Skip to main content Accessibility help
×
Home
Hostname: page-component-5959bf8d4d-kpqxq Total loading time: 0.895 Render date: 2022-12-10T04:50:43.293Z Has data issue: true Feature Flags: { "useRatesEcommerce": false } hasContentIssue true

Context v. algorithm: evidence that a transdiagnostic framework of contextual clinical characterization is of more clinical value than categorical diagnosis

Published online by Cambridge University Press:  25 August 2021

Jim van Os*
Affiliation:
Department of Psychiatry and Neuropsychology, School for Mental Health and Neuroscience, Maastricht University Medical Centre, Maastricht, The Netherlands Department of Psychiatry, UMC Utrecht Brain Center, University Medical Center Utrecht, Utrecht University, Utrecht, The Netherlands Department of Psychosis Studies, Institute of Psychiatry, Psychology & Neuroscience, King's College London, London, UK
Lotta-Katrin Pries
Affiliation:
Department of Psychiatry, UMC Utrecht Brain Center, University Medical Center Utrecht, Utrecht University, Utrecht, The Netherlands
Margreet ten Have
Affiliation:
Department of Epidemiology, Netherlands Institute of Mental Health and Addiction, Utrecht, The Netherlands
Ron de Graaf
Affiliation:
Department of Epidemiology, Netherlands Institute of Mental Health and Addiction, Utrecht, The Netherlands
Saskia van Dorsselaer
Affiliation:
Department of Epidemiology, Netherlands Institute of Mental Health and Addiction, Utrecht, The Netherlands
Maarten Bak
Affiliation:
Department of Psychiatry, UMC Utrecht Brain Center, University Medical Center Utrecht, Utrecht University, Utrecht, The Netherlands FACT, Mondriaan Mental Health, Maastricht, The Netherlands
Gunter Kenis
Affiliation:
Department of Psychiatry and Neuropsychology, School for Mental Health and Neuroscience, Maastricht University Medical Centre, Maastricht, The Netherlands
Bochao D. Lin
Affiliation:
Department of Translational Neuroscience, UMC Utrecht Brain Center, University Medical Center Utrecht, Utrecht University, Utrecht, The Netherlands
Nicole Gunther
Affiliation:
School of Psychology, Open University, Heerlen, The Netherlands
Jurjen J. Luykx
Affiliation:
Department of Translational Neuroscience, UMC Utrecht Brain Center, University Medical Center Utrecht, Utrecht University, Utrecht, The Netherlands Department of Neurology, Brain Centre Rudolf Magnus, University Medical Centre Utrecht, Utrecht, The Netherlands GGNet Mental Health, Apeldoorn, The Netherlands
Bart P. F. Rutten
Affiliation:
Department of Psychiatry, UMC Utrecht Brain Center, University Medical Center Utrecht, Utrecht University, Utrecht, The Netherlands
Sinan Guloksuz
Affiliation:
Department of Psychiatry, UMC Utrecht Brain Center, University Medical Center Utrecht, Utrecht University, Utrecht, The Netherlands Department of Psychiatry, Yale University School of Medicine, New Haven, CT, USA
*
Author for correspondence: Jim van Os, E-mail: j.j.vanos-2@umcutrecht.nl
Rights & Permissions[Opens in a new window]

Abstract

Background

A transdiagnostic and contextual framework of ‘clinical characterization’, combining clinical, psychopathological, sociodemographic, etiological, and other personal contextual data, may add clinical value over and above categorical algorithm-based diagnosis.

Methods

Prediction of need for care and health care outcomes was examined prospectively as a function of the contextual clinical characterization diagnostic framework in a prospective general population cohort (n = 6646 at baseline), interviewed four times between 2007 and 2018 (NEMESIS-2). Measures of need, service use, and use of medication were predicted as a function of any of 13 DSM-IV diagnoses, both separately and in combination with clinical characterization across multiple domains: social circumstances/demographics, symptom dimensions, physical health, clinical/etiological factors, staging, and polygenic risk scores (PRS). Effect sizes were expressed as population attributable fractions.

Results

Any prediction of DSM-diagnosis in relation to need and outcome in separate models was entirely reducible to components of contextual clinical characterization in joint models, particularly the component of transdiagnostic symptom dimensions (a simple score of the number of anxiety, depression, mania, and psychosis symptoms) and staging (subthreshold, incidence, persistence), and to a lesser degree clinical factors (early adversity, family history, suicidality, slowness at interview, neuroticism, and extraversion), and sociodemographic factors. Clinical characterization components in combination predicted more than any component in isolation. PRS did not meaningfully contribute to any clinical characterization model.

Conclusion

A transdiagnostic framework of contextual clinical characterization is of more value to patients than a categorical system of algorithmic ordering of psychopathology.

Type
Original Article
Creative Commons
Creative Common License - CCCreative Common License - BY
This is an Open Access article, distributed under the terms of the Creative Commons Attribution licence (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted re-use, distribution and reproduction, provided the original article is properly cited.
Copyright
Copyright © The Author(s), 2021. Published by Cambridge University Press

Introduction

Diagnosis in psychiatry represents an unresolved issue (Guloksuz & van Os, Reference Guloksuz and van Os2020). Although the practice of classification according to ICD and DSM criteria remains firmly rooted in clinical practice, it is widely recognized that the classical diagnostic functions of predicting the need for care and health care outcome are not well served in the ICD/DSM diagnostic framework (Mullins-Sweatt, Lengel, & DeShong, Reference Mullins-Sweatt, Lengel and DeShong2016). Research shows that individuals in the same ICD/DSM diagnostic category are different with regard to the need for care and health care outcome, and as a group show only weak differences in risk factors, psychopathology, need for care, and outcome compared to patients in other diagnostic categories, as evidenced by low diagnostic likelihood ratios (Allardyce, McCreadie, Morrison, & van Os, Reference Allardyce, McCreadie, Morrison and van Os2007; Van Os, Reference Van Os2010; Van Os et al., Reference Van Os, Gilvarry, Bale, Van-Horn, Tattan, White and Murray1999, Reference Van Os, Gilvarry, Bale, van-Horn, Tattan, White and Murray2000).

In fact, it is increasingly recognized that mental suffering may be transdiagnostic and contextual. ‘Transdiagnostic’ refers to the growing awareness that symptoms are not specific to mental disorders, occurring in highly personal clusters that differ from person to person. ‘Contextual’ refers to the growing awareness that psychopathology arises across a range of contexts including social, existential, somatic, and temporal factors, that also require ‘diagnosing’ to guide clinical approaches (van Os, Delespaul, Wigman, Myin-Germeys, & Wichers, Reference van Os, Delespaul, Wigman, Myin-Germeys and Wichers2013). Thus taking into account transdiagnostic and contextual perspectives in diagnosis opens up a new way to advance clinical practice beyond the traditional psychiatric taxa (Dalgleish, Black, Johnston, & Bevan, Reference Dalgleish, Black, Johnston and Bevan2020). Although a range of alternative, more ‘transdiagnostic’, diagnostic formulations have been proposed, for example, dimensional alternatives such as Hierarchical Taxonomy of Psychopathology (HiTOP) (Kotov et al., Reference Kotov, Krueger, Watson, Achenbach, Althoff, Bagby and Zimmerman2017) and Research Domain Criteria (RDoC) (Cuthbert & Insel, Reference Cuthbert and Insel2010), as well as network models of psychopathology (Borsboom & Cramer, Reference Borsboom and Cramer2013), these were developed and researched mostly from the perspective of contribution to academic theory, not of contribution to patient value, i.e. adding to the diagnostic prediction of care need and health care outcome. In addition, they do not offer a contextual perspective. Thus, diagnostic innovation with a specific focus on adding patient value in the ‘moral’ era of medicine and health care (Berwick, Reference Berwick2016) may require a different approach (van Os, Guloksuz, Vijn, Hafkenscheid, & Delespaul, Reference van Os, Guloksuz, Vijn, Hafkenscheid and Delespaul2019).

A more patient-centered and contextual approach toward diagnostic innovation is the ‘clinical characterization’ diagnostic framework (Maj, Reference Maj2020). Clinical characterization is focused on specifically increasing patient value by combining or replacing categorical ICD/DSM diagnosis with transdiagnostic, personalized, and contextual information on symptoms, clinical severity, clinical staging (McGorry & van Os, Reference McGorry and van Os2013), and antecedent and concomitant variables (Maj, Reference Maj2020). While clinical characterization will always form part of the assessment process, the specific inclusion of clinical characterization in the diagnostic framework opens up a way of specifically researching and validating modes of clinical characterization, and how these may maximize patient value and reduce the self-perpetuated oversized importance attached to context-less categorical and algorithm-based classification per se (van Os et al., Reference van Os, Guloksuz, Vijn, Hafkenscheid and Delespaul2019).

Here, we wished to examine the value of the diagnostic framework of clinical characterization by specifically examining the transdiagnostic contextual components of the framework in terms of contribution to predicting care need and health care outcome, in comparison with the categorical algorithm-based approach. To this end, we prospectively modeled the occurrence of care needs and outcomes associated with transdiagnostic psychopathology in a unique representative general population cohort interviewed four times over a period of 9 years. As there is much speculation about including biological information in the clinical characterization component, particularly genetic information in the form of polygenic risk scores (PRS; Murray et al., Reference Murray, Lin, Austin, McGrath, Hickie and Wray2021), these were also included in the analyses.

Method

Sample

All four waves of the Netherlands Mental Health Survey and Incidence Study-2 (NEMESIS-2) were used. NEMESIS-2 was conducted to study the prevalence, incidence, course, and consequences of mental disorders in the Dutch general population (n = 6646 at baseline). The baseline data of NEMESIS-2 were collected from 2007 to 2009, follow-up was until 2018. Non-clinician, trained interviewers applied the Composite International Diagnostic Interview (CIDI) version 3.0 (Alonso et al., Reference Alonso, Angermeyer, Bernert, Bruffaerts, Brugha and Bryson2004; de Graaf, ten Have, Burger, & Buist-Bouwman, Reference de Graaf, ten Have, Burger, Buist-Bouwman, Ustun and Kessler2008) and additional questionnaires during home visits. Further details are provided in the online Supplementary material.

Assessment of DSM-IV disorders

The following 13 CIDI, version 3.0, DSM-IV diagnoses were assessed: major depression, dysthymia, bipolar disorder, panic disorder, agoraphobia, social phobia, specific phobia, GAD, alcohol abuse and dependence, drug abuse and dependence, and any clinical psychosis. Further details of the diagnostic procedure are provided in the online Supplemental material.

Outcomes to be predicted by diagnostic framework

Mental health service use

Service use was measured based on the service use section of NEMESIS (Bijl & Ravelli, Reference Bijl and Ravelli2000) at each interview wave. A variable of receiving any mental health care was defined on the basis of any care received for mental health or addiction problems by psychiatrists, psychologists, psychotherapists, and professionals working in mental health and addiction care, in the last 12 months (T0) or since the last interview (T1–T3).

Unmet need for care and medication use

Unmet need for care was defined as the person reporting, at each interview wave, that he or she needed care for a mental problem but had not sought such care, or not enough, in the last 12 months. At each wave, a binary rating was made for use of any medication for mental problems or addiction in the last 12 months.

Clinical characterization components

Staging variables

Staging variables were constructed to create a variable providing a temporal sub-characterization to the psychopathological variables in the model. Four mutually exclusive staging variables were constructed, at each wave, indicating no disorder, subthreshold syndrome, incident disorder, and persistent disorder. Variables were mutually exclusive in that persistent disorder trumped incident and subthreshold, and incident disorder trumped subthreshold syndrome. Significant subthreshold psychopathology was considered present if individuals had rated positive on any CIDI 3.0 core screening symptom (for each disorder, CIDI 3.0 has a two-stage process of core symptom screening questions which, if positive, result in follow-up symptom questions) of any anxiety disorder, depression, or bipolar disorder, or had any self-reported psychotic experience in the absence of any diagnosis of, respectively, any anxiety disorder, depression, bipolar disorder, or clinical psychosis. Incident disorder was defined as the first occurrence of any of the 13 disorders, defined above, over the NEMESIS-2 period of observation. Baseline occurrence was considered as ‘incident’ if the first onset had transpired in the 6 months before baseline interview, based on the age of onset questions about the disorder. Persistence was defined as any occurrence of a DSM-IV disorder persisting from one interview wave to the next.

Sociodemographic variables

Demographic variables included were sex (0 = male, 1 = female), age in years, and dichotomous ethnic minority status (Moroccan, Turkish, Surinamese, Antillean, Indonesian, or other non-Western ethnic groups). Age was analyzed as a dichotomous variable, defining a younger age group encompassing the range most at risk of the onset of a psychotic disorder (18–35 years, 24% at baseline) v. the older group, consistent with previous work in this sample (Hasmi et al., Reference Hasmi, Pries, Ten Have, de Graaf, van Dorsselaer, Bak and van Os2021). Marital status at each interview was coded married/widowed v. divorced/never married. Unemployment at each interview was coded as being unemployed or disabled v. employed/student/homemaker/retired. Educational level at baseline was a two-level variable (primary, lower, and higher secondary v. higher professional/university education). Income at each interview was net annual household income (i.e. individual and, if applicable, partner), rated on a scale from 1 to 14 (not rated at one interview and predicted linearly from the values at the interviews before and after) and modeled as ‘low income’ v. other, by dichotomizing around the 25% percentile. Having ever been on disability benefit over the period of observation was analyzed as a binary variable. The variable ‘living alone’ at each interview indicated that the participant was the only person in the household. The variable ‘children at home’ indicated the presence of one or more children in the household. The perceived status gap was assessed at T1, T2, and T3 using two questions. First, the MacArthur Scale of Subjective Social Status (Adler, Epel, Castellazzo, & Ickovics, Reference Adler, Epel, Castellazzo and Ickovics2000) was used to rate subjective social status. In an easy pictorial format, it presents a ‘social ladder’ with 10 levels and asks individuals to place an ‘X’ on the rung on which they feel they stand. The second question was about a similar ladder, but this time with regard to the desired level of social status. The difference between the subjective desired and actual social status was used as an independent variable in the analyses. The mean value of T1–T3 was used to replace missing values at T0. In the analyses, it was dichotomized around the perception of being more than one level below desired social status. Level of urbanicity of current residence, assessed at each interview wave, was defined at five levels based on the Dutch classification of residence topography or population density: (1) countryside (distances to facilities is larger), (2) village (<25 000 inhabitants), (3) small city (25 000–50 000 inhabitants), (4) medium city (50 000–100 000 inhabitants), (5) large city (>100 000 inhabitants). Consistent with previous work, the cut-off of at least >50 000 inhabitants was used to define the binary variable of urban area (Guloksuz et al., Reference Guloksuz, van Nierop, Lieb, van Winkel, Wittchen and van Os2015).

Clinical variables

Childhood adversity was assessed at T0 using a questionnaire based on the NEMESIS trauma questionnaire (de Graaf, Ten Have, & van Dorsselaer, Reference de Graaf, Ten Have and van Dorsselaer2010). Whenever a subject reported having experienced one of five types of childhood adversity before the age of 16 years [emotional neglect, physical abuse, psychological abuse, peer victimization (bullying), and sexual abuse], they were asked to state how often it had occurred. From this, as described previously, a childhood adversity sum score was derived and dichotomized at the 80th percentile (Heins et al., Reference Heins, Simons, Lataster, Pfeifer, Versmissen, Lardinois and Myin-Germeys2011; van Dam et al., Reference van Dam, van Nierop, Viechtbauer, Velthorst, van Winkel, Genetic and Wiersma2015; van Os, Marsman, van Dam, Simons, & Investigators, Reference van Os, Marsman, van Dam, Simons and Investigators2017). Based on the ‘Brugha Life events section’ (Brugha, Bebbington, Tennant, & Hurry, Reference Brugha, Bebbington, Tennant and Hurry1985), participants were asked at each interview whether they experienced one of 10 negative life events within the last 12 months at T0 and T3, and some additional pregnancy- and illness-related negative life events at T1–T2. Examples of items are serious sickness, death of a family member or close friend, and serious financial problems. A dichotomous exposure was created around at least one negative life event in the last year. Family history was assessed as a person-level binary variable in two stages, as described previously (Radhakrishnan et al., Reference Radhakrishnan, Guloksuz, Ten Have, de Graaf, van Dorsselaer, Gunther and van Os2019) and detailed in the online Supplementary material. One of the interviewer observations that was rated at each interview concerned the degree to which the person had understood the interview questions or was ‘slow to understand’ (very good, reasonable, poor). Any rating of understanding the questions less than very good at any of the interviews was rated as a dichotomous variable of ‘slow to understand’. For each interview, a binary variable ‘any suicidality’ was rated positive if the participant had admitted to any suicidal plans, thoughts, or attempts at the corresponding items of the CIDI interview. Two 12 binary item subscales of the Eysenck Personality Questionnaire – Revised Short Scale (EPQ-RSS) were used at baseline to measure neuroticism (i.e. emotional instability, worrying, nervousness) and extraversion (i.e. impulsivity, outgoing, lively) (Schotanus-Dijkstra et al., Reference Schotanus-Dijkstra, Pieterse, Drossaert, Westerhof, de Graaf, ten Have and Bohlmeijer2016). For each personality trait, a total score was computed with higher scores indicating higher levels of the trait ranging from 0 to 12 (Cronbach's α 0.80 for neuroticism and 0.83 for extraversion), analyzed as binary variables, dichotomized around the 75th percentile.

Symptoms scores

Scores of depressive (28 symptoms), anxiety (43 symptoms), mania (18 symptoms), and psychosis symptoms (20 symptoms) were based on adding the relevant CIDI 3.0 symptoms as described previously in detail (van Nierop et al., Reference van Nierop, Viechtbauer, Gunther, van Zelst, de Graaf, Ten Have and van Winkel2015).

Physical health variables

Hearing impairment and visual impairment were assessed during the face-to-face interview at each wave, by asking whether participants had experienced deafness or serious hearing impairment, or serious visual impairment in the past 12 months. Ratings were yes (1) or no (0). High BMI, assessed at each interview, was defined as BMI > 25. Physical exercise, assessed at each interview, was coded ‘1’ if the person met the Dutch health criterion for sufficient physical activity and otherwise coded ‘0’. Any smoking of tobacco, assessed at each interview, was coded ‘1’ and otherwise coded ‘0’. Binary presence of pain was assessed at each interview with The Short-Form-36 Health Survey (SF-36) (Stewart & Ware, Reference Stewart and Ware1992) pain module, dichotomized around the 75% percentile. The presence of 15 groups of somatic disorder was assessed at each interview and coded as ‘1’ for any presence of one or more somatic disorders – and otherwise coded ‘0’.

Polygenic risk scores for schizophrenia, bipolar disorder, depression, and IQ

The polygenic risk score for schizophrenia (PRS-SZ) was created from best-estimate genotypes at six different p thresholds (0.5, 0.1, 0.05, 5.10−3, 5.10−5, 5.10−8). For our primary analyses, we used the p threshold of <0.05, as this threshold explained most variation in liability in the Psychiatric Genomics Consortium analysis (Schizophrenia Working Group of the Psychiatric Genomics Consortium, 2014) and was shown to perform well for the current phenotype of SF-36 mental health (Pries et al., Reference Pries, van Os, Ten Have, de Graaf, van Dorsselaer, Bak and Guloksuz2020). Further details on the genotyping procedure and PRS calculation were described previously indetail (Guloksuz et al., Reference Guloksuz, Pries, Ten Have, de Graaf, van Dorsselaer, Klingenberg and van Os2020; Pries et al., Reference Pries, Guloksuz, Ten Have, de Graaf, van Dorsselaer, Gunther and van Os2018). Consistent with previous analyses, statistical analyses with PRS-SZ were adjusted for three principal components (Pries et al., Reference Pries, van Os, Ten Have, de Graaf, van Dorsselaer, Bak and Guloksuz2020).

Material for DNA analysis of sufficient quality, and hence for PRS calculation, was available for 3104 individuals (47%) at T0. Excluding individuals who at interview has been assessed as a member of an ethnic minority, given lack of generalizability of PRS to this group, and individuals diagnosed with a clinical psychosis, left 3052 for PRS calculation.

DSM-IV diagnosis

CIDI 3.0 lifetime diagnoses (T0) and interval diagnoses (T1-T3) of the 13 disorders described above served as the diagnostic variables. Rates presented are not weighted and therefore may slightly differ from previous results (de Graaf, ten Have, van Gool, & van Dorsselaer, Reference de Graaf, ten Have, van Gool and van Dorsselaer2012).

Analysis

Logistic regression models of three dependent variables and 46 independent variables representing six domains of clinical characterization

All analyses were performed using Stata, version 16 (StataCorp, 2019). p < 0.05 (two-tailed) was considered nominally statistically significant. We tested the contributions of 46 independent variables, clustered into six clinical characterization domains, entering the variables pertaining to each specific group (DSM diagnosis, symptoms, social, clinical, physical, and staging), in logistic regression models of incidence of three binary dependent variables relevant for diagnostic function: unmet need for care, use of mental health service, and use of medication.

We first calculated the contribution of the independent variables pertaining to the six domains of clinical characterization separately, in separate logistic regression models for each of the three dependent variables. This was followed by testing the contributions of the independent variables pertaining to the six domains entered jointly into a single model, for each of the three dependent variables. This latter full logistic regression model is visualized in Fig. 1.

Fig. 1. Main logistic regression model of three dependent variables and 46 independent variables, representing six domains of clinical characterization.

In order to understand the interplay between specifically the domain of DSM diagnosis in relation to each of the other five domains, we separately modeled the joint contribution of the diagnosis domain and each of the other five domains.

Definition of ‘incidence’ of the three dependent variables

The NEMESIS-2 repeated cross-sectional design (i.e. each person contributing four observations over time) allowed fitting multiple prediction models for binary outcomes (i.e. care need, medication use, care use) using logistic regression to model the incidence of the outcome at time point t. The outcome was considered incident if there was the absence of the outcome at the previous wave t–1, and the presence of the outcome at time point t.

The risk set for the three outcomes can be seen in Table 1 (left side; incidence).

Table 1. Distribution of need for care and health care outcomes, per interview wave, in NEMESIS-2 cohort, incident, and prevalent

a Incidence was defined as the absence of the outcome at the previous wave t–1, and the presence of the outcome at time point t, over the four interview waves in NEMESIS-2.

The Stata cluster option was used to take into account intra-group correlations occasioned by clustering of observations within individuals.

Testing the contribution of variables clustered in six domains of clinical characterization

Contributions to the model were evaluated, for each of the clusters of variables representing a domain of clinical characterization, with the population attributable fraction (PAF) for binary outcomes. The Stata PUNAF command was used (Newson, Reference Newson2013) to estimate the attributable fraction and the PAF with 95% CIs for each cluster of variables representing a domain of clinical characterization. Under the assumption that the different risk groups are causally associated with the outcome, the PAF indicates the proportion of the care need (or other outcome used) that might be prevented if the risk were eliminated (Levine, Reference Levine2007).

PRS subsample

Given the fact that polygenic risk was only present in a subsample, separate analyses were conducted with the five clinical characterization variable groups with polygenic risk variables added as a seventh group (polygenic risk, diagnosis, symptom, social, clinical, physical, and staging) on the same three outcomes, in order to evaluate the contribution of PRS. Models including PRS were adjusted for three principal components, conform previous work (Guloksuz et al., Reference Guloksuz, Pries, Ten Have, de Graaf, van Dorsselaer, Klingenberg and van Os2020; Marsman et al., Reference Marsman, Pries, Ten Have, de Graaf, van Dorsselaer, Bak and van Os2020; Pries et al., Reference Pries, van Os, Ten Have, de Graaf, van Dorsselaer, Bak and Guloksuz2020).

Overfitting and multicollinearity

Given the use of up to 46 independent variables in the logistic regression model, we examined the possibility of overfitting and multicollinearity that may ensue. This is described further in the online Supplementary material.

Results

Distributions of the clinical characterization variables over the four waves are shown in online Supplementary Tables S1–S5. The distribution of the outcomes, incident and prevalent, is shown in Table 1. In the subsample of 3052 with PRS data, the mean value of PRSZ-SZ was −131.2 (s.d. = 4.3).

In the models of clinical characterization domains separately predicting need, medication use, and mental health service use, diagnosis and clinical characterization groups all contributed, with somewhat weaker contribution of the social and physical clinical characterization variables. In the models of clinical characterization domains jointly predicting need, medication use, and mental health service use, particularly symptom and staging variables contributed to all predictions whereas diagnosis variables did not predict any outcome and other variables had more specific contributions (e.g. social variables to need and service use, and clinical variables to need and medication use). The physical domain did not contribute meaningfully (Table 2). Also, for all models, the combined contribution of clinical characterization variables in the models examining the contributions of the domains jointly was around 2–2.5 times larger than the contribution of categorical diagnosis in the models examining the contributions of the domains separately (Table 2).

Table 2. Contribution of diagnosis and clinical characterization variable groups, modeled separately and jointly, to models of need incidence (need for care, medication, service use)

a Population attributable fraction: proportion of the outcome that could have been prevented had the area of clinical characterization been reduced to zero.

b PAF separately – the variable group of each domain was modeled separately, yielding PAF for each domain separately.

c CI, confidence interval.

d PAF jointly – the variable groups of all domains were modeled jointly, yielding mutually adjusted PAF.

e The following six clinical characterization variable clusters were examined for contribution to the model:

Diagnostic: cluster of 13 DSM-IV diagnoses (clinical psychosis, bipolar disorder, major depression, dysthymia, social phobia, specific phobia, panic disorder, generalized anxiety disorder, agoraphobia, alcohol dependence, alcohol abuse, drug dependence, drug abuse).

Social: cluster of variables indexing: living alone, single/divorced, unemployed, low income, low education, status gap, disability pension, young age, female sex, ethnic minority, urban residence, children at home.

Symptoms: cluster of variables indexing: psychosis, anxiety, depression, mania.

Clinical: cluster of variables indexing: family history, childhood adversity, life event, suicidality, slow at interview, high neuroticism, high extraversion.

Physical: cluster of variables indexing: somatic disorder, high pain, high BMI, lack of movement, smoking, hearing impairment, visual impairment.

Staging: three dummy variables that help to sub-characterize psychopathology into mutually exclusive categories of: subthreshold syndrome, incident disorder, and persistent disorder.

In the models examining the contribution of the diagnosis domain in combination with each of the other four domains separately (Table 3), it is revealed that any contribution of the diagnosis domain was reducible to the symptom domain. Similarly, the diagnosis domain did not contribute much over and above the contribution of the staging domain. The contribution of the diagnosis domain was mostly complementary to the social and the clinical domain, and trumped the contribution of the physical domain (Table 3).

Table 3. Clinical characterization domain of DSM diagnosis modeled jointly with each of the other five domains, for three outcomes

a The following six clinical characterization variable clusters were examined for contribution to the model:

Diagnostic: cluster of 13 DSM-IV diagnoses (clinical psychosis, bipolar disorder, major depression, dysthymia, social phobia, specific phobia, panic disorder, generalized anxiety disorder, agoraphobia, alcohol dependence, alcohol abuse, drug dependence, drug abuse).

Social: cluster of variables indexing: living alone, single/divorced, unemployed, low income, low education, status gap, disability pension, young age, female sex, ethnic minority, urban residence, children at home.

Symptoms: cluster of variables indexing: psychosis, anxiety, depression, mania.

Clinical: cluster of variables indexing: family history, childhood adversity, life event, suicidality, slow at interview, high neuroticism, high extraversion.

Physical: cluster of variables indexing: somatic disorder, high pain, high BMI, lack of movement, smoking, hearing impairment, visual impairment.

Staging: Dummy variables that help to sub-characterize psychopathology into mutually exclusive categories of: subthreshold syndrome, incident disorder, and persistent disorder.

b Population attributable fraction: proportion of the outcome that could have been prevented had the area of clinical characterization been reduced to zero.

c CI, confidence interval.

d Full model: combined contribution to the model of domain of DSM diagnosis with the other domain in the model.

e Contribution to the model of one domain of clinical characterization, adjusted for the other domain in the model.

PRS-SZ added to clinical characterization

Analyses with the PRS variable added to the clinical characterization framework showed similar results for the non-PRS clinical characterization variables and only one significant but marginal contribution (PAF = 4%) across the models with the different components included separately and no meaningful or significant contribution in any of the models with the different components jointly included (maximum PAF = 2%; data not shown).

Discussion

Findings

The results indicate that although formal categorical diagnoses predict care need and health care outcomes (service use and medication use), this is explained entirely by contextual clinical characterization at the level of the person. In other words, categorical diagnosis in the presence of adequate contextual characterization conveys no additional information whereas the reverse does not hold. It could be argued, therefore, that the diagnostic process should focus on contextual clinical characterization first, rendering the process of formal diagnosis of secondary, administrative importance from the perspective of clinical utility, as the elements in the clinical characterization framework can be translated directly to clinical interventions in a way that formal diagnosis cannot ‘compete’ with (Spitzer, Reference Spitzer1998). For example, exposure to childhood adversity, excessive cannabis use, or loss of a relative can directly inform on a specific treatment need in a way that the diagnosis of ‘schizophrenia’ or ‘depression’ cannot.

The results are in agreement with a growing body of evidence that the etiology, treatment, and expression of mental illness is largely transdiagnostic, i.e. better described as overlapping and dynamically evolving personal symptom constellations that transcend classic diagnostic boundaries, arising from overlapping liabilities with little diagnostic specificity (Brainstorm Consortium et al., Reference Anttila, Bulik-Sullivan, Finucane, Walters, Bras and Murray2018) and managed with largely non-specific treatment approaches (Dalgleish et al., Reference Dalgleish, Black, Johnston and Bevan2020). In such a transdiagnostic era of psychopathology, clinical characterization is the most fitting and efficient way to predict treatment need and service use outcomes, thus providing value for patients and clinicians.

While it may be argued that good clinical care will always focus on accurate clinical characterization, actual clinical practice in psychiatry continues to attach outsized importance to the ‘correct’ diagnosis, often subjecting each patient to lengthy structured interviews that yield a diagnosis through a series of computerized algorithms – yet are not equipped to yield even simple dimensional scores of major symptom domains. In fact, the influence of formal categorical diagnosis in psychiatry is such that it has been described as an ‘epistemic prison’ (Hyman, Reference Hyman2010). In contrast, a structured format for accurate contextual clinical characterization at the level of the person in routine clinical practice has yet to see the light. Academic studies on how to approach the issue of clinical characterization for major syndromes have only recently begun to appear (Maj et al., Reference Maj, Stein, Parker, Zimmerman, Fava, De Hert and Wittchen2020).

Components of contextual clinical characterization

The results suggest that simple contextual clinical measures account for a large proportion of the outcome whereas omnibus measures of common variants of genetic liability do not significantly contribute at all. Although there is much speculation about the clinical use of PRS (Murray et al., Reference Murray, Lin, Austin, McGrath, Hickie and Wray2021), these results are in line with a previous study in the same sample, showing that transdiagnostic PRS did not contribute meaningfully to the onset of psychopathology (Marsman et al., Reference Marsman, Pries, Ten Have, de Graaf, van Dorsselaer, Bak and van Os2020). The finding that PRS did not contribute to the clinical characterization framework is disappointing, albeit not unexpected, given that genetic liability to mental illness is broadly distributed, each individual carrying thousands of small-effect genetic risk variants. Although there have been reports of PRS being associated with clinical outcomes, these associations generally are characterized by statistical significance but tiny effect sizes (Murray et al., Reference Murray, Lin, Austin, McGrath, Hickie and Wray2021). The staging component of clinical characterization had the largest impact on need and outcome and much of the effect of the diagnostic component was reducible to staging. This is not surprising, as persistence of psychopathology over a 3-year follow-up period provides a much greater window of opportunity to seek help and develop need than incident disorder or subsyndromal manifestations. Nevertheless, as a simple measure providing a temporal perspective to psychopathology, it contributes considerably to the utility of clinical characterization. Symptoms clearly ‘trumped’ categorical diagnosis in predicting need and outcome, in line with a large body of research showing quantitative measures of psychopathology outperform categories in representing psychopathology (Haslam, McGrath, Viechtbauer, & Kuppens, Reference Haslam, McGrath, Viechtbauer and Kuppens2020). Symptoms also contributed most persistently across the three outcomes, whereas other factors, such as clinical factors and social factors, had more modest and/or less consistent contributions. This suggests that the components of the clinical characterization model may contribute differentially. However, this requires replication across different samples. Physical health variables had the least impact on the models. However, as the range of outcomes under study was limited, these cannot be considered without value. Indeed, somatic health is an important outcome in the mentally ill, and physical variables likely are diagnostically relevant in this domain.

Psychopathology components of clinical characterization

We included three psychopathology-based domains of clinical characterization that were all derived, albeit in a different fashion, from the same CIDI3.0 items: diagnosis, symptoms, and staging. The findings were consistent in showing poor performance of psychopathology represented as ex-cathedra algorithmic ordering, i.e. DSM diagnoses. Compared to the complex rules of algorithmic decision making, a simple dimensional representation of four-symptom dimensions performed much better, as did the measure of psychopathology in relation to temporality. These findings suggest that the elaborate and time-consuming iterating cycle of development and revision of algorithm-based systems for the diagnosis of psychopathology may not be necessary. Instead, it may be more useful to focus on a system describing contextual components of clinical characterization with direct use for clinical practice.

Methodological issues

The results should be interpreted in light of several methodological issues. Although the study was prospective, outcomes used were crude as was the measure of staging. On the other hand, this would mimic the actual situation of routine clinical practice in many countries. The analytical paradigm was to analyze the incidence of care need and service use rather than the incidence of psychopathology, instead including psychopathology as a ‘staged’ measure including subthreshold state, incidence, and persistence. This analytical framework was chosen to reproduce the pathway to care as it evolves in natural settings, with the degree of psychopathology, together with a range of moderators as captured by the clinical characterization framework, determine help-seeking behavior, and, eventually, service use (Goldberg & Huxley, Reference Goldberg and Huxley1980). The study was conducted in a small high-income country which will impact outcomes such as service use and use of medication. Therefore, the contribution of clinical characterization variables to these outcomes requires further study in other countries with different health care coverages, particularly low-income and intermediate-income countries.

Supplementary material

The supplementary material for this article can be found at https://doi.org/10.1017/S0033291721003445.

Financial support

NEMESIS-2 is conducted by the Netherlands Institute of Mental Health and Addiction (Trimbos Institute) in Utrecht. Financial support has been received from the Ministry of Health, Welfare and Sport, with supplementary support from the Netherlands Organization for Health Research and Development (ZonMw). This funding source had no further role in study design; in the collection, analysis, and interpretation of data; in the writing of the report; or in the decision to submit the paper for publication. Bart PF Rutten was funded by a VIDI award number 91718336 from the Netherlands Scientific Organisation. Drs Guloksuz and van Os are supported by the Ophelia research project, ZonMw grant number: 636340001.

Conflict of interest

None.

References

Schizophrenia Working Group of the Psychiatric Genomics Consortium (2014). Biological insights from 108 schizophrenia-associated genetic loci. Nature, 511(7510), 421427. doi: 10.1038/nature13595CrossRefGoogle Scholar
StataCorp (2019). STATA statistical software: Release 16. Texas: College Station: StataCorp.Google Scholar
Adler, N. E., Epel, E. S., Castellazzo, G., & Ickovics, J. R. (2000). Relationship of subjective and objective social status with psychological and physiological functioning: Preliminary data in healthy white women. Health Psychology, 19(6), 586592. doi: 10.1037//0278-6133.19.6.586CrossRefGoogle ScholarPubMed
Allardyce, J., McCreadie, R. G., Morrison, G., & van Os, J. (2007). Do symptom dimensions or categorical diagnoses best discriminate between known risk factors for psychosis? Social Psychiatry and Psychiatric Epidemiology, 42(6), 429437. doi: 10.1007/s00127-007-0179-yCrossRefGoogle ScholarPubMed
Alonso, J., Angermeyer, M. C., Bernert, S., Bruffaerts, R., Brugha, T. S., Bryson, H., … EsemeD/Mhedea Investigators – European Study of the Epidemiology of Mental Disorders Project (2004). Sampling and methods of the European Study of the Epidemiology of Mental Disorders (ESEMeD) project. Acta Psychiatrica Scandinavica Supplementum (420), 820. doi:10.1111/j.1600-0047.2004.00326Google ScholarPubMed
Brainstorm Consortium, Anttila, V., Bulik-Sullivan, B., Finucane, H. K., Walters, R. K., Bras, J., … Murray, R. (2018). Analysis of shared heritability in common disorders of the brain. Science (New York, N.Y.), 360(6395). doi:10.1126/science.aap8757Google ScholarPubMed
Berwick, D. M. (2016). Era 3 for medicine and health care. JAMA, 315(13), 13291330. doi: 10.1001/jama.2016.1509CrossRefGoogle ScholarPubMed
Bijl, R. V., & Ravelli, A. (2000). Psychiatric morbidity, service use, and need for care in the general population: Results of The Netherlands Mental Health Survey and Incidence Study. American Journal of Public Health, 90(4), 602.Google ScholarPubMed
Borsboom, D., & Cramer, A. O. (2013). Network analysis: An integrative approach to the structure of psychopathology. Annual Review of Clinical Psychology, 9, 91121. doi: 10.1146/annurev-clinpsy-050212-185608CrossRefGoogle ScholarPubMed
Brugha, T., Bebbington, P., Tennant, C., & Hurry, J. (1985). The list of threatening experiences: A subset of 12 life event categories with considerable long-term contextual threat. Psychological Medicine, 15(1), 189194. doi: 10.1017/s003329170002105xCrossRefGoogle ScholarPubMed
Cuthbert, B. N., & Insel, T. R. (2010). Toward new approaches to psychotic disorders: The NIMH Research Domain Criteria project. Schizophrenia Bulletin, 36(6), 10611062. doi: 10.1093/schbul/sbq108CrossRefGoogle ScholarPubMed
Dalgleish, T., Black, M., Johnston, D., & Bevan, A. (2020). Transdiagnostic approaches to mental health problems: Current status and future directions. Journal of Consulting and Clinical Psychology, 88(3), 179195. doi: 10.1037/ccp0000482CrossRefGoogle ScholarPubMed
de Graaf, R., ten Have, M., Burger, H., & Buist-Bouwman, M. (2008). Mental disorders and service use in the Netherlands. Results from the European Study of the Epidemiology of Mental Disorders (ESEMeD). In Ustun, T. & Kessler, R. (Eds.), The WHO world mental health surveys: Global perspectives on the epidemiology of mental disorders (pp. 388405). New York: Cambridge University Press.Google Scholar
de Graaf, R., Ten Have, M., & van Dorsselaer, S. (2010). The Netherlands Mental Health Survey and Incidence Study-2 (NEMESIS-2): Design and methods. International Journal of Methods in Psychiatric Research, 19(3), 125141. doi: 10.1002/mpr.317CrossRefGoogle ScholarPubMed
de Graaf, R., ten Have, M., van Gool, C., & van Dorsselaer, S. (2012). Prevalence of mental disorders and trends from 1996 to 2009. Results from the Netherlands Mental Health Survey and Incidence Study-2. Social Psychiatry and Psychiatric Epidemiology, 47(2), 203213. doi: 10.1007/s00127-010-0334-8CrossRefGoogle ScholarPubMed
Goldberg, D., & Huxley, P. (1980). Mental illness in the community. London: Tavistock Publications.Google Scholar
Guloksuz, S., Pries, L. K., Ten Have, M., de Graaf, R., van Dorsselaer, S., Klingenberg, B., … van Os, J. (2020). Association of preceding psychosis risk states and non-psychotic mental disorders with incidence of clinical psychosis in the general population: A prospective study in the NEMESIS-2 cohort. World Psychiatry, 19(2), 199205. doi: 10.1002/wps.20755CrossRefGoogle ScholarPubMed
Guloksuz, S., van Nierop, M., Lieb, R., van Winkel, R., Wittchen, H. U., & van Os, J. (2015). Evidence that the presence of psychosis in non-psychotic disorder is environment-dependent and mediated by severity of non-psychotic psychopathology. Psychological Medicine, 45(11), 23892401. doi: 10.1017/S0033291715000380CrossRefGoogle ScholarPubMed
Guloksuz, S., & van Os, J. (2020). Dr Strangelove, or how we learned to stop worrying and love uncertainty. World Psychiatry, 19(3), 395396. doi: 10.1002/wps.20794CrossRefGoogle ScholarPubMed
Haslam, N., McGrath, M. J., Viechtbauer, W., & Kuppens, P. (2020). Dimensions over categories: A meta-analysis of taxometric research. Psychological Medicine, 50(9), 14181432. doi: 10.1017/S003329172000183XCrossRefGoogle ScholarPubMed
Hasmi, L., Pries, L. K., Ten Have, M., de Graaf, R., van Dorsselaer, S., Bak, M., … van Os, J. (2021). What makes the psychosis ‘clinical high risk’ state risky: Psychosis itself or the co-presence of a non-psychotic disorder? Epidemiology and Psychiatric Sciences, 30, e53. doi: 10.1017/S204579602100041XCrossRefGoogle ScholarPubMed
Heins, M., Simons, C., Lataster, T., Pfeifer, S., Versmissen, D., Lardinois, M., … Myin-Germeys, I. (2011). Childhood trauma and psychosis: A case-control and case-sibling comparison across different levels of genetic liability, psychopathology, and type of trauma. American Journal of Psychiatry, 168(12), 12861294. doi: 10.1176/appi.ajp.2011.10101531CrossRefGoogle ScholarPubMed
Hyman, S. E. (2010). The diagnosis of mental disorders: The problem of reification. Annual Review of Clinical Psychology, 6, 155179. doi: 10.1146/annurev.clinpsy.3.022806.091532CrossRefGoogle Scholar
Kotov, R., Krueger, R. F., Watson, D., Achenbach, T. M., Althoff, R. R., Bagby, R. M., … Zimmerman, M. (2017). The Hierarchical Taxonomy of Psychopathology (HiTOP): A dimensional alternative to traditional nosologies. Journal of Abnormal Psychology, 126(4), 454477. doi: 10.1037/abn0000258CrossRefGoogle ScholarPubMed
Levine, B. (2007). What does the population attributable fraction mean? Preventing Chronic Disease, 4(1), A14.Google ScholarPubMed
Maj, M. (2020). Beyond diagnosis in psychiatric practice. Annals of General Psychiatry, 19, 27. doi: 10.1186/s12991-020-00279-2CrossRefGoogle ScholarPubMed
Maj, M., Stein, D. J., Parker, G., Zimmerman, M., Fava, G. A., De Hert, M., … Wittchen, H. U. (2020). The clinical characterization of the adult patient with depression aimed at personalization of management. World Psychiatry, 19(3), 269293. doi: 10.1002/wps.20771CrossRefGoogle ScholarPubMed
Marsman, A., Pries, L. K., Ten Have, M., de Graaf, R., van Dorsselaer, S., Bak, M., … van Os, J. (2020). Do current measures of polygenic risk for mental disorders contribute to population variance in mental health? Schizophrenia Bulletin, 46(6), 13531362. doi: 10.1093/schbul/sbaa086CrossRefGoogle ScholarPubMed
McGorry, P., & van Os, J. (2013). Redeeming diagnosis in psychiatry: Timing versus specificity. Lancet (London, England), 381(9863), 343345. doi: 10.1016/S0140-6736(12)61268-9CrossRefGoogle ScholarPubMed
Mullins-Sweatt, S. N., Lengel, G. J., & DeShong, H. L. (2016). The importance of considering clinical utility in the construction of a diagnostic manual. Annual Review of Clinical Psychology, 12, 133155. doi: 10.1146/annurev-clinpsy-021815-092954CrossRefGoogle ScholarPubMed
Murray, G. K., Lin, T., Austin, J., McGrath, J. J., Hickie, I. B., & Wray, N. R. (2021). Could polygenic risk scores be useful in psychiatry?: A review. JAMA Psychiatry, 78(2), 210219. 10.1001/jamapsychiatry.2020.3042.CrossRefGoogle ScholarPubMed
Newson, J. B. (2013). Attributable and unattributable risks and fractions and other scenario comparisons. The Stata Journal, 13, 672698.CrossRefGoogle Scholar
Pries, L. K., Guloksuz, S., Ten Have, M., de Graaf, R., van Dorsselaer, S., Gunther, N., … van Os, J. (2018). Evidence that environmental and familial risks for psychosis additively impact a multidimensional subthreshold psychosis syndrome. Schizophrenia Bulletin, 44(4), 710719. doi: 10.1093/schbul/sby051CrossRefGoogle ScholarPubMed
Pries, L. K., van Os, J., Ten Have, M., de Graaf, R., van Dorsselaer, S., Bak, M., … Guloksuz, S. (2020). Association of recent stressful life events with mental and physical health in the context of genomic and exposomic liability for schizophrenia. JAMA Psychiatry, 77(12), 12961304. doi: 10.1001/jamapsychiatry.2020.2304CrossRefGoogle Scholar
Radhakrishnan, R., Guloksuz, S., Ten Have, M., de Graaf, R., van Dorsselaer, S., Gunther, N., … van Os, J. (2019). Interaction between environmental and familial affective risk impacts psychosis admixture in states of affective dysregulation. Psychological Medicine, 49(11), 18791889. doi: 10.1017/S0033291718002635CrossRefGoogle ScholarPubMed
Schotanus-Dijkstra, M., Pieterse, M. E., Drossaert, C. H. C., Westerhof, G. J., de Graaf, R., ten Have, M., … Bohlmeijer, E. T. (2016). What factors are associated with flourishing? Results from a large representative national sample. Journal of Happiness Studies, 17(4), 13511370. doi: 10.1007/s10902-015-9647-3CrossRefGoogle Scholar
Spitzer, R. L. (1998). Diagnosis and need for treatment are not the same. Archives of General Psychiatry, 55(2), 120.CrossRefGoogle Scholar
Stewart, A. L., & Ware, J. E. (1992). Measuring functioning and well-being: The medical outcomes study approach. Durham and London: Duke University Press.Google Scholar
van Dam, D. S., van Nierop, M., Viechtbauer, W., Velthorst, E., van Winkel, R., Genetic, R., … Wiersma, D. (2015). Childhood abuse and neglect in relation to the presence and persistence of psychotic and depressive symptomatology. Psychological Medicine, 45(7), 13631377. doi: 10.1017/S0033291714001561CrossRefGoogle ScholarPubMed
van Nierop, M., Viechtbauer, W., Gunther, N., van Zelst, C., de Graaf, R., Ten Have, M., … van Winkel, R. (2015). Childhood trauma is associated with a specific admixture of affective, anxiety, and psychosis symptoms cutting across traditional diagnostic boundaries. Psychological Medicine, 45(6), 12771288. doi: 10.1017/S0033291714002372CrossRefGoogle ScholarPubMed
Van Os, J. (2010). Are psychiatric diagnoses of psychosis scientific and useful? The case of schizophrenia. Journal of Mental Health (Abingdon, England), 19(4), 305317. doi: 10.3109/09638237.2010.492417CrossRefGoogle ScholarPubMed
van Os, J., Delespaul, P., Wigman, J., Myin-Germeys, I., & Wichers, M. (2013). Beyond DSM and ICD: Introducing ‘precision diagnosis’ for psychiatry using momentary assessment technology. World Psychiatry, 12(2), 113117. doi: 10.1002/wps.20046CrossRefGoogle Scholar
Van Os, J., Gilvarry, C., Bale, R., Van-Horn, E., Tattan, T., White, I., & Murray, R. (1999). A comparison of the utility of dimensional and categorical representations of psychosis. UK700 Group. Psychological Medicine, 29(3), 595606.CrossRefGoogle ScholarPubMed
Van Os, J., Gilvarry, C., Bale, R., van-Horn, E., Tattan, T., White, I., & Murray, R. (2000). Diagnostic value of the DSM and ICD categories of psychosis: An evidence-based approach. Social Psychiatry and Psychiatric Epidemiology, 35(7), 305311.CrossRefGoogle Scholar
van Os, J., Guloksuz, S., Vijn, T. W., Hafkenscheid, A., & Delespaul, P. (2019). The evidence-based group-level symptom-reduction model as the organizing principle for mental health care: Time for change? World Psychiatry, 18(1), 8896. doi: 10.1002/wps.20609CrossRefGoogle Scholar
van Os, J., Marsman, A., van Dam, D., Simons, C. J., & Investigators, G. (2017). Evidence that the impact of childhood trauma on IQ is substantial in controls, moderate in siblings, and absent in patients with psychotic disorder. Schizophrenia Bulletin, 43(2), 316324. doi: 10.1093/schbul/sbw177CrossRefGoogle ScholarPubMed
Figure 0

Fig. 1. Main logistic regression model of three dependent variables and 46 independent variables, representing six domains of clinical characterization.

Figure 1

Table 1. Distribution of need for care and health care outcomes, per interview wave, in NEMESIS-2 cohort, incident, and prevalent

Figure 2

Table 2. Contribution of diagnosis and clinical characterization variable groups, modeled separately and jointly, to models of need incidence (need for care, medication, service use)

Figure 3

Table 3. Clinical characterization domain of DSM diagnosis modeled jointly with each of the other five domains, for three outcomes

Supplementary material: PDF

van Os et al. supplementary material

van Os et al. supplementary material

Download van Os et al. supplementary material(PDF)
PDF 183 KB
You have Access Open access
2
Cited by

Save article to Kindle

To save this article to your Kindle, first ensure coreplatform@cambridge.org is added to your Approved Personal Document E-mail List under your Personal Document Settings on the Manage Your Content and Devices page of your Amazon account. Then enter the ‘name’ part of your Kindle email address below. Find out more about saving to your Kindle.

Note you can select to save to either the @free.kindle.com or @kindle.com variations. ‘@free.kindle.com’ emails are free but can only be saved to your device when it is connected to wi-fi. ‘@kindle.com’ emails can be delivered even when you are not connected to wi-fi, but note that service fees apply.

Find out more about the Kindle Personal Document Service.

Context v. algorithm: evidence that a transdiagnostic framework of contextual clinical characterization is of more clinical value than categorical diagnosis
Available formats
×

Save article to Dropbox

To save this article to your Dropbox account, please select one or more formats and confirm that you agree to abide by our usage policies. If this is the first time you used this feature, you will be asked to authorise Cambridge Core to connect with your Dropbox account. Find out more about saving content to Dropbox.

Context v. algorithm: evidence that a transdiagnostic framework of contextual clinical characterization is of more clinical value than categorical diagnosis
Available formats
×

Save article to Google Drive

To save this article to your Google Drive account, please select one or more formats and confirm that you agree to abide by our usage policies. If this is the first time you used this feature, you will be asked to authorise Cambridge Core to connect with your Google Drive account. Find out more about saving content to Google Drive.

Context v. algorithm: evidence that a transdiagnostic framework of contextual clinical characterization is of more clinical value than categorical diagnosis
Available formats
×
×

Reply to: Submit a response

Please enter your response.

Your details

Please enter a valid email address.

Conflicting interests

Do you have any conflicting interests? *