Browse the corpus

Child and youth mental health (CYMH) problems have intensified and are now considered a public health crisis (). Nearly 15% of youth ages 10–19 years have experienced a mental illness, accounting for 13% of the global burden of disease within this population (). As many as 1 in 5 people are diagnosed with a mental illness before the age of 25, with 70% experiencing their first symptoms before the age of 18 (). Emerging evidence demonstrates a rising prevalence and severity of mental illness, particularly anxiety and mood disorders, as reflected in increasing rates of mental health service utilization and psychotropic medication use (). The burden of CYMH problems is projected to intensify in the coming years, driven by the rise of social media consumption, enduring consequences of the COVID-19 pandemic, heightened experiences of loneliness and social isolation, and mounting concerns about a future shaped by global instability and climate change (, ).

In recent years, a growing proportion of CYMH cases have been identified as having ‘complex mental health needs’, requiring higher intensity services and more frequent care with sustained involvement from CYMH agencies (). Emergency departments (EDs) are becoming the default sites for CYMH care (), with ED visits surpassing increases in CYMH outpatient visits (). This is problematic as EDs have limited capacity to provide such care (, ). In addition, nearly a quarter of CYMH patients re-visit EDs for more mental healthcare within six months (, ). This suggests that many CYMH patients cannot get their needs met, highlighting the inability of current healthcare systems to provide treatment in an effective or timely manner ().

The limitations of current CYMH care have received increased scrutiny (). For instance, Bickman et al. () called into question the utility, reliability, and validity of psychiatric diagnoses in CYMH care, determining in a three-part study of affective and behavior disorders of children and youth that: few CYMH outcomes were diagnosed more consistently than a random selection of symptoms; there was low diagnostic inter-rater agreement between parents, youth, and clinicians; and that comorbidities posed a significant barrier to clinician-based diagnoses (). Insel further questioned the status quo of CYMH care, highlighting the need to pivot away from reliance on behavioral symptoms, the predominant method for diagnosis of CYMH disorders, to instead create a neurodevelopmental framework (). It is clear that transformation of CYMH care and the research that drives it is necessary. The Precision Child and Youth Mental Health (PCYMH) paradigm has the potential to accomplish this (, ). Such a transformation will not be easy and must not discard the advances made in evidence-based care (, ), but the standard ‘one size fits all’ approach, doesn't work for all ().

Furthermore, the growing popularity of the PCYMH paradigm is, in part, a response to the problems stemming for questionable validity of CYMH diagnoses, concerns about underlying heterogeneity in patients labeled with the current nomenclature, and the difficulty of applying statistical mean-based results from randomized controlled trials (RCTs) to the individual needs of children and youth (, ). Advances in PCYMH have already yielded several promising avenues for tailoring mental health services (). For instance, at the Children’s Hospital of Eastern Ontario (CHEO), the creation of a participatory logic model demonstrated how the synthesis of implementation science and artificial intelligence (AI) data science can be used to plan a PCYMH research and clinical care transformation program (). Another example is the development of a clinical decision support system by a Norwegian research team, that has shown how big data analytics, and the electronic health record (EHR) can be used to create an Individualized Digital Decision Assist System (IDDEAS) to enhance precision and timeliness of medical decisions via clinician decision-making support based on targeted clinical knowledge and patient health information (, ).

However, many developments in the growing field of PCYMH research remain fragmented and siloed. For PCYMH methods to be smoothly integrated into the wider corpus of CYMH care and research, it is essential to catalogue the work done to date. Such efforts are the first step towards establishing consistent definitions and terminology within the field and may help guide funders on strategic allocation of their resources. To this end, we conducted a scoping review aiming to provide a ‘lay of the literature’ review of the emerging field of PCYMH. Our goal was to provide a snapshot of current scientific work published under the rubric of PCYMH or its synonyms and make recommendations for future research.

We formulated the research question using the Population, Concept, Context (PCC) framework (): What are the characteristics or features of all published research studies involving precision child and youth mental health? This exercise facilitated development of the search strategy, guided creation of inclusion and exclusion criteria, and structured the data collection. The review took a broad view, rather than focusing on a specific disorder or method. Our study was structured according to the Preferred Reporting Items for Systematic Reviews and Meta-analysis (PRISMA-ScR) (see for framework components' locations in the paper) (). We searched PubMed and Embase, in consultation with an in-house librarian. Together, these two databases provide extensive coverage of medical and biomedical academic literature from the past eight decades, with each database comprising over 25 million records. Because the concept of precision medicine had its roots earlier than 1999 (), the usual landmark date cited, we conducted our search from January 1, 1980 to November 30, 2023, updating it on November 1, 2024.

The search strategy was mapped onto the following three domains with synonyms: (1) precision health (e.g., ‘precision medicine’, ‘precision health’, ‘personalized medicine’); (2) child and youth (e.g., ‘child’, ‘adolescent’, ‘youth’); and (3) mental health (e.g., ‘mental health’, ‘behavioral health’, ‘child psychiatry’). Because child and youth mental health research often takes place in pediatric settings, we set the age range at 0 to <18 years. Relevant keywords and synonyms for each conceptual construct were joined using ‘OR’, with ‘AND’ being used to join the three constructs into a single search strategy. The search strategy and results for PubMed and Embase are presented in , respectively.

PCYMH research was defined as studies with aims or goals that addressed precision diagnosis, treatment, prognosis, or prevention of a mental health, psychiatric, or behavioral health condition based on differences in individuals’ biological characteristics, lifestyle, and environment, in addition to symptoms. Studies were included if they: (1) investigated the topic of precision mental health; (2) had a primary study population between the ages of 0 and up to 18 years old or, if in a study of adults and children/youth, separate findings on the 0-18-year-olds were provided; (3) were published in English; and 4) had “precision” or one of its synonyms in the title or abstract. Documents and publications not presenting published original research were excluded, e.g., conference abstracts, commentaries, literature reviews, meta-analyses, book chapters, and reports.

Prior to starting the screening process, inclusion and exclusion criteria were iteratively refined by JSL and KP through a series of consultations with other members of the research team. The research team pointed out that our initial inclusion criteria of “0–18 years of age” could be interpreted as those “up to” or “including the 18 th year of life (which would make them 19)”. They gave examples from other studies showing that many papers with this age range often were included in studies of adults, without any separate results for the 18- year-olds. Therefore, this inclusion criterion was adjusted to specify “up to, but not including, the 18 th year of life”.

Covidence () was used to process papers retrieved from the search. Screening comprised two steps, both of which were carried out following the eligibility criteria by pairs of ten raters. Ratings were conducted blind to ratings by others. The first step was title and abstract screening. Articles remaining eligible after this step were next subjected to full-text review, again conducted by pairs of raters blind to each other’s work. For both steps, reviewer discrepancies were resolved through deliberation between KP and JSL. All papers still meeting criteria after full-text review underwent final data extraction by two blinded raters, using the rubric in . In addition to data more commonly collected on study characteristics, we also collected information on “type of PCYMH study” (defined by the aim(s) or objective(s)), each study’s use of PCYMH tools, e.g., use of big data, AI, omics (, ),and whether or not a reporting guideline was used to structure the paper.

Quantitative data were summarized with descriptive statistics. Qualitative analysis, particularly looking for patterns in clinical disorders or problems, type of PCYMH study, and use of PCYMH tools was based on the approach to latent pattern content analysis outlined by Potter & Levine-Donnerstein (). As per this approach, analysis comprised the following four stages: (1) decontextualization – identification of recurrent units of meaning; (2) recontextualization – organization of the data into codes; (3) categorization – grouping of codes into categories and sub-categories of shared meaning; and (4) compilation – final refinement of categories and sub-categories. This process was iterative and recursive, with the findings generated during each stage of the analysis informing the coder’s approach to subsequent stages. All steps of the qualitative analysis were independently conducted by the first and senior authors.

The search yielded 1,266 studies (see for the PRISMA diagram). Of these studies, 272 were flagged by Covidence as duplicates and removed. All automatic duplications were manually verified by the first author. The remaining 995 studies underwent title and abstract screening, during which 600 studies were dropped as not meeting criteria. Full text screening identified 271 more that did not meet criteria. The remaining 124 (–) articles comprised the final set for review.

Dates of publication for the articles are displayed in by four-year blocks of time from pre-2000 to 2024. Only one paper meeting our criteria was published before 2000. This was followed by a slow but steady rise in publication rate until an inflection point occurred between 2010-2014, signaling a sharp uptick in the publication rates every four years over the subsequent decade. The highest rate to date was between 2020-2024, during which 48% (60/124) (–, , , , , , , , , , –, –, , –, –, , , –, , , , –, , –, , , , , –, , , –) of the entire set was published. Articles were from 26 countries, with multi-site study papers coded according to the primary site, determined through descriptions of the study setting, the senior author’s address, and locations of funding sources. As can be seen on the map in , 51% (63/124) (, , , , , , , –, , –, , , , , , , –, , –, –, , , , , –, –, , , , –, –, , , ) of the papers were from the United States, with China and the UK representing 8% (10/124) (, , , , , , , –) and 6% (7/124) (, , , , , , ), respectively. The remaining countries each contributed 1–6 papers.

None of the included studies used a reporting guideline. provides an overview of the studies included, organized by first author’s last name and displaying CYMH focus, type of PCYMH study as defined by aim, PCYMH tool(s), and key findings. - present more details on each study. The most common study design category was observational: cohort studies (35% (43/124)) (–, , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , –); case-control studies (32% (40/124)) (, , –, , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ); cross-sectional studies (5% (6/124)) (, , , , , ); and case series (3% (4/124)) , , , . RCTs were used in 19% (24/124) (, , , , , –, , , , , , , , , , , , , , , , ) of the studies. Two studies used non-randomized controlled trials (, ), and one was a qualitative study ().

The clinical focus of a study was defined as either a primary mental health diagnosis, e.g., attention deficit hyperactivity disorder (ADHD) or mental health problem, e.g., suicidality. There were 11 clinical foci, as can be seen in the tree map diagram in . The most frequent studies were about ADHD (28% (35/124)) (–, , , , , , –, , –, , , , , , , , –, , , , , , , , , , ) autism spectrum disorders (ASD) (28% (35/124)) (, , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , –, ); and depression (13% (16/124)) (–, , , , , , , , , , , , , ). The fewest studies were on bipolar disorder (), aggression (, ), psychosis (, ) and post-traumatic stress disorder (, ), each of which only comprised 2% of the study’s set. A total of 110,386 participants were studied, the count excluding publications which used the same sample more than once. Sample sizes ranged from 6 to 26,055 children and youth, with a mean sample size and standard deviation of 912 ± 3120 and a median of 110 subjects with an interquartile range of 61 to 221. It was found that 12% (15/124) of studies included more than 1,000 participants. In contrast, 10% (12/124) of studies had 20 or fewer participants.

Children were defined as those between the ages of 0 and 10 years, while youth or adolescents were defined as those between the ages of 10 and up to, but not including, 18 years. Samples were further categorized as children only 30% (37/124), youth only 19% (24/124), or mixed 51% (63/124). Girl-only populations were investigated in less than 1% (1/124) () of the studies, boy-only populations in 3% (4/124) (, , , ) and mixed-gender populations in 91% (113/124) (–, –, –, –, –, , , –, –, –, –, –). The remaining 6% (7/124) (, , , , , , ) of studies did not adequately describe the gender composition of the populations. Constructed from the qualitative content analysis were four categories of PCYMH research foci aiming to develop more precise diagnoses, mental health problem definition (e.g., suicidality) prognoses, or prediction of treatment response. Studies collecting data from any biological system with any method were labeled the Biomarker category and comprised (68% (84/124)) of the publications (, , –, , –, –, –, –, , , –, , , , , , , –, –, –, , –, , , –, , , , , , , , , , –, ).

The Non-Biological Markers category investigated potential markers consisting of specific symptoms of disorders, demographic information, or family history (without genetics data) as predictors in addition to components of the general external exposome, defined by Neufcourt, et al () as exposures outside the body, such as social, cultural, and ecological contexts without have specific biological effects. These constituted 17% (22/124) of the papers (, , , , , , , , , , , , , , , , –, , , ). The third most commonly used PCYMH focus was categorized as Implementation of PCYMH interventions, defined as studies of the feasibility, effectiveness, or acceptability of a novel PCYMH-driven intervention in a clinical setting. These constituted 14% (17/124) of the total dataset (, , , , , , , , , , , , , , , , ). The smallest category was Predictive Algorithms (5% (6/124)) (, , , , , ). Most used machine learning or other AI-assisted modelling to investigate the likelihood that an individual might develop a specific diagnosis, suicidality, or present with repeat emergency department visits.

Categories were not mutually exclusive. Several studies (4% (5/124)) (, , , , ) fell into two categories, three of which were in the Biomarker and Non-Biological Marker categories (, , ), one which was in the Implementation and Biomarker categories () and one which was in the Implementation and Predictive Algorithm categories (). As a result, the percentages presented above do not necessarily add up to 100%. presents the sub-categories for each of the main categories of PCYMH focus. For instance, of the 84 studies investigating biomarkers, 46% (39/84) (–, , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ) examined neural biomarkers, 33% (28/84) (, , , , , , , , –, , , , , , –, , , , , , , , , ) investigated genetic biomarkers, and 8% (7/84) (, , , , , , ) investigated metabolite biomarkers. The remaining sub-categories accounted for less than 10% of the studies, including eye tracking, gut biome, peptide, protein, antibody, voice analysis, sleep, circadian rhythm, reaction time and skin conductance biomarkers. Again, the sub-categories were not mutually exclusive, and as such five studies (, , , , ) fell into multiple sub-categories.

In contrast, the 22 studies investigating non-biological markers were relatively evenly distributed across the category’s composite sub-categories. Of the studies investigating non-biological markers, 68% (15/22) (, , , , , , , , , , , , , , ) investigated phenotypic, i.e. symptoms markers, 45% (10/22) (, , , , , , , , , ) investigated behavioral markers, and 36% (8/22) (, , , , , , , ) investigated sociodemographic markers. As 36% (8/22) (, , , , , , , ) of studies employed multiple types of non-biological markers in tandem, the percentages do not necessarily add up to 100%. Neither the Biomarker nor the Non-Biological Marker categories contained work beyond Phase 1, i.e. identifying a putative marker, in the process of marker development for clinical use (). A list of PCYMH tools was developed using key foundational literature (, ). Of our study set, we found that 19% (23/124) (, , , , , , , –, , , , –, , , , , , , ) used -omics methods, 17% (21/124) (, , , , , , , , , , , , , , , , , , , , ) employed machine learning techniques, 8% (10/124) (, –, , , , , , ) used multimodal profiling, 9% (11/124) (, –, , , , , , , ) worked with big data, less than 3% (3/124) (, , ) employed the use of digital health data and less than1% (1/124) () employed virtual populations.

To our knowledge, this is the first scoping review to comprehensively map the PCYMH literature. We retrieved publications using PCYMH and synonymous terms in the title/abstract, advancing understanding of the field’s current scope. Although several position papers and commentaries have outlined the potential of the PCYMH paradigm to improve child and youth mental health (, , ), the empirical evidence remains limited (, ). This is reflected in a relatively small number of eligible studies, the absence of replication, few clinical validation efforts, and a scarcity of implementation-focused PCYMH research. Even the CYMH diagnosis categories with the highest number of articles (ADHD and ASD) only had 1–3 papers about each biomarker or non-biological marker and few implementation studies. Finally, none of the included studies used a reporting guideline, highlighting the lack of standardization across this emerging field.

The recent explosion in publications (49% since 2020) indicates the field is just coming into its own, but the dominance of biomarker studies (67%) compared to the scarcity of implementation studies (14%) tells us that the focus remains on basic discovery rather than clinical application. These features, in combination with only ten studies using individualized multimodal profiles in PCYMH research or care, and the absence of reporting guidelines, leads us to conclude that PCYMH research is in the infancy of its development, with great potential, with many hopes still unrealized. We know of no other PCYMH reviews with which to compare our results. However, adult precision MH research does appear to be further along. For example, in a systematic review of research through 2019 about precision health or medicine in adults with myriad mental health conditions such as psychotic, mood, anxiety, and substance use disorders, Salazar de Pablo et al. () were able to identify 584 prediction modeling studies, estimating individualized risks for diagnosis, prognosis, or treatment response. We suspect that the field of PCYMH may just need more time, as in our dataset, the majority of studies weren’t published until after 2015 (and most in the last five years), when President Obama declared the beginning of the precision medicine era ().

Another comparator is precision medicine or health research in the non-CYMH pediatric population. Here we also find a more well-developed corpus of work, especially studies identifying molecular targets for treatment in pediatric cancer (), cardiac disorders (), and rare diseases (). Why isn’t PCYMH research as advanced as precision research in other pediatric disorders? We speculate that PCYMH, like previous research in CYMH diagnosis, treatment, and prevention, is particularly difficult because causation is complex and non-singular (). Complicating this issue is the transmutation of symptoms and problems within the context of evolving child and youth development (, ). Precision health research in non-CYMH pediatric disorders also wrestles with constant change in the individual from normal development, but most pediatric medical disorders have objective markers of pathology, the lack of which has always been a major challenge in CYMH research and care. However, the tools now available in precision health are, for the first time, enabling biological, lifestyle, and environmental discoveries about psychopathology beyond the description of symptoms and behaviours, giving reason for optimism about the future transformational power of PCYMH ().

In addition, the effects of social determinants of health, including exposure to adverse events, makes the development of PCYMH more complicated (). It is now accepted that these determinants derail normal child and adolescent development and have consequences well into adulthood, but their effects on CYMH problems are particularly severe and consequential for an individual’s entire life (). Furthermore, correlates or possible etiologic factors such as poverty, abuse, neglect, or out of home placement are complicated to “treat” and not under the control of any one societal system, including health care. Precision mental health or psychiatry has been touted as the next scientific revolution (, ), but potential barriers to widespread operationalization and uptake are significant, including possible psychological harm to patients, unknown economic consequences, potential increases in mental healthcare disparities, failure to deliver on promises of increased treatment success, and inadequately trained clinical staff, with poor integration of research into care ().

However, if PCYMH is in its infancy, this is the ideal time to shape this body of research to facilitate maximal opportunity for success. First, standardization of how PCYMH research is planned, conducted, and reported would enhance the quality of studies and enhance inter-study comparison, allowing better of results from multiple studies. The first reporting guideline was published in 1996 () and this line of work has steadily expanded in scope and impact. Using such guidelines could have obviated publication of the papers we found with inadequate information on samples and methods.

Second, while developing studies, it is always important to consider and mitigate biases. However, in the context of PCYMH, where the main objective is typically to identify sub-groups needing different diagnosis, treatments or prevention interventions than the “the average” research participant or patient, issues such as selection bias run the risk of significantly undermining the validity of any observations made. Studies using AI, e.g., various types of machine learning to make relevant clinical predictions, are particularly vulnerable to myriad biases (). Since these studies have the potential to have a large impact on care, identifying and mitigating AI-related bias is particularly important. Evidence-based medicine has significantly advanced CYMH research and it remains relevant and essential for translating PCYMH research findings into real-world impact for children and youth (). AI-generated algorithms, biomarker, and non-biomarker studies all need to be replicated and validated and a large proportion of this work requires evidence-based medicine methods. Very few studies in our dataset have progressed beyond identification of markers or early development of Ai generated predictive algorithms and this dearth of such studies will undermine advancement of PCYMH (). Furthermore, validation of algorithms, biomarkers, and non-biological markers by clinicians and persons with lived experience who represent the population of interest, whether a clinical condition or the community, is critical to close the standard multi-year gap between research discoveries and impact. PCYMH will not succeed if results do not penetrate clinical care settings or the community.