Pathways Linking Birth Weight and Insulin Sensitivity in Early Adolescence: A Double Mediation Analysis

Abstract

Purpose

We examined pathways linking birth weight, weight gain from 0 to 2 years, and adiposity during childhood with insulin sensitivity in early adolescence.

Methods

Data were from a longitudinal cohort of 630 Quebec white children with a parental history of obesity (Quebec Adipose and Lifestyle Investigation in Youth study). In a subsample of children born at term (n = 395), weight-for-length z score (zWFL) from 0 to 2 years were computed. At 8 to 10 years, the percentage of body fat was assessed using dual energy X-ray absorptiometry. At 10 to 12 years, the Matsuda insulin sensitivity index (ISI) and the homeostasis model assessment for insulin resistance were determined. A linear regression-based approach for mediation analysis was used to estimate the distinct pathways linking zWFL at 0 to 2 years to insulin sensitivity.

Results

Every additional unit in zWFL at birth was associated with a 10% (95% CI, 5.26% to 14.85%) increase in the Matsuda ISI in early adolescence, independently of the weight at 0 to 2 years and in childhood. An indirect effect of zWFL at birth on the Matsuda ISI was also observed but in the opposite direction (−4.44; 95% CI, −7.91 to −1.05). This relation was mediated by childhood adiposity but not by weight gain from 0 to 2 years. The indirect effect of weight gain from 0 to 2 years, via childhood adiposity, also led to lower insulin sensitivity (−4.83%, 95% CI, −7.34 to −2.53). The findings were similar for the homeostasis model assessment for insulin resistance or when restricted to children with appropriate-for-gestational-age birth weights.

Conclusions

A greater birth weight-for-length resulted in improved insulin sensitivity in early adolescence. However, in the presence of excess childhood adiposity, both a greater birth weight and a faster rate of weight gain from 0 to 2 years resulted in lower insulin sensitivity.

Among term infants, both a greater birth weight and faster weight gain from birth to 2 years resulted in lower insulin sensitivity in early adolescence via their effects on childhood adiposity.

Several studies have shown that birth weight is inversely associated with type 2 diabetes (T2D) in adults (1, 2). Studies focusing on children have shown a complex pattern of association between birthweight and risk factors for T2D (3–5). Low birth weight, a marker of fetal undernutrition, has been suggested as a key contributor to the etiology of T2D (6–8). In contrast, other studies have shown that a high birth weight, especially when not followed by a postnatal “catch-down” (9), is associated with subsequent obesity (10), and obesity is an important risk factor for T2D in youth (11). Postnatal patterns of weight gain during infancy and childhood might underlie the distinct associations between birth weight and risk factors for T2D such as insulin resistance (12). The effect of birth weight on risk factors T2D might in fact be different depending on the postnatal weight gain patterns. It might be that specific trajectories of weight from birth and into childhood are associated with lower insulin sensitivity and that other trajectories have more favorable effects on insulin sensitivity. Few studies have considered simultaneously the effects of birth weight, postnatal weight gain from birth to 2 years, and adiposity during childhood on the subsequent risk factors for T2D. A formal mediation analysis has seldom been used to test distinct pathways (12).

The objectives of the present study were to determine whether the association between birth weight and insulin sensitivity during early adolescence (age, 10 to 12 years) is mediated by postnatal weight gain from 0 to 2 years) and/or adiposity during childhood (age, 8 to 10 years) We also sought to determine whether the association between weight gain from birth to 2 years and insulin sensitivity in early adolescence is mediated by adiposity during childhood (Fig. 1).

Figure 1.

Pathways linking birth weight to insulin sensitivity in early adolescence via two potential mediators: the rate of weight gain from birth to 2 y [mediator 1 (M1)] and adiposity during childhood [mediator 2 (M2)]. The indirect effect of birth weight-for-length on outcomes via M1 (rate of weight gain from birth to 2 y) corresponded to (a*e). The indirect effect of birth weight-for-length on outcomes via M2 (adiposity during childhood) corresponded to (b*f). The indirect effect of birth weight-for-length on outcomes via M1 (rate of weight gain from birth to 2 y) and M2 (adiposity during childhood) corresponded to (a*d*f). The indirect effect of M1 (rate of weight gain from birth to 2 y) on outcomes via M2 (adiposity during childhood) corresponded to (d*f). The direct effect of birth weight-for-length on outcomes corresponded to coefficient c. The direct effect of rate of weight gain from birth to 2 y on outcomes corresponded to coefficient e.

Subjects and Methods

The participants were drawn from the Quebec Adipose and Lifestyle Investigation in Youth (QUALITY) study, an ongoing longitudinal investigation of the natural history of obesity and cardiovascular risk factors in white youth. The children were recruited through elementary schools located within three major urban centers in Quebec, Canada: Montreal, Quebec City, and Sherbrooke. The participants were required to be white, to be aged 8 to 10 years at recruitment, and to have both biological parents available to participate in baseline data collection, with at least one parent having obesity [i.e., body mass index (BMI) ≥30 kg/m² and/or waist circumference ≥102 cm in men and ≥88 cm in women]. At baseline, 630 families participated in a clinic visit during which questionnaires were completed, and biological and physiological measurements were obtained. A similar assessment was conducted 2 years later, when the children were aged 10 to 12 years (n = 564). The parents provided written informed consent, and the children provided assent. The ethics review boards of the CHU Sainte-Justine and the Quebec Heart and Lung Institute approved the present study. A detailed description of the study design and data collection methods have been previously reported (13).

For the present analysis, we used a subsample of the QUALITY participants (n = 395) born at term (i.e., 37 to <42 weeks of gestation), with complete anthropometric data available from birth to 24 months (i.e., weight and length measured at birth, at least once from 1 to 12 months old, and at least once from 12.1 to 24 months old), and who had completed a follow-up visit at age 10 to 12 years. The participants included in the analysis were similar to those excluded for all characteristics, except for having a lower likelihood of having been exposed to tobacco in utero (P = 0.0097; Supplemental Table 1). The analyses were performed using data from three time points: (1) 0 to 2 years; (2) 8 to 10 years; and (3) 10 to 12 years of age.

Measurements at 0 to 2 years

Health booklets are issued to all Quebec parents of newborn children; growth and other health-related information is subsequently entered by health professionals (nurses or physicians) during routine medical visits. From these, the birth weight and length, gestational age, and measurements of weight and length collected at 1 to 24 months of age were extracted. The participants were required to have a minimum of three time points with measurements during the first 24 months of life (median, 6; range, 3 to 10 entries). The measures of weight and length were transformed to sex-specific weight-for-length z score (zWFL) using the World Health Organization reference values (14). Individual slopes (rate of increase or decrease) for zWFL from birth to 24 months were used to estimate the early postnatal weight gain. Simple linear regressions were fitted to measurements of each participant, with age in months as the independent variable and zWFL as the dependent variable. The slope variables thus represent the estimated rate of zWFL gain per month during the first 2 years of life. To facilitate interpretation of subsequent regression analyses, we rescaled the variable such that one unit corresponded to a change of 0.04 zWFL unit per month, equivalent to approximately one SD (0.96 z score) during a 24-month period. The latter also corresponded to a biologically plausible rate of weight gain in the first 2 years of life.

Finally, size at birth was determined from Canadian reference values for gestational age and sex-specific percentiles: small birth weight for gestational age was defined as a birth weight <10th percentile, large birth size for gestational age as a birth weight > 90th percentile, and appropriate birth size for gestational age (AGA) otherwise (15).

Measurements at 8 to 10 years

Body composition was measured using dual-energy X-ray absorptiometry (Prodigy Bone Densitometer System, DF+14664; GE Lunar Corporation, Madison, WI) when the participants were aged 8 to 10 years. Childhood adiposity was estimated using the percentage of body fat mass, calculated as the total fat mass/total body mass × 100. Data on potential confounders were collected in questionnaires completed by the parents. These include gestational diabetes, history of maternal smoking or hypertension during pregnancy, mother’s age at conception of the child, and whether the child had been breastfed (never vs ever). The maternal and paternal BMI were computed from weight and height measured using standard protocols during the baseline visit, when the children were aged 8 to 10 years (13).

Measurements at 10 to 12 years

All participants underwent a 2-hour oral glucose tolerance test (OGTT) after a 12-hour overnight fast. Blood samples were collected during fasting and at 30-, 60-, 90-, and 120-minute intervals after an oral glucose dose of 1.75 g/kg of body weight (maximum dose, 75 g). The blood samples were centrifuged, separated into aliquots, and stored at −80°C until analysis. Plasma insulin was measured using the ultrasensitive Access immunoassay system (Beckman Coulter, Inc.), which has no cross-reactivity with proinsulin or C-peptide (16). The plasma glucose concentrations were computed using the Beckman Coulter Synchron LX20 automat (LX20; Beckman Coulter, Brea, CA) and the glucose oxidase method. Analyses were performed in batches at the Centre Hospitalier Universitaire Sainte-Justine Clinical Biochemistry laboratory twice monthly. A fasting-based index [homeostasis model assessment for insulin resistance (HOMA-IR)] and an OGTT-derived index [Matsuda insulin sensitivity index (ISI)] were used to measure insulin sensitivity. The HOMA-IR index was used as a measure of fasting insulin resistance (17). The Matsuda ISI, derived from the OGTT, was computed as follows: 10,000/[(fasting glucose × fasting insulin) × (mean OGTT glucose × mean OGTT insulin)] (18). Both measures of insulin sensitivity have previously been validated in children of similar ages as those in the QUALITY study (19).

The participants’ physical activity was assessed during a 7-day period using an Actigraph LS 7164 activity monitor (Actigraph, Pensacola, FL). Accelerometry data were downloaded as 1-minute epochs and underwent standardized quality control and data reduction procedures (20). The data from the participants with a minimum of 4 days and a minimum of 10 hours of wear time were retained for analyses (21). Nonwear time was defined as any period of ≥60 minutes of 0 counts, accepting 1 to 2 consecutive minutes in which the count values were >0 and ≤100 (22). Moderate-to-vigorous physical activity was computed by summing the total minutes spent daily in moderate and vigorous physical activity and averaging for the total number of valid days of wear (23). Screen time was assessed using an interviewer-administered questionnaire, documenting the daily hours of television viewing and leisure computer or video game use on a typical weekday and weekend day. The average daily hours of screen time were computed. Finally, the pubertal development stage was assessed by a trained nurse by visual inspection of the genitalia and pubic hair development in boys and breast and pubic hair development in girls using the 5-stage Tanner scale (24, 25). Pubertal development was dichotomized as prepubertal (Tanner 1) vs puberty initiated (Tanner >1).

Statistical analysis

The mean ± SD or proportions were used to describe the participants. The insulin outcome variables were transformed (100 × natural logarithm of variable) to normalize their distribution. Thus, β-coefficients represent the percentage of change in each of the dependent variables for a 1-unit increase in the independent variable (26). A linear regression-based approach for double mediation analysis (27) was used to estimate the adjusted direct and indirect effects of zWFL at birth on insulin sensitivity in early adolescence by the rate of zWFL gain from birth to 2 years (mediator 1) and the percentage of fat mass during childhood (mediator 2). Therefore, we estimated six linear regression models to test associations between (1) the birth zWFL and rate of zWFL gain from birth to 2 years; (2) birth zWFL and percentage of fat mass during childhood, (3) birth zWFL and insulin sensitivity in early adolescence, (4) rate of zWFL gain from birth to 2 years and percentage of fat mass during childhood, (5) rate of zWFL gain from birth to 2 years and insulin sensitivity, and, finally, (6) percentage of fat mass during childhood and insulin sensitivity (Fig. 1). The models were adjusted for potential confounders in the association between the main exposure (i.e., birth zWFL) and outcomes and for confounders of the exposure to mediator association and mediator to outcome association. The control variables were sex, exact age at outcome measurement, in utero exposure to gestational diabetes, maternal hypertension and tobacco use during pregnancy, whether the child had ever been breastfed, and mean daily minutes of moderate-to-vigorous physical activity and mean daily hours of screen time at 10 to 12 years. The models were also adjusted for the percentage of fat mass at 8 to 10 years (except in model 2). Further adjustment for Tanner stage did not change our findings (Supplemental Table 2). Given that pubertal development was likely not a confounder and that it could be in the causal pathway for the associations of interest (i.e., if children who grow more rapidly experience earlier puberty), we opted to not include the Tanner stage in our final models. The parameter estimates from models 1 to 6 were combined to estimate the indirect and direct effects (Fig. 1) (27). A nonparametric bootstrapping method (with n = 5000 bootstrap resamples) was used to obtain standard errors using the percentile method, and 95% CIs were determined using the 2.5th and 97.5th percentile of parameter estimates distribution.

Sensitivity analyses were conducted. First, to account for missing data, multiple imputation via chained equations was used to create 20 imputed data sets (using Proc MI), and the results from case complete analyses were compared with those from the imputed analyses. Second, given that mechanisms linking early growth and later risk factors for T2D might be different for subjects born with extreme birth weights, we repeated all analyses on a sample restricted to children born with AGA (n = 318). Third, we adjusted associations for the mother’s and father’s BMI. All analyses were performed in SAS, version 9.4 (SAS Institute, Cary, NC).

Results

The characteristics of participants are listed in Table 1. The mean ± SD birth weight and birth length were 3.6 ± 0.5 kg and 51.4 ± 2.3 cm, respectively. The correlation between the birth zWFL and rate of zWFL gain from birth to 2 years was −0.54; 20% of the participants had a negative zWFL slope (catch-down) from birth to 2 years. Of the children, 7% and 13% were born small birth weight for gestational age and large birth size for gestational age, respectively. At 8 to 10 years, 54% of the children were of normal weight, 22% were overweight, and 24% had obesity. The observed prevalence of overweight and obesity in our sample reflects the greater prevalence of excess weight associated with a parental history of obesity. The insulin sensitivity outcomes were measured when children were aged an average of 11.6 ± 0.9 years.

Table 1.

Characteristics of 395 Participants Included in Present Analysis, QUALITY Cohort Study

Variable	Value	% (n) Missinga
Children at age 0–2 y
Male sex, % (n)	57.0 (225)
Birth weight, kg	3.6 ± 0.5
Birth length, cm	51.4 ± 2.3
Birth zWFL	−0.3 ± 1.3
Postnatal growth estimated from slope of zWFL from birth to 24 mo	0.06 ± 0.08b
Postnatal growth slope <0, % (n)	19.7 (78)
Gestational age, wk
Median	40.0
IQR	39.0–40.0
Birth size, % (n)
SGA	6.6 (26)
AGA	80.5 (318)
LGA	12.9 (51)
Ever breastfed, % (n)	80.5 (313)	1.5 (6)
Children at age 8–10 y
Child’s age, y	9.6 ± 0.9
Puberty initiated (Tanner >1), % (n)	21.3 (84)
Total body fat mass, %	25.9 ± 10.7	1.3 (5)
BMI category, % (n)
Normal weight	54.2 (214)
Overweight	21.5 (85)
Obese	24.3 (96)
MVPA, min/d		11.6 (46)
Median	48.0
IQR	30.8–64.6
Screen time, h/d		0.5 (2)
Median	2.3
IQR	1.3–3.7
Children at age 10–12 y
Child’s age, y	11.6 (0.9)
Puberty initiated (Tanner >1), % (n)	66.8 (264)
BMI category, % (n)
Normal weight	50.1 (198)
Overweight	24.1 (95)
Obese	25.8 (102)
MVPA, min/d		20.5 (81)
Median	38.8
IQR	25.0–54.4
Screen time, h/d		0.3 (1)
Median	2.9
IQR	1.9–4.4
HOMA-IR		2.5 (10)
Median	1.3
IQR	0.8–1.9
Matsuda ISI		4.6 (18)
Median	6.3
IQR	4.4–9.5
Maternal characteristics
History of gestational diabetes, % (n)	17.2 (68)
History of hypertension during pregnancy, % (n)	9.9 (39)
Maternal smoking during pregnancy, % (n)	12.4 (49)	0.3 (1)
Mother’s BMI when child aged 8–10 y, kg/m2	29.2 ± 6.5	0.3 (1)

Abbreviations: IQR, interquartile range; LGA, large birth size for gestational age; MVPA, moderate-to-vigorous physical activity; SGA, small birth weight for gestational age.

^aA greater proportion of missing data were observed for measures of moderate-to-vigorous physical activity; given that it was considered a valid measure, participants were required to have a minimum of 4 d with a minimum of 10 h of accelerometry wear time.

^bOn average, participants gained 0.06 zWFL per month between birth and 2 y of age.

Indirect and direct effects on measures of insulin sensitivity are listed in Table 2. With respect to the indirect effects, childhood adiposity, but not the rate of zWFL gain from birth to 2 years, was found to mediate the association between zWFL at birth and insulin sensitivity. Every additional unit of zWFL at birth resulted in a 4.74% increase (95% CI, 1.13% to 8.41%) in HOMA-IR and a 4.44% decrease (95% CI, −7.91 to −1.05) in the Matsuda ISI in early adolescence via its effect on childhood adiposity. In contrast, the direct effects of zWFL at birth on insulin sensitivity, adjusted for weight gain from birth to 2 years and childhood adiposity, were in the opposite direction, such that a higher zWFL at birth resulted in a decrease in HOMA-IR (−6.81; 95% CI, −11.89 to −1.87) and an increase in the Matsuda ISI (9.99; 95% CI, 5.26 to 14.85) in early adolescence (Table 2). A double-mediated indirect effect of birth zWFL via both the rate of zWFL gain from birth to 2 years and adiposity in childhood was not observed. Although the rate of zWFL from birth to 2 years was not directly associated with the outcomes, indirect effects were observed. Every additional increase in zWFL from birth to 2 years of age resulted in a 5.16% increase (95% CI, 2.72% to 7.78%) in the HOMA-IR and a 4.83% decrease (95% CI, −7.34 to −2.53) in the Matsuda ISI in early adolescence via its effect on adiposity in childhood. When restricting the analysis to children who were born with AGA, the associations remained very similar, with the exception of the direct effect of birth zWFL on HOMA-IR, which became nonstatistically significant (−4.34; 95% CI, −9.93 to 1.20; Table 3).

Table 2.

Indirect and Direct Effects (β-Coefficients^a and 95% CIs) of Birth zWFL and Rate of zWFL Gain From Birth to 2 Years on Measures of Insulin Sensitivity

Variable	Case Complete Analysis		Analysis on Imputed Data
	HOMA-IR (n = 295)	Matsuda ISI (n = 289)	HOMA-IR (n = 383)	Matsuda ISI (n = 371)
Indirect effects
Birth weight via postnatal weight gain (M1)	0.65 (−1.88 to 3.18)	−1.99 (−4.66 to 0.56)	0.13 (−1.94 to 2.41)	−1.76 (−4.12 to 0.26)
Birth weight via childhood adiposity (M2)	4.74 (1.13 to 8.41)	−4.44 (−7.91 to −1.05)	4.60 (1.61 to 7.62)	−4.21 (−7.05 to −1.43)
Birth weight via postnatal weight gain (M1) and childhood adiposity (M2)	1.25 (−2.77 to 5.39)	−2.54 (−6.60 to 1.23)	0.76 (−2.71 to 4.24)	−2.33 (−5.70 to 0.92)
Postnatal weight gain via childhood adiposity (M2)	5.16 (2.72 to 7.78)	−4.83 (−7.34 to −2.53)	4.87 (2.66 to 7.17)	−4.46 (−6.71 to −2.38)
Direct effects
Birth weight	−6.81 (−11.89 to −1.87)	9.99 (5.26 to 14.85)	−5.03 (−9.40 to −1.14)	7.51 (3.86 to 11.64)
Postnatal weight gain	−0.84 (−4.04 to 2.28)	2.50 (−0.67 to 5.74)	−0.18 (−2.97 to 2.35)	2.18 (−0.29 to 5.20)

Effects were adjusted for child’s age at follow-up, sex, moderate-to-vigorous physical activity at follow-up (min/d), screen time at follow-up (h/d), in utero exposure to gestational diabetes, maternal hypertension and tobacco use, and whether the child was ever breastfed.

Abbreviations: M1, mediator 1; M2, mediator 2.

^a β-Coefficients represent the percentage of increase or decrease in the outcome for a 1-unit increase in the corresponding independent variable via the specified mediator for indirect effects or independent of weight at other time points for direct effects.

Table 3.

Indirect and Direct Effects (β-Coefficients^a and 95% CIs) of Birth zWFL and Rate of zWFL Gain From Birth to 2 Years on Measures Of Insulin Sensitivity Among Children Born With AGA

Variable	Case Complete Analysis Among Children Born With AGA		Analysis of Imputed Data Among Children Born With AGA
	HOMA-IR (n = 239)	Matsuda ISI (n = 235)	HOMA-IR (n = 314)	Matsuda-ISI (n = 306)
Indirect effects
Birth weight via postnatal weight gain (M1)	0.007 (−2.92 to 2.92)	−1.48 (−4.31 to 1.30)	−0.72 (−3.37 to 1.71)	−1.11 (−3.66 to 1.25)
Birth weight via childhood adiposity (M2)	5.18 (0.30 to 10.06)	−4.91 (−9.56 to −0.28)	4.43 (0.93 to 8.18)	−4.14 (−7.59 to −0.87)
Birth weight via postnatal weight gain (M1) and childhood adiposity (M2)	0.96 (−4.04 to 6.13)	−2.39 (−7.29 to 2.37)	−0.03 (−3.84 to 4.02)	−1.76 (−5.96 to 1.87)
Postnatal weight gain via childhood adiposity (M2)	5.28 (2.02 to 8.68)	−5.00 (−8.24 to −1.88)	4.65 (1.82 to 7.44)	−4.35 (−7.01 to −1.56)
Direct effects
Birth weight	−4.34 (−9.93 to 1.20)	7.59 (2.53 to 12.94)	−2.43 (−6.83 to 3.15)	5.06 (−0.10 to 8.88)
Postnatal weight gain	−0.03 (−3.73 to 3.56)	1.87 (−1.62 to 5.43)	0.88 (−2.34 to 4.02)	1.41 (−1.74 to 4.56)

Effects were adjusted for child’s age at follow-up, sex, moderate-to-vigorous physical activity at follow-up (min/d), screen time at follow-up (h/d), in utero exposure to gestational diabetes, maternal hypertension and tobacco use, and whether the child was ever breastfed.

Abbreviations: M1, mediator 1; M2, mediator 2.

^a β-Coefficients represent the percentage of increase or decrease in the outcome for a 1-unit increase in the corresponding independent variable via the specified mediator for indirect effects or independent of weight at other time points for direct effects.

A sensitivity analysis, using multiple imputations to account for missing data, resulted in similar findings (Tables 2 and 3). Further adjustment for the effects of the mother’s and/or father’s BMI measured when the children were aged 8 to 10 years did not change our findings (Supplemental Tables 3 and 4).

Discussion

The results of our study have clarified the mechanisms linking birth weight, rate of weight gain from birth to 2 years, adiposity during childhood, and risk factors for T2D in early adolescence among children who were born at term and have a family history of obesity. We found evidence of both a positive direct effect and a negative indirect effect via childhood adiposity of birth zWFL on insulin sensitivity in early adolescence. Although the rate of weight gain adjusted for the growth in length from birth to 2 years was not directly associated with insulin sensitivity, a faster rate of zWFL gain in the first 2 years of life was positively associated with adiposity at age 8 to 10 years, which, in turn, resulted in lower insulin sensitivity 2 years later. The same findings were reached when the analyses were restricted to children who were born with a healthy birth weight (AGA). Thus, among children who were born at term and who had a family history of obesity, a high birth weight was beneficial to insulin sensitivity in early adolescence. However, this was true only if the children remained within their growth percentiles and did not gain excessive weight from birth to 2 years or during childhood.

Our findings are in line with those from other studies reporting a positive association between birth weight and insulin sensitivity, with the associations strengthening after accounting for the contemporaneous weight status (4, 5), with, hence, a favorable direct effect (4, 5, 28). In our study, this protective direct association was found even when we restricted the analysis to participants born with AGA, suggesting that any increase in birth weight confers metabolic protection, independently of any subsequent weight gain. Our study also revealed a distinct pathway through which an increased birth weight might have a deleterious effect on insulin sensitivity. Children born with a higher zWFL and experiencing greater levels of adiposity in childhood were found to have lower insulin sensitivity in early adolescence. This deleterious indirect effect was also found when restricting the analyses to participants born with AGA. These results suggest the possible coexistence of a positive direct effect (independent of weight in the first 2 years of life and in childhood) and a negative indirect effect (via childhood adiposity) of birth weight on insulin sensitivity. This could explain why some studies have reported the lack of a total effect (sum of the direct and indirect effects) for the association between birth weight and risk factors for T2D in pediatric populations (5). A small number of studies have shown a greater birth weight to be associated with disrupted glucose homeostasis, independently of the current weight (29, 30), suggesting that, at least in some populations, a greater birth weight might be associated with an increased risk of T2D. Although a greater birth weight might confer metabolic protection, such as was observed in the present study, exposure to obesogenic postnatal environments could reverse this protection.

Most (3, 28, 31–33), but not all (34), studies have reported associations between an increased rate of postnatal weight gain and lower insulin sensitivity or impaired fasting glucose among children. Studies that examined weight gain at multiple points in time during childhood have shown that the body size measured most proximal to the outcome is more strongly associated with insulin sensitivity (5, 28). Few studies, however, have used formal mediation analysis to examine the direct and indirect effects of early weight gain on insulin sensitivity, with one exception. Slining et al. (12) studied 1409 Filipino participants and reported that faster postnatal weight gain measured in grams per month of weight gain during the first 2 years of life was positively associated with the BMI and waist circumference at 22 years of age, which was positively associated with the HOMA-IR. However, that study only examined the indirect effect of the rate of weight gain from birth to 2 years, via childhood adiposity, without examining the role of the starting point (i.e., the birth weight). Moreover, a substantial proportion of participants in the study by Slining et al. (12) were underweight and/or stunted at birth and at 2 years of age, limiting the generalizability of their findings to Western populations. Despite important differences in the study populations and design, our results are comparable regarding the unfavorable indirect effect of faster weight gain in the first 2 years of life on insulin sensitivity mediated by childhood adiposity.

Most studies examining associations between early weight gain and risk factors for T2D have focused on measurements of weight (e.g., weight-for-age z score or weight in kilograms). We used zWFL, which reflects the body weight in proportion to the attained growth in length and which has been proposed as a better predictor of overweight/obesity than the use of weight alone (35, 36). It has been suggested that a greater lean mass at birth instead of fat mass provides the link between greater birth weight and improved cardiometabolic outcomes (37–39). Newborns can be heavy for genetic reasons (e.g., longer babies), in which case their greater birth weight might be attributable to increased lean mass, which confers lower metabolic risk. Alternatively, they could be heavy because of an increased fat mass (e.g.,in utero exposure to gestational diabetes), which would lead to an increased metabolic risk. The use of weight at birth or weight gain unadjusted for length might yield biased estimates of the association between early weight gain and the risk of T2D.

In sensitivity analyses, we found very similar results when restricting the sample to children born with AGA, and the associations between the birth zWFL and insulin sensitivity outcomes were linear. These findings suggest that associations might be present across the spectrum of birth weights, with no specific threshold effect (40). These findings also suggest that the beneficial direct effect of birth weight on insulin sensitivity and its detrimental indirect effect are not driven by children whose birth weight was at the extremes of the distribution.

Our study sample included white children born at term from families with a relatively high income and education. We were able to estimate the direct effects (i.e., independently of adiposity) and indirect effects (i.e., mediated by the rate of weight gain from birth to 2 years and/or adiposity during childhood). In addition to our analytical approach, the strengths of the present study included the availability of repeated measures of weight and length from birth to 2 years of age and that of adiposity measures from dual energy X-ray absorptiometry scans obtained 2 years before the measurement of insulin sensitivity. Our study had some limitations. First, the weight and length data were obtained from the health booklets and were not collected for research purposes; thus, variability in measurement precision is likely. To mitigate this, the growth charts plotted for each participant were examined visually by two pediatricians and a total of 209 measurements (7%) of weight or height were removed (e.g., physiologically implausible measurements). Second, we did not have measures of body composition from birth to age 2 years. Third, other than breastfeeding, we did not have nutritional data for the infancy period (e.g., age at the introduction of solid foods). Thus, confounding by early nutrition on the associations of interest could not be investigated. Fourth, data on the prepregnancy maternal weight status or weight gain during pregnancy were not available. However, further adjustment of associations for the present maternal weight (i.e., when the child was aged 8 to 10 years) did not change our findings. Fifth, our sample did not include a large enough number of participants born at small birth weight for gestational age or large birth size for gestational age to examine the associations in these subgroups. Mechanisms linking birth weight and insulin sensitivity might be distinct across categories of size for gestational age at birth (6, 9). Finally, given that our sample included white youth with at least one obese parent, our findings might not be generalizable to other groups.

Conclusions

We have shown that in addition to the previously described protective direct effect of birth weight on insulin sensitivity (1), birth weight is also an indirect determinant of diminished insulin sensitivity, via childhood adiposity. Similarly, a faster rate of weight gain from birth to 2 years is an indirect determinant of diminished insulin sensitivity, via childhood adiposity. Both a higher birth weight and faster weight gain in the first 2 years of life, even among children born with AGA, might set the stage for the development of increased adiposity in childhood, which, in turn, results in lower insulin sensitivity and greater insulin resistance. For clinicians, our findings suggest that children with a lower birth weight should be monitored for metabolic complications, just as should children with a higher birth weight or a greater rate of weight gain early in life when followed by excess gain in adiposity during childhood. These efforts might contribute to the prevention of childhood obesity and its deleterious metabolic consequences.

Glossary

Abbreviations:

AGA

appropriate birth size for gestational age

BMI

body mass index

HOMA-IR

homeostasis model assessment for insulin resistance

ISI

insulin sensitivity index

OGTT

oral glucose tolerance test

QUALITY

Quebec Adipose and Lifestyle Investigation in Youth

T2D

type 2 diabetes

zWFL

weight-for-length z score