Introduction
Epidemiological data show that most psychiatric diagnoses in childhood have high comorbidity and limited temporal stability [
1-
6]. This means that during a young person’s engagement with Child and Adolescent Mental Health Services (CAMHS), a diagnosis once received can transition into or be supplemented by different diagnostic classifications. Such diagnostic shifts have potentially profound implications for young people and families, given diagnostic labels’ significance for making sense of emotional and behavioural difficulties [
7]. Knowledge of the frequency and directions of diagnostic adjustments in CAMHS is a precondition for supporting clinicians and service users with any challenges these clinical experiences may present. Understanding diagnostic trajectories will also assist clinicians and policy makers in anticipating likely prognoses and future service needs. However, few studies have either investigated the prevalence of diagnostic adjustments in CAMHS or established the typical patterns through which they occur. The current study explores these issues using data from a London-based mental health case register.
The inter-relations between diagnostic categories are typically considered in terms of concurrent comorbidity—i.e., children who qualify for different diagnoses at the same time point [
3,
4,
8-
10]. Beyond concurrent comorbidity, an issue that has received less empirical attention is the way diagnostic classifications evolve across time. The limited research investigating youth diagnostic trajectories reveals that longitudinal transitions between different diagnostic categories occur frequently and at above-chance levels. For instance, a study of British school-age children found only half of children who met criteria for a psychiatric disorder at baseline retained that diagnosis at 3-year follow-up [
11]. Another study, which pooled data from three large community studies, showed “all disorders predicted multiple disorder categories later in development and all disorders were predicted by at least three disorder categories at the prior developmental period” [
1]. Although most diagnostic categories are subject to diagnostic transitions [
1], some diagnostic sequences appear particularly prevalent. Early depression predicts later anxiety disorder and vice versa [
1,
2,
12,
13]. There is also temporal cross-over between ADHD and oppositional defiant and conduct disorders [
2,
10]. Early conduct problems can prefigure later anxiety and depression [
10,
12,
13]. Many of these associations recapitulate those identified in the comorbidity literature [
3,
4,
8-
10]. However, a longitudinal lens reveals some unique nuances, for instance, while early conduct difficulties predict later mood and anxiety disorders, the reverse does not seem true [
1].
There are numerous outstanding questions in the literature on longitudinal diagnostic trajectories. Minimal research has explored whether diagnostic trajectories are related to socio-demographic or clinical variables. Ford et al. [
11] found that the predictors of homotypic continuity (i.e. stability of the same diagnosis across time) differed across diagnostic class: persistence of ADHD and anxiety disorders was predicted by peer relationship problems; persistent conduct disorder was predicted by intellectual disability, poor family function, socio-economic deprivation and baseline psychopathology; while no factors significantly predicted persistence of depression. Very little research enlightens the socio-demographic or clinical variables that predict heterotypic continuity, i.e. change to (rather than persistence of) particular diagnostic categories. Costello et al. [
2] observed that almost all cases of heterotypic continuity (i.e. longitudinal progression between different diagnostic categories) in their sample occurred in girls, and suggested this indicated the DSM-IV diagnostic criteria used were biased towards male phenotypes. An alternative explanation is that disorders with higher female incidence (e.g., mood and anxiety disorders) also have higher diagnostic instability. Analytic strategies that isolate the independent effects of diagnostic category and socio-demographic variables such as gender are necessary to clarify the profile of service users most likely to experience diagnostic adjustments.
A further feature of the literature on diagnostic trajectories is the predominant use of community samples. While such designs have important advantages for drawing population-level conclusions, community cohort studies may not accurately reflect the diagnostic patterns that occur in clinical practice. Most community studies apply research-defined diagnostic criteria rather than recording the diagnoses children have actually received. Agreement between clinician and research diagnoses is often poor, with clinical settings showing more conservative diagnostic practice [
14,
15]. Additionally, diagnostic decisions in real clinical contexts may be influenced by patient preferences, clinician biases and pragmatic considerations [
16-
18], to which the structured assessment criteria in research studies are less sensitive. Understanding the diagnostic trajectories that occur in clinical settings is important for numerous reasons. First, it is the diagnoses actually recorded, rather than those for which children hypothetically qualify, that influence service planning and resource distribution. Second, if recorded diagnostic trajectories differ from those revealed by community studies, it may indicate certain biases or shortcomings in clinical practice (e.g., failure to regularly re-assess young service users), or alternatively that structured research assessments lack important information gleaned from nuanced clinical interviews. Third, it is clinical diagnoses that influence the ways children and families make sense of a young person’s difficulties. Research shows psychiatric diagnoses affect young people’s self-identity in diverse ways [
7]. The revision or supplementation of a diagnosis may have important social and emotional repercussions for service users.
A small number of studies have investigated diagnostic stability in clinical child and adolescent populations [
19-
23]. These studies indicate that psychotic disorders typically have highest temporal stability, followed by internalising disorders, externalising disorders and personality disorders [
20,
22,
23]. Overall, temporal reliability of diagnosis in clinical practice tends to be low. For instance, a study in a Canadian hospital found poor correspondence (average ϰ = 0.23) between the first and last primary diagnoses recorded in children’s clinical files: reliability was higher with shorter (0–1 year) intervals between first and last diagnoses (ϰ = 0.34), but extremely poor when the interval stretched to 4 years (ϰ = 0.08) [
21]. A US study of diagnostic stability in children who experienced multiple hospitalisations over a 9-year period found similarly poor reliability across diagnostic episodes (ϰ = 0.37), despite patients typically being assigned to the same clinical team across hospitalisations [
23]. These studies confirm diagnostic adjustments are common in clinical practice. However, the literature on diagnostic reliability in clinical contexts shows numerous limitations, with the few studies that exist relying on small (
n < 100) samples [
20,
22,
24], recruiting participants from a single clinical (usually inpatient) setting [
20-
23], and/or exclusively focusing on one diagnostic class [
19,
24-
26]. Finally, individual studies define and measure stability in different ways, usually in terms of prospective concordance, retrospective concordance or kappa coefficients. While these measures offer useful information [
23], none facilitates easy inference of the proportion of CAMHS attendees who experience modification of their diagnosis during their service engagement. Furthermore, the extant literature on diagnostic stability in child and adolescent mental healthcare does not provide detailed information on typical diagnostic sequences, i.e. the diagnoses that tend to precede and follow each other. Acquiring this information is important for prognosis and service planning.
The current study represents the first large-scale study of diagnostic trajectories in child psychiatric clinical records, which incorporates data from diverse clinical settings and on multiple diagnostic classes. The analysis sought to answer the following questions:
1.
What proportion of CAMHS service users undergo a longitudinal diagnostic adjustment?
2.
What are the typical diagnostic trajectories that occur?
3.
Are diagnostic trajectories related to any socio-demographic or clinical variables?
a.
What predicts the occurrence of any diagnostic adjustment?
b.
What predicts addition of specific diagnostic classes?
Discussion
This analysis established that 19.3% of children attending London-based mental health services underwent a diagnostic adjustment, i.e. received a subsequent additional diagnosis, different from and at least 30 days after their first recorded diagnosis. This is likely an underestimate of the true prevalence of diagnostic adjustments, as the structured diagnostic fields used for this analysis may not capture all diagnostic activity pertaining to a child (e.g., diagnoses registered outside of SLaM or in CRIS free-text notes). Nevertheless, the study facilitates the first conservative estimate of the prevalence of diagnostic adjustments in CAMHS: approximately one in five young service users had their diagnoses adjusted during their engagement with services.
The study suggests the overall likelihood of experiencing a diagnostic adjustment is unrelated to gender, area deprivation or age of first diagnosis. The absence of an independent effect of gender conflicts with previous observations that most diagnostic discontinuity occurs in girls [
2,
11]. The current analysis suggests such gender differences may not be due to gender per se, but because disorders with higher incidence among females (mood and anxiety disorders) account for most cases of diagnostic adjustment. The analysis indicates children identified as White/British are more likely to experience a diagnostic adjustment. This cannot be attributed to higher engagement with mental health services, since the ethnicity effect persisted independent of measures of service engagement. It is possible that Black or Minority Ethnic (BME) families interact with services in distinct ways that render diagnostic modification less likely (e.g., language difficulties or reluctance to query diagnosis). Alternatively, implicit racial/ethnic bias among clinical staff [
32] could affect their interactions with BME families and likelihood of revisiting diagnostic formulations. Further research is required to enlighten the pathways by which ethnicity affects diagnostic practice. Similar to Ford et al. [
11], particular patterns of diagnostic sequencing had unique predictors, with clinical variables generally more predictive than socio-demographic characteristics.
The diagnostic trajectories revealed in this clinical records study are largely consistent with previous epidemiological research. The study validates Copeland et al. [
1] finding that all disorder categories predicted different disorders later in development and vice versa. The analysis also replicated previous findings of common diagnostic sequences. The study confirms affective and anxiety disorders longitudinally predict each other [
1,
2,
12,
13], as do hyperkinetic and conduct disorders [
2,
10]. Consistent with previous findings of concurrent comorbidity of ADHD and autism [
9,
33], the analysis established this relationship also manifests longitudinally. Importantly, the current analysis confirms these cross-diagnostic sequences persist (1) in a clinical setting, and (2) when socio-demographic and clinical variables are controlled.
The analysis indicates minimal longitudinal overlap between (a) pervasive developmental and anxiety/stress-related, affective, or conduct/impulse disorders; (b) hyperkinetic and affective or anxiety/stress-related disorders; and (c) conduct/impulse and affective or anxiety/stress-related disorders. The latter finding coheres with Copeland et al. [
1] but contradicts indications from other studies that early conduct problems predict later anxiety and depression [
10,
12,
13]. The divergence from previous epidemiological studies may suggest children with conduct diagnoses have distinct service experiences. For instance, it is possible these young people are more likely to disengage and hence not be re-assessed, or that staff direct less clinical attention to the emotional symptoms they experience. The lack of longitudinal progression between pervasive developmental and anxiety/stress-related disorders is also notable given previous evidence of high anxiety symptoms in ASD populations [
9,
33]. This may reflect a tendency towards ‘diagnostic overshadowing’ in ASD, whereby anxiety symptoms are ascribed to an existing ASD diagnosis rather than distinct mental health disorders [
34,
35].
Research on diagnostic stability and instability is complicated by inconsistent methodological approaches regarding the degree of diagnostic drift that represents a meaningful diagnostic discontinuity [
11]. The current analysis adopted a conservative operational definition of diagnostic adjustment, which required recording of a diagnosis from a different high-level diagnostic class. Shifts between specific diagnoses within each class (for example, between bipolar and unipolar depression [
36] or obsessive compulsive disorder and generalised anxiety disorder [
19]) were not classified as diagnostic adjustments in this study and should be a focus of future research. On the other hand, the current study may be seen as liberal in adopting a minimum time period for diagnostic adjustment of just 30 days (although the average interval between first and second diagnosis was much longer, at almost 2 years). Future research should clarify how such methodological parameters affect estimates of stability. The field would benefit from more consensus regarding the most appropriate ways to define and measure diagnostic (in)stability in clinical data.
The analysis was subject to a number of limitations. The 30-day interval criterion meant the dataset did not include complete information on young people’s diagnostic histories, as additional diagnoses that may have been applied at less than 30 days from the original index diagnosis were excluded, as were comorbid secondary diagnoses applied contemporaneously with the index diagnosis. Further, the clinical validity of the diagnoses recorded in this dataset is unclear. However, the aim of this research was to explore patterns of actual diagnostic practice, rather than psychopathological development per se. It is recorded diagnoses, even if invalid, that influence service delivery and service user understanding. The fact the trajectories identified in this study overlap with those from previous epidemiological research provides some confidence in the overall validity of these clinical diagnoses. A case-by-case evaluation of the clinical validity of diagnostic adjustments may yield interesting findings that enlighten the reasons behind these clinical decisions.
It is important to note the paucity of data on factors pertaining to the service, rather than child, that may predict diagnostic adjustments. For instance, as data on the individual clinicians who made diagnoses were not available, it was not possible to explore whether diagnostic practice varied across professional disciplines. Notably, 61% of second diagnoses were recorded by a different clinical team than had recorded the first diagnosis. This is partly due to referral to specialist teams for assessment and diagnosis of particular disorders (e.g., ASD). However, it may also indicate diagnostic adjustments are likely when a child enters new clinical environments with different staff, cultures and agendas. Future research should expand investigation of how variables such as team culture, institutional priorities and professional discipline may contribute to variation in diagnostic practice.
A further and related limitation is that, due to the reliance on structured diagnostic fields, the clinical rationale behind diagnostic adjustments remains opaque. As previously discussed, the data available did not discriminate between diagnostic adjustments that were intended to replace or supplement prior diagnoses. There are various reasons why new diagnoses may be ascribed. New diagnoses could reflect genuine change in symptomatology, revelation of previously undetected symptoms, or correction of prior diagnostic errors. Diagnostic adjustments may also represent indeterminate or atypical cases that induce diagnostic dilemmas. They can reflect evolutions in clinical knowledge or diagnostic instruments. A new diagnosis can be applied for pragmatic reasons, such as securing access to educational resources. Finally, diagnostic adjustments could reflect individual clinicians’ different personal, cultural and professional outlooks. Qualitative work with the textual clinical notes held by CRIS is ongoing to explore the various reasons new diagnoses are added to clinical files.
Further limitations of this study are common across research with case registers. Although registers have the advantage of reflecting real clinical activity, data quality depends on clinicians’ recording practices, which are often inconsistent and incomplete [
37]. The SLaM electronic records system was implemented in 2006 and is now well embedded within the service, with audits showing the structured diagnostic fields used for this analysis are well populated [
29]. Further concerns relate to generalisability. Extrapolation from the data is limited by its generation by a single service provider. This notwithstanding, SLaM provides a large and diverse range of clinical services, and CRIS represents one of the largest psychiatric case registers globally [
37]. While the diverse demographic profile of SLaM’s catchment area is representative of many urban populations [
29], international generalisation is impeded by the unique UK healthcare structure, with most services state-provided. However, this policy context is also an empirical advantage as the relative scarcity of private services means CRIS claims near-total coverage of mental healthcare provision in its catchment area [
37]. This offers confidence the data collected comprehensively document children’s pathways through services.