Cohort Profile: The National Longitudinal Study of Adolescent to Adult Health (Add Health)

The National Longitudinal Study of Adolescent Health (Add Health) was developed in the 1990s in response to a mandate from the United States Congress to fund a study of adolescent health, and was designed by a team of multidisciplinary investigators from the social, behavioural and biomedical sciences. The original purpose of Add Health was to understand the causes of adolescent health and health behaviour, with special emphasis on the multiple contexts of adolescent life. To achieve this scientific goal, Add Health sampled the school and family environments in which young people live their lives, which included data on peer relationship dyads, parents, siblings, neighbourhoods and communities, and provides independent and direct measurement of these complex environments over time. As the cohort transitioned into adulthood, research objectives turned to understanding how adolescent experiences, behaviours and contexts are linked to health and achievement outcomes in adulthood, and the name of the study was officially changed to The National Longitudinal Study of Adolescent to Adult Health in 2014. Add Health is housed at the Carolina Population Center at the University of North Carolina (UNC) and has been led by two principal investigators and project directors: J Richard Udry from 1994–2004; and Kathleen Mullan Harris from 2004 to the present. Add Health is a nationally representative cohort study of more than 20 000 adolescents in grades 7–12 (aged 12–19) in the USA in 1994–95, who have been followed through adolescence and into adulthood with five in-home interviews in 1995 (Wave I), 1996 (Wave II), 2001–02 (Wave III), 2008–09 (Wave IV) and 2016–18 (Wave V).1Figure 1 displays the sampling design for selecting the original cohort. A school-based design selected 80 high schools and a paired feeder school from a list of all high schools in the USA in 1994. An in-school questionnaire was administered to more than 90 000 students in grades 7–12, who attended these schools during the 1994–95 school year, and school administrators also filled out a questionnaire about the school. Sampling structure. School rosters from the 1993–94 school year provided the sampling frame for a second level of sampling for a 90-min in-home interview with an adolescent and a 30-min interview with one parent. A grade- and gender-stratified core sample was selected from each school pair, representing a self-weighting nationally representative sample of 12 105 American adolescents in grades 7–12 in 1995. Based on responses to the in-school survey, specific subpopulations were oversampled for purposes of providing sufficient numbers for research on vulnerable and otherwise rare populations, including ethnic (Cuban, Puerto Rican and Chinese), genetic relatedness to siblings (identical/fraternal twins, full/half siblings and unrelated adolescents living in the same household), adoption status and disability samples. Black adolescents with highly educated parents were also oversampled. For two large schools and fourteen small schools, interviews with all enrolled students were attempted to create a special saturation sample. The core sample plus the special samples yield a total of 20 745 adolescents. This Wave I in-home sample represents the national cohort of adolescents in grades 7–12 in the USA in 1995, which is followed prospectively. Because school rosters from the year preceding sample selection were used as the sampling frame for the prospective cohort, high school dropouts over 2 years (e.g. 1993–94; 1994–95) were eligible for sample selection, resulting in little bias due to high school dropouts.2 For more details on design, see Harris 2010 and Harris et al., 2013.3,4 Figure 2 shows the longitudinal design of Add Health. The Wave I in-home adolescent cohort has been followed up with four subsequent waves spanning 20+ years. In 1996, all adolescents in grades 7 through 11 in Wave I (plus 12th graders who were part of the genetic and adopted sample) were re-interviewed for Wave II (n = 14 738); the decision not to follow up the seniors who were in grade 12 at Wave I was design-based. The Wave II sample were in grades 8 through 12. A follow-up school administrator interview measured change in school context from 1995 to 1996. Longitudinal design. The original cohort was followed through their transition to early adulthood with a Wave III in-home interview in 2001–02 when the sample was aged 18–26 years (n = 15 197). A sample of 1507 partners were randomly selected during the in-home interview and interviewed, filling quota samples of about 500 married, 500 cohabiting and 500 dating partners. Wave IV re-interviewed the original cohort as they settled into young adulthood in 2008–09 when the cohort was aged 24–32 years (n = 15 701). Wave V followed the cohort to the end of young adulthood when they were aged 32–42, with continuous interviewing using a mixed-mode protocol during 2016–18. Finally, the Add Health Parent Study completed a 20-year follow-up of a subset of the parents of Add Health respondents during 2015–17 (n = 3006). Add Health uses state-of-the-art methods and techniques for panel maintenance and tracing to locate and schedule an interview with all living eligible respondents, including those who may have been non-responsive in a preceding wave. Table 1 presents response rates for the eligible sample at each completed wave of interviews. Response rates have been quite high, highest when the interval between interview waves is short but remarkably high even when the interval is over 5 years at Waves III and IV. The transition from adolescence to early adulthood and the young adult period is an especially transient phase of the life course and it is difficult to track and locate young people. Add Health has done exceptionally well, with response rates of 77.4% and 80.3% at Waves III and IV. Response rates in Add Health for the eligible sample at each wave of in-home interviews By design, respondents who were in the 12th grade at Wave I and who were not part of the genetic sample were not interviewed at Wave II. Response rate at Wave III is based on 15 170 respondents who had data at Wave I. An additional 27 respondents without Wave I data were included at Wave III as part of the genetic subsample. Wave V response rates are not provided because data collection is ongoing, and only a subset of Wave V respondents completed in-home interviews. Response rates in Add Health for the eligible sample at each wave of in-home interviews By design, respondents who were in the 12th grade at Wave I and who were not part of the genetic sample were not interviewed at Wave II. Response rate at Wave III is based on 15 170 respondents who had data at Wave I. An additional 27 respondents without Wave I data were included at Wave III as part of the genetic subsample. Wave V response rates are not provided because data collection is ongoing, and only a subset of Wave V respondents completed in-home interviews. There has been differential attrition by gender, age, socioeconomic status, urban residence, immigrant status and race across time, with higher response rates for female, younger, higher socioeconomic status, urban, native-born and White respondents at Waves III and IV. These attrition patterns are consistent with most longitudinal cohort studies. Add Health response rates exceed other national studies with multiple year intervals between waves (e.g. National Survey of Families and Households 2001–03 wave had a 55% response rate; Midlife in the United States 2004–06 interview had a 75% retention rate).5,6 At each wave, Add Health analysed whether patterns of attrition pose any bias to estimates of survey outcomes.7–10 In general, non-response analyses compare respondents and non-respondents on a range of demographic, health, behavioural and attitudinal indicators measured at baseline, and estimate the extent to which differences between respondents and non-respondents introduce bias in study results. Results indicated that total and relative biases, remaining after study estimates were adjusted with final sampling weights, were minimal and that the sample at each wave adequately represented the same population as the Wave I sample. Analysis of bias due to attrition at Wave IV indicated low rates of bias that rarely exceeded 1%, which is small relative to the 20% to 80% prevalence rates for most of the baseline indicators. Despite common patterns of attrition over time, the design strategy to re-interview the original Wave I cohort at each follow-up wave minimizes non-response bias and continues to adequately represent the original cohort of 7-12th graders in US schools in 1995. Add Health contains unprecedented environmental, behavioural, psychosocial, biological and genetic data from early adolescence into adulthood on a large, nationally representative sample with extensive racial, ethnic, socioeconomic and geographical diversity.4 Longitudinal survey data on respondents’ social, economic, psychological and physical well-being is combined with contextual data on family, neighbourhood, community, school, friendships, peer groups and romantic relationships, providing unique opportunities to study how psychological characteristics, social environments and behaviours beginning in early adolescence are linked to health and well-being in adulthood. Extensive longitudinal life histories of health-related behaviour are available, including physical activity, risk behaviour, substance use, sexual behaviour, civic engagement, education and multiple longitudinal indicators of health status, such as general health, chronic illness, overweight status and obesity, mental health, disability, health promotion and sleep. Objective measures of health were collected across all waves, including anthropometrics, sexually transmitted infection (STI) test results including human immunodeficiency virus (HIV), DNA and an expanded set of biomarkers in adulthood (Waves IV and V) (including blood pressure and pulse, measures of glucose homeostasis, lipid metabolism, inflammation, immune and renal function and a medications inventory). Below we describe the innovative multilevel data that have provided unprecedented research opportunities for a multidisciplinary scientific community. The clustered design of Add Health makes possible unique contextual levels of measurement, shown in Table 2. School-context data come from school administrator reports on school policies, health services and other school characteristics and from the in-school interviews of students whose aggregated responses represent school census measures. From respondent reports of colleges attended, college context data have been linked to individual records. Family-context data come from parent questionnaires, adolescent in-school and in-home questionnaires and interviews with siblings and additional adolescents living in the same household. Contextual levels of measurement in Add Health W1 = adolescent in- school and in-home, parent and school administrator surveys, geocodes. W2 = adolescent in-home, school administrator and geocodes. W3 = young adult in-home and partner sample surveys, geocodes, biomarkers. W4 = young adult in-home, biomarkers, geocodes. W5 = adult survey, biomarkers, geocodes. Additional variables from administrative datasets (e.g. US Census, Centers for Disease Control and Prevention, National Center for Health Statistics, Federal Bureau of Investigation, National Council of Churches, Common Core of Data, Private School Survey). , variable available; o, planned variable construction. Contextual levels of measurement in Add Health W1 = adolescent in- school and in-home, parent and school administrator surveys, geocodes. W2 = adolescent in-home, school administrator and geocodes. W3 = young adult in-home and partner sample surveys, geocodes, biomarkers. W4 = young adult in-home, biomarkers, geocodes. W5 = adult survey, biomarkers, geocodes. Additional variables from administrative datasets (e.g. US Census, Centers for Disease Control and Prevention, National Center for Health Statistics, Federal Bureau of Investigation, National Council of Churches, Common Core of Data, Private School Survey). , variable available; o, planned variable construction. Adolescents were asked to nominate friends and sexual and romantic partners from the school rosters in the in-school and in-home surveys at Waves I and II. Peer networks characteristics can be constructed by linking friends’ data and constructing variables based on friends’ responses, and similar measures can be constructed for linked romantic and sexual partners. These peer- and dyad-context measures constitute the social network data, including information on friendship networks, sexual networks and friendship and relationship dyads. Respondents’ home residences have been geocoded at each interview wave, and contextual data on the neighbourhood, community and state have been merged to all individual records. Nearly 12 000 environmental data elements at multiple geographical levels are available across waves. This includes such information as race, ethnic, foreign-born and religious denomination composition, poverty rates, crime statistics, STI prevalence, divorce and child support laws, welfare policies, cigarette taxes, the proximity and number of parks, sidewalks, recreation centres, fast food restaurants, alcohol outlets and other physical and social characteristics of the environments in which young people live. Table 3 shows the array of survey and biological data in Add Health. The top panel lists the domains covered by the survey instruments at each wave, including individual-level data on household and family structure, personality, religiosity and spirituality, relationships, sexual behaviour, contraception, pregnancy, children and parenting, sleep patterns, physical activity, diet, substance use/abuse, violence, delinquency, involvement with the criminal justice system, education history, work experiences, military service, chronic and disabling conditions, injury, mental health, suicide and health service access and use. Even though respondents were first interviewed in early adolescence, there are data on infancy (birthweight) and childhood (e.g. maltreatment, chronic conditions, attention deficit hyperactivity disorder) and complete data on fertility outcomes (there were more than 14 500 births to Add Health respondents by Wave IV). Survey and biomarker domains across Waves I-V in Add Health Survey and biomarker domains across Waves I-V in Add Health The bottom panel of Table 3 shows the biological measures available across waves. The original study design included important features for understanding biological processes in health and developmental trajectories across the life course, including an embedded genetic sample with more than 3000 pairs of adolescents with varying biological resemblance (see Figure 1) and measurement of height and weight to track the obesity epidemic. At Wave III, urine and saliva samples were collected to test for STI and HIV,11,12 and buccal cell saliva was collected from twins and full siblings in the genetic subsample for DNA extraction.13 An expanded set of biological measures were collected at Wave IV, including biomarkers of cardiovascular health (blood pressure, pulse), metabolic processes (waist circumference, glycosylated haemoglobin, blood glucose, lipids), immune function (Epstein-Barr virus), inflammation (C-reactive protein) and a medications inventory. Repeat biomarker measures were collected at Wave V, including new markers of renal disease. Saliva DNA was collected from the full sample at Wave IV. Candidate loci in the dopamine and serotonin pathways have been genotyped and disseminated to the scientific community.14 Genome-wide genotyping was completed on 10 974 Wave IV respondents who consented to archive their specimens for further testing, and genome-wide association study (GWAS) data are available from the database of Genotypes and Phenotypes (dbGaP). Add Health maintains a biospecimen archive available for ancillary studies. Add Health has a large and multidisciplinary user base of more than 50 000 researchers around the world, who have published over 3500 peer-reviewed articles in more than 750 different disciplinary journals, and has been the data source for more than 800 master’s theses and dissertations. Publications are listed at https://www.cpc.unc.edu/projects/addhealth/publications. Early publications focused on the role of social context in the development of adolescent health, behaviour, expectations and attainment, finding important influences of family and school connectedness,15 peer influence,16–18 romantic relationships,19 and neighbourhoods.20–22 For example, adolescents with a greater number and higher quality of connections to their school and family had better physical and mental health and higher attainment than youth with Adolescents whose friendship networks included friends with highly parents were than those whose friends had parents to studies that romantic in adolescence can adolescent and is with higher rates of obesity and weight in publications an of chronic young including a prevalence of and prevalence of the longitudinal data, researchers have the developmental and health pathways to young adult Add Health data support longitudinal studies of partner substance and health during the early life course from adolescence into young Add Health has the obesity and outcomes for adolescents. In adolescence of the sample were in 2001–02 when the cohort was aged the to in of the cohort at 24–32 was on these longitudinal data, The and the of obesity early in that adolescents were more to obesity in young adulthood with overweight by a risk of to individual obesity trajectories from adolescence at Wave II to young adulthood at Wave III into not who were who during the transition to young adulthood, and who were adolescence and young adulthood, As shown in Figure greater to obesity during adolescence and young adulthood is with a higher of high and sleep in adulthood. from adolescence to young adulthood with multiple health outcomes in adulthood (n The unique design and of the sample possible health research on special including the adopted youth living with parents sexual and adopted adolescents are more to suicide than their adolescents are at higher health risk on a range of indicators with adolescents who only one and more and than There is a large and of research that the genetic data with the longitudinal environmental data to the of and in health and behavioural of genetic research articles have these on a range of including risk substance sexual and friendship and Genome-wide association study (GWAS) data were for the pairs and Wave IV archive the of a large number of These the new differences in the education and of school environments in the education association with and family and education role in social and cohort differences in the genetic relationship between education and In innovative new research is providing human of genetic in which the of individual behaviour, for and Figure for social genetic of and friends on attainment, and The the of school and of for attainment in the top in the second panel and height in the The is the baseline of on outcomes in a with other and the represent the adjusted for individual-level The results that the of an and friends the attainment, an height is with the height of Add Health has in for new including attainment, height and alcohol genetic Add Health research uses biomarker data to social and behavioural with measures of research has the between a childhood and STI social status and and and life course of social and life for The role of social and social context in pathways was for the first young the for in the early in life and biological are to family and urban young from social is with mental health but physical health for young with The of Add Health from contextual and national design. The adolescent social context and peer network data, in are unique because they not on to an of an national of people who live in all 50 and come from race, ethnic, geographical and socioeconomic and ethnic with sufficient numbers to of Puerto and and and vulnerable including with children and adopted and sexual genetic sample of over 3000 pairs of with varying biological and longitudinal data from respondents and their longitudinal social, behavioural and biological data beginning in early adolescence and into extensive longitudinal multilevel data beginning in early adolescence on life and social and physical including family, school, neighbourhood, community and social measures of health including blood pressure, glucose, virus circumference, and and DNA on 000 and genome-wide genotyping on the full sample at Wave and collection of DNA on twins and full data be available in the including and data (see Table a of the of survey data measurement of specific and a for survey and biomarker measures. are available to researchers in representing a subset of and high which are only to which can only be used in a data to Add Health data to other and high school data, which are available in data access the and of respondents data access to a range of including data access study and can be at data can be Add Health is an longitudinal study of a nationally representative US cohort of 20 745 adolescents in grades (aged in includes four in-home interviews in 1996, 2016–18. attrition has been with response rates from to across follow-up waves, and attrition bias has been The study unprecedented environmental, behavioural, psychosocial, biological and genetic data from early adolescence into adulthood with extensive racial, ethnic, socioeconomic and geographical Add Health has a large, multidisciplinary user base of over 50 000 researchers around the world, who have published over 3500 research the Add Health cohort at the of the obesity with for health and social genetic of and on health and Add Health datasets are to a data to the of information and in of the see Add Health is by the National of Health and with from other and The full list of can be at This research uses data from Add a project by Kathleen Mullan Harris and designed by J Richard and Kathleen Mullan at the University of North Carolina at and by from the National of Health and with from other and is due to and for in the original design. also to data from National of Health and to Kathleen Mullan Harris and to Kathleen Mullan and of