW is likely to be a main contributor to the inverse association of BW with all-cause mortality; an association supported by convincing evidence in our assessment [42]. The higher incidence of perinatal mortality in the low BW group is also likely contributing to the all-cause mortality association with low BW, but only to a small extent. Babies born with a BW below 2500 g had increased perinatal mortality, an association supported by a very large summary effect estimate and a very small P value [46]. However, the meta-analysis on perinatal mortality was focused exclusively on developing countries. Therefore, the effect estimate might be exaggerated due to lack of neonatal intensive care units or difficult access to specialized healthcare facilities in these countries [47]. These data could not be generalised to other settings where high-quality healthcare is available.

The association between low BW and low bone mineral concentration in later life is less well studied compared to other outcomes and current data stem from six studies contributing to the meta-analysis [48]. Despite the fact that the association with bone mineral concentration in hip showed convincing evidence, cautious interpretation is required as data on osteoporotic fractures has not been reviewed and meta-analyses on other anatomical sites (e.g. lumbar spine) showed evidence for excess significance bias and no convincing associations.

Comparisons between BW and later overweight and obesity do not support a detrimental health effect of low BW. BW less than 2500 g was found to be protective for being overweight or obese, whereas BW greater than 4000 g was linked with an increased risk for being overweight or obese in adult life [43]. These associations were supported by highly suggestive evidence, but they also displayed very large between-study heterogeneity. Heterogeneity could be due to biased results in some of the included studies, but it could also reflect genuine differences across studies [35]. BW distributions are remarkably different across developed and developing countries [49], and the associations between BW and later adiposity may differ in these populations, contributing to the heterogeneity of the observed results. High BW is potentially causally associated with maternal BMI and glucose levels [50, 51]; however, the extent to which it could be modified through lifestyle or pharmacological interventions merits further investigation, particularly through long-term follow-up of interventions during pregnancy, which will strengthen and enhance the available evidence, particularly between high BW and subsequent risk of childhood and adulthood obesity [52–54].

Although 29 associations focused on outcomes related to different types of cancer, high BW was found to be a risk factor only for developing leukaemia [13]. The associated summary effect estimate might be inflated by the presence of small-study effects and excess significance bias. However, the statistical heterogeneity was not large, the 95 % prediction interval excluded the null value and the association was highly significant. Similarly, despite diabetes being central in the “fetal origin hypothesis” [7], its association with high and low BW has weak evidence in the literature and is only suggestive of a direct association with high BW in line with the obesity-associated evidence.

Despite intensive research on BW reflected by the large number of meta-analyses identified, there were only three papers that performed meta-analyses of studies assessing low BW in relation to gestational age [40, 55, 56], whereas no single meta-analysis on large-for-gestational-age neonates was identified. As BW and gestational age are highly correlated, analyses which consider size-for-gestational-age rather than BW adjusted for gestational age have been proposed as a more appropriate alternative [57, 58]. Among the examined phenotypes in relation to small-for-gestational-age, the association between small-for-gestational-age without low BW and childhood stunting in low- and middle-income countries showed convincing evidence. However, those results require cautious interpretation as the analyses were stratified by BW and the association between small-for-gestational age with low BW and childhood stunting showed a much weaker effect estimate and was only supported by weak evidence. Additionally, those analyses focused on low- and middle-income countries, limiting the generalisability of those results but at the same time also highlighting the need for interventions during the pregnancy period in these populations [40]. The remaining meta-analyses included a mixture of studies that adjusted or not their analyses for gestational age and, hence, the current literature is inconclusive on the effects of BW relative to gestational age.

In the present study, we applied the umbrella review approach summarising data from already published systematic reviews and meta-analyses. This approach takes full advantage of the existing meta-analyses to perform a standardised methodological process for the assessment of the epidemiological credibility of the findings. However, our study has some caveats. First, the Egger test and excess statistical significance test offer hints of bias, and not proof thereof, while the Egger test is difficult to interpret when the between-study heterogeneity is large. Further, our excess significance estimates were based on the largest study of each meta-analysis and they might be conservative, because often these studies were not necessarily very large or might have had inherent biases themselves. Furthermore, we did not appraise the quality of the primary studies, because this was beyond the scope of this umbrella review. This should be the aim of the original systematic reviews and meta-analyses, which should examine the methodological characteristics of the component studies.