A prospective protocol for remotely investigating brain-behaviour-genetics associations in adolescent patients in a paediatric health system with pre-existing clinical brain MRIs

Abstract

Introduction

Adolescence is a critical period marked by rapid brain development and the onset of many mental health disorders. Brain MRI studies during adolescence, especially when paired with behavioural phenotypes and information about genetic risk factors, hold promise to advance early identification of mental health risk and spur the creation of targeted treatments to improve patient function, prognosis and quality of life. However, prospective neuroimaging is costly and time-intensive, and individuals who participate may not be reflective of the general population. These challenges are compounded when examining adolescents, as many families lack the time, energy or resources to participate in studies that use research-grade imaging. Repurposing clinical MRIs obviates many of the challenges of neuroimaging research. Here, we describe the brain-behaviour-genetics study protocol. This protocol describes procedures used to recruit participants with recent high-quality clinical brain MRIs and prospectively acquire genetic and sociobehavioural data, resulting in a highly cost-efficient design that harnesses a vast and underused neuroscientific resource.

Methods and analysis

The brain-behaviour-genetics protocol aims to recruit 1000 adolescents who have clinical brain MRIs contained in Children’s Hospital of Philadelphia’s electronic health record. One or both parents of the adolescent proband will be recruited when possible. Parents and adolescents will complete a series of self-report scales spanning the domains of mental health, trauma, risk and resilience. Saliva samples will be collected from the adolescent and at least one biological parent, using an at-home saliva collection kit. Subsequent analysis will examine associations between brain development, genetics and behavioural measures in adolescence.

Ethics and dissemination

Approval for the study had been obtained from the Children’s Hospital of Philadelphia’s institutional review board (IRB #23–0 20 851). Results will be published in peer-reviewed journals.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• This study establishes a scalable protocol for a brain-behavior-genetics study that harnesses pre-existing clinical brain MRIs.

• The protocol identifies adolescents with brain MRIs conducted in the last year using the electronic health record of a large paediatric hospital system.

• Youth with brain MRIs without imaging pathology are prospectively recruited.

• Computerised self-assessments are used to measure psychopathology.

• Genetic samples from youth and parents are obtained using mail-out saliva kits returned directly to an institutional biobank.

Introduction

In adolescence, individuals undergo significant brain, cognitive and behavioural development. Developmental differences in this period may impact mental health throughout the remainder of the lifespan. Across mental health disorders, brain MRI is a promising tool for screening and early identification and a putative biomarker that may facilitate creation of targeted treatments to improve patient function, prognosis and quality of life. Given that neuropsychiatric disorders are highly heritable and highly polygenic, combined imaging-genetic studies hold particular promise in characterising relations among brain development, genetic variants and symptom manifestations.^{1 2} Accelerated research in psychiatric imaging-genetics is needed to elucidate these complex developmental pathways and identify at-risk individuals.

Extant prospective brain MRI studies are limited in terms of their scope, scalability and generalisability. Factoring in scan time, study personnel and participant fees, each scan generally costs over US$1000.³ The cost is magnified by the scientific necessity to amass large sample sizes to ensure replicability of brain-gene and brain-behaviour associations.⁴ Although smaller sample sizes are feasible with optimised experimental designs and analytic approaches, a principal challenge in brain MRI research continues to be recruiting a sample large enough to support valid scientific inference and generalisability. Prospective brain-behaviour-genetics studies that recruit patients and healthy volunteers may struggle to recruit representative cohorts. Indeed, individuals who face structural barriers such as health issues or economic precarity often lack the resources necessary to participate in these prospective studies, leading to representation of ‘wealthier and healthier’ in samples rather than the broader population.^{5 6}

Considering these practical limitations on prospective large-scale brain MRI studies and the increased reliance in medicine on electronic health records (EHR), the ability to leverage clinically acquired brain MRIs is a promising complementary approach to investigate associations among brain, behaviour and genetics. As of 2025, the Children’s Hospital of Philadelphia (CHOP) EHR contains more than 250 000 brain MRIs obtained during paediatric healthcare encounters.⁷ These clinically obtained brain MRIs can be linked with clinical, demographic, genetic and assessment data collected both retrospectively via their EHR and prospectively in a research setting.^{8 9} Combining clinically acquired brain MRIs with prospective data collection (e.g., genetic samples, behavioural and exposomic phenotypes) would simultaneously advance scientific discovery while reducing cost and participant burden.

Below, we describe our approach to conducting such an imaging-genetic study within one of the nation’s largest paediatric hospital settings. Our premise is that retrospective analyses of clinically acquired brain MRIs are of sufficient quality for scientific discovery and that leveraging pre-existing healthcare systems data will supplement costly, prospective neuroimaging studies. From this perspective, if they meet basic quality requirements, the millions of brain MRIs acquired each year in clinical settings across the world are a valuable and vastly underused resource for investigating neurodevelopment. We overcome prior limitations that hindered the utilisation of these clinical scans, addressing challenges related to clinical informatics and image curation, processing, harmonisation and statistics. The current protocol harnesses clinical brain MRIs available at CHOP with prospectively collected sociobehavioural and genetic data: (1) to demonstrate the feasibility of prospective studies based on recruiting patients with brain MRI scans without serious imaging pathology; and (2) to investigate associations between neuroanatomical differences quantified from clinical MRIs, behavioural measures and genetic variation.

Methods and analysis

Identification of brain MRI scans with limited imaging pathology cohort

Before initiating recruitment, a cohort of clinically acquired MRIs without gross pathology, referred to as scans with limited imaging pathology (SLIP), is identified. While we use the language ‘limited imaging pathology’, it is important to note that patients recruited from a clinical sample cannot necessarily be considered ‘healthy.’ Moreover, graders are instructed to consider all aspects of the report, including the ‘indication’ and ‘history’ sections in which pertinent conditions or diagnoses may appear. To be included, these MRIs are obtained at a CHOP location. Clinical brain MRIs typically include T1-weighted, T2-weighted, fluid attenuated inversion recovery, diffusion-weighted and susceptibility-weighted images. Although there are no specific inclusion criteria based on slice thickness or voxel size, a large proportion of CHOP’s clinical brain MRIs include an approximately 1 mm isotropic T1-weighted magnetisation prepared rapid acquisition gradient echo sequence. The process of determining the SLIP cohort is outlined in detail in a companion protocol paper (Zimmerman et al)¹⁰ but in brief, it involves the following steps. Prior to the scan review, a research assistant without clinical expertise (i.e., ‘grader’) undergoes a training and reliability phase to systematically rate clinical radiology reports. The grading procedures are based on standards developed with expert neuroradiologists.¹⁰ Once reliability is achieved through a trial grading process which generates a Cohen’s kappa score, graders manually review the signed radiology reports (written during clinical care by licensed paediatric neuroradiologists) from the EHR and assign the scan a grade (see figure 1). A three-point grading system is used to reflect the level of pathology noted in the radiology reports. SLIP scans must also be free of gross image artefacts due to motion, orthodontic hardware or other factors highlighted in the radiology report as typically noted within the clinical indication and findings section of the radiology report. SLIP criteria are met based on aggregated scores across independent graders.

Figure 1. SLIP grading flowchart. Clinical MRIs are graded on a 0, 1, 2 scale by at least two independent graders. EHR, electronic health record; SLIP, scans with limited imaging pathology.

Study design

The Brain-Behavior-Genetics study aims to enrol approximately 1000 adolescents with clinically acquired brain scans in the preceding year. Recruitment will take place concurrent to data collection, spanning 5 years from the study start (7 July 24–31 May 29). Participant eligibility requires a clinically obtained MRI of the brain without significant pathology in CHOP medical records. An institutional review b Institutional Review Board (IRB)-approved third party provides the study team with contact information and demographics for eligible participants. Following study consent, in which participants consent to the use of their retrospective brain MRIs along with prospective study procedures, behavioural and exposomic phenotypic data are collected virtually from parents and children who complete a series of behavioural self-report scales, described further below. Scales are completed via REDCap and take less than 1 hour to complete in total (see figure 2). Next, a biospecimen is collected for genetic analysis. Biospecimens could include residual specimens leftover from clinical care like blood leftover from clinical testing or tissue leftover from clinical analysis or a specimen collected specifically for research, like blood or saliva. In most cases, we anticipate biospecimen collection will be completed using Oragene Discover OGR 600 DNA saliva kits. These kits are shipped to the participant’s home with instructions detailing how to provide a sample. In addition to the adolescent proband, saliva samples are collected from one or both biological parents, when possible, for trio analysis. Though participants are solely responsible for producing viable samples, our biological sample collection process is quite straightforward. Clear, printed instructions along with colour-coded tubes labelled ‘mother,’ ‘child’ and ‘father’ have ensured that approximately 99% of families have correctly packaged their DNA. Once saliva collection procedures are complete, participants return samples using enclosed packaging materials and a packing slip. The samples are sent directly to the CHOP Biorepository Resource Center, an institutional core lab that specialises in high quality sample processing, storage and tracking, with the capacity for approximately 2–3 million samples. Saliva samples are incubated for at least 2 hours at 50°C, then two 1800 µL aliquots are prepared for each sample and stored indefinitely in −80°C freezers. After data collection is complete for the study, low-depth whole genome sequencing will be conducted on DNA samples. Participants receive monetary compensation. All procedures were approved by the IRB of CHOP and the University of Pennsylvania. Signed consent (and assent for minors) is obtained from all participants.

Figure 2. Panel A illustrates a flowchart of study procedures and Panel B is the templated recruitment text infographic sent to all families during the recruitment process.

Behavioural self-report scales

The adolescent proband and one parent (‘parent 1’, typically the primary caregiver) complete a series of self-report scales spanning domains of mental health, behaviour, trauma, risk and resilience (see table 1 for complete list of proband scales and table 2 for a list of parent 1 scales). When necessary, scales were modified to exclude questions which may engender reportable events given the asynchronous nature of the protocol. The use of self-administered scales presents a challenge for ensuring study completion, particularly in adolescent populations. Therefore, dedicated participant follow-up and outreach is employed to ensure that the majority of adolescents complete their scales in full. Of note, the proband self-reports their mental health symptoms; parents report family environment and the parents’ own mental health symptoms. Scale selection for the current procedures prioritised commonly used, well-validated scales. Adolescent participants complete a set of six core scales: (1) the Youth Self Report (YSR) version of the Child Behavior Checklist (CBCL) is a 112-item scale used to measure emotional and behavioural problems in adolescents along internalising and externalising domains, modified in this case to remove items relating to self-harm or endangering others¹¹; (2) the Patient Health Questionnaire-8 (PHQ-8A) is an 8-item scale to evaluate depression symptoms¹²; (3) the Prevention through Risk, Identification, Management and Education-5 Screen is a brief, 5-item age-normed screener for subthreshold psychosis symptoms in youth¹³; (4) the Child and Adolescent Trauma Screen is a 9-item scale used to measure traumatic or stressful life events in youth, modified in this case to exclude sensitive questions regarding domestic abuse¹⁴; (5) the Patient-Reported Measurement Information System Polysymptomatic Distress Sleep Disturbances Scale is a 4-item subscale used to measure sleep problems in adolescents¹⁵; (6) the Tanner Pubertal Scale is a 15-item scale used to assess the progression of adolescent pubertal development.¹⁶

Table 1. Adolescent self-report scales.

Construct	Scale	# of Items
Core scales
Total emotional and behavioural problems	CBCL-YSR11	112
Depression	PHQ-8A12	8
Psychosis	PRIME-5 Screen13	5
Trauma	CATS14	9
Sleep	PROMIS PSD15	4
Pubertal status	Tanner16	15
Optional scales
Emotion regulation	RRB Emotion Regulation Subscale17	5
Reward/approach	BIS/BAS-Reward Subscale18	5
Anxiety	GAD-719	7
Victimisation/bullying	ABCD Cyber Bullying20	12
Discrimination	ABCD Perceived Discrimination21	11
Screen time/media use	ABCD Screen Media Activity22	16

ABCD Cyber Bullying, ABCD Cyber Bullying and Peer Victimisation and Perpetration; CATS, Child and Adolescent Trauma Screen; CBCL-YSR, Child Behaviour Checklist Youth Self Report; ABCD Perceived Discrimination, ABCD Perceived Discrimination Measure; GAD-7, Generalised Anxiety Disorder-7; PHQ-8A, Patient Health Questionnaire-8; PROMIS PSD, Patient-Reported Outcomes Measurement Information System Polysymptomatic Distress Scale Paediatric Sleep Disturbances; RRB Emotion Regulation, Risk and Resilience Battery; PRIME-5 Screen, Prevention through Risk, Identification, Management and Education-5 Screen; BIS/BAS-Reward Subscale, Behavioural Inhibition System and Behavioural Activation System Reward Subscale; Tanner, Tanner Pubertal Status Scale Girls and Boys.

Table 2. Parent 1 self-report scales.

Construct	Scale	# of Items
Stability of resources	AHC-HRSN23	10
Neighbourhood environment	Neighbourhood Safety and Cohesion24	4
Depression	PHQ-225	2
Anxiety	GAD-226	2
Emotion regulation	RRB Emotion Regulation Subscale-Parent17	5
Stress	PSS-427	4

AHC-HRSN, Accountable Health Communities Health-Related Social Needs Screening Tool; GAD-2, Generalized Anxiety Disorder-2; PHQ-2, Patient Health Questionnaire 2-item; PSS-4, Perceived Stress Scale; RRB Emotion Regulation-Parent, Risk and Resilience Battery Parent.

Participants are asked to complete additional optional scales but are informed these are ‘extra’ and not required for study compensation. This approach was chosen to minimise participant burden while allowing additional information to be collected in some participants. The six optional scales are listed in table 1 and include: (1) the Emotion Regulation Scale from the Risk and Resilience Battery (RRB Emotion Regulation Subscale) is a 5-item subscale which assesses emotion dysregulation. Items are reverse scored so that higher scores reflect better emotion regulation¹⁷; (2) the Behavioral Inhibition System and Behavioral Activation System Reward Subscale is a 5-item subscale used to measure reward responsiveness¹⁸; (3) the Generalized Anxiety Disorder-7 (GAD-7) is a 7-item scale which measures severity of anxiety¹⁹; (4) the ABCD Cyber Bullying and Peer Victimisation and Perpetration Questions is a 12-item instrument used to assess the extent to which an adolescent has been the perpetrator or victim of cyberbullying²⁰; (5) the ABCD Perceived Discrimination Measure is an 11-item scale used to measure adolescent experiences of discrimination based on multiple domains²¹; (6) the ABCD Screen Media Activity Questionnaire is a 16-item scale used to assess the frequency and duration of different types of screen usage by adolescents.²²

Parent 1 is asked to complete self-report scales for an assessment of the adolescent proband’s environmental exposures. These six scales are listed in table 2 and include: (1) the Accountable Health Communities Health-Related Social Needs Screening Tool (AHC-HRSN) is a 10-item scale used to measure family needs and available resources²³; (2) Neighbourhood Safety and Cohesion is a 4-item subscale of the Risk and Resilience Battery which assesses neighbourhood environment²⁴; (3) the Patient Health Questionnaire is a two-item scale which surveys depression symptoms²⁵; (4) GAD-2 is a two-item scale used to measure anxiety²⁶; (5) RRB Emotion Regulation Subscale¹⁷; (6) the Perceived Stress Scale is a four-item scale used to measure severity of stress.²⁷ In the future, we hope to integrate this protocol with others in a larger database, adding more prospective assessments. We particularly hope to add measures which assess executive function or social communication difficulties, such as the Behavior Rating Inventory of Executive Function Self-Report and the Social Responsiveness Scale (2nd edition) Adolescent Self-Report. We additionally note that as we recruit participants from CHOP’s EHR, there is the potential to link previously collected mental health screeners with our prospectively collected data to widen the breadth and scope of our scales.

Brain-Behavior-Genetics study inclusion

Eligible youths are 12–20 years old at the time of the clinical brain MRI, which must have been completed within the CHOP healthcare network in the past year. The CHOP healthcare system includes two hospitals in Philadelphia and King of Prussia, Pennsylvania, USA, as well as outpatient facilities throughout Eastern Pennsylvania and Southern New Jersey. As described above, study inclusion also requires that the brain MRI must be graded as a SLIP. The benefit of this process is to recruit patients who would generally not be prohibited from participating as controls in prospective studies. For demographic details on this potential cohort, please see table 3. For parental inclusion, only biological parents complete the saliva samples; biological or non-biological primary caregivers can complete the ‘parent 1’ scales.

Table 3. Demographic information of patients with scans meeting study eligibility criteria from 1 March 2024–2025.

	(N=1887)
Sex
Female	1163 (61.6%)
Male	724 (38.4%)
Age (years)
Mean (SD)	15.5 (1.87)
Median (min, max)	15.6(12.0, 20.0)
Race
White	1124 (59.6%)
Black or African American	267 (14.1%)
Asian	77 (4.1%)
Multiple races	73 (3.9%)
Other/unknown	346 (18.3%)
Ethnicity
Non-Hispanic or non-Latino	1553 (82.3%)
Hispanic or Latino	205 (10.9%)
Other/unknown	129 (6.8%)

Brain-Behavior-Genetics study exclusion

Youths with brain MRIs with significant pathology or imaging artefacts are excluded. Youths with neurological conditions or conditions that would interfere with the completion or comprehension of study procedures are excluded, based on phecodes derived from lifetime International Classification of Diseases diagnostic codes in the EHR (see online supplemental table 1 for full list of excluded phecodes). Patients with neurological conditions such as epilepsy are excluded from the prospective study even if no gross pathology is observed on their brain MRI. Non-biological parents are excluded from providing saliva samples.

Data analysis

Analysis of neuroimaging data will follow previously published protocols.²⁸ MRIs inDigital Imaging and Communications in Medicine format retrieved through the CHOP picture archiving and identification system will be deidentified with LOCUTUS, converted to NIfTI format and curated into Brain Imaging Data Structure using CuBIDS.729,31 Multiple image processing approaches will be used to derive quantitative imaging features, including ‘recon-all-clinical’ (FreeSurfer V.7.4.1) and Synthseg+,32,34 deep-learning based segmentation approaches that do not require additional preprocessing steps. Imaging-derived phenotypes will include cortical and subcortical brain volumes defined using the ‘aseg’ and ‘Desikan Killiany’ atlases. Alternative image processing approaches will be considered prior to data analysis given rapid technical advancements in this evolving field. COMBAT-LS will be used to harmonise imaging-derived phenotypes across MRI scanners and sequence type, while protecting demographic information such as age and sex.³⁵ We anticipate that approximately 15 scanners will be included. Quantified neuroanatomical differences will be benchmarked against brain charts constructed from a larger cohort of CHOP patients with retrospective SLIP MRIs (anticipated n, approximately 10 000 youth) including scans obtained contemporaneously with those of study participants. These brain charts are constructed using Generalized Additive Models for Location, Shape and Scale, with promising preliminary evidence suggesting a high degree of similarity between SLIP MRI brain charts and brain charts derived from a mega-analysis of research control participants by the Lifespan Brain Chart Consortium.^{28 36}

Our initial analysis of prospectively collected data will prioritise general measures of psychopathology, environmental exposure and genetic risk. The overall psychopathology score (‘p-factor’) will be calculated from the CBCL-YSR using a bifactor model from item-level responses as in our prior work.³⁷ A general ‘exposome’ score will also be estimated as in our prior work, combining item-level responses with geocoded neighbourhood data.³⁸ Finally, after genetic samples are collected, we will conduct whole genome sequencing to identify rare sequence variants and copy number variations and to compute polygenic indices. The brain-gene-behavioural analysis may unfold in a myriad of directions, unlocking a multitude of scientific questions for the research community. Associations between developmental imaging phenotype centiles and phenotypic battery scores will be explored to identify axes of multivariate centile deviation, including cortical and subcortical regional volumes, that correlate with axes of clinical vulnerability. We will test the hypothesis that deviations from brain growth charts will be associated with overall psychopathology, environmental exposure and genetic risk. These analyses will characterise the complex interplay between genetic and environmental risk factors as they influence brain structure and how this relationship can impact risk for psychopathology. A better understanding of these risk factors will pave the way for developing more targeted treatment and prevention strategies, especially for vulnerable populations such as those with environmental stressors.

Discussion

The present Brain-Behavior-Genetics Study protocol will couple retrospective, clinically acquired brain MRI scans with prospectively acquired behavioural and genetic data in a large cohort of adolescents. This study is one part of a broader effort to harness underused clinical brain MRI data for research purposes, with multiple ongoing and future directions. While these clinical MRIs have more variability when compared with research-grade brain MRIs, such as those acquired in the ABCD cohort, they nonetheless represent a promising resource for recruiting populations who otherwise would not have the bandwidth to participate in brain-behaviour research studies. In this way, feasibility and ease of using these previously acquired scans may help to facilitate more representational research cohorts. Indeed, within prospective studies of brain and behaviour, clinical MRIs have the potential for many creative applications. The distribution of a brain phenotype in a large clinical MRI sample can be queried even prior to prospective recruitment for a behavioural study. Oversampling recruitment from the extremes of the MRI phenotype has the potential to increase power and replicability for brain-behaviour studies without requiring extremely large sample sizes.³⁹ Regarding genetic testing, paired recruitment into genetic biobanks, pulling across hospital systems, could rival well-powered samples for imaging genetics. In this way, the protocol provides a pathway towards future multisite projects, pulling clinical data across various health systems. We hope to use these data to demonstrate the feasibility of using clinically acquired brain MRIs in research settings and advance understanding of how genetic makeup and neuroanatomical differences may contribute to psychiatric disorders.

Reproducibility

Data gathered over the course of this study will be made available to researchers as part of a data-sharing agreement with the National Institute of Mental Health. All code for data analysis will be shared on GitHub (https://github.com/BGDlab) with relevant publications.