Abuse and Neglect in Childhood and Diabetes in Adulthood
Abuse and Neglect in Childhood and Diabetes in Adulthood
This study used restricted-use data from 14,493 (46.1% male) participants in Wave IV of Add Health. We used variables drawn from participant responses to Waves I, III, and IV in-home interviews. At Wave I, conducted during the 1994–95 school year, a nationally representative sample of 20,745 adolescents in grades 7 through 12 completed in-home interviews. Waves III (2001–02) and IV (2008–09) included all Wave I in-home respondents who could be located, yielding a sample of 15,197 adults aged 18 to 28 years at Wave III, and 15,701 aged 24 to 34 at Wave IV. In addition, at Wave IV, researchers measured height and weight and collected blood for DNA and various biomarker analyses, including fasting or nonfasting blood glucose and HbA1c levels. Details regarding Add Health are available elsewhere. Because previous investigations have observed sex-specific associations between childhood maltreatment and obesity, analyses were stratified by sex.
We coded diabetes status from Add Health Wave IV biospecimen data and modeled it as a 3-level variable: 1) diabetes (defined as any of the following: HbA1c ≥6.5%, fasting glucose ≥126 mg/dL, nonfasting blood glucose ≥200 mg/dL, self-reported taking antidiabetic medication, and/or report of receiving a diagnosis of diabetes or high blood glucose by a health care provider); 2) prediabetes or impaired glucose tolerance (HbA1c 5.7%–6.4% and/or fasting glucose 100–125 mg/dL); or 3) no diabetes. We did not use nonfasting blood glucose alone for classification of prediabetes because the American Diabetes Association does not provide guidelines for doing so.
We coded child maltreatment variables from Wave III and Wave IV in-home interview questions (Box) regarding how often respondents experienced specific forms of child maltreatment by adult caregivers. Questions had 5 response options, from "this never happened" to "more than 10 times." Add Health assessed child neglect at Wave III only and emotional abuse at Wave IV only. Although assessments used identical descriptions of childhood sexual abuse and physical abuse at Wave III and Wave IV, questions at Wave III asked about events occurring before the respondent was in 6th grade; Wave IV questions asked about events occurring before age 18. In both interviews, an additional question about the respondent's age when the event first occurred followed positive responses. Because the focus of the current study was child abuse and neglect, we coded events reported at Wave IV as positive only if they first happened before age 12. To distinguish recurrent abuse from abuse that occurred only once or twice, we operationalized each type of maltreatment as a 3-level variable: 3 or more times, 1 to 2 times, or never. These cutpoints coincided with the median number of incidents for respondents who had ever experienced physical abuse, sexual abuse, or neglect.
Body mass index (BMI [kg/m]) was calculated from height and weight measured at Wave IV. We categorized BMI into 5 levels: obese classes III (≥40), II (35.0–39.9), and I (30.0–34.9); overweight (25.0–29.9); and normal weight (<25.0). Because of low numbers, underweight adults (BMI <18.5) were included in the normal-weight category (n = 191; 1.3% of total sample).
We also included covariates that were known to be associated with both childhood maltreatment and diabetes that were not likely to be in the causal pathway between childhood maltreatment and diabetes and were available in the data set. We modeled the 6-category race/ethnicity preconstructed variable from the Wave I data set (ie, white, black, Latino, Asian/Pacific Islander, American Indian/Native American, and other) as a set of indicator variables, with white as the reference category. We dichotomized self-report of highest education attained at Wave IV as receiving versus not receiving a 4-year college degree. We coded financial insecurity in adolescence from the question in the parental interview: "Do you have enough money to pay your bills?" Because 15.0% of respondents did not have parental interview data, we modeled this variable as a set of indicator variables: enough money to pay bills, not enough money to pay bills (the reference category), or parental data missing. Such subjective measures of social status have been identified as strong predictors of health and, for some measures, are more predictive of health than objective measures such as income and education. Furthermore, without information about household size or region, estimates of income would not be accurate. We obtained information on whether respondents had ever smoked daily from the Wave IV interview.
We analyzed data from 14,493 Add Health Wave IV participants with biomarker data by using survey procedures in Stata version 9.2 (Stata Corp LP) to account for Add Health's complex survey design, stratifying all analyses by sex. First, we used χ analyses to assess bivariate associations of the 3-category diabetes dependent variable (ie, diabetes, prediabetes, or no diabetes) with the 4 child maltreatment variables (ie, sexual abuse, physical abuse, neglect, and emotional abuse) and BMI category and other potential covariates (Table 1). Next, we estimated separate multinomial logistic regression models with 3-category diabetes status as the dependent variable (no diabetes as reference category) for each form of child maltreatment, separately in men and women (models 1–4 [Table 2]). We then estimated a model with all 4 forms of child maltreatment as independent variables (Model 5). To this model, we added the following covariates: age, race/ethnicity, college degree, daily smoking, and childhood financial insecurity (Model 6). Finally, we added BMI category to the model (Model 7) and compared the odds ratios (ORs) of Models 6 and 7. In all models, we conducted post-hoc tests to evaluate differences between the ORs for 1 to 2 versus 3 or more childhood maltreatment incidents for each type of maltreatment.
Methods
This study used restricted-use data from 14,493 (46.1% male) participants in Wave IV of Add Health. We used variables drawn from participant responses to Waves I, III, and IV in-home interviews. At Wave I, conducted during the 1994–95 school year, a nationally representative sample of 20,745 adolescents in grades 7 through 12 completed in-home interviews. Waves III (2001–02) and IV (2008–09) included all Wave I in-home respondents who could be located, yielding a sample of 15,197 adults aged 18 to 28 years at Wave III, and 15,701 aged 24 to 34 at Wave IV. In addition, at Wave IV, researchers measured height and weight and collected blood for DNA and various biomarker analyses, including fasting or nonfasting blood glucose and HbA1c levels. Details regarding Add Health are available elsewhere. Because previous investigations have observed sex-specific associations between childhood maltreatment and obesity, analyses were stratified by sex.
Key Variables
We coded diabetes status from Add Health Wave IV biospecimen data and modeled it as a 3-level variable: 1) diabetes (defined as any of the following: HbA1c ≥6.5%, fasting glucose ≥126 mg/dL, nonfasting blood glucose ≥200 mg/dL, self-reported taking antidiabetic medication, and/or report of receiving a diagnosis of diabetes or high blood glucose by a health care provider); 2) prediabetes or impaired glucose tolerance (HbA1c 5.7%–6.4% and/or fasting glucose 100–125 mg/dL); or 3) no diabetes. We did not use nonfasting blood glucose alone for classification of prediabetes because the American Diabetes Association does not provide guidelines for doing so.
We coded child maltreatment variables from Wave III and Wave IV in-home interview questions (Box) regarding how often respondents experienced specific forms of child maltreatment by adult caregivers. Questions had 5 response options, from "this never happened" to "more than 10 times." Add Health assessed child neglect at Wave III only and emotional abuse at Wave IV only. Although assessments used identical descriptions of childhood sexual abuse and physical abuse at Wave III and Wave IV, questions at Wave III asked about events occurring before the respondent was in 6th grade; Wave IV questions asked about events occurring before age 18. In both interviews, an additional question about the respondent's age when the event first occurred followed positive responses. Because the focus of the current study was child abuse and neglect, we coded events reported at Wave IV as positive only if they first happened before age 12. To distinguish recurrent abuse from abuse that occurred only once or twice, we operationalized each type of maltreatment as a 3-level variable: 3 or more times, 1 to 2 times, or never. These cutpoints coincided with the median number of incidents for respondents who had ever experienced physical abuse, sexual abuse, or neglect.
Body mass index (BMI [kg/m]) was calculated from height and weight measured at Wave IV. We categorized BMI into 5 levels: obese classes III (≥40), II (35.0–39.9), and I (30.0–34.9); overweight (25.0–29.9); and normal weight (<25.0). Because of low numbers, underweight adults (BMI <18.5) were included in the normal-weight category (n = 191; 1.3% of total sample).
We also included covariates that were known to be associated with both childhood maltreatment and diabetes that were not likely to be in the causal pathway between childhood maltreatment and diabetes and were available in the data set. We modeled the 6-category race/ethnicity preconstructed variable from the Wave I data set (ie, white, black, Latino, Asian/Pacific Islander, American Indian/Native American, and other) as a set of indicator variables, with white as the reference category. We dichotomized self-report of highest education attained at Wave IV as receiving versus not receiving a 4-year college degree. We coded financial insecurity in adolescence from the question in the parental interview: "Do you have enough money to pay your bills?" Because 15.0% of respondents did not have parental interview data, we modeled this variable as a set of indicator variables: enough money to pay bills, not enough money to pay bills (the reference category), or parental data missing. Such subjective measures of social status have been identified as strong predictors of health and, for some measures, are more predictive of health than objective measures such as income and education. Furthermore, without information about household size or region, estimates of income would not be accurate. We obtained information on whether respondents had ever smoked daily from the Wave IV interview.
Data Analysis
We analyzed data from 14,493 Add Health Wave IV participants with biomarker data by using survey procedures in Stata version 9.2 (Stata Corp LP) to account for Add Health's complex survey design, stratifying all analyses by sex. First, we used χ analyses to assess bivariate associations of the 3-category diabetes dependent variable (ie, diabetes, prediabetes, or no diabetes) with the 4 child maltreatment variables (ie, sexual abuse, physical abuse, neglect, and emotional abuse) and BMI category and other potential covariates (Table 1). Next, we estimated separate multinomial logistic regression models with 3-category diabetes status as the dependent variable (no diabetes as reference category) for each form of child maltreatment, separately in men and women (models 1–4 [Table 2]). We then estimated a model with all 4 forms of child maltreatment as independent variables (Model 5). To this model, we added the following covariates: age, race/ethnicity, college degree, daily smoking, and childhood financial insecurity (Model 6). Finally, we added BMI category to the model (Model 7) and compared the odds ratios (ORs) of Models 6 and 7. In all models, we conducted post-hoc tests to evaluate differences between the ORs for 1 to 2 versus 3 or more childhood maltreatment incidents for each type of maltreatment.
Source...