Birth Weight and Risk of Type 2 Diabetes
A Systematic Review
- Peter H. Whincup, PhD;
- Samantha J. Kaye, MSc;
- Christopher G. Owen, PhD;
- Rachel Huxley, PhD;
- Derek G. Cook, PhD;
- Sonoko Anazawa, MD;
- Elizabeth Barrett-Connor, MD;
- Santosh K. Bhargava, MD;
- Bryndís E. Birgisdottir, PhD;
- Sofia Carlsson, PhD;
- Susanne R. de Rooij, PhD;
- Roland F. Dyck, MD;
- Johan G. Eriksson, MD;
- Bonita Falkner, MD;
- Caroline Fall, DM;
- Tom Forsén, MD;
- Valdemar Grill, MD;
- Vilmundur Gudnason, MD;
- Sonia Hulman, MD;
- Elina Hyppönen, PhD;
- Mona Jeffreys, PhD;
- Debbie A. Lawlor, PhD;
- David A. Leon, PhD;
- Junichi Minami, MD;
- Gita Mishra, PhD;
- Clive Osmond, PhD;
- Chris Power, PhD;
- Janet W. Rich-Edwards, ScD;
- Tessa J. Roseboom, PhD;
- Harshpal Singh Sachdev, MD;
- Holly Syddall, MSc;
- Inga Thorsdottir, PhD;
- Mauno Vanhala, MD;
- Michael Wadsworth, PhD;
- Donald E. Yarbrough, MD
- Author Affiliations: Division of Community Health Sciences, St George’s, University of London (Drs Whincup, Owen, and Cook and Ms Kaye), Medical Research Council (MRC) Centre of Epidemiology for Child Health, University College London Institute of Child Health (Drs Hyppönen and Power), London School of Hygiene and Tropical Medicine (Dr Leon), and MRC Unit for Lifelong Health and Ageing, Department of Epidemiology and Public Health, University College London (Drs Mishra and Wadsworth), London, England; the George Institute for International Health, University of Sydney, Sydney, Australia (Dr Huxley); Department of Internal Medicine, Saiseikai Central Hospital, Tokyo, Japan (Dr Anazawa); Division of Epidemiology, Department of Family and Preventive Medicine, University of California San Diego School of Medicine, San Diego (Dr Barrett-Connor); Department of Paediatrics, S. L. Jain Hospital, Delhi, India (Dr Bhargava); Unit for Nutrition Research, Faculty of Food Science and Nutrition, University of Iceland and Landspitali-University Hospital (Drs Birgisdottir and Thorsdottir), and Icelandic Heart Association Research Institute and University of Iceland (Dr Gudnason), Reykjavik; Division of Epidemiology, Stockholm Centre of Public Health, and Department of Epidemiology, Institute of Environmental Medicine, Karolinska Institutet, Stockholm, Sweden (Dr Carlsson); Department of Clinical Epidemiology, Biostatistics, and Bioinformatics, Academic Medical Centre, University of Amsterdam, Amsterdam, the Netherlands (Drs de Rooij and Roseboom); Department of Medicine, University of Saskatchewan, Saskatoon, Saskatchewan, Canada (Dr Dyck); Diabetes Unit, Department of Health Promotion and Chronic Disease Prevention, National Public Health Institute, and Department of General Practice and Primary Health Care, University of Helsinki, Helsinki, Finland (Drs Eriksson and Forsén); Vasa Central Hospital, Vasa, Finland (Drs Eriksson and Forsén); Department of Medicine and Paediatrics, Thomas Jefferson University, Philadelphia, Pennsylvania (Dr Falkner); MRC Epidemiology Resource Centre, University of Southampton, Southampton, England (Drs Fall and Osmond and Ms Syddall); Department of Cancer Research and Molecular Biology, Norwegian University of Science and Technology, and Department of Internal Medicine, University Hospital of Trondheim, Trondheim, Norway (Dr Grill); Division of Neonatology, Crozer Chester Medical Center, Crozer-Keystone Health System, Upland, Pennsylvania (Dr Hulman); Department of Social Medicine (Dr Jeffreys) and MRC Centre for Causal Analyses in Translational Epidemiology (Dr Lawlor), University of Bristol, Bristol, England; Department of Hypertension and Cardiorenal Medicine, Dokkyo University School of Medicine, Tochigi, Japan (Dr Minami); Division of Women's Health and Connors Center for Women's Health and Gender Biology, Harvard Medical School, and Department of Medicine, Brigham and Women's Hospital, Boston, Massachusetts (Dr Rich-Edwards); Department of Paediatrics and Clinical Epidemiology, Sitaram Bhartia Institute of Science and Research, New Delhi, India (Dr Sachdev); Central Hospital of Central Finland, Jyväskylä, Finland (Dr Vanhala); Unit of Family Practice, Central Hospital of Middle Finland and Kuopio University Hospital, University of Kuopio, Kuopio, Finland (Dr Vanhala); and Sacred Heart Medical Center, Eugene, Oregon (Dr Yarbrough).
-
Corresponding Author: Peter H. Whincup, PhD, Division of Community Health Sciences, St George’s, University of London, Cranmer Terrace, London SW17 0RE, England (p.whincup@sgul.ac.uk).
Abstract
Context Low birth weight is implicated as a risk factor for type 2 diabetes. However, the strength, consistency, independence, and shape of the association have not been systematically examined.
Objective To conduct a quantitative systematic review examining published evidence on the association of birth weight and type 2 diabetes in adults.
Data Sources and Study Selection Relevant studies published by June 2008 were identified through literature searches using EMBASE (from 1980), MEDLINE (from 1950), and Web of Science (from 1980), with a combination of text words and Medical Subject Headings. Studies with either quantitative or qualitative estimates of the association between birth weight and type 2 diabetes were included.
Data Extraction Estimates of association (odds ratio [OR] per kilogram of increase in birth weight) were obtained from authors or from published reports in models that allowed the effects of adjustment (for body mass index and socioeconomic status) and the effects of exclusion (for macrosomia and maternal diabetes) to be examined. Estimates were pooled using random-effects models, allowing for the possibility that true associations differed between populations.
Data Synthesis Of 327 reports identified, 31 were found to be relevant. Data were obtained from 30 of these reports (31 populations; 6090 diabetes cases; 152 084 individuals). Inverse birth weight–type 2 diabetes associations were observed in 23 populations (9 of which were statistically significant) and positive associations were found in 8 (2 of which were statistically significant). Appreciable heterogeneity between populations (I2 = 66%; 95% confidence interval [CI], 51%-77%) was largely explained by positive associations in 2 native North American populations with high prevalences of maternal diabetes and in 1 other population of young adults. In the remaining 28 populations, the pooled OR of type 2 diabetes, adjusted for age and sex, was 0.75 (95% CI, 0.70-0.81) per kilogram. The shape of the birth weight–type 2 diabetes association was strongly graded, particularly at birth weights of 3 kg or less. Adjustment for current body mass index slightly strengthened the association (OR, 0.76 [95% CI, 0.70-0.82] before adjustment and 0.70 [95% CI, 0.65-0.76] after adjustment). Adjustment for socioeconomic status did not materially affect the association (OR, 0.77 [95% CI, 0.70-0.84] before adjustment and 0.78 [95% CI, 0.72-0.84] after adjustment). There was no strong evidence of publication or small study bias.
Conclusion In most populations studied, birth weight was inversely related to type 2 diabetes risk.
- KEYWORDS:
- BIRTH WEIGHT
- DIABETES MELLITUS TYPE 02
- INFANT
- INFANT, LOW BIRTH WEIGHT
- MOTHERS
- RISK FACTORS
Type 2 diabetes is a major global public health problem, increasing in prevalence in all regions of the world.1,2 The development of effective primary prevention strategies for the control of type 2 diabetes is crucial.3 Type 2 diabetes has a multifactorial etiology, with important contributions from obesity, lifestyle, and genetic factors.4,5 In 1991, Hales et al6 reported a graded inverse association between birth weight and risk of type 2 diabetes, with the highest risks of type 2 diabetes occurring at the lowest levels of birth weight, and suggested that fetal undernutrition might be important in the etiology of type 2 diabetes; the association between birth weight and risk of type 2 diabetes has subsequently been examined in a large number of published reports.7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37 However, several important questions remain about the association that have not been systematically addressed. These include its strength and consistency and its independence of confounding (particularly by socioeconomic status, which is related both to birth weight and to risks of type 2 diabetes). Early studies reported the association adjusted for adult body size (particularly for body mass index, which is itself strongly related to risk of type 2 diabetes38,39). The validity of such adjustment is controversial40 and its impact not fully documented. Moreover, the influences of maternal diabetes in pregnancy and high birth weight (macrosomia) on the association between birth weight and type 2 diabetes have not been systematically examined. Maternal diabetes in pregnancy is associated with an increased risk of macrosomia41; both factors are related to increased risks of type 2 diabetes.15,18
Earlier reviews of the association between birth weight and type 2 diabetes have been limited by being qualitative rather than quantitative42 or by including only a limited number of published studies, focusing on extreme birth weights rather than the birth weight distribution as a whole, and not taking account of potential confounding.43 We therefore carried out a systematic quantitative review of published studies reporting on the association between birth weight and type 2 diabetes in adult population to establish the overall strength, consistency, and independence of the association. To take account of the possibility that the association is nonlinear,43 we formally examined both the shape of the birth weight–type 2 diabetes association and the effects of excluding participants with macrosomia and with maternal diabetes. We also examined whether the association differed among populations, notably native North American populations, in which the prevalence of maternal diabetes is exceptionally high.15,28
METHODS
We followed a modified version of the Meta-analysis of Observational Studies in Epidemiology (MOOSE) guidelines for the conduct of systematic reviews and meta-analyses of observational studies.44 Studies published between 1950 and November 2007 were identified through the MEDLINE (from 1950 onward), EMBASE (from 1980 onward), and Web of Science (from 1980 onward) databases with a search strategy that combined text word and Medical Subject Headings identifying reports relating birth weight and type 2 diabetes and glucose and insulin levels (full search strategy available from the authors). The literature search was updated to June 2008 using automatic Ovid alerts. The reference lists of identified reports were also scanned to identify any other relevant studies. Studies were included if they reported either quantitative or qualitative estimates of the association between birth weight and type 2 diabetes. Studies were excluded if the population consisted of abnormal subgroups (for example, very low-birth-weight infants), were based on twins, or reported outcomes primarily in children (mean age at outcome <16 years).
In total, 327 articles were identified and reviewed in detail. Of these, 31 articles included unduplicated information on the association between birth weight and type 2 diabetes in adults and were included in the review. The other 296 were excluded because they either had no such information or because they duplicated data from other reports (Figure 1); a full list of these reports is available from the authors. For the 31 articles included, the first or corresponding author was approached and asked to provide either a limited individual participant data set to allow the investigators to carry out prespecified analyses or (if preferred) the results of the prespecified analyses directly. When it was not possible to obtain information from study authors, information was extracted from published reports whenever possible. The data set requested from authors included current age (years), sex, birth weight, current body mass index (calculated as weight in kilograms divided by height in meters squared), socioeconomic status (based on the occupation and/or education of the individual if possible), history of maternal diabetes, type 2 diabetes status, and, if available, fasting and/or postload blood glucose values. When possible, a combination of self-reported diagnosis and measured glucose concentrations using cutoffs based on standard World Health Organization criteria45 were used to define diabetes.
We also extracted information on key indicators of study quality, identified a priori. These included the method of ascertainment of birth weight (whether recorded from medical records or obtained by recall), method of ascertainment of type 2 diabetes (whether based on the results of blood measurements, medical records, participant recall, or a combination of these), and study response rates (both participation rates and follow-up rates where appropriate). The availability of important potential confounding variables (age, sex, and social class) and other modifying cofactors (body mass index, maternal diabetes) were also documented as described above.
Prespecified analyses were carried out either by the reviewers using individual participant data sets or by the relevant study investigators. Unmatched logistic regression models were used to provide estimates of the odds ratio (OR) of type 2 diabetes (with 95% confidence intervals [CIs]) associated with a 1-kg increase in birth weight, fitted as a continuous variable. Whenever possible, models were fitted (1) unadjusted; (2) adjusted for age and (when appropriate) sex; (3) adjusted for age, sex, and individual socioeconomic status; and (4) adjusted for age, sex, and current body mass index. Whenever possible, model 2 was repeated excluding participants with birth weights in excess of 4 kg and then also excluding participants with a maternal history of diabetes. In studies in which birth weight data were available only in predefined groups rather than as exact measurements, participants were allocated a mid-point birth weight for the relevant birth weight group, similar to that used in the published analysis of those studies.24,31 In studies for which the only information available was from published reports, this was generally in the form of logistic regression coefficients for predefined birth weight groups.13,17,26,34,46 Estimates of the continuous relationships of birth weight to type 2 diabetes were obtained for these studies using log-linear dose-response regression.47
Meta-analysis was carried out using Stata/SE, version 9 (Stata Corp, College Station, Texas). An a priori decision was made to pool ORs using random-effects models, specifically to allow for the possibility of variation in the associations between birth weight and type 2 diabetes between different populations. Statistical tests to assess the extent of heterogeneity were carried out using the Cochran Q test.48 Formal descriptions of the extent of heterogeneity were provided by calculating the I2 statistic and its 95% CI; these define the percentage of variation between studies due to heterogeneity rather than chance.49 Funnel plots were used to assess whether small studies yielded larger effect estimates than larger studies, suggesting the presence of publication bias.50 Tests described by Begg and Mazumdar51 and Egger et al52 for publication bias and small study bias, respectively, were also performed.
Meta-regression was used to examine whether differences in study size and statistical weight (expressed as the inverse of the squared standard error of the logistic regression coefficient), age at outcome, and year of birth influenced the associations seen or explained heterogeneity across the studies. Meta-regression was also used to examine whether key indicators of study quality (particularly method of ascertainment of birth weight, method of ascertainment of type 2 diabetes, and response rate [all defined a priori]) influenced the associations seen or accounted for heterogeneity across the studies.
Individual participant data were used to examine the shape of the association between birth weight and type 2 diabetes. Random-effects logistic regression analyses were conducted with prespecified birth weight groupings to achieve this. These groupings were also used to explore the association between birth weight (fitted as a continuous variable) and type 2 diabetes in each fifth of the distribution of current body mass index (defined for each study individually), fitting logistic regression models adjusted for age and sex and testing for the statistical significance of an interaction term ordered across the 5 body mass index groups.
RESULTS
A total of 31 relevant published reports were identified relating birth weight to risk of type 2 diabetes, involving 6260 cases of diabetes in 152 594 individuals in 32 populations (1 report from Saskatchewan included separate studies in native North Americans and in a general population of predominantly white European origin15) (Table 1). The investigators of 17 studies provided individual participant data for analysis.8,9,18,20,21,22,23,24,25,27,29,30,31,32,33,35,37 The investigators of a further 8 studies provided the results of relevant analyses on the strength of the birth weight–type 2 diabetes association with varying degrees of adjustment.6,10,12,14,15,16,19,36 For 5 published studies, it was possible to obtain ORs from published reports.13,17,26,28,34 For the 1 remaining study, it was not possible to obtain quantitative estimates of the association either from the authors or from the published report.11
Table 1. Studies Included in the Meta-analysis
In total, therefore, the analyses include data from 30 studies in 31 populations, involving 6090 cases of diabetes among 152 084 individuals (Table 1). Most of the studies were from populations in Western Europe (18 studies) and North America (6 studies). Two studies were based in India,9,18 2 in China,29,34 and 2 in Japan.8,33 One Western European study was based on famine offspring.14 Two of the North American studies included native populations15,28; 1 of these also included a general population sample.15 Among these 30 studies, birth weight was based on information recorded at the time of birth in 23 studies, by recall in 6 studies, and by a combination of both methods in 1 study (Table 1).
Studies used more than 1 method to define type 2 diabetes in many cases (Table 1). Information on investigator-measured fasting plasma glucose concentration was used in 15 studies,9,10,12,18,20,21,25,26,27,29,32,33,34,35,37 information on investigator-measured postload glucose concentration in 17 studies,6,9,10,12,14,18,20,21,26,27,28,29,30,32,34,35,37 and information on hemoglobin A1c in 1 study.8 Information on physician-diagnosed diabetes by individual recall was used in 21 studies8,9,10,13,14,17,18,21,22,23,24,25,26,27,29,30,31,34,35,36,37 and information on physician-diagnosed diabetes from medical registers or other routine sources by 3 studies.15,16,19 In 14 studies, the diagnosis of diabetes was based on a combination of recall of physician-diagnosed diabetes and measured glucose or hemoglobin A1c concentrations.8,9,10,14,18,21,25,26,27,29,30,34,35,37 In 6 studies, the diagnosis of diabetes was based entirely on the measurement of blood glucose.6,12,20,28,32,33 In 10 studies, the ascertainment of type 2 diabetes was based on physician diagnosis, either by recall (7 studies)13,17,22,23,24,31,36 or from registers or other routine health information sources.15,16,19
Odds Ratios Relating Birth Weight and Type 2 Diabetes in Different Populations
The ORs for type 2 diabetes for a 1-kg increase in birth weight, adjusted for age and sex, are shown for all study populations, ranked by age at outcome, in Figure 2. Among the 31 populations, 8 showed positive associations (2 of which were statistically significant at P < .05) and 23 showed inverse associations (9 of which were statistically significant at P < .05). There was strong evidence of heterogeneity among the populations studied (30 populations; I2 = 66%; 95% CI, 51%-77%; P < .001). Two large studies of native North American populations in Arizona and Saskatchewan showed positive birth weight–type 2 diabetes associations.15,28 Removal of these populations reduced the heterogeneity slightly (28 populations; I2 = 50%; 95% CI, 24%-68%; P = .001). Much of the remaining heterogeneity was accounted for by a study in young adults from the Saskatchewan general population (predominantly consisting of individuals of white European origin), which also showed a positive association between birth weight and type 2 diabetes15; removal of this study population greatly reduced the heterogeneity (27 populations; I2 = 20%; 95% CI, 0-50%; P = .17). The combined OR, including all 31 populations and based on a random-effects model, was 0.80 (95% CI, 0.72-0.89). Exclusion of the 2 native North American populations strengthened the estimate slightly in the remaining 29 populations (OR, 0.77; 95% CI, 0.70-0.84). Additional exclusion of the Saskatchewan general population study had little effect on the strength of association in the remaining 28 populations (OR, 0.75; 95% CI, 0.70-0.81). There was no strong evidence of marked heterogeneity between the 2 native North American populations (P = .12); the pooled OR relating birth weight and type 2 diabetes in these 2 populations was 1.21 (95% CI, 0.96-1.53).
CI indicates confidence interval. Values less than 1 indicate an inverse association of birth weight to type 2 diabetes. Markers
are proportional in size to study weight (denoted by the inverse of the variance). The mean age (in years) of each study's
participants is shown in descending order of age at which type 2 diabetes was ascertained.
aNative North American population.
bGeneral population (predominantly white origin).
Across all 31 study populations, the association of birth weight with type 2 diabetes tended to be increasingly inverse with increasing age at outcome. In a formal meta-regression analysis, a 10-year increase in age at outcome was associated with a difference of −0.09 (95% CI, −0.01 to −0.16; P = .02) in the log OR of type 2 diabetes per 1-kg increase in birth weight. However, this evidence of a difference in the strength of the birth weight–type 2 diabetes association at different ages of outcome depended strongly on the presence of the 2 native North American populations, both consisting predominantly of young adults.15,28 The removal of these populations reduced the influence of age at outcome markedly to −0.04 (95% CI, −0.12 to 0.05; P = .40). Year of birth showed a modest positive association with the log OR of type 2 diabetes per 1-kg increase in birth weight, with a change of 0.06 in the log OR (95% CI, −0.01 to 0.12; P = .11) per decade for all 31 populations. Again, however, the removal of the 2 native North American populations reduced this association markedly, to 0.02 (95% CI, −0.05 to 0.09; P = .60). In 10 populations,9,18,20,24,27,29,30,31,32,33 it was possible to distinguish term and preterm births. The restriction of analyses to term births had little effect on the strength of the birth weight–type 2 diabetes association (OR, 0.80; 95% CI, 0.72-0.90 before restriction and OR, 0.78; 95% CI, 0.69-0.88 after restriction).
Influence of Study Size and Publication and Reporting Bias
We examined the influence of study size, publication bias, and reporting bias among 29 populations, excluding the 2 native North American populations because of their obvious difference from the other populations studied.15,28 Studies were divided into 3 approximately equal groups by their statistical weight (determined as the inverse of the variance of the logistic regression coefficient; low: <10.3; medium: 10.3-31; high: >31).
The OR of type 2 diabetes for a 1-kg increase in birth weight did not differ consistently between studies with a high statistical weight,6,10,12,13,15,16,19,25,31 studies with an intermediate weight,8,17,18,22,23,24,26,34,36,37 and studies with a low statistical weight9,14,20,21,27,29,30,32,33,35; ORs were 0.82 (95% CI, 0.73-0.91), 0.68 (95% CI, 0.57-0.80), and 0.78 (95% CI, 0.55-1.11), respectively. Differences in statistical weight between these groups did not account for an appreciable amount of between-study heterogeneity (2 populations; I2 = 39%; 95% CI, 0-81%; P = .19). The exclusion of the largest single study,31 which accounted for 9.9% of the statistical weight among these 29 populations, increased the OR only slightly, to 0.75 (95% CI, 0.67-0.84). A funnel plot provided no strong evidence of publication bias. A Begg test for publication bias and an Egger test for small study bias both yielded statistically nonsignificant results (P = .97 and P = .17, respectively). The only study for which no published estimate could be obtained, that of Burke et al,11 observed an inverse association between birth weight and type 2 diabetes risk, with an OR of 2.1 (95% CI, 1.3-3.4) in participants with a birth weight of less than 6.5 lb (2.95 kg) compared with participants with a birth weight of 6.5 to 8.5 lb (2.95-3.86 kg).
Shape of Association Between Birth Weight and Type 2 Diabetes
We were able to examine the shape of the birth weight–type 2 diabetes association in 17 studies for which the authors provided individual participant data8,9,18,20,21,22,23,24,25,27,29,30,31,32,33,35,37 and in 1 other large study, the Health Professionals Follow-up Study,13 using published data. In 2 of these studies, the Nurses' Health Study31 and the Health Professionals Follow-up Study,13 birth weight was only available in grouped categories. The inverse association between birth weight and type 2 diabetes risk appeared graded in all studies, particularly at birth weights of 3 kg or less (Figure 3). At higher levels of birth weight, the association still appeared to be inverse, though the degree of imprecision in the estimate meant that it was not possible to exclude a modest positive association.
Results are based on individual participant data from 16 studies8,9,18,20,21,22,23,24,25,27,29,30,32,33,35,37 and data from the Nurses' Health Study31 (NHS) and the Health Professionals Follow-up Study (HPFS).13 Data from the NHS and the HPFS are presented separately because birth weight data were available only in grouped form. Mean birth weights for each group were calculated directly from data for the 16 studies, based on estimates provided in the published report (NHS),31 and estimated using published data on birth weight distribution (HPFS).13
In a meta-analysis of the 16 studies with individual, ungrouped birth weight data, the OR of type 2 diabetes associated with a 1-kg increase in birth weight at birth weights above 3 kg was 0.85 (95% CI, 0.68-1.05). In a corresponding analysis of birth weights above 4 kg, the association was positive, though with wide CIs; the OR was 1.35 (95% CI, 0.67-2.72).
We also examined the shape of the birth weight–type 2 diabetes associations in the 2 populations consisting of native North Americans15,28 and in the associated Saskatchewan general population study, which also showed an overall positive birth weight–diabetes association.15 All 3 showed strong U-shaped associations, which were dominated by marked increases in diabetes risk among participants with birth weights of more than 4 kg (Figure 4). Appreciably U-shaped associations were not observed for any of the other populations studied.
Results are based on data from Pima Indians,28 native North Americans,15 and the Canadian general population.15 Mean birth weights for each group were calculated directly from data for the native North Americans15 and the Canadian general population15 and estimated using published data on birth weight distribution (Pima Indians).28
Exclusion of Macrosomic Participants
We examined the influence of excluding macrosomic participants (with birth weights of more than 4 kg) on the strength of the birth weight–type 2 diabetes association. In 21 studies of populations other than native North Americans (but including the Saskatchewan general population study),6,8,9,10,15,16,18,19,20,21,22,23,24,25,27,29,30,31,32,35,37 the exclusion of macrosomic participants increased the OR relating birth weight and type 2 diabetes by a small amount, from 0.80 (95%CI, 0.72-0.90) to 0.78 (95% CI, 0.69-0.88) (Table 2). In 11 studies that had data both on macrosomia and maternal diabetes,8,12,13,14,18,25,29,31,35,36,37 the exclusion of both macrosomic participants and those with maternal diabetes appeared to strengthen the inverse association between birth weight and type 2 diabetes appreciably, from 0.81 (95% CI, 0.77-0.86) to 0.67 (95% CI, 0.61-0.73). In the 2 native North American populations,15,28 the exclusion of participants with birth weights greater than 4 kg effectively abolished the positive association between birth weight and type 2 diabetes described earlier (pooled OR after exclusion of macrosomic patients, 0.93; 95% CI, 0.76-1.13).
Table 2. Odds Ratios of Type 2 Diabetes Associated With a 1-kg Increase in Birth Weight, Showing Effects of Adjustment and Exclusion
Adjustment for Socioeconomic Status and Body Mass Index
We examined the effect of adjustment for adult socioeconomic status in 19 studies with available data6,9,10,14,16,18,19,22,23,24,25,27,29,30,31,32,35,36,37; such data were not available for either of the native North American populations. Adjustment for socioeconomic status had no material effect on the strength of the birth weight–type 2 diabetes association, which was 0.77 (95% CI, 0.70-0.84) before adjustment and 0.78 (95% CI, 0.72-0.84) after adjustment (Table 2). Examining the effect of social class adjustment in the presence of adult body mass index yielded very similar results (data not presented). We examined the effect of adjustment for adult body mass index in 25 studies with available data6,8,9,10,14,16,17,18,19,20,21,22,23,24,25,26,27,29,30,31,32,33,35,36,37; again, data were not available for native North American populations. Such adjustment increased the strength of the inverse association between birth weight and type 2 diabetes from an OR of 0.76 (95% CI, 0.70-0.82) to 0.70 (95% CI, 0.65-0.76). In 16 studies with individual participant data including exact (rather than grouped) birth weight data,8,9,18,20,21,22,23,24,25,27,29,30,32,33,35,37 we examined the strength of the birth weight–type 2 diabetes association in 5 equal current body mass index groups. The inverse association between birth weight and type 2 diabetes diminished slightly in strength with increasing body mass index. The ORs of type 2 diabetes per 1-kg increase in birth weight were 0.68 (for lowest body mass index), 0.72, 0.78, 0.74, and 0.83 (for highest body mass index). A test for trend in the strength of the birth weight–type 2 diabetes association across body mass index groups was statistically significant (χ21 test for trend, P = .13).
Analyses of Study Quality and Other Factors
Analyses of study quality were carried out in 29 populations, excluding the 2 populations based on native North American participants. In 22 of these 29 study populations, birth weight was ascertained from information recorded at the time of birth (Table 1). However, the strength of the birth weight–type 2 diabetes association did not differ appreciably between these 22 study populations (OR, 0.75; 95% CI, 0.65-0.88) and the 7 study populations in which birth weight was based exclusively on recall by study participants (OR, 0.78; 95% CI, 0.70-0.87) (test for heterogeneity, χ21 = 0.13; P = .72). In 13 study populations, type 2 diabetes was ascertained by a combination of blood glucose measurements and physician diagnosis; in 10 populations it was based on physician diagnosis alone and in 6 populations on blood glucose measurements alone.
However, the strength of the birth weight–type 2 diabetes association did not differ appreciably among the 3 groups (OR for combined methods, 0.76 [95% CI, 0.66-0.88]; OR for physician diagnosis alone, 0.76 [95% CI, 0.67-0.88]; and OR for blood glucose measurement alone, 0.79 [95% CI, 0.58-1.06]) (test for heterogeneity, χ22 = 0.04; P = .98). There was no evidence that the birth weight–type 2 diabetes association was related to variations in response rates (either when based on participation rates in populations recruited in adult life or follow-up rates in populations followed from birth). The estimates of the birth weight–type 2 diabetes association were little affected by the exclusion of 4 study populations in which estimates were extracted from published articles13,17,26,34; the OR in the remaining 25 study populations was 0.79 (95% CI, 0.71-0.87).
COMMENT
In most of the middle-aged and older populations studied in this review, there were inverse, graded, and independent associations between birth weight and risk of type 2 diabetes. These associations were not dependent on adjustment for current body size nor confounded by adult socioeconomic status and corresponded to a reduction in type 2 diabetes risk of about one-fifth per 1-kg increase in birth weight. Birth weight–type 2 diabetes associations among native North American populations tended to be positive. The present review includes data from almost all relevant published studies, particularly for the main analyses, and examined the effects of adjustments and exclusions in a systematic way. Data on most covariates appeared reliable, though information on gestational age was limited to term/preterm in many studies, and few studies distinguished between gestational diabetes and other forms of maternal diabetes. Birth weight was ascertained directly from birth records in most reports; only a minority depended on long-term participant recall of birth weight. Individual participant analyses examined both the overall strength and shape of associations. The impact of excluding macrosomic participants was systematically examined, though the cutoff for macrosomia (4 kg) may have been high for populations with low average birth weights (for example, the Indian populations).
Two previous reviews have examined the association between birth weight and subsequent diabetes risk. The review by Newsome et al42 was essentially qualitative and provided no information on the shape, strength, or independence of the birth weight–diabetes association. The more recent review by Harder et al43 included only 14 relevant studies and reported that the birth weight–diabetes association was U-shaped, with increases in type 2 diabetes risk at both ends of the distribution. Our analyses are consistent with those of Harder et al in suggesting that low birth weights (<2.5 kg) are associated with an increase in type 2 diabetes risk. However, our results emphasize that the inverse pattern of association between birth weight and type 2 diabetes is the dominant one in most populations and extends to higher birth weights (at least to 3 kg), though we cannot exclude a modest positive birth weight–type 2 diabetes association at higher birth weights (>4 kg). This would be biologically plausible given the recognized association of both prepregnancy type 2 diabetes and gestational diabetes to macrosomia.28
The 2 studies of native North American populations show overall positive birth weight–type 2 diabetes associations. These populations have exceptionally high prevalences of type 2 diabetes and obesity from early ages, coupled with an exceptionally high population prevalence of gestational diabetes, affecting more than 10% of pregnancies both in the Pima Indians53 and the Saskatchewan native North American population.54 Earlier reports suggest that gestational diabetes explains most if not all of the excess in type 2 diabetes risk observed at birth weights of 4.5 kg or more in Pima Indians.28 The only other population with a robust positive birth weight–type 2 diabetes association in the current review included young adult Canadians from Saskatchewan who were predominantly of white European origin.15 The reason for this exceptional finding remains uncertain. This study population had an appreciable prevalence of macrosomia (8.8%), though the prevalence of maternal diabetes in this population was not appreciably raised (3%-4%)54 and there was less than 5% native North American admixture.55 The results of the meta-regression analyses examining the influence of age at outcome and year of birth on the birth weight–type 2 diabetes association suggest that the positive association in this population of young adults may reflect a wider pattern of increasingly positive birth weight–type 2 diabetes associations among younger populations from later birth cohorts. However, published data on the pattern of birth weight–type 2 diabetes associations from young adult populations other than native North Americans remain too limited at present to establish whether this is definitely the case.
Earlier reviewers have suggested that inverse associations between birth weight and adult chronic disease risk might reflect confounding, particularly by socioeconomic status.56,57,58 However, in the current review, adjustment for adult socioeconomic status had no appreciable impact on the birth weight–type 2 diabetes association, a finding similar to that for the association between birth weight and coronary heart disease.59 Individual socioeconomic markers often fail to capture all dimensions of social position across the life course, so residual socioeconomic confounding remains possible. However, individual studies examining the impact of adjustment for childhood socioeconomic status21,25,31 or parental socioeconomic status6,12,19,24 showed no evidence of confounding by these markers. Maternal smoking in pregnancy could be a confounder, being associated with low birth weight60,61 and with risk of offspring smoking in later life,62 which in turn is related to risk of type 2 diabetes.63 None of the individual studies reviewed here reported the effect of adjustment for maternal smoking, which may be an important issue for future investigation.
The potential importance of fetal growth and nutrition in the prevention of chronic diseases has been much debated.64,65,66,67 Type 2 diabetes has a multifactorial etiology, including strong contributions from adult obesity and lifestyle as well as genetic factors,4,5 and, therefore, offers scope for prevention by reduction of several important exposures. The public health implications of the inverse association between birth weight and type 2 diabetes depend on the precise nature of the underlying causal exposure and its amenability to change. If birth weight is itself the causal exposure, the public health impact of birth weight modification (based on the results of the current review) could be worthwhile, though modest. Populationwide interventions increasing birth weight (mostly through changes in maternal nutrition or maternal smoking habit) have generally increased birth weight by up to 100 g,68 with larger increases (up to 200 g) in populations with marginal nutrition.69 Such changes could translate into reductions in type 2 diabetes risk of the order of 5% to 10%. Although it is possible that low birth weight has a particularly strong influence on type 2 diabetes risk among overweight and obese adults6,70 this possibility was not supported by the individual participant data analyses conducted within the present review. If the relevant causal exposure were not birth weight itself but an underlying disturbance of fetal health or nutrition, the public health impact could be greater. Results of the Dutch Hunger Winter Studies, showing that extreme maternal undernutrition had metabolic effects independent of impact on birth weight and other measures of birth size,71,72 are consistent with this possibility. However, relevant underlying exposures have still to be identified in practice and their influence on type 2 diabetes risk demonstrated.
The present review provides strong evidence that most middle-aged and older populations studied in the later 20th century have inverse birth weight–type 2 diabetes associations. However, whether this pattern will persist in the future remains uncertain. The secular increases in prevalence of adult overweight and obesity now occurring in many countries73 will lead to increased risks of type 2 diabetes and, importantly, to an increasing prevalence of abnormal glucose tolerance during gestation.1 Both gestational diabetes and lesser degrees of hyperglycemia are associated with increased birth weight and with increased offspring risk of type 2 diabetes.53,74,75,76 It is therefore possible that the positive or U-shaped birth weight–type 2 diabetes associations observed in native North American populations with high maternal diabetes prevalence and in 1 other population of young adults will become an increasingly common pattern. This is borne out by recent findings in Taiwanese children and adolescents growing up in a setting where both obesity and type 2 diabetes are common from an early age77 and in whom the birth weight–type 2 diabetes association is strongly U-shaped, further emphasizing the influence of maternal diabetes on the birth weight–type 2 diabetes association in contemporary populations.78 The results of the present review suggest that maternal diabetes has a strong influence on the birth weight–type 2 diabetes association and raise the possibility that the birth weight–type 2 diabetes association is less strongly inverse at younger ages. These observations suggest that high, rather than low, birth weight could become an increasingly important influence on type 2 diabetes risk in the future. More data on the contemporary associations between birth weight and type 2 diabetes in young adults from a range of different populations are needed to inform strategic priorities for the early prevention of type 2 diabetes over the next generation.
Author Contributions: Dr Whincup had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Whincup, Kaye, Owen, Huxley, Cook.
Acquisition of data: Whincup, Kaye, Owen, Huxley, Cook, Anazawa, Barrett-Connor, Bhargava, Birgisdottir, Carlsson, de Rooij, Dyck, Eriksson, Falkner, Forsén, Grill, Gudnason, Hulman, Hyppönen, Jeffreys, Lawlor, Leon, Minami, Mishra, Osmond, Power, Rich-Edwards, Roseboom, Sachdev, Syddall, Thorsdottir, Vanhala, Wadsworth, Yarbrough.
Analysis and interpretation of data: Whincup, Kaye, Owen, Huxley, Cook, Anazawa, Barrett-Connor, Bhargava, Birgisdottir, Carlsson, de Rooij, Dyck, Eriksson, Falkner, Forsén, Grill, Gudnason, Hulman, Hyppönen, Jeffreys, Lawlor, Leon, Minami, Mishra, Osmond, Power, Rich-Edwards, Roseboom, Sachdev, Syddall, Thorsdottir, Vanhala, Wadsworth, Yarbrough.
Drafting of the manuscript: Whincup, Kaye, Owen, Huxley, Cook.
Critical revision of the manuscript for important intellectual content: Whincup, Kaye, Owen, Huxley, Cook, Anazawa, Barrett-Connor, Bhargava, Birgisdottir, Carlsson, de Rooij, Dyck, Eriksson, Falkner, Forsén, Grill, Gudnason, Hulman, Hyppönen, Jeffreys, Lawlor, Leon, Minami, Mishra, Osmond, Power, Rich-Edwards, Roseboom, Sachdev, Syddall, Thorsdottir, Vanhala, Wadsworth, Yarbrough.
Statistical analysis: Whincup, Kaye, Owen, Cook, Carlsson, de Rooij, Dyck, Mishra, Osmond, Rich-Edwards, Thorsdottir.
Obtained funding: Whincup, Owen, Cook.
Administrative, technical, or material support: Whincup, Kaye, Owen, Cook.
Study supervision: Whincup, Owen, Huxley, Cook.
Financial Disclosures: None reported.
Funding/Support: This work was supported by a research grant from Diabetes UK (grant RD 05-0003099), with additional support from the British Heart Foundation (grant PG/04/072).
Role of the Sponsor: The agencies that supported this research did not have any role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; or preparation, review, or approval of the manuscript.
Additional Contributions: We thank the following individuals who helped to provide data for this research: Doris Campbell, MD, University of Aberdeen, Aberdeen, Scotland; Paul J. Cascagnette, BSc, Saskatchewan Health Quality Council, Saskatoon, Saskatchewan, Canada; Heather Clark, MSc, Dugald Baird Centre, Aberdeen, Scotland; Diane Feskanich, ScD, Channing Laboratory, Department of Medicine, Brigham and Women's Hospital and Harvard Medical School, Boston, Massachusetts; Catherine Law, FFPH, MRC Centre of Epidemiology for Child Health, Institute of Child Health, University College London, London, England; Jie Mi, MD, Department of Epidemiology, Capital Institute of Pediatrics, Beijing, China; and Alistair Shiell, PhD, MRC Epidemiology Resource Centre, Southampton, England. Mr Cascagnette received financial compensation for his contribution; the other individuals listed did not receive any compensation for their contributions to the study.
REFERENCES
- 1.
- 2.
- 3.
- 4.
- 5.
- 6.
- 7.
- 8.
- 9.
- 10.
- 11.
- 12.
- 13.
- 14.
- 15.
- 16.
- 17.
- 18.
- 19.
- 20.
- 21.
- 22.
- 23.
- 24.
- 25.
- 26.
- 27.
- 28.
- 29.
- 30.
- 31.
- 32.
- 33.
- 34.
- 35.
- 36.
- 37.
- 38.
- 39.
- 40.
- 41.
- 42.
- 43.
- 44.
- 45.
- 46.
- 47.
- 48.
- 49.
- 50.
- 51.
- 52.
- 53.
- 54.
- 55.
- 56.
- 57.
- 58.
- 59.
- 60.
- 61.
- 62.
- 63.
- 64.
- 65.
- 66.
- 67.
- 68.
- 69.
- 70.
- 71.
- 72.
- 73.
- 74.
- 75.
- 76.
- 77.
- 78.
