Study setting/site
The BT20 Cohort was established in South Africa (SA) in the 1988 with the aim of conducting a longitudinal study to assess the health of children in the Johannesburg area (15). The study enrolled women in their second and third trimester of pregnancy. Singleton children (n=3273) who were residents of Soweto-Johannesburg and born between April and June 1990 were enrolled into the birth cohort and were followed from birth to date (16). The entry criterion included that mother and baby remain in the Soweto-Johannesburg area until the child was at least six months old. Attrition rate in the first two decades of the cohort has also been relatively low (30%), with most occurring in infancy and early childhood. At the time of writing, there are just over 2,000 children and families participating in the study (17).
Study population
A sub-cohort (approximately n=500) from the birth cohort study was formed at age 9 years to investigate the in-depth longitudinal changes in body composition and whole-body, lumbar spine and hip-bone mass during adolescence and into adulthood (Bone Health Cohort). The current study sample was comprised of young adult (23 and 24 years) males (n=53) and females (n=47) recruited from Bone Health Cohort in Johannesburg, South Africa.
Study variables
Bone-lead concentration was the exposure variable in this study. Tibia-lead concentration was measured in the study participants using K-shell X-ray fluorescence (KXRF), a non-invasive procedure (18). Results from KXRF spectrometry with humans have been used for decades in dozens of studies as a biomarker to assess cumulative lead exposure levels (19). XRF uses 109Cd as the source that emits 88.035 keV photons to fluoresce x-rays from the lead atoms stored in bone. The silver x-rays that also accompany the decay of 109Cd are filtered by copper, minimizing participant radiation dose (effective dose is equivalent to less than ten minutes of natural background radiation for an adult) (20). Backscattered photons and fluoresced Pb x-rays are recorded with a spectroscopy system (intrinsic germanium detector, preamplifier and digital signal processor). The spectrum (distribution of photons against energy) then undergoes non-linear least-squares fitting to extract the areas of the lead x-ray peaks seen atop the Compton scattering background. Coherent scatter normalization, matrix correction and comparison to calibration measurements, made of lead-doped plaster of Paris (CaSO4.2H2O) calibration standards, yield in vivo concentrations in micrograms of lead per gram of bone mineral (21).
Measurement of Aggression
Aggression was the outcome variable in the study. The Buss-Perry Aggression Questionnaire (BPAQ) was administered to study participants to measure aggression as a score (22). The BPAQ is a validated tool that has been used in studies primarily in low- to middle-income countries (23, 24). The questionnaire consists of twenty-nine items that measure four components of aggression: physical and verbal aggression, hostility and anger. Physical aggression consisted of nine questions, and the scoring from this item ranged from 18 to 38. Verbal aggression had five items, with the scoring ranging from 10 to 25. Anger consisted of seven items, with a scoring range of 14 to 35; and hostility consisted of eight items, with a scoring range of 10 to 37 (22). The level of aggression in the questionnaire was rated on a five-point Likert scale, presented as 1 (extremely uncharacteristic of me), 2 (somewhat uncharacteristic of me), 3 (neither uncharacteristic nor characteristic of me), 4 (somewhat characteristic of me) and 5 (extremely characteristic of me). The total aggression scores of the twenty-nine items were also calculated, and these were also used in the analyses. A Cronbach’s α reliability coefficient was used to determine the reliability of the items. An alpha coefficient of 0.72 was obtained for the nine items of physical aggression, 0.6684 for the seven items of anger, 0.7150 for the eight items of hostility and 0.5640 for the five items of verbal aggression, indicating acceptable reliability among the items. The 29 items reported a scale reliability coefficient of 0.8364, similar to that reported previously (24).
Study confounders
A separate questionnaire was administered to the study participants to obtain information on demographics, socio-economic and psychosocial factors. A confounding variable in the study was defined as a variable that is a risk factor for aggression or is associated with, but is not a consequence of, bone lead concentration. The following variables were therefore considered as study confounders and potential predictors of aggression: age; sex; level of schooling (categorized into three: Grade 5 or less, grade 6-12 and tertiary education); presence of both parents at home; home environment; neighbourhood crime; profile of illegal substance abuse (use of drugs such as dagga and glue); use of alcohol; and socio-economic factors (maternal education, type of housing and occupation status). Information on the participant’s home environment (referred to in this analysis as “history of family violence”) was obtained by asking the participants to respond to the following statements: “We argue a lot in our family, “people in my family hardly ever lose their temper” and “people in my family sometimes hit each other when they are angry”. Participant were required to agreed or responded with a “Yes” (coded as 1) or disagreed or responded with a “No” (coded as 2). To obtain information on neighbourhood factors, the participants were asked how they generally feel in their neighbourhood: a “feeling of somewhat unsafe or very unsafe” coded as 1 and “somewhat safe or very safe” coded as 2. Questions on whether the participants had personally experienced crime and violence in the neighbourhood were asked: “ever in your life experienced any crime”, with “Yes” coded as 1 and “No” coded as 2. Socioeconomic status was measured by considering three levels: (1) maternal level of education (categorized into four levels: no formal education, primary schooling, secondary schooling and post-school education); (2) type of housing: formal (such as free-standing house, townhouse or hostel) and informal (shack, squat and any other informal room); and (3) participant’s level of education and occupational status as proxies.
Ethical consideration
Ethical approval was obtained for this study from the University of the Witwatersrand, Human Ethics Research Committee (M 191116).
Statistical Analysis
All data cleaning and analysis were performed using Stata version 15 (StataCorp. 2017; Stata Statistical Software: Release 15. College Station, TX: StataCorp LP). Data were checked for duplicates and missing values. Study participants’ psychosocial and demographic characteristics were described. The categorical variables were presented as frequencies and proportions. Data were stratified by sex. The Pearson chi-squared test was used to assess the association between categorical variables. Continuous variables in the study were bone-lead concentration and the four aggression level scores (physical, verbal, anger and hostility). Tibia-lead concentrations were analysed in the study as a continuous variable to retain all values, including values below the detection limit and values lower than zero (22). The distribution of the continuous variables was checked for normality. Bone-lead concentrations were summarized as mean and SD, median, 25th and 75th Interquartile ranges (IQR), as appropriate. For continuous variables that were normally distributed, an independent Student’s t-test was conducted to test for the difference in the mean. There was no further post hoc correction analysis required. Testing was set at the 0.05 level of significance. The geometric mean, median and ranges for the aggression scales were described, and a Student’s t-test was conducted to test for the differences in means, stratified by sex.
To assess the association between bone-lead concentration and aggression, a linear regression model was fitted. In the univariate analysis, a simple linear regression was fitted, with each aggression scale as the outcome variable and with bone lead as the main explanatory variable. A backward elimination, using a liberal p-value of 0.20, was used to include variables in the multivariate model. Variables with p ≤0.001 were reported as highly significant, and those with p ≤0.08, also retained in the final model, were reported as marginally significant. Age and sex were retained as study confounders in all multivariable models. Goodness of fit was assessed via regression diagnostics, and residuals were assessed to check the assumptions of linearity, normality and constant variance and the adequacy of the final models.
To further quantify and assess the direction of the relationship between bone lead levels and aggression, structural equation modelling (SEM) was performed. In the SEM model, aggression was the latent variable, which, as previously indicated, was assessed via four observed variables: physical aggression, verbal aggression, hostility and anger. All observed variables were denoted by rectangular boxes, and latent variables (unobserved) were denoted in ovals. In model I, pathways between educational level, age, sex, type of housing, occupational status and maternal education were created, as these variables were identified as determinants of bone-lead levels (25). Subsequently, the variables (educational level, age, sex, type of housing, occupational status and maternal education) created indirect pathways to aggression via (continuous) bone-lead concentration. Direct predictors for aggression in model II were: a history of family violence, exposure to crime, growing up with a single parent and use of drugs and alcohol. To assess model fit, SEM fit indices that included: the root mean square error (RMSE), standardized root mean square residual (SRMR), comparative fit index (CFI) and Tucker Lewis index (TLI). A RMSE below 0.05, P-close greater than 0.05, SRMR greater than 0.08, CFI and TLI value of 0.95 and above indicated a good fitting model (26). Where necessary, the model was checked for improvement using modification of indices.