Caffeine is the world’s most widely consumed central nervous system stimulant. It occurs naturally or is added to foods and beverages, with coffee and tea as the most common and major sources.1 After ingestion, caffeine is readily absorbed into the bloodstream and distributed to the tissues. It is metabolised in the liver by the microsomal cytochrome P450.2 During pregnancy, elimination of caffeine is prolonged and it rapidly passes all biological membranes, including the blood-brain and placenta barriers, resulting in exposure of the fetus.3 A maximum intake level of caffeine for pregnant women has been stipulated by several authorities, most of which agree that it should not exceed 200 mg/day, based on the evidence of its adverse effects on miscarriage rates and fetal growth restriction.1 4 The negative effects of caffeine consumption during pregnancy on fetal growth have been well documented in epidemiological studies, including the Norwegian Mother and Child Cohort Study (MoBa).5 In a recent meta-analysis, the highest, compared with the lowest, maternal caffeine intake level was associated with a 38% increased risk of low birth weight (<2.5 kg).6
Fetal growth and growth in infancy are important determinants for the development of obesity and for long-term cardiometabolic health.7–9 Excess infant growth programmes later obesity, fat mass and risk of adult disease, independent of intrauterine growth.10–15 The prevalence of metabolic disorders, including obesity, cardiovascular disease and type 2 diabetes is rapidly growing across the globe, with the number of obese people risen worldwide from 105 million in 1975 to 641 million in 2014.16 This trend indicates that the probability of reaching the WHO global obesity target, of no rise in obesity by 2025, is close to zero.16There is compelling human and animal evidence supporting the ‘fetal programming’ hypothesis, according to which in utero exposures permanently alter an organism’s physiology and metabolism, leading to susceptibility to subsequent disease, including obesity and metabolic disorders, with transgenerational effects.17 18
In utero exposure to caffeine has been related to an increased risk of overweight and higher body fat in childhood, in two previous epidemiological studies.19 20 However, the link between in utero caffeine exposure and excess growth in infancy is yet to be studied, even though excess infant growth is an established risk factor in the aetiology of obesity and cardiometabolic disease.13 15 21 22
Based on our previous findings on the association of prenatal caffeine exposure with fetal growth restriction5 and the fetal programming hypothesis,23 we hypothesised that prenatal caffeine exposure might affect postnatal growth. Thus, the objective of this study was to investigate the associations between maternal caffeine intake in pregnancy and child growth and risk of overweight up to age 8 years in a large prospective population-based cohort.
Our study was conducted within the Norwegian Mother and Child Cohort Study (MoBa), a prospective population-based pregnancy cohort study conducted by the Norwegian Institute of Public Health.24 Pregnant women from all over Norway were recruited during 1999–2008 and 40.6% of the invited women consented to participate. The cohort now includes 114 500 children, 95 200 mothers and 75 200 fathers. Follow-up of the participants, after delivery, have been conducted at 6 months, 18 months, 36 months, 5 years, 7 years and 8 years. Data used in this study are based on V.8 of the quality-assured data files, released for research in February 2014, with linkage to the Medical Birth Registry of Norway. The data collection in MoBa was licensed by the Norwegian Data Inspectorate. All MoBA participants have provided a signed consent form.
After exclusion of multiple gestations, stillbirths, malformations and chromosomal abnormalities, 96 875 live-born singletons remained. Of these, 78 819 pregnant women had answered the Food Frequency Questionnaire (FFQ) developed and validated for MoBa and in use from 2002 and onwards. The eligible study population, with available information on maternal caffeine intake and all relevant covariates, constituted 62 034 mother-child pairs. Our final study population consisted of 50 943 mother-child pairs with additional information on small for gestational age (SGA) and at least one postnatal measurement of weight or length/height. . After 5 years, approximately 40% of the study population returned the questionnaire and had information on weight and height, while the distribution of mothers by caffeine intake level did not differ by follow-up age, meaning that loss to follow-up was not related to maternal caffeine intake in pregnancy.
Maternal caffeine intake estimation in MoBa has been described in detail previously by Sengpiel et al.5 In brief, self-reported intake of 255 dietary items was assessed at pregnancy week 22 with a FFQ developed and validated for MoBa.25 This is a semiquantitative FFQ designed to record dietary habits during the first 4–5 months of gestation. Average, daily caffeine intake was calculated as the aggregated intake (in mg/day) from all available sources, including several types of coffee, black tea, caffeinated soft drinks, energy drinks, chocolate, chocolate milk, and sandwich spread, desserts, cakes and sweets containing cocoa. Online supplementary table 2 includes more details on the estimation of maternal caffeine intake. The median (25th−75th centiles) caffeine intake was 57 mg/day (23–120 mg/day) for the included population and 64 mg/day (25–129 mg/day) for the non-included population with available caffeine information (n=11 091 mothers) (P<0.001 for Mann-Whitney test). We categorised caffeine intake, based on the calculated median as well as national and international recommendations for caffeine consumption during pregnancy, in four levels of caffeine intake: low (0–49 mg/day), average (50–199 mg/day), high (200–299 mg/day) or very high (≥300 mg/day).
Weight and length/height measurements at eleven age points (6 weeks, 3 months, 6 months and 8 months and 1 year, 1.5 years, 2 years, 3 years, 5 years, 7 years and 8 years) were reported. Up to 18 months the reported measurements were as documented in the child’s health card, while for measurements from 2 years to 8 years no specification was provided. Implausible anthropometrics were identified and excluded by separately implementing three different methods: (1) by comparing with the WHO Growth Standards, as a weight-for-age or height-for-age z-score <6 SD below or >6 SD (5 SD for weight) above the mean,26 (2) by identifying measured values with a >|5SD| difference from the predicted value as derived from the Jenss-Bayley growth curve model, and (3) by the conditional growth centiles.27 After exclusion of implausible values, 464 343 and 452 980 measurements of weight and height/length were reported for our study population. Seven repeated measurements per child were available on average, for both anthropometrics
First, we assessed excess infant weight gain by calculating the difference in gender-adjusted WHO weight-for-age z-scores between birth and age 1 year, using reported weights.26 A z-score gain of >0.67 represents an upward crossing of the centile line,28 referred to as excess growth.29
Second, we determined childhood overweight, including obesity, at two preschool-age (3 years and 5 years) and one school-age (8 years) time points, using the International Obesity Task Force criteria.30 Used body mass index (BMI) cut-offs and overweight prevalences are presented in online supplementary table 3.
BMI was derived by growth models. Individual growth trajectories for weight and length/height were obtained by modelling the overall growth from age 1 month to age 8 years, using the Jenss-Bayley growth curve model, a structural growth model based on a basic functional form of growth. This four-parameter, non-linear model is suitable for describing growth of both weight and length/height during infancy and early childhood, up to age 8 years,31 before growth starts to accelerate again at puberty. To assess individual growth trajectories, we applied a mixed-effect approach using the stochastic approximation of expectation-maximisation (SAEM) algorithm.32 33 We then calculated weight and length/height, BMI (weight (kg) divided by squared height (m)), as well as weight and height gain velocities at 14 age points (1 month, 2 months, 3 months, 6 months, 9 months, 12 months and 18 months and 2 years, 3 years, 4 years, 5 years, 6 years, 7 years and 8 years), using the growth model derivatives. These predicted anthropometrics were also assessed as outcomes.
As including birth weight in the model may influence the estimated trajectories, and in order to assess the effect of caffeine on early growth independently of its effect on birth size,5 we did not include birth weight and length in the growth models.
We used logistic regression models to examine associations between maternal caffeine intake in categories and excess growth in infancy and childhood overweight. Low caffeine intake (0–49 mg/day) was the reference group. Similar analysis was performed after modelling caffeine by restricted cubic splines with four knots at centiles 5, 35, 65 and 95, as recommended by Harrell,34 and corresponding to caffeine intakes of 6 mg/day, 34 mg/day, 91 mg/day and 253 mg/day, respectively. The reference level of caffeine intake was set at 50 mg/day, corresponding to the median intake in our study population. The associations were described graphically. Finally, we used mixed-effect linear regression models with random intercept by child and a random slope for age to analyse associations between predicted weight, length/height, BMI, weight and height gain velocities from ages 1 month to 8 years (14 age points: 1 month, 2 months, 3 months, 6 months, 9 months, 12 months and 18 months and 2 years, 3 years, 4 years, 5 years, 6 years, 7 years and 8 years). Covariates’ effects have been models as fixed in the mixed-effect models. All regression models were adjusted for random effects of sibling clusters since some mothers participated with more than one pregnancy.
Logistic and linear mixed models were adjusted for variables related to both maternal caffeine intake and excess growth by bivariate analysis: maternal age, maternal education, parity, prepregnancy BMI, paternal BMI, maternal and paternal smoking during pregnancy, maternal energy intake and nausea/vomiting during pregnancy. Gestational age and child’s gender were also included in the models as a priori covariates Maternal height, paternal weight, paternal alcohol consumption and gestational diabetes (yes/no) were also considered but not included in the final models as they did not meet the criteria. Our main analysis consists of complete case analysis of 38 338 mother-child pairs for the risk of excess growth and of 50 943 mother-child pairs for all other growth outcomes. The cohort attrition due to loss to follow-up was addressed by the use of predicted anthropometric measurements. The correlation between measured and predicted anthropometrics ranged from 0.85 to 0.99 for weight and from 0.95 to 0.98 for length/height (data not shown).
In separate sensitivity analyses, (1) we excluded SGA neonates (SGA was defined as birth weight below the 10th centile, according to population curves as described by Skjaerven et al 35), (2) we excluded smokers during pregnancy, (3) we adjusted for birth weight; only the overweight models and not the excess growth model, because birth weight is included in the excess growth calculation formula, (4) we explored caffeine intake by three main sources (ie, from black coffee, black tea and soda drinks), (5) we excluded very high caffeine consumers and (6) we assessed the association between maternal caffeine intake and childhood overweight, using the measured instead of predicted anthropometric data to define the outcome. Possible interactions with SGA and birth weight were tested with all studied outcomes. Since the associations between the outcomes and the interaction terms were not significant and the inclusion of the interaction term did not modify our results, we have not included these analyses in the manuscript.
Finally, we performed negative control analysis, using paternal caffeine intake as the negative control. Negative control analysis is a suggested method to test for the possibility of unmeasured confounding. We have assumed that there is no direct association between the father’s exposure during the pregnancy period and the child’s outcome, and that the shared confounders are equally associated with the mother’s and the father’s exposures.36 37 We have calculated the caffeine intake of the father using the caffeine concentrations and serving sizes as used for the mother’s calculationsfor five food items: filtered coffee, boiled coffee, espresso coffee, caffeinated soft-drink with sugar or artificially sweetened. Only 16 455 (32%) fathers had available information.
The main analyses were performed with the Stata V.14 statistical software (Stata Corporation, College Station, Texas, USA) and R V.3.2.238 was used for the growth models.
No patients were involved in setting the research question or the outcome measures, nor were they involved in developing plans for design or implementation of the study. No patients were asked to advise on interpretation or writing up of results. There are no plans to disseminate the results of the research to study participants or the relevant patient community.
In our study population, 7.13% (n=3633) and 3.21% (n=1634) of women reported caffeine intake higher than 200 mg/day and 300 mg/day, respectively. The distribution of women not included in the analysis, by caffeine intake level, was similar to the included (low: 43%, average: 46%, high: 8% and very high: 3%). The higher the caffeine intake, the higher the likelihood of a mother being older than 30 years, being multiparous, having a daily energy intake in the upper tertile, being a smoker during pregnancy and not suffering nausea and/or vomiting during pregnancy. Moreover, women with very high caffeine intake were more likely to have low education, have been obese before pregnancy and have partners who were obese and smokers, compared with those consuming less caffeine per day
Paternal median (5th−95th centiles) intake was 193 mg/day (0–493 mg/day), with caffeine from coffee being the main contributor (median: 187 mg/day). Fathers were consuming statistical significantly more caffeine than their partners (P<0.001 for Wilcoxon matched-pairs signed-ranks test). The spearman correlation coefficient between maternal and paternal caffeine intakes was 0.15 (P value <0.0001). However, paternal intake was increasing by increasing levels of maternal intake and 45% of mothers with very high intake were with partners in the highest quartile of caffeine intake.