The Behavior and Mind Health (BeMIND) study: Methods, design and baseline sample characteristics of a cohort study among adolescents and young adults

Abstract Objectives The Behavior and Mind Health (BeMIND) study is a population‐based cohort study of adolescents and young adults from Dresden, Germany. The aim is to investigate psychological and behavioral factors linked to a range of mental disorders and health behaviors and their interaction with social‐environmental and genetic/biologic factors. Methods A random sample of 14–21 year olds was drawn from the population registry in 2015. The baseline investigation was completed 11/2015–12/2016 (N = 1,180). Assessments include standardized diagnostic interview, cognitive‐affective tasks, questionnaires, biosamples, and ecologic momentary assessment in real life with combined actigraphic/geographic monitoring. In the family study component, parents completed similar assessments and provided information on child's early development. Results The participation rate (minimum response proportion) was 21.7%; the cooperation rate was 43.4%. Acceptance and completion of study components were high. General health data indicate that more than 80% reported no or only mild impairment due to mental or somatic health problems in the past year; about 20% ever sought treatment for mental health problems or chronic somatic illnesses, respectively. Conclusions Data from BeMIND baseline and follow‐up investigations will provide novel insights into contributors to health and disease as adolescents grow into adulthood.

Although behavioral factors, as defined within a larger psychological or behavioral science perspective, are deemed a core contributor to almost all ill-health conditions, their consideration and objective assessment in epidemiological studies have so far been limited (Wittchen, Knappe, Andersson, et al., 2014). Improved knowledge on the behavioral and psychological determinants, including cognitiveaffective factors and decision-making processes, in the evolution of mental disorders and health risk behaviors contributing to somatic disease, and the interplay of these factors with genetic/biological and environmental factors, may improve etiopathogenetic models and targeted interventions suitable for changing disease trajectories.
We therefore launched a prospective-longitudinal epidemiological study focusing on mental disorders and health risk behaviors in adolescents and young adults, in which traditional subjective, retrospective assessments of mental and behavioral health and disorders, as

| Study design
The Behavior and Mind Health (BeMIND) study is designed as a cohort study in a general population sample of adolescents and young adults from Dresden, a major city in the eastern part of Germany. The study comprises a baseline investigation and 1-and 3-year follow-up investigations to examine developmental trajectories of mental disorders and health risk behaviors related to noncommunicable somatic disease ( Figure 1). In addition, the study includes a family study component.
To increase the overall sample size and to allow for replications of exploratory findings, a second smaller baseline-cohort has been independently sampled approximately 2 years after the original baselinecohort (not detailed herein). The study protocol and its amendments were approved by the ethics committee of the Technische Universität Dresden (TUD; EK38110214).

| Sampling
The 14-to 21-year-old population living in Dresden, Germany, represents the study's target population. An age-and sex-stratified random sample of 14-21 year olds was drawn from the population registry of the city of Dresden in 2015 with the aim to recruit~1,000 adolescents and young adults to become part of the prospective-longitudinal BeMIND study. Because it was deemed more important to ensure a sufficient sample size for each age group than to resemble the age/sex distribution of the target population (considerably more young adults than adolescents live in Dresden due to two large higher education institutions), younger individuals were oversampled. Sample size was determined based on a priori power calculation; Data S1A

| Field work and procedures
Address lists of randomly selected adolescents and young adults with primary living address in Dresden were provided by the city's resident registry office. In case of minors, names and address of the legal guardians were also provided. A personal invitation letter was sent by the BeMIND study team with information about the study, a response sheet, and a postage-paid return envelope. In the case of minors, letters were addressed to both subjects and parents. The information covered aims, approach, and comprehensiveness of the BeMIND study program; 50€ were offered for participating in all baseline study components (overall 6-10 hr on two assessment days and during 4 days in real life; details below); parents were offered 30€.
Individuals/families indicated their interest to participate and contact information or the reasons for nonparticipation on the response sheet.

| Assessments and measures
An overview of the BeMIND baseline assessments is provided in

| Ecological momentary assessment
In the EMA assessment (Shiffman, Stone, & Hufford, 2008), index subjects answered questions presented via a self-developed study smartphone app on eight occasions per day over the course of four consecutive days (two weekdays and the weekend). Three survey sets were configured (one morning, six midday, and one evening assessment), each containing 203-248 items, most of which related to the time window since the previous assessment. Implemented branching rules allowed an adaptive answering of the questions so that the study load and time burden could be reduced (~3 min per assessment).
The assessments covered current mood and emotions, perceived stress, substance use, daily activities, approach/avoidance behaviors, eating behavior, and physical activity (Table 1) On the day before the first assessment, three questionnaire sets were presented in order to familiarize participants with the EMA modality. On the day after the last assessment, a postassessment questionnaire asked for specific impressions and difficulties during the assessment period.
Collected data were stored locally on the smartphone and HRV device and transferred to the study server after the participant returned the equipment to the study center. GPS data were recorded on the smartphone (sampling frequency 0.2 Hz), and HRV (in millisecond accuracy) and physical activity data (three-axis acceleration sensor motion data with a sampling frequency of 12.5 Hz) were recorded continuously over the 4-day period. Participants were asked to wear the HRV sensor over the entire EMA period (including the night), which was attached to the skin with electrodes on the upper body and only to be taken off before contact with water. Replacement electrodes were made available to the subjects in sufficient quantities.
On the two weekdays of the EMA period, the awakening reaction and the daily variation of cortisol levels was determined by collecting saliva samples immediately after awakening, 30 min later, and 30 min before bedtime. Subjects were shown how to provide the saliva samples using Salivettes (Sarstedt, Nümbrecht, Germany) and received reminders by the smartphone. The actual saliva sampling time was recorded with the use of MEMSCAPS (MEMS 6 TrackCap container, Aardex Ltd., Switzerland). Participants were asked to store the saliva samples immediately in their freezer before returning them at the second personal appointment to the study personnel, who immediately stored the samples in a laboratory freezer. After thawing, saliva samples were centrifuged for 10 min at 4,000 rpm and cortisol concentrations were determined using a commercially available chemiluminescence assay (CLIA, IBL-Hamburg, Germany).
Participating parents completed a paper-and-pencil-based EMA twice a day (morning and evening) over the course of 4 days (two weekdays and weekend) and also provided three saliva samples on the two weekdays.

| Biosamples and anthropometric measurements
Besides saliva samples over the course of two EMA days for diurnal cortisol analyses, a hair sample was taken by trained study personnel for assessment of long-term cortisol secretion both from index subjects and participating parents (Kirschbaum, Tietze, Skoluda, & Dettenborn, 2009). Two to three 3-cm-long hair samples were collected, scalp-near from a posterior vertex position. The selected hair was about the diameter of a pencil (~3 mm) and stored in foil. After the hair sample, participants filled in questions about hair treatments (e.g., dyeing), alcohol consumption, smoking, sporting activities, and recent illnesses or stressful situations (Stalder & Kirschbaum, 2012).
In addition, two 9-ml EDTA blood samples were collected by venipuncture from index subjects and participating parents using the vacuum method for DNA analyses (S-Monovettes; Sarstedt, Nümbrecht) and were stored at −80 C. For participants not concurring with the blood withdrawal, buccal swabs (Biozyme, Wien) for sampling buccal mucosa were offered.
Standard digital scales were used with index-subjects and participating parents for anthropometric measures (weight, height, and waist circumference). Systolic and diastolic blood pressure were measured after a resting period of at least 5 min, and three times on the arm without blood sampling of seated subjects using an oscillometric digital blood pressure monitor (705IT, OMRON).

| Questionnaire assessments
Index subjects completed a range of questionnaires during the personal assessments in the study center and during a web-based online assessment between the two personal appointments.
Questionnaires assessed putative distal and proximal individual and environmental risk and protective factors covering constructs from various domains (see Table 1). The total time for filling in the questionnaires was approximately 60-90 min online and 30-60 min on site.
Participating parents completed a shorter version of the questionnaire assessment and provided additional information on pregnancy, birth, and early development of the index child.

| Behavioral tests and tasks
Index subjects completed paper-and-pencil tests and computerized tasks on executive functioning, cognitive control, and decision making during the second personal appointment. The total time for this behavioral experimental assessment was~70 min.
Three paper-and-pencil tests were conducted by trained examiners.
For a nonlanguage-based measure of intelligence ("speed of processing"), the number connection test (German ZVT; Oswald & Roth, 1987) was conducted. The ZVT is comparable with the trailmaking test (Reitan, 1955), which was also applied. To measure working-memory's number storage capacity, a digit span task was used (Richardson, 2007).
The matlab-programmed task battery included five tasks that were presented in the same order to every subject and that started automatically. Subjects were introduced to the set up by trained study personnel. During task battery performance, subjects were alone in a quiet room and allowed to do self-paced breaks between the individual tasks. The tasks included the number Stroop task (Stroop, 1932) to measure inhibition (using a mouse-click version in contrast to common keyboard-based response), the emotional face approach-avoidance task (Face-AAT; Heuer, Rinck, & Becker, 2007) to investigate reactions to emotional faces, the AX continuous performance task (AX-CP; Cohen, Barch, Carter, & Servan-Schreiber, 1999) to measure context processing, goal maintenance and updating, a novel variant of an intertemporal choice task (ITC, Scherbaum, Dshemuchadse, Leiberg, & Goschke, 2013) to measure individually determined advantageous and disadvantageous choices, and a go-nogo task (Wolff et al., 2016) to measure inhibition.
Participating parents completed the same paper-and-pencil tests and a shortened task battery (Face-AAT, AX-CP, and ITC).

| Quality assurance
To ensure high data quality different steps were followed including Prior to the study, all assessments and study procedures were tested for feasibility, practicability, and time requirements. Novel assessments were also checked for reliability and validity. The entire study procedure was piloted with 20 subjects from the general population. To avoid false or missing values and for a time-economic study, execution value ranges were defined and branching rules were applied. A comprehensive IT infrastructure has been developed for several study aspects such as recruitment, participant management, scheduling appointments, data entry, and integration.

| Recruitment flow and study population
Invitation letters were subsequently sent to overall 6,321 sampled individuals/families. Of these, 14.1% were ineligible, mostly due to the fact that they were not residing under the provided address ( Figure 2). Of the remaining 5,428 individuals, 1,180 were assessed resulting in a minimum response (participation) proportion of 21.7% (AAPOR, 2016: formula RR1). The cooperation rate among those with known eligibility was 43.4% (formula COOP1). The main reason for nonparticipation was refusal, mostly due to lack of time or interest, followed by failure to contact and arranging suitable appointments; 42.8% of all invited individuals/families did not answer the invitation letter, two reminder letters, and an anonymous nonresponder questionnaire. Assuming that the proportion of eligible subjects in these cases with unknown eligibility is the same as the proportion of eligible subjects among those with known eligibility, the estimated overall response proportion was 24.8% (formula RR3) and the overall cooperation rate was 49.5% (formula COOP3). Table 2 shows the sample and the total population of 14-to 21-year-old people in Dresden by age and sex. Participation was generally higher in females than males.
In order to improve representativeness of the sample, we apply sample weights. The sample is divided into 16 strata according to the 16 possible combinations of sex and age. The sample weights are calculated so that after weighting adjustment, the relative sample frequencies of these groups equal the corresponding relative frequencies of the strata in the population of the 14-to 21-yearold people of Dresden (the target population). Note that this accounts for (a) intended (sampling probabilities differing over age groups, by design) and (b) unintended discrepancies. The distribution of other determinants of participation is also adjusted for to an extent that these are associated with sex and age. The age and sex distribution of the target population were taken from the Registration Office of the city of Dresden (Landeshauptstadt Dresden-Kommunale Statistikstelle, 2016b) and can be considered highly accurate.

| Sample characteristics of index subjects
The mean age of the N = 1,180 index participants at baseline was 17.3 years (SD 2.3) and was similar for boys (17.1, SD 2.3) and girls (17.4, SD 2.2). As shown in Table 3, most participants had German Nationality (97.5%); 61.2% still went to school, 19.6% were university students; 99.3% of the index subjects had never been married, and 74.7% lived with a parent.  Table 4 reveals that the proportion of people with high education level is higher in the BeMIND cohort than in the Microcensus-sample for F I G U R E 2 Flow of participant recruitment in the BeMIND study the 14-to 21-year-old Dresden population. Additional weighting procedures accounting for the overrepresentation of high education in the BeMIND sample will, therefore, be used for sensitivity analyses in the future. Furthermore, the proportion of people living with parents and the proportion of people attaining school is higher in the BeMIND cohort than in the Microcensus sample. To a certain extent, this just reflects the fact that at age 16-18, people attaining higher secondary education at school (assigned to the high education group) were more likely to participate in the BeMIND study than people already attaining job training (usually assigned to middle education level). vs. 11.9%) and reported to be a current smoker (20.7% vs. 11.9%).

| Family study
Overall, 709 parents of 549 index participants were directly assessed with similar procedures. Further, 207 parents completed at least a minor assessment via online questionnaire. Thus, 916 parents of 677 index subjects provided data.

| Completion of baseline study assessments
Data S1B provides an overview of the numbers of subjects who provided data to the individual baseline study components. Participation T A B L E 2 Demographic distribution of the adolescent/young adult population in the city of Dresden, the potentially eligible sample, and the participants unweighted and weighted in each study component was high with around 90% or more of the subjects providing some or even complete data.

| DISCUSSION
The For analysis and interpretation of BeMIND study data, several limitations need to be considered. First of all, the response rate is relatively low (AAPOR-RR1: 21.7%; RR3: 24.8%). The cooperation rate was more favorable, yet still below 50%. Decreasing participation is a general problem in epidemiologic study and survey research (Galea & Tracy, 2007), a trend that has been continuing in more recent health studies in Germany and other European countries (Loeffler et al., 2015;Scheidt-Nave et al., 2012;Scholtens et al., 2015;Volkert et al., 2017) with particularly low participation of adolescents and young adults (Keeble, Baxter, Barber, & Law, 2016;Lange et al., 2017).   below one third of the BeMIND sample size). Therefore, we decided to weight by default for age and sex as a consequence of our sampling scheme only and not additionally for education. However, we will use such a related weighting variable for sensitivity analysis. No comprehensive epidemiologic study can ever achieve complete representativeness, that is, a sample that represents the source or even target population in every aspect. However, bias due to selection strongly depends on the target parameter to be estimated (e.g., prevalence/incidence rate, association, and causal effect). This bias occurs only if determinants of participation are related to the target parameter (e.g., if education is a moderator of the association between an exposure such as a life event and an outcome such as depression). Then the amount of this bias depends on the sign and magnitude of that association and the difference of the determinant's distribution between sample and target population. Roughly speaking, larger bias is more likely when estimating marginal parameters (prevalences/incidences) than when estimating associations or effects (Little, Lewitzky, Heeringa, Lepkowski, & Kessler, 1997); and bias might be further smaller when moderators of associations are investigated. This assumes that the individual variation in a parameter (heterogeneity) decreases with an increasing number of variables involved. In the BeMIND study, these determinants could include concern and experience of mental health issues (with exposed individuals expected to have higher participation rates). Percentages of main categories are percentages from those with available information, percentage of category "unknown" is raw percentage from the complete sample.
Missing data within the sample is another limitation and source of potential bias. Among BeMIND study participants, completion of individual assessments was generally high. We will check types of missing data (at random/not at random) and apply appropriate methods (e.g., imputation techniques, sensitivity analyses; Pedersen et al., 2017).
Another limitation refers to generalization of the BeMIND results to adolescents and young adults in Germany (or other countries).
Given the regionally restricted sample (Dresden, Germany), it is important to consider how Dresden compares with other German regions.
The city of Dresden is the capital of one of the 16 states of Germany, located in the east of Germany. Dresden has a total population of 548,800 inhabitants (in 2015 when sampling occurred). In contrast to other, mostly rural areas in the eastern part of Germany, the mean age of the population is relatively low (mean: 42.9 years) and as such rather comparable with other large cities in eastern Germany (such as Berlin) and to most regions in western Germany. Similar applies to population density. Compared with other large German cities, there is a relatively low proportion of migrants (6.2%, Landeshauptstadt Dresden-Kommunale Statistikstelle, 2016b). The unemployment rate is in the medium range (7.4%, Landeshauptstadt Dresden-Kommunale Statistikstelle, 2016a). In eastern Germany, including the city of Dresden, there is relatively high population movement, particularly among young adults. For the large-scale follow-up assessment, resources for travel are therefore budgeted. The address information is kept up to date by reminding subjects during the regular newsletters to return their new living information to the study center. New addresses will also be obtained from other information sources as available contact persons and-if needed-from the population registry.
To contrast the limitations of the BeMIND study with the strengths, the EMA of mood, emotions, and behaviors in real life with combined objective measures of activity and stress, as well as the laboratory-based behavioral indicators of cognitive-affective functioning and decision making go beyond traditional assessments of epidemiological surveys on mental and behavioral health. They allow for novel insights into the dynamic networks of symptoms and behaviors (Borsboom, 2017;Bringmann et al., 2016) as well as their predictors and predictive potential. Thus, the BeMIND study will advance our knowledge on the behavioral and psychological factors contributing to health and disease as adolescents grow into adulthood and provide new avenues for early detection and personalized interventions.