024 5408 4630 4180 21.8 mean (SD) 15.9 (1.63) 17.3 (2.41) 20.3 (2.72) 23.1 (2.90) 25.6 (4.04) 27.8 (4.37) 28.1 (4.81) ( ) 1.40 1.44 1.55 2.27 11.0 25.3 27.5 (IQR) (15.0,16.7) (15.8,18.2) (18.5,21.4) (21.2,24.5) (23.1,27.5) (25.0,30.1) (24.9,30.4) 7501 N 6422 6114 5229 6057 5562 4677 4139 N 6422 6114 5229 6057 5562 4677 4139 18.5 mean (SD) 15.9 (1.91) 17.6 (2.70) 21.0 (2.96) 22.1 (3.25) 24.6 (4.97) 27.0 (5.62) 26.8 (5.60) ( ) 2.44 1.64 1.49 3.07 12.0 23.7 23.0 (IQR) (14.6,16.7) (15.8,18.9) (19.0,22.5) (20.0,23.4) (21.4,26.5) (23.1,29.7) (22.9,29.5) 4621 4622 4621 4621 5.95 8.29 0.48 10.7 4687 4687 4687 4687 6.14 11.7 2.71 13.9 N ( ) N ( ) FemalesTable based on observed data, N varies due to missing data. *Neglect at 7y and/or 11y (if one missing, other age used).IQR = inter-quartile range defined as BMI!20.63 at 7y, 25.10 at 11y, 28.88kg/m2 at 16y for males; !20.51, 25.42 and 29.4kg/m2 respectively in females; in adulthood BMI!30kg/m2. doi:10.1371/journal.pone.0119985.tStatistical analysisIn preliminary analyses we examined simple associations between childhood maltreatments and BMI at different ages in order to obtain a life-course overview before proceeding to longitudinal analyses of BMI trajectories. Specifically for our first aim, which was to assess associations between childhood maltreatments and BMI at different life-stages, we conducted unadjusted linear regression analysis separately for each maltreatment and BMI at each age 7y to 50y; likewise, using logistic regression we assessed associations with obesity at each age. Patterns of association were similar for neglect at 7y and 11y, hence we report results for the summary variable, neglect at 7y and/or 11y. A unit (kg/m2) of BMI has different implications in child and adulthood, so in subsequent analyses of BMI as a continuous variable, we internally standardised BMI to age and sex-specific SD scores to facilitate comparison across age. In main analyses to establish whether child maltreatment groups had differing lifetime BMI trajectories from the non-maltreated, we performed longitudinal analyses of child to midadulthood BMI. Specifically, to examine associations between each maltreatment and zBMI trajectory we used multilevel (mixed-effects) models that allow for correlation betweenPLOS ONE | DOI:10.1371/journal.pone.0119985 March 26,4 /Child Maltreatment and BMI Trajectoriesrepeated zBMI measures on the same individuals over time. Models were fitted by assuming an unstructured variance-covariance matrix among repeated measures. Although the mean trajectory for zBMI for the PXD101 side effects cohort approximates a horizontal line FT011 site centred at zero, models include a random intercept (zBMI at 7y, as age was centred at 7y) and random slope for zBMI (coefficient for linear age term, i.e. rate of change in zBMI/year), a maltreatment group and its interaction with age (as a fixed effect), to test whether the rate of change in zBMI varied by maltreatment. The quadratic term of age (i.e. age2) and its interaction with maltreatment measure were further added to the model to test whether the difference in average zBMI by maltreatment was a non-linear function of age. This latter term was significant and retained in models for neglect at 7/11y only. Thus, with age centred at 7y, the average rate of change over time is (coefficient for age + coefficient for neglect?age) +2?(coefficient for age2 + coefficient for neglect?age2) for those with neglect and (coefficient for age) +2?(coefficient for age2) for tho.024 5408 4630 4180 21.8 mean (SD) 15.9 (1.63) 17.3 (2.41) 20.3 (2.72) 23.1 (2.90) 25.6 (4.04) 27.8 (4.37) 28.1 (4.81) ( ) 1.40 1.44 1.55 2.27 11.0 25.3 27.5 (IQR) (15.0,16.7) (15.8,18.2) (18.5,21.4) (21.2,24.5) (23.1,27.5) (25.0,30.1) (24.9,30.4) 7501 N 6422 6114 5229 6057 5562 4677 4139 N 6422 6114 5229 6057 5562 4677 4139 18.5 mean (SD) 15.9 (1.91) 17.6 (2.70) 21.0 (2.96) 22.1 (3.25) 24.6 (4.97) 27.0 (5.62) 26.8 (5.60) ( ) 2.44 1.64 1.49 3.07 12.0 23.7 23.0 (IQR) (14.6,16.7) (15.8,18.9) (19.0,22.5) (20.0,23.4) (21.4,26.5) (23.1,29.7) (22.9,29.5) 4621 4622 4621 4621 5.95 8.29 0.48 10.7 4687 4687 4687 4687 6.14 11.7 2.71 13.9 N ( ) N ( ) FemalesTable based on observed data, N varies due to missing data. *Neglect at 7y and/or 11y (if one missing, other age used).IQR = inter-quartile range defined as BMI!20.63 at 7y, 25.10 at 11y, 28.88kg/m2 at 16y for males; !20.51, 25.42 and 29.4kg/m2 respectively in females; in adulthood BMI!30kg/m2. doi:10.1371/journal.pone.0119985.tStatistical analysisIn preliminary analyses we examined simple associations between childhood maltreatments and BMI at different ages in order to obtain a life-course overview before proceeding to longitudinal analyses of BMI trajectories. Specifically for our first aim, which was to assess associations between childhood maltreatments and BMI at different life-stages, we conducted unadjusted linear regression analysis separately for each maltreatment and BMI at each age 7y to 50y; likewise, using logistic regression we assessed associations with obesity at each age. Patterns of association were similar for neglect at 7y and 11y, hence we report results for the summary variable, neglect at 7y and/or 11y. A unit (kg/m2) of BMI has different implications in child and adulthood, so in subsequent analyses of BMI as a continuous variable, we internally standardised BMI to age and sex-specific SD scores to facilitate comparison across age. In main analyses to establish whether child maltreatment groups had differing lifetime BMI trajectories from the non-maltreated, we performed longitudinal analyses of child to midadulthood BMI. Specifically, to examine associations between each maltreatment and zBMI trajectory we used multilevel (mixed-effects) models that allow for correlation betweenPLOS ONE | DOI:10.1371/journal.pone.0119985 March 26,4 /Child Maltreatment and BMI Trajectoriesrepeated zBMI measures on the same individuals over time. Models were fitted by assuming an unstructured variance-covariance matrix among repeated measures. Although the mean trajectory for zBMI for the cohort approximates a horizontal line centred at zero, models include a random intercept (zBMI at 7y, as age was centred at 7y) and random slope for zBMI (coefficient for linear age term, i.e. rate of change in zBMI/year), a maltreatment group and its interaction with age (as a fixed effect), to test whether the rate of change in zBMI varied by maltreatment. The quadratic term of age (i.e. age2) and its interaction with maltreatment measure were further added to the model to test whether the difference in average zBMI by maltreatment was a non-linear function of age. This latter term was significant and retained in models for neglect at 7/11y only. Thus, with age centred at 7y, the average rate of change over time is (coefficient for age + coefficient for neglect?age) +2?(coefficient for age2 + coefficient for neglect?age2) for those with neglect and (coefficient for age) +2?(coefficient for age2) for tho.
