OUP user menu

Life course body mass index, birthweight and lipid levels in mid-adulthood: a nationwide birth cohort study

Snehal M. Pinto Pereira, Chris Power
DOI: http://dx.doi.org/10.1093/eurheartj/ehs333 1215-1224 First published online: 12 December 2012


Aims Improvement in lipid profiles is an important public health and clinical goal for which a better understanding is needed of biological pathways and influences. Evidence is scant on the role of growth, including trajectories of body mass index (BMI), so we aimed to determine whether particular life stages from birth to adulthood are important for lipid levels in mid-adulthood (45 years).

Methods and results In the 1958 British birth cohort (n = 3927 men; 3897 women), weight and height were recorded at: birth (weight only), 7, 11, 16, 23, 33, and 45 years. Birthweight was inversely associated with triglycerides and in women with total- and non-high-density lipoprotein cholesterol; associations were little affected by adjustment for 7-year BMI. Associations between lipids and BMI strengthened with age, e.g. in women, adult (45-year) triglycerides were elevated by 1.54% (95% confidence interval: 0.87–2.21%) and 3.57% (3.29–3.86%), respectively, per kg/m² higher BMI at 11 and 45 years. Body mass index gain was related to lipids, with strongest associations for the interval between 33 and 45 years, where a kg/m² gain in BMI was associated with ∼0.6% higher total cholesterol and ∼5.3% higher triglycerides. Associations between 45-year BMI and lipids were stronger for those with lowest than highest BMI at younger ages (P for interaction ≤0.05). A long duration of obesity and obesity in childhood but not thereafter were unrelated to adult lipid levels.

Conclusions Our findings from a large population-based cohort highlight detrimental consequences of high adult BMI for lipids as most pronounced for those with a lower BMI at earlier life stages.

  • Body mass index
  • Cardiovascular diseases
  • Cholesterol
  • Growth


Cardiovascular disease (CVD) is a major health problem causing over two million deaths and costing the European Union ∼€192 billion each year.1 Trials have shown that lowering total- and/or low-density lipoprotein (LDL) cholesterol reduces the risk of stroke2 and coronary heart disease.3 Therefore, to reduce the burden of CVD, improvement in lipid profiles is an important goal. To achieve this goal, a better understanding is needed of the biological pathways affecting lipid levels. For some CVD risk factors, e.g. blood pressure,4,5 there have been studies on the role of growth, including trajectories of body mass index (BMI) across the life course, but little has been done for lipids. Lipids track from childhood to adulthood,6 implying that influences on adult cholesterol begin at young ages. Associations between birthweight and later cholesterol levels have been reported, albeit of small magnitude,7,8 and while there is scant research on postnatal growth and subsequent lipid levels, associations have been seen with adult leg length, a marker of pre-pubertal growth.9

Overweight and obesity are associated with abnormal lipid levels in youths10,11 and adults,12,13 yet little attention has been given to changes in body size and subsequent lipids. One study suggested that BMI gains from adolescence were associated with adverse lipid levels in adulthood and fast BMI gains during the pubertal period were associated with lower high-density lipoprotein (HDL) cholesterol levels.14 Most studies investigating change in body size and subsequent lipid levels cover wide time intervals, for example from birth to adulthood, with few intervening measures;15,16 and while effects of catch-up growth have been documented, studies are based on small samples15,17 or long intervals between birth and growth measurement.15 Consequently, it is unclear whether there are sensitive periods of BMI gain for later lipid levels. It is also unknown whether associations between concurrent body size and lipids are modified by body size at younger ages, or whether the duration of obesity from childhood affects adult lipid levels.

Using data from the 1958 British Birth Cohort, we aimed to determine whether there are phases of the life course, from birth to mid-adulthood, that are particularly important for adult lipid levels. Specifically, we assessed whether (i) birthweight was associated with 45-year lipid levels and whether any association was due to catch-up growth from birth to 7 years; (ii) there were particular periods during childhood or adulthood when BMI, or change in BMI, influenced subsequent lipid levels; (iii) the association between current BMI and lipid levels was modified by BMI at younger ages; and (iv) the duration of obesity was associated with lipid levels.


Study sample

The 1958 cohort consists of over 17 000 participants followed-up since birth during 1 week in March 1958 across the UK and contacted at 7, 11, 16, 23, 33, 42, and 45 years.18 At 45 years, 11 971 cohort members were invited to participate in a biomedical survey; the 9377 (78%) respondents were broadly representative of the total surviving cohort;19 7824 respondents had lipid measures.


All 45-year measurements were taken using standardized protocols. Non-fasting venous blood samples were obtained and posted to a central laboratory, where lipid levels were measured. Total cholesterol, HDL cholesterol, and triglycerides were analysed by an autoanalyzer (Olympus AU640, Japan) using enzymatic methods (further details in Supplementary material online). Non-HDL cholesterol was calculated by subtracting HDL cholesterol from total cholesterol. Lipid-lowering medications were identified from packaging of currently prescribed medications.

Birthweight was measured in ounces and converted into kilograms. Gestational age was estimated from the date of the mothers' last menstrual period. Sex-specific birthweight standardized within each gestational week was derived. Height and weight were measured using standard protocols at 7, 11, 16, 33, and 45 years, and were self-reported at 23 years. BMI Body mass index was calculated as weight/height² (kg/m²) at each age. Overweight and obesity were defined in childhood using international BMI cut-offs;20 and in adulthood as ≥25 and ≥30 kg/m², respectively. Age of obesity onset was identified as the first age when BMI was defined as obese; and in a similar manner for overweight onset (details in table footnotes). At 45years waist circumference (cm) was measured midway between the lower ribs and iliac crest.

Covariates for the analysis included: social class at birth and in adulthood, classified using the Registrar General's Social Classification into four levels from professional to unskilled; 42-year education, classified into five categories from none to degree; 42-year smoking, classified as never, ex or current smoker; 45-year alcohol use, measured using the Alcohol Use Disorders Identification Test questionnaire21 and categorized into five levels from none to very heavy drinker; 45-year menopausal status (women only); and 45-year hypertension, defined as systolic blood pressure ≥140 mmHg, diastolic blood pressure ≥90 mmHg or medication for hypertension.22

Statistical analysis

A unit of BMI has different implications in child and adulthood, hence to compare across ages BMI was converted to (sex-specific) standard deviation (SD) scores, i.e. the difference between an individual's BMI and mean BMI, divided by SD of BMI at a given age. A BMI SD score (zBMI) represents an individual's rank on a standardized scale at a particular age. Similarly, we calculated sex-specific birthweight SD scores.

Using linear regression we examined associations between birthweight and 45-year lipids, adding 7-year zBMI to the model and, then including an interaction term between birthweight and 7-year zBMI to assess whether birthweight associations were mediated or modified by catch-up growth. To determine whether there were particular periods in childhood or adulthood when BMI, or change in BMI (i.e. changes in the relative position in the BMI distribution) influenced subsequent lipids we investigated (i) associations with BMI at different ages separately and (ii) BMI change over specific age intervals. First, we examined simple associations between lipids and zBMI at each age. Next we assessed associations between lipids and change in zBMI between each age interval, conditioned on zBMI at the previous age. For example, for the interval of BMI gain 7–11 years: lipid measure = a + bzBMI7 + c(zBMI11 − zBMI7), where c represents mean difference in the lipid level per SD increase in BMI 7–11 years, given 7-year zBMI. To determine whether associations between BMI and lipid levels at 45 years were modified by birthweight or BMI at younger ages, we tested interactions between 45-year zBMI and birthweight or zBMI at each previous age. For significant interactions (P ≤ 0.05), we examined associations between 45-year zBMI and lipid levels stratified by BMI tertiles at the previous age. To further validate whether associations between adiposity and lipid levels at 45 years were modified by adiposity at younger ages, we tested interactions of zBMI at each previous age (or birthweight) with 45-year waist circumference on 45-year lipid levels. For illustrative purposes, we present some results as kg/m² (BMI units) and cm (waist circumference units). All analyses for birthweight were repeated using birthweight for gestational age; results were broadly similar to those for birthweight and hence not shown. To assess the role of duration of obesity, we examined associations between obesity onset and lipids using ‘‘never obese’’ as the reference group. Analyses were repeated for overweight (including obesity) onset.

For all models, lipid levels were log-transformed and multiplied by 100, whereby regression coefficients can be interpreted as symmetric percentage difference in means.23 Ignoring or excluding participants on medication can bias associations.24 For example, excluding participants on treatment can lead to ‘shrinkage bias’, i.e. under-estimation of effects. We found some evidence for such dampening of associations in analyses that excluded those on lipid-lowering drugs (n = 126) compared with the recommended method of correcting for treatment.24 Hence, we present analyses based on corrected values for those on treatment (+25% for total cholesterol; +18% for triglycerides; and −5% for HDL cholesterol25). Because non-fasted blood samples were used, we examined the impact of recent food consumption on our results. In stratified analyses of those eating or drinking >2 or <2 h before blood collection, results were broadly similar to those for all participants combined (Supplementary material online, Tables S1–3). Hence, analyses shown here are based on all participants. We examined the impact of fieldwork factors, including examination month, time of day, and sample delivery time to the laboratory. All showed negligible influences on associations of interest (data not shown).

The analysis consisted of 3927 males and 3897 females with data on lipids. To minimize data loss missing covariates were imputed using multiple imputation chained equations, following current guidelines.26 Imputation models included all model variables plus previously identified key predictors of missingness: cognitive ability and behaviour at 7 years, social class at birth, and in adulthood.19 Regression analyses were run across 10 imputed data sets. Imputed results were broadly similar to those using observed values; only the former are presented. Analyses were sex specific, conducted using STATA (Version 11, Stata Corp) and two-sided testing was performed using a significance level of 0.05.


As the cohort aged, BMI and the variation in BMI increased: e.g. mean (SD) BMI in males at 7 and 45 years was 15.90 (1.51) kg/m² and 27.74 (4.18) kg/m², respectively. Overweight prevalence was fairly constant in childhood, at ∼6% for males, but increased thereafter to 50% at 45 years. Similar patterns were observed for obesity and females (Table 1).

View this table:
Table 1

Life course anthropometric measures and 45-year lipid levelsa

nMean, birthweight(g)/ BMI(kg/m²) (SD)n (%), overweightn (%), obesenMean, birthweight(g)/ BMI(kg/m²) (SD)n (%), overweightn (%), obese
Anthropometric measures
Birth36033416.18 (514.81)35843263.95 (500.43)
7b318515.90 (1.51)205 (6.44)31 (0.97)313415.81 (1.83)275 (8.77)72 (2.30)
11b309717.18 (2.28)194 (6.26)40 (1.29)310217.55 (2.59)285 (9.19)42 (1.35)
16b285420.21 (2.63)177 (6.20)37 (1.30)284520.93 (2.80)257 (9.03)34 (1.20)
23c329222.93 (2.72)513 (15.58)64 (1.94)337122.02 (3.08)343 (10.18)83 (2.46)
33c337125.41 (3.62)1355 (40.20)301 (8.93)348924.36 (4.53)813 (23.30)349 (10.00)
45c391627.74 (4.18)1955 (49.92)967 (24.69)389026.88 (5.52)1278 (32.85)894 (22.98)
nMean (SD)nMean (SD)
Lipid levels
Total cholesterol (mmol/L)39276.07 (1.14)38965.70 (1.00)
HDL cholesterol (mmol/L)39141.43 (0.34)38931.69 (0.41)
Non-HDL cholesterol (mmol/L)39144.62 (1.11)38924.00 (1.02)
Triglycerides (mmol/L)d39122.09 (2.05–2.13)38861.36 (1.34–1.39)
  • aCohort members with data on 45-year lipids. n varies dues to variation in missing data.

  • bBMI classified using international age and sex-specific cut-offs20: overweight, respectively, for ages 7, 11, and 16 years, as ≥17.92, ≥20.55, and ≥23.90 kg/m² for boys and ≥17.75, ≥20.74, and ≥24.37 kg/m² for girls; obesity, respectively, as ≥20.63, ≥25.10, and ≥28.88 kg/m² for boys and ≥20.51, ≥25.42, and ≥29.43 kg/m² for girls.

  • cOverweight ≥25 kg/m², obesity ≥30 kg/m².

  • dGeometric mean (95% Confidence Interval).


There was no association between birthweight and HDL cholesterol in either sex or with total cholesterol and non-HDL cholesterol in men. Per SD higher birthweight, triglycerides were lower by 2.87% (95% confidence interval: 0.95–4.79%) in men and 4.44% (2.70–6.18%) in women; total cholesterol and non-HDL cholesterol were lower by 1.25% (0.68–1.82%) and 1.74% (0.90–2.59%), respectively, in women. Associations were little affected by adjustment for 7-year zBMI, and there was no interaction between birthweight and 7-year zBMI (P ≥ 0.22 for all).

Body mass index at specific ages

Adult zBMI at 33 and 45 years were consistently associated with all lipids; with strongest associations for 45-year zBMI (Table 2). Total cholesterol was higher by ∼1 and ∼2% per SD BMI at 33 and 45 years, respectively; triglycerides were higher by ∼14 and ∼19%, respectively. At all life stages from 7 years, zBMI was negatively associated with HDL cholesterol (both sexes). Associations with triglycerides were seen from 11 years in women and 16 years in men, e.g. per SD higher BMI at 11 and 45 years triglycerides were elevated by 4.01% (2.26–5.75%) and 19.75% (18.18–21.32%), respectively, in women. These estimates equate to 1.54% (0.87–2.21%) and 3.57% (3.29–3.86%) elevated triglycerides per 1kg/m² higher BMI at 11 and 45 years, respectively. With few exceptions, adjustment for covariates reduced but did not eliminate associations (Table 2).

View this table:
Table 2

Mean % (95% Confidence Interval) difference in lipid levels per standard deviation higher body mass index at ages 7 to 45 years

BMI at:Total cholesterolHDL cholesterolNon-HDL cholesterolTriglycerides
7 years0.04 (−0.59, 0.67)−0.04 (−0.66, 0.58)−0.91 (−1.70, −0.11)*−0.80 (−1.56, −0.04)*0.21 (−0.63, 1.04)0.07 (−0.77, 0.90)1.19 (−0.70, 3.08)0.79 (−1.07, 2.65)
11 years0.26 (−0.40, 0.92)0.10 (−0.55, 0.76)−1.24 (−2.01, −0.46)*−0.89 (−1.62, −0.17)*0.70 (−0.14, 1.55)0.38 (−0.46, 1.22)1.83 (−0.07, 3.73)0.84 (−1.05, 2.73)
16 years−0.13 (−0.76, 0.51)−0.31 (−0.95, 0.33)−1.77 (−2.56, −0.98)*−1.39 (−2.13, −0.65)*0.36 (−0.48, 1.21)−0.01 (−0.86, 0.84)2.46 (0.48, 4.44)*1.23 (−0.75, 3.21)
23 years0.42 (−0.19, 1.02)0.17 (−0.44, 0.78)−3.43 (−4.17, −2.70)*−3.18 (−3.89, −2.47)*1.60 (0.81, 2.40)*1.20 (0.39, 2.01)*7.87 (6.01, 9.72)*6.31 (4.43, 8.19)*
33 years1.22 (0.62, 1.82)*1.01 (0.40, 1.62)*−5.10 (−5.81, −4.39)*−4.82 (−5.53, −4.12)*3.26 (2.47, 4.05)*2.89 (2.09, 3.70)*12.49 (10.65, 14.33)*11.00 (9.13, 12.88)*
45 years2.72 (2.14, 3.31)*2.59 (1.99, 3.19)*−6.80 (−7.49, −6.12)*−6.96 (−7.63, −6.29)*5.75 (4.99, 6.50)*5.60 (4.81, 6.38)*18.42 (16.68, 20.16)*17.39 (15.59, 19.19)*
7 years−0.53 (−1.14, 0.08)−0.70 (−1.30, −0.11)−1.66 (−2.44, −0.88)*−1.59 (−2.31, −0.86)*−0.02 (−0.93, 0.89)−0.26 (−1.14, 0.61)0.53 (−1.37, 2.42)−0.15 (−1.96, 1.66)
11 years0.22 (−0.37, 0.82)−0.01 (−0.61, 0.58)−3.20 (−3.98, −2.43)*−2.69 (−3.44, −1.94)*1.62 (0.76, 2.47)*1.10 (0.24, 1.95)*4.01 (2.26, 5.75)*2.75 (1.05, 4.45)*
16 years0.03 (−0.59, 0.65)−0.38 (−1.02, 0.26)−3.96 (−4.80, −3.11)*−3.14 (−3.93, −2.35)*1.66 (0.73, 2.59)*0.77 (−0.17, 1.70)5.75 (3.97, 7.53)*3.64 (1.92, 5.35)*
23 years1.05 (0.47, 1.64)*0.51 (−0.09, 1.11)−6.14 (−6.89, −5.40)*−4.95 (−5.68, −4.23)*4.05 (3.20, 4.90)*2.81 (1.95, 3.67)*10.67 (8.95, 12.39)*7.49 (5.77, 9.21)*
33 years1.35 (0.79, 1.90)*0.94 (0.36, 1.51)*−7.81 (−8.53, −7.09)*−6.72 (−7.43, −6.01)*5.17 (4.36, 5.98)*4.14 (3.31, 4.96)*14.76 (13.11, 16.41)*12.22 (10.55, 13.89)*
45 years2.28 (1.73, 2.83)*1.92 (1.35, 2.49)*−9.29 (−9.98, −8.61)*−8.58 (−9.27, −7.89)*7.25 (6.46, 8.03)*6.45 (5.64, 7.27)*19.75 (18.18, 21.32)*17.61 (15.99, 19.23)*
  • aFor social class at birth, adult social class, smoking, education, alcohol consumption, hypertension and for women menopausal status.

  • *P ≤ 0.05.

Duration of obesity

Nearly a quarter of the population were obese at 45 years (Table 1), most from mid-adulthood (16.1% males, 13.5% females) and few with childhood onset (∼2%, Table 3). There was no trend of worsening lipid levels with an increasing duration of obesity (Table 3) or overweight (data not shown). However, HDL cholesterol was lower in those who were overweight from childhood or young-adulthood than those with later onset, e.g. by 4.30% (1.99–6.61%) after adjustment, for women with childhood onset. Obesity or overweight in childhood but not thereafter was unrelated to lipids compared with the never obese. An exception was seen for triglycerides, with 20.56% (1.96–39.16%) lower levels in females with obesity in childhood only (Table 3), but this was not observed for females who were overweight in childhood only (data not shown).

View this table:
Table 3

Mean % (95% Confidence Interval) difference in lipid levels by age of obesity onseta

n (%)bTotal cholesterolHDL cholesterolNon-HDL cholesterolTriglycerides
  Never2880 (73.34)
  Childhood only27 (0.69)−4.16 (−11.37, 3.05)−2.81 (−9.92, 4.30)−0.29 (−9.14, 8.55)2.75 (−5.52, 11.03)−5.78 (−15.34, 3.78)−2.33 (−11.62, 6.96)−6.89 (−29.74, 15.96)−5.94 (−28.36, 16.48)
  Childhood59 (1.50)1.90 (−2.95, 6.75)1.64 (−3.19, 6.46)−15.71 (−21.57, −9.84)*−13.21 (−18.79, −7.63)*7.49 (1.13, 13.86)*11.46 (8.68, 14.23)*34.47 (19.38, 49.55)*30.14 (15.17, 45.10)*
  Young-adulthood327 (8.33)2.12 (−0.15, 4.38)1.60 (−0.70, 3.91)−13.01 (−15.64, −10.38)*−12.09 (−14.65, −9.53)*6.50 (3.54, 9.47)*12.20 (10.01, 14.39)*30.35 (23.70, 36.99)*26.53 (19.85, 33.21)*
  Mid-adulthood634 (16.14)4.11 (2.46, 5.75)*3.69 (2.04, 5.34)*−12.66 (−14.60, −10.72)*−12.10 (−13.96, −10.23)*8.88 (6.71, 11.04)*13.59 (11.28, 15.90)*32.97 (28.01, 37.93)*30.23 (25.27, 35.19)*
  Never2918 (74.88)
  Childhood only30 (0.77)−3.02 (−9.45, 3.41)−3.31 (−9.68, 3.07)1.47 (−7.24, 10.19)1.69 (−6.46, 9.85)−4.98 (−14.41, 4.44)−3.94 (−9.76, 1.89)−18.70 (−37.84, 0.44)−20.56 (−39.16, −1.96)*
  Childhood89 (2.28)1.89 (−1.92, 5.69)0.47 (−3.29, 4.23)−19.93 (−24.74, −15.13)*−16.85 (−21.47, −12.23)*11.44 (5.90, 16.99)*11.02 (8.62, 13.42)*27.52 (16.48, 38.56)*20.17 (9.32, 31.03)*
  Young-adulthood335 (8.60)2.27 (0.21, 4.32)*1.09 (−0.99, 3.18)−20.58 (−23.16, −18.00)*−16.81 (−19.34, −14.29)*11.40 (8.42, 14.37)*13.26 (11.02, 15.49)*38.60 (32.54, 44.66)*30.92 (24.83, 37.01)*
  Mid-adulthood525 (13.47)4.99 (3.34, 6.64)*4.19 (2.55, 5.83)*−15.57 (−17.67, −13.48)*−14.08 (−16.11, −12.06)*13.77 (11.38, 16.15)*11.14 (8.95, 13.33)*33.91 (29.03, 38.78)*29.58 (24.82, 34.34)*
  • aObesity onset: never obese; obese in childhood/adolescence only (BMI ≥cut-off at 7, 11 y, or 16 years but not at 23, 33, and 45 years); childhood or adolescence onset (BMI ≥cut-off at 7, 11, or 16 years plus BMI ≥30 kg/m² at 23, 33, or 45 years); young-adulthood onset (BMI ≥30 kg/m² at 23 or 33 years, but not in childhood) and mid-adulthood onset (BMI ≥30 kg/m² at 45 years). Overweight onset (including obesity) defined similarly.

  • bAveraged across imputed data sets.

  • cFor social class at birth, adult social class, smoking, education, alcohol consumption, hypertension, and for women menopausal status.

  • *P ≤ 0.05.

Changes in body mass index

Body mass index gains (i.e. in relative rank position) in adulthood were associated with all lipids, with strongest associations observed for the most recent 33–45 years interval (Table 4). In women for example, triglycerides were higher by 21.80% (19.14–24.46%) per SD gain in BMI 33–45 years (conditional on 33-year zBMI). For HDL cholesterol and non-HDL cholesterol, triglycerides and, in women total cholesterol, associations were seen with BMI gains 16–23 years. Gains during childhood/adolescence were associated only with HDL cholesterol in both sexes and, in women, with triglycerides and non-HDL cholesterol. With few exceptions, adjustment for covariates reduced, but did not eliminate associations (Table 4).

View this table:
Table 4

Mean % (95% Confidence Interval) difference in lipid levels per standard deviation increase in body mass index during different age intervals adjusted for ‘‘baseline’’ body mass indexa

Total cholesterolHDL cholesterolNon-HDL cholesterolTriglycerides
BMI change between
7–11 years0.37 (−0.54, 1.28)0.20 (−0.69, 1.09)−1.08 (−2.16, −0.01)*−0.65 (−1.65, 0.35)0.90 (−0.26, 2.05)0.53 (−0.62, 1.67)1.74 (−0.73, 4.21)0.57 (−1.87, 3.01)
11–16 years−0.66 (−1.75, 0.43)−0.80 (−1.90, 0.29)−1.82 (−3.19, −0.45)*−1.56 (−2.79, −0.32)*−0.30 (−1.74, 1.14)−0.59 (−2.04, 0.86)2.39 (−1.03, 5.80)1.31 (−2.08, 4.71)
16–23 years0.84 (−0.003, 1.69)0.63 (−0.23, 1.49)−3.90 (−4.93, −2.87)*−3.84 (−4.82, −2.85)*2.32 (1.21, 3.42)*2.03 (0.91, 3.16)*10.66 (8.23, 13.09)*9.30 (6.82, 11.79)*
23–33 years1.85 (0.98, 2.72)*1.72 (0.84, 2.60)*−5.37 (−6.40, −4.33)*−5.06 (−6.09, −4.02)*4.27 (3.14, 5.41)*3.98 (2.85, 5.11)*13.92 (11.25, 16.60)*12.81 (10.16, 15.45)*
33–45 years4.51 (3.54, 5.49)*4.38 (3.40, 5.36)*−7.19 (−8.32, −6.07)*−7.79 (−8.88, −6.71)*8.15 (6.88, 9.43)*8.13 (6.86, 9.41)*22.11 (19.26, 24.96)*21.48 (18.61, 24.34)*
7–11 years1.06 (0.20, 1.92)*0.85 (−0.02, 1.71)−3.83 (−5.01, −2.65)*−2.96 (−4.09, −1.83)*2.99 (1.79, 4.20)*2.33 (1.12, 3.54)*6.70 (4.17, 9.24)*5.22 (2.73, 7.72)*
11–16 years−0.27 (−1.40, 0.85)−0.77 (−1.94, 0.39)−3.44 (−4.75, −2.12)*−2.47 (−3.71, −1.22)*1.03 (−0.65, 2.72)−0.06 (−1.75, 1.63)5.95 (3.16, 8.75)*3.40 (0.73, 6.08)*
16–23 years1.85 (0.99, 2.72)*1.35 (0.46, 2.24)*−6.29 (−7.35, −5.23)*−5.03 (−6.06, −4.00)*5.28 (4.01, 6.55)*4.05 (2.78, 5.33)*12.23 (9.75, 14.71)*8.90 (6.45, 11.35)*
23–33 years1.24 (0.36, 2.13)*1.17 (0.28, 2.07)*−7.14 (−8.27, −6.01)*−6.47 (−7.55, −5.39)*4.75 (3.46, 6.03)*4.35 (3.06, 5.64)*14.96 (12.41, 17.52)*14.03 (11.44, 16.61)*
33–45 years3.28 (2.37, 4.19)*2.99 (2.07, 3.92)*−8.39 (−9.55, −7.24)*−8.34 (−9.46, −7.21)*8.52 (7.21, 9.83)*8.10 (6.78, 9.42)*21.80 (19.14, 24.46)*20.27 (17.62, 22.92)*
  • aMean % difference in lipid level per SD increase in BMI between each age interval, conditional on zBMI at the previous age.

  • bFor social class at birth, adult social class, smoking, education, alcohol consumption, hypertension, and for women menopausal status.

  • *P ≤ 0.05.

Examining whether relationships between 45-year zBMI and lipids were modified by earlier zBMI, we found that those in the lowest BMI tertile at younger ages consistently had stronger associations with lipids than those in the highest tertile. In terms of per unit (kg/m²) BMI at 45 years, total cholesterol was elevated by 2.02% (1.58–2.45%) compared with 0.33% (0.07–0.59%) for males in the lowest vs. highest tertile of BMI at 33 years. Corresponding estimates were 1.16% (0.83–1.49%) and 0.53% (0.29–0.77%), respectively, for the lowest vs. the highest tertiles at 16 years. Figure 1 illustrates these associations (Pinteraction 45-year zBMI with zBMI at 16, 23, and 33 years: 0.05, <0.001 and <0.001, respectively). Similar results for women were evident from 7 years (data not shown) and, for both sexes the general patterns of interaction were replicated using 45-year waist circumference instead of 45-year BMI. For example, per unit (cm) waist circumference at 45 years, total cholesterol was elevated by 0.61% (0.47–0.76%) compared with 0.13% (0.03–0.22%) for males in the lowest vs. the highest tertile of BMI at 33 years. Triglycerides and non-HDL cholesterol showed similar patterns of effect-modification to those observed for total cholesterol; and for HDL cholesterol there was effect-modification by adult zBMI from 33 years in men and 23 years in women. There was no interaction of birthweight and 45-year zBMI or waist circumference on lipids (P ≥ 0.17 for all). Adjustment for covariates had little effect on associations (data not shown).

Figure 1

Association* between total cholesterol and 45-year BMI, stratified by tertiles of BMI at 16, 23, and 33 years, Men. *45-year BMI centred on 27.74 kg/m² (i.e. mean 45-year BMI). Modelled separately for each tertile of BMI at 16, 23, and 33 years, respectively.


Our findings in this nationwide cohort highlight the importance of current BMI and recent BMI gains for lipid levels in mid-adulthood. In terms of BMI units (kg/m²), these associations ranged from ∼0.4 to ∼4.4% per unit higher 45-year BMI, and from ∼0.6 to ∼5.3% per 33–45 year kg/m² gain for total cholesterol and triglycerides, respectively. Body mass index at younger ages was also important, particularly with regard to its modifying effect on the association between current adiposity and lipids. Consistently, associations with current BMI (or waist circumference) were stronger among those in the lowest tertile of BMI at younger ages (i.e. those gaining the most adiposity) compared with those with the highest BMI. For some lipids, this effect-modification was seen at multiple ages from childhood. A long duration of obesity from childhood was no more detrimental to 45-year lipid levels than obesity onset in mid-adulthood, and, generally, obesity or overweight in childhood but not thereafter was unrelated to adult lipid levels. Finally, we found that those with lower birthweight had higher triglycerides, and in women higher total cholesterol and non-HDL cholesterol; these associations were not explained by catch-up growth to 7 years.

Methodological considerations

A major study strength is the availability of repeated prospective BMI measures over approximately four decades of life. Few other studies can examine associations between body size at multiple ages from birth and lipids in mid-adulthood when risk factors for CVD are reaching clinically significant levels. A further strength includes our ability to adjust for several socio-economic and life-style characteristics from across the life course, recorded prospectively. Nonetheless, study weaknesses are acknowledged. Ages of follow-up may not always capture important phases of BMI change, possibly accounting for our lack of a mediating or modifying effect of 7-year zBMI on the birthweight–lipids association, while others suggest that early life catch-up growth is important.17,27 Birthweight has limitations as a proxy for prenatal growth, yet results were similar for birthweight adjusted for gestational age. Body mass index, although widely used, does not completely reflect body fatness and waist circumference was available at only a single age. Use of non-fasted blood samples may be considered a study limitation. While total cholesterol and HDL cholesterol are little affected by fasting status, triglycerides are lower after fasting and vary by duration of fasting and time of day. Fasting and non-fasting triglycerides are positively correlated,28 and a meta-analysis found no variation in results for triglycerides by fasting status,29 suggesting that such measures are useful in large epidemiological studies.29 Although our approach is supported by sensitivity analyses (Supplementary material online, Tables S1–3), we cannot entirely discount the possibility that associations are affected by recent dietary intake and associated fasting status. As in previous studies that did not have direct measurements of LDL cholesterol,27 we used non-HDL cholesterol. Sample attrition had occurred over follow-up to 45 years, although 45-year respondents are broadly representative of the surviving cohort.19 Multiple imputation was used to avoid further reduction in the sample due to missing information.

Comparison with other studies

Of the few studies available on child to adulthood body size and adult lipids, all highlight the importance of concurrent adult BMI for lipids,14,16,30 as shown in our study. For example, one study found both 3- and 20-year BMI to be positively correlated with 20-year total cholesterol, but more strongly for concurrent BMI.16 In an older generation with a different BMI trajectory to the 1958 cohort,31 BMI gains from adolescence/young-adulthood were associated with higher total cholesterol  and LDL cholesterol, and gains from even younger ages were associated with lower HDL cholesterol,14 consistent with our study. Our finding of stronger associations for current BMI among those with a lower vs. a higher BMI at younger ages is noteworthy. While previous studies have shown similar associations for childhood BMI at 9–13 years in smaller regional populations from UK30 and USA32 our nationwide population study highlights the detrimental influence on lipids of BMI gain across different ages, in some instances, from childhood to mid-adulthood. The general pattern of effect-modification was robust using 45-year waist circumference instead of BMI, suggesting that the stronger associations for current BMI among those with a lower vs. a higher BMI at younger ages is unlikely to be due to residual confounding by adult fatness.30 Underlying explanations for these associations, including, for example, non-alcoholic fatty liver disease, dietary, or genetic influences, are unclear. Yet, the pattern of associations shows consistency in that our analyses of obesity onset from childhood showed no greater effect on 45-year lipids than onset in mid-adulthood. This agrees with a four population pooled study showing similar adverse lipid levels for those who were overweight/obese in childhood and obese in adulthood (i.e. long duration) and those obese only in adulthood (i.e. short duration),33 but not with another study, restricted to overweight duration in adulthood only, using retrospective recall and binary outcomes.34 Our related finding that obesity in childhood but not thereafter was not associated with lipids in mid-adulthood agrees with the scant evidence available to date.33 Such evidence supports the hypothesis that pathways through which obesity duration affects CVD mortality are independent of lipids, as suggested elsewhere.35 While obesity rates in childhood are low in our cohort compared with today,36 our findings are relevant to understanding how obesity and lifetime changes in adiposity affect morbidity and mortality.33,35 Findings for birthweight, reported previously37 agree with other studies7,8 in showing the negative association with total cholesterol to be of small magnitude and varying by gender, possibly due to chance.37 Extending previous work, we found that birthweight associations were not mediated or modified by catch-up growth to 7 years. This is consistent with the observation that 7-year BMI did not mediate or modify the association between birthweight and coronary heart disease.38 Whereas others suggest that an association between low birthweight and high adult cholesterol exists only among the obese,15,39 we found no interaction between birthweight and 45-year BMI.

Conclusions and implications

We have demonstrated the importance of both BMI and BMI changes in adulthood for lipid levels. A one kg/m² gain in BMI between, for example, 33 and 45 years was associated with ∼1.5% higher non-HDL cholesterol; conversely a kg/m2 lower BMI was associated with reduced non-HDL cholesterol by ∼1.5%. If the non-HDL cholesterol–CVD association is causal, such small effects could have an appreciable impact on CVD reduction at the population level. Given that a large proportion of CVD events occur among individuals with low to intermediate risk, therapies targeted at high-risk individuals do little to reduce the overall population burden of CVD.40 Thus our findings are of clinical relevance and indicate the hazard of BMI gain in adulthood with regard to lipid profiles. Our finding suggesting that childhood obesity was not particularly detrimental to adult lipids might be interpreted as argument against prevention strategies aimed at childhood overweight and obesity. However, high childhood BMI is associated with adulthood obesity,41 suggesting the need for interventions to prevent childhood overweight and obesity. The importance of childhood BMI was further highlighted because it modified the association between concurrent BMI and lipids. All associations of current BMI with lipids were stronger among those who gained the most compared with the least BMI from younger ages. This is of public health relevance in countries undergoing the obesity epidemic, as the generation now gaining BMI in adulthood had low or normal BMI in childhood. In conclusion, our findings suggest that life course trajectories of body size influence adult lipid levels, with the consequences of high adult BMI for adult lipid levels being particularly adverse for those with lower BMI at earlier life stages.

Supplementary material

Supplementary material is available at European Heart Journal online.


This work was supported by the Department of Health Policy Research Programme through the Public Health Research Consortium. The views expressed in the publication are those of the authors and not necessarily those of the Department of Health. Information about the wider programme of the PHRC is available from http://phrc.lshtm.ac.uk. The Medical Research Council funded the 45y biomedical survey (grant: G0000934). The GOSH/UCL Institute of Child Health was supported in part by the Department of Health's NIHR Biomedical Research Centre. The Centre for Paediatric Epidemiology and Biostatistics was supported in part by the Medical Research Council in its capacity as the MRC Centre of Epidemiology for Child Health. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Conflict of interest: none declared. 


View Abstract