Body mass index and height over three generations: evidence from the Lifeways cross-generational cohort study

Participants

This prospective longitudinal study was established in an 18 month period of 2001-2003 when a random sample of 1124 expectant Irish-born mothers were recruited while attending their first ante-natal visit [19, 20]. The cohort is comprised of the index mother and child, and where agreeable, the father, and at least one grandparent randomly selected on a rotated basis from a full list of all available grandparents, maternal grandmother (MGM), maternal grandfather (MGF), paternal grandmother (PGM), and paternal grandfather (PGF) [21]. At recruitment mothers were asked to complete a questionnaire with sections relating to health, diet, lifestyle factors, demographic, occupation, social, and living characteristics [19]. Mothers were asked to report their height and weight before they became pregnant and BMI (weight (kg)/height (m²)) was subsequently calculated. Fathers and grandparents were asked to complete a shorter version of the same questionnaire where they also reported their height and weight [19].

A follow-up study of the families was conducted in a 9 month period of 2007-2008 when the children were aged 5 years on average. Of the 1124 mothers who consented to the study at baseline, 1082 families with 1094 live infants were invited to participate in the follow-up. This number included 12 children, born with varying degrees of birth abnormalities, who were contacted separately. Of the 1082 families, 669 of mothers (62%) responded to the follow-up. Mothers who remained in the study did not report any significant difference in BMI at baseline from those who were no longer in the study [22]. Mothers completed a follow-up questionnaire from baseline which was expanded to provide information on their child's health, physical activity, and diet [22].

Children's weight and height were measured at home to the nearest 0.1 kg and 1 cm by a team of researchers trained using standardized protocols [23, 24]. Height was measured using a Leicester portable stadiometer and weight a Tanita Digital Weighing Scale Model HD305 (both sourced and calibrated by Chasmors Ltd., Camden High Street, London). Height and weight measures were used to calculate BMI. Age and sex specific BMI standards and the International Obesity Taskforce cut-offs that correspond to BMI of 25 or 30 kg/m² at age 18 years were used to identify overweight and obese children [25, 26]. In adults overweight was defined as a BMI between 25 and 29.9 kg/m² and obesity was defined as a BMI greater than 30 kg/m². Weight, height and BMI were the outcomes of interest. The sample used in this analysis was the cohort of children at follow-up who had complete questionnaire information and measurement information (n = 585). Twins (n = 10) and children whose mothers were pregnant or up to 6 weeks postpartum at follow-up were excluded from the analysis resulting in a final sample of n = 529 children.

Risk factors

The risk factors were selected for the main analysis on the basis that the same variable, derived from the baseline questionnaire, was available for each family member. Indicators of socioeconomic status used were level of education (None/primary, Second level, Third level) and medical card status (No medical card versus medical card holder where eligibility to free medical care is on an income basis and was previously used as a good indicator of social disadvantage in an Irish sample) [27]. These two variables were used in duplicate for both mother and child as, at this age, the child may not have started formal education and would be dependent on the mother's general medical services. Dietary intakes were assessed using a standardized Food Frequency Questionnaire (FFQ) which contains 149 food items and responders are asked to report their average use of each food item over the previous year or, in the case of mothers, since they became pregnant [19]. Two proxy measures for general healthy diet were created from the FFQ on the basis of adherence to dietary guidelines. At present, the recommendation for fruit and vegetable intake is 5 or more portions per day and intake of oily fish at least once a week. The proxy variable for fruit and vegetables was created using a sum of average daily intakes which was then dichotomised into those who did or did not meet the recommendations (fruitveg). Average daily intakes of oily fish were also computed and responses categorised into those who were regular consumers (at least once as week), occasional consumers and those who never consumed oily fish. Other variables included in the analysis were self reported physical activity level which was created from responders engagement in light, moderate or heavy physical exercise. Each individual was then graded on their frequency and level of physical activity (based on a score of 0 to 18) and then categorised into one of four groups (none, low, medium, and high levels of physical activity). Finally self-rated health (SRH) was used as an indicator of overall health status and the original five response options were collapsed to those who reported excellent or very good health, good or fair health, and poor health.

Statistical analyses

Independent t-tests and χ² tests were used to investigate for the presence of patterns in the missing values which might influence quantitative outcomes.

We carried out descriptive analyses of weight, height, and BMI and t-test for gender differences among children in each age group. Correlation coefficients between family members were calculated for weight, height and BMI. Univariate linear regression analysis was conducted for child weight, height, BMI, and family data. Parent-offspring linear regression analyses were further employed to estimate to what extent the phenotypic variance (V_P) in BMI was attributable to genetic (V_G) versus environmental factors (V_E). The proportion of the phenotypic variance attributable to additive genetic variance (V_A) was estimated using narrow-sense heritability (h²) where h² = V_A/V_P. Heritability estimates were calculated by conducting a linear regression with each child and parent/grandparent pair. The slope of the regression line or the regression co-efficient is used to estimate the association between the two parents and their offspring, Therefore twice the regression co-efficient was used to approximate offspring-parent heritability of height, weight and BMI [28].

A mixed model analysis was conducted for, at a minimum, both child and mother (n = 454). The maximum number of individuals in a family was seven: child, mother, father, maternal grandmother (MGM), maternal grandfather (MGF), paternal grandmother (PGM), paternal grandfather (PGF), or groups numbered 1 to 7. Each family had a unique identifier (family id). Each family member provided at least one record for the dataset and the resulting dataset had n = 3703 records. Thus, it was possible to include data on incomplete families in the models. BMI was the outcome measure in mixed model. Family id was fitted as a random effect and BMI measures on individuals in the same family were assumed correlated i.e. BMI measures between family members of the different groups were assumed correlated. Different correlation structures were considered: unstructured (un), compound symmetry (cs), heterogenous compound symmetry (csh), un(1), which fits different variances in each group but zero correlation between family members, and the best one chosen using Akaike's Information Criterion (AIC) or a likelihood ratio test in the case of nested structures.

The fixed effect variables considered for modelling were: gender, group, age, self-rated health (SRH), education, medical card holder (GMS), fruit & vegetables and fish consumption, physical activity level, and interactions of these variables. Stepwise procedures were used to find the fixed effects that provided the best model. P-values < 0.05 are regarded as significant.

The estimated correlations between family members from the models represent the correlations when fixed effects common to all individuals are removed. The Bonferroni correction was used to adjust p-values arising from multiple correlations between groups. Confidence intervals based on Fisher's z transform, rather than standard errors are reported for correlations as correlation estimates typically are not symmetrically distributed.

A similar model was fitted with height as the outcome variable. In addition, each individual was classified as obese/not obese and a model fitted to this binary outcome variable using the same fixed and random effects as in the models described above.

Models were fitted using the SAS version 9.1.3 statistical procedures Mixed and Glimmix.

Ethical approval for the Lifeways Cohort and follow-up was granted by ethics committees in the National University of Ireland, Galway; The Coombe Women's Hospital, Dublin; University College Hospital, Galway; The Irish College of General Practitioners.

Mixed models of familial BMI

The variables medical card holder (GMS), fish consumption, and physical activity level were not significant in this model or any of the later models.

Mixed models of familial height

Strengths and limitations

The advantages of this study are the prospective design with BMI information collected prior to pregnancy and in early childhood, in addition to data from three generations of one family. To our knowledge this is the first prospective study to show intergenerational associations between BMI in the youngest generation and BMI in the previous two generations, with individual measurements for all four grandparents.

Ideally, the strength of our findings could be improved by a higher response rate and larger cohort numbers. Nevertheless, in a previous analysis we did not find any differences in pre-pregnancy BMI of the mothers who responded to the follow-up compared to non-responders. Furthermore, an investigation into the pattern of missing data did not demonstrate any systematic variability. A potential limitation of this study was that self-reported adult height and weight measures were used and, while such measures are acceptable, may lead to underreporting bias particularly in women [35, 36]. This means in fact we may be under-estimating the strength of the effects found since the actual BMI of mothers and grandmothers is likely to be higher.

In epidemiological terms this is a relatively small cohort however there are no comparable cohorts with seven members of the one family. Limited families within the study have complete information on all family members but the mixed model maximises the use of the available data. Furthermore, despite the smaller sample numbers of fathers and grandparents the analyses have sufficient power to demonstrate different familial relationships for BMI and height.