The Collaborative Psychiatric Epidemiology Studies (CPES), sponsored by the National Institute of Mental Health, includes three nationally representative surveys of U.S. adults. These surveys were designed to provide epidemiologic data regarding mental disorders using standardized diagnostic criteria. The three surveys include the National Comorbidity Survey Replication (NCS-R), the National Study of American Life (NSAL) and the National Latino and Asian American Study of Mental Health (NLAAS) (17
). Although independent surveys, the CPES were designed to allow combining of the three datasets by having shared scientific objectives, measurement tools, implementation procedures, and sampling methodologies (18
). The core CPES questionnaire was based on the World Health Organization’s expanded version of the Composite International Diagnostic Interview (CIDI). The present study included only the NCS-R and the NSAL because the NLAAS did not assess ADHD.
The NCS-R was a cross-sectional survey designed to provide representative data of the non-institutionalized population of English-speaking U.S. adults age 18 and over between 2001 and 2003 (19
). The NCS-R used a four-stage probability sampling method, first sampling Metropolitan Statistical Areas and counties, then area segments, followed by housing units within the area segments, and finally eligible respondents within the housing units (18
). The recruitment method included a mailed introductory letter and study brochure followed by a household visit by a trained interviewer to identify eligible respondents and provide informed consent. The survey consisted of face-to-face computer-assisted personal interviews (CAPI). The interview included two parts. Part 1 included the core diagnostic assessment of all 9,282 persons participating in the study (response rate 70.9%). Part 2 was administered to a subset of 5,692 respondents, including all who met lifetime diagnosis of a disorder and a probability sample of all other respondents. Part 2 included questions related to disorders of secondary interest. Diagnostic questions for ADHD were included in Part 2, and were asked only of persons age 18 to 44 (n=3,197) (1
The NSAL was a cross-sectional survey conducted between 2001 and 2003 that targeted English-speaking U.S. adults who self-identified as African American (n=3570), of Caribbean descent (n=1,621) and non-Hispanic Whites living in predominantly black neighborhoods (n=891). A sampling method similar to that of the NCS-R was used (18
). Interviewers made household visits to obtain consent and complete the CAPI. The majority of CAPIs (86%) were completed face-to-face and the remaining by phone. The response rate was 71.5%. The entire sample (n=6,082) completed all components of the CAPI. Analyses were restricted to respondents aged 18 to 44 (n=3,538) to be consistent with the NCS-R. Analyses were performed using SAS version 9.1.2 and Stata version 10.
The analytic sample for this study included 6,735 respondents aged 18 to 44 years old, 3,197 from the NCS-R and 3,538 from the NSAL. In the NCS-R, only a randomly selected sub-sample completed questions related to eating disorders, of whom 1,672 were 18 to 44 years old. Thus, analyses including the binge eating disorder (BED) variable were conducted only among 5,210 respondents, 1,672 from the NCS-R and 3,538 from the NSAL. Because data were missing at random, imputation methods for missing data were not performed. This study was approved by the University of Massachusetts Medical School Institutional Review Board.
Respondents self-reported their height and weight as part of the core interview. BMI was calculated using the standard formula (weight in kilograms/height in meters squared). BMI was included in the CPES as a recoded six category variable: underweight (BMI≤18.4); normal weight (BMI 18.5.–24.9), overweight (BMI 25.0–29.9), class I obesity (BMI 30.0–34.9), class II obesity (BMI 35.0–39.9) and class III obesity (BMI≥ 40.0). For the purposes of the present study, we recoded BMI into a three category variable: normal/underweight, overweight, and obese.
As described elsewhere (1
), childhood ADHD was assessed using a retrospective version of the Diagnostic Interview Schedule for DSM-IV (21
). Respondents who reported 6 or more symptoms of inattention or hyperactivity were administered follow-up questions regarding the remaining DSM-IV criteria. If a respondent met criteria for childhood ADHD, they were then asked whether problems persisted to the present. Clinical reappraisal interviews were performed by clinical interviewers on a subsample of participants using the Adult ADHD Clinical Diagnostic Scale (22
), the ADHD Rating Scale (24
), and an adaptation of the ADHD Rating Scale. Respondents were diagnosed with adult ADHD if they endorsed six or more symptoms of either inattention or hyperactivity within the last six months (DSM-IV Criterion A), two or more Criterion A symptoms before age seven (Criterion B), impairment in two or more areas of living within the past six months (Criterion C), and clinically significant impairment in at least one of these areas (Criterion D). No attempt was made to operationalize DSM-IV diagnostic hierarchy rules (Criterion E). Inter-rater reliability for diagnosis was .78. The clinical reappraisal subsample was weighted to be representative of the U.S. population in the age range of the sample (18–44 years). A detailed description of the clinical reappraisal procedures is reported elsewhere (1
). We grouped the ADHD diagnoses into three mutually exclusive categories: never met diagnostic criteria, met full childhood criteria with no current symptoms (childhood ADHD), and met full childhood criteria with current symptoms (adult ADHD).
Major depressive disorder was diagnosed using the CIDI. Diagnoses were based on DSM-IV criteria. Two non-overlapping probability samples were selected, one to complete a clinical reappraisal interview to evaluate lifetime diagnoses, and a second to complete a clinical reappraisal interview to evaluate 12-month diagnoses. The Structured Clinical Interview for DSM-IV (25
) was used for clinical reappraisal. Respondents were classified using three mutually exclusive categories: lifetime (meeting criteria for a major depressive disorder at some point during their life), past 12 months (meeting criteria at some point in the past year), or never having met criteria.
The CIDI was also used to assign diagnosis of BED; however, while DSM-IV requires binges to persist for 6 months or more, the CIDI required only 3 months or more. In the present study, respondents were classified into three categories: 1) lifetime, meeting criteria for BED at some point during their life; 2) past 12 months, meeting criteria at some point in the past year; or 2) never having met criteria.
Respondents were asked to bring in medication bottles for all medications and medication status was recorded. Because use of stimulants and antidepressants could potentially influence the results, the presence/absence of each medication type was dummy coded and entered as covariates in the analyses. Antidepressants were coded in two ways: those with a weight gain side effect and all others. Demographic factors assessed included age, gender, race/ethnicity and education level, and smoking status.
Distributions of BMI categories according to major depressive disorder, BED, and ADHD status were compared using Chi-square statistics. Multinomial logistic regression was used to determine the odds of overweight and obesity compared to normal BMI among persons with childhood ADHD and adult ADHD compared to those with no ADHD. A crude model was computed followed by a multivariate model adjusting for age, gender, race/ethnicity, education and smoking status. To explore the possible impact of major depressive disorder and BED on the association between ADHD and BMI, mediation models were conducted via methods developed by Baron and Kenny (26
). Accordingly, mediation may be present if the following four criteria are met: 1) the independent variable (ADHD) is associated with the outcome (BMI), 2) the independent variable is associated with the hypothesized mediator (major depressive disorder and BED), 3) the mediator is associated with the outcome, controlling for the independent variable and 4) the association between the independent variable and the outcome variable becomes attenuated or not statistically different from zero when controlling for the mediator. Mediation models were derived separately for BED and major depressive disorder. Each model controlled for demographic factors and smoking status. Although originally proposed for longitudinal designs, mediational methods (26
) may also be valid for cross-sectional data, although inferences of causality are limited (27
The CPES dataset included respondent sampling probability weights to allow inference to the general U.S. population of 18–44 years. The weights took into consideration the multi-stage sampling methods used in each study and the over-sampling of Part I NCS-R respondents for Part II of the questionnaire. Weights for individual datasets and combinations of datasets were included. The analyses use weights for the combined NCS-R Part II and NSAL dataset.