Perinatal Risk Factors for Diabetes in Later Life

The study cohort has been described in more detail previously (14). Within a well-defined source population, we manually examined ∼250,000 births records from 1925 through 1949 and identified an exposed cohort by identifying all newborn infants with a gestational duration <35 weeks and/or a birth weight <2,000 g for girls and <2,100 g for boys. Subjects who emigrated or deceased before 1987 were excluded. For unexposed cohort members, we selected subjects who were not born preterm or with low birth weight. For convenience, we selected the first child born of the same sex and hospital of birth as each exposed subject.

We used the date of the mother's last menstrual period to estimate gestational duration. Birth weight for gestational age was used as the measure of fetal growth. We used the Swedish reference curve for normal fetal growth (18) and categorized birth weights for gestational age into five groups according to their distance from the mean birth weight for gestational age (−2 SD or less, above −2 to −1 SD, above −1 to 0 SD, above 0 to 1 SD, and above 1 SD). Subjects whose birth weight was more than 4 SDs above or below the mean birth weight for gestational age were excluded.

The socioeconomic status of the family was assessed by the father's occupation, or by the mother's occupation in single-parent families, using three categories: high (college education), medium (white-collar workers and farm owners with no college education), and low (blue-collar workers and farmhands).

The follow-up started on 1 January 1987 and continued to 31 December 2006. We used the Swedish Register of Population and Population Changes to ascertain emigration and the Cause of Death Register to ascertain deaths. Diagnoses of diabetes were determined from the Hospital Discharge Register, which lists one main diagnosis and up to seven additional diagnoses. During the period 1987–1996, diagnoses were categorized according to the ICD-9 and thereafter according to the ICD-10. We considered cohort subjects as cases if they had a main or additional diagnosis of diabetes with diagnostic code 250 before 1997 and diagnostic code E11 after 1997.

Data were modeled through conditional Cox regression, using the TPHREG procedure in SAS Statistical Software (version 9.1; SAS Institute, Cary, NC). Analyses were conditioned by calendar period of birth, socioeconomic status, and sex. Additional adjustments for gestational duration and fetal growth were also obtained by conditioning on these factors. Missing information on socioeconomic status was treated as a separate category when the analysis was stratified on this variable. The study was approved by the research ethics committee of the Karolinska Institutet.

At the start of follow-up, there were 6,425 subjects in the cohort. The distributions of birth weight, gestational duration, and fetal growth are presented in further detail in Table 1. During follow-up, 508 subjects were treated as inpatients with a main or an additional diagnosis of diabetes. Calendar period of birth, socioeconomic status, and sex were all associated with risk of diabetes (Table 2), and all further analyses were adjusted for these three variables.

Birth weight was strongly associated with risk of diabetes (Table 3). Furthermore, both fetal growth and gestational duration were independently associated with risk of diabetes (Table 3). To evaluate whether our choice of reference curves affected the association between fetal growth and risk of diabetes, we also analyzed the data with all growth category limits shifted 50 and 100 g downwards and upwards, but those alterations had no effect on the results (data not shown).

The association of fetal growth and gestational duration with risk of diabetes was not affected by calendar period of birth (Table 3). Likewise, none of the variables for maternal age, hypertensive diseases during pregnancy, twin status, or breast-feeding at time of hospital discharge were associated with risk when fetal growth or gestational duration were included in the model (data not shown).

Table 4 describes an analysis of interaction between fetal growth and gestational age on risk of diabetes. The association between poor fetal growth and diabetes was independent of gestational age, and likewise the association between low gestational age and diabetes was independent of fetal growth.

This is the first study large enough to assess the individual contributions of fetal growth and length of gestation on risk of diabetes in adult life with appropriate statistical precision. Our data confirm the previously reported inverse association between birth weight and adult diabetes (2–12) and add the evidence that short gestational duration and fetal growth restriction are both independently associated with an increased risk of diabetes.

Our study is population based, information about birth characteristics was based on prospectively collected data, and follow-up was uniform across comparison groups through the use of register data. It is therefore unlikely, if not precluded, that measurement errors should vary between subjects with and without diabetes.

One limitation is the expanse of time between the subjects’ dates of birth and the beginning of the follow-up. Study subjects were born between 1925 and 1949, but the follow-up did not start until 1987. The cohort therefore consists of subjects born >50 years ago who survived to 37–62 years of age. Another limitation was our inability to test whether the association between birth weight and diabetes might be explained by a common genetic etiology (19). A third limitation of our study is the lack of structured data on the neonatal care of these subjects, and so there can only be speculation as to what extent the knowledge derived from this cohort from the early 1900s applies to today's newborns.

In our cohort, ∼70% of the subjects born at <33 weeks of gestation had a birth weight for gestational age that was above average. Selective survival favoring non–growth-restricted subjects among those born preterm may partly explain this skewness (20) and may have biased the obtained associations. However, childhood mortality declined steeply in Sweden from 1925 to 1949, whereas our risk estimates were fairly stable across birth cohorts. Thus, we think that selective survival is an unlikely explanation for our findings.

We used information on the mother's last menstrual period to estimate gestational age, which may lead to some misclassification. By excluding all subjects whose birth weight deviated more than 4 SDs from the mean birth weight for gestational age category, misclassification was reduced. Fetal growth was estimated through growth curves from ultrasonographically dated pregnancies from the 1990s. Reassuringly, our results remained unchanged when we altered the cutoff levels for the definition of fetal growth rate, indicating that the choice of reference curves does not affect the association.

Subjects in our cohort were registered as diabetic if they had required hospitalization, and because diabetes is frequently diagnosed in open clinics, we are likely to have underestimated the true incidence of diabetes. The absolute occurrence of diabetes in the study is therefore not interpretable. The relative risks of diabetes, however, could not be inflated but, rather, could be biased toward unity by such underestimation.

In this study, follow-up did not cover the entire life span of the subjects. For subjects born in 1925, follow-up started at the age of 61 years, whereas for subjects born in 1949, it stopped at the age of 57 years. Because risk estimates were similar regardless of birth period, we find it unlikely that this truncated follow-up has had more than a marginal influence on our results.

The fetal origins hypothesis suggests that fetal undernutrition during middle and late gestation triggers lasting hormonal and metabolic adaptations that ultimately lead to insulin resistance and diabetes (13,19,21). Our finding that not only fetal growth restriction but also short gestational duration contribute to the risk of diabetes is not entirely compatible with that hypothesis. It suggests, rather, that the association between low birth weight and diabetes has two components: one component mediated through poor fetal growth and the other through preterm birth. It cannot be ruled out that the effect on glucose metabolism of preterm birth and fetal growth share a common mechanistic pathway. It could, for example, be speculated that the association between low birth weight and diabetes could be due to postnatal nutritional exposures, both in those born small for gestational age and those born preterm. Unfortunately, we were unable to assess postnatal diet and infant growth in the present data.

Previous studies have found results consistent with ours. In a study on insulin resistance, Hofman et al. (16) found that infants born preterm and with appropriate weight for gestational age faced the same increase in risk of insulin resistance as did infants born small for gestational age at term. In a recent study of Hovi at al. (17), the authors found that preterm infants with very low birth weight (<1,500 g) showed similar patterns of glucose intolerance and insulin resistance regardless of whether they were born small or appropriately sized for gestational age.

In a recent meta-analysis on the association between birth weight and diabetes, Harder et al. (22) reported an additional association between high birth weight and risk of diabetes. We did not find such an association in our data, but our study was designed to study the risks in the lower end of the birth-weight spectrum, and consequently we lacked the power to assess the association between high birth weight and risk of diabetes with precision.

Consistent with the fetal origins hypothesis, we have previously found that the association between low birth weight and risk of both ischemic heart disease and hypertension was entirely mediated through poor fetal growth (14,15). The risk patterns in the present analysis are, however, distinct from our findings on heart disease and hypertension, suggesting that different mechanisms are involved in the perinatal origins of these diseases.

In conclusion, we have found that the association between low birth weight and risk for diabetes seems to be mediated through both poor fetal growth and preterm birth. The underlying programming mechanisms seem to involve not only prenatal but also postnatal factors.

Funding of data collection for this research was provided by U.S. Army Grant DAMD17-98-1-8117 and by the King Gustaf V Jubilee Foundation.

No potential conflicts of interest relevant to this article were reported.