It is frustrating that diagnoses are based largely on descriptive phenomena. Often they vary from practitioner to practitioner, and may well change markedly over the course of illness (for example, non-psychotic to psychotic syndrome or unipolar to bipolar mood disorder). As diagnoses are made at non-specific time points along complex illness pathways that evolve from risk to onset and progression to chronic ill-health, they often relate poorly to the actual stage of illness. For the major anxiety, mood or psychotic disorders, the illness process typically has its onset in late childhood or early puberty and then recurs or continues progressively into adult life 29 30 31 . Although 75% of major mental disorders begin before the age of 25 years 32 , our diagnostic criteria are derived largely from the experiences reported by middle-aged persons with established illness. These phenotypes often map poorly onto earlier and often less specific phases of the illness experience 22 25 33 34 .

The current systems also assume the concept of multiple parallel pathways each leading to distinct diagnostic categories - an assumption that is not readily supported by modern family, genetic and neurobiological risk factor studies 20 35 36 . Criterion-based symptom sets (which then give rise to specific and ‘independent’ categories) prioritize phenomena such as delusions, hallucinations, periods of elevated mood or increased energy, psychomotor slowing, emotional blunting, or cognitive slowing for disorders such as schizophrenia, bipolar disorder or severe depression. Data from recent community studies that assess patients longitudinally from childhood or adolescence 29 37 38 39 , however, emphasize the extent to which many of these phenomena are shared across disorders 40 . Prototypically, anxiety disorders that are evident in children before the age of 12 years predict later depressive, bipolar and psychotic disorders 41 .

With regards to adult-type disorders, persistence or recurrence of symptoms appears to have greater predictive significance than cross-sectional observation of specific symptoms 39 42 43 44 45 . Hence, the great clinical challenge is to derive new diagnostic systems that are not only consistent with developmental epidemiology and neurobiology but also useful when applied in everyday clinical practice.

The great virtue of the research-based classification systems of the late 1970s was that they promoted the pursuit of reliable diagnoses. To achieve diagnostic reliability, a small number of dimensions in human behavior (for example, anxiety, depression, impaired cognition or psychotic phenomena) were organized into a large number of discrete and separate ‘disorders’ - on the basis of the presence or absence of set numbers of key symptoms. Inevitably, this gave rise to a checklist approach to diagnostic practice and the proliferation of diagnostic ‘categories’.

However, reliability was oversold as the necessary precursor to validation of those entities 16 . While the ‘atheoretical’ and reliability-driven approach of DSM-III did set clinical research free from the previous psychological, behavioral or medical models, it did not result in a new era of preventive and therapeutic strategies. In retrospect, it appears that the DSM-III-derived disorders or independent categories relied too heavily on descriptive psychopathology, historical practice or clinical consensus 46 47 .

The post-DSM-III era did enhance our capacity to conduct large, multi-site and international research and ensured that diagnostic concepts were less constrained by local history, culture, religion or social customs, or fashion. Such aspirations were essential to promoting the international ‘science’ base of clinical psychiatry and psychology and for forging links to key areas of emerging neuroscience, especially molecular genetics and neuroimaging. Further, this greatly assisted the move away from highly idiosyncratic practice or ‘diagnostic systems’ that were used to prop up the delivery of poorly evaluated (or intrinsically harmful or discriminatory) forms of mental health care. Mental health practice is one area of social endeavor that has benefited greatly from globalization and greater transparency, communication and scrutiny 2 .

When linked to national epidemiological surveys, the international classification systems have underpinned more accurate estimates of relative disease burden, access to care and impacts of mental disorders on broader health and social systems 1 32 48 . This has led to more forward thinking about how to best support broad social settings that enhance mental health and wellbeing 49 . Thus, in many ways, the great successes of the post-DSM-III diagnostic era are in public health and related health system developments. It is rather frustrating that, despite the best efforts of scientists and practitioners, the stated aims of advancing clinical research have not been so successful.

Given the relative failure to validate the DSM- or ICD-derived categories, it is time to set new goals (Table 1) - and propose a range of possible, and rather diverse, strategies to support those goals (Table 2). An over-riding concern in mental health has become the reduction of the population-health burden by adoption of various early intervention strategies 28 50 51 . These focus clinical service development and related research on identifying early forms of illness not only to reduce current morbidity but also to prevent progression to more severe or chronic illness types and associated role impairment 52 53 54 55 56 57 . Although the fundamental empirical work in this area has focused largely on psychotic disorders, there is now a much broader clinical database emerging that is targeted at the early phases of the more prevalent anxiety and depressive disorders 22 53 58 59 60 61 .

CBT: cognitive behavioral therapy; SNRI: selective norepinephrine reuptake inhibitor; SSRI: selective serotonin reuptake inhibitor.

This movement is similar in conceptualization with that being promoted in other related clinical areas of medicine that have very large impacts on current and future health costs, notably diabetes and related metabolic syndromes 125 126 127 . As with the move to define pre-diabetes (an intermediate state between normal and clearly raised blood glucose concentrations), it is entirely possible to define emerging mental disorders by their intermediate symptom levels, impacts on function, patterns of persistence over time or predictive capacity 21 22 56 57 . Further, it is possible to design new health care platforms to increase the access to care for young people presenting with these conditions 53 55 .

Importantly, the same types of concerns arise in the arenas for both pre-diabetes and early intervention in mental health 51 128 . That is, as the number of individuals identified by health systems grows, and those with less severe forms come into active clinical care, the extent to which early forms of illness can be effectively managed largely by non-pharmacological or other lifestyle-based approaches is a key consideration 66 129 130 . Critics of this area 11 131 all too readily see an over-medicalization or a pharmaceutical-industry-driven conspiracy to be at the heart of such genuine public health movements. Yet the evidence from the clinical staging of ischemic heart disease largely demonstrates the opposite. The greatest public health emphasis was placed on reducing smoking rates, promoting exercise and addressing other modifiable risks. High cost, high risk interventions (for example, revascularization) were reserved for those experiencing major events (for example, first myocardial infarct), those with recurring episodes (for example, coronary by-pass surgery), or following a progression to chronic disease (for example, pharmacotherapy of heart failure) 132 133 .

Currently, the key difference between the diabetes and early mental disorder arenas is the extent to which the diagnosis of the ‘at-risk’ or ‘prodromal’ state for diabetes relies on an independent laboratory test - and the extent to which that laboratory test has predictive value for later poor health outcomes. Therefore, concurrently with the development and evaluation of earlier diagnostic thresholds based on descriptive phenomena and related disability estimates 22 52 53 , we must intensify our search for better neuropsychological, brain imaging, circadian, neurophysiological, immune or other markers of these early states 58 59 72 76 99 106 134 135 136 and design studies to test their predictive capacity.

These new goals respond directly to the challenges posed by novel preventive and longitudinal approaches that are targeted at the recognition of early phenotypes. For example, for depressive disorders, there is international recognition that the extraordinary premature death and disability costs attributable to these conditions 48 137 reflect their early age-of-onset, high current and lifetime population prevalence, typical illness patterns of recurrence and chronicity, and likely comorbidity with alcohol and other substance misuse as well as physical ill-health, most notably in the form of premature cardiovascular disease 90 107 138 139 140 141 . The ways in which incorporating these new goals might impact on the classification of depressive disorders is outlined here (see Table 1).

As importantly, it is necessary to state those factors which are not the primary goals of a clinical or research classification system. These include: categorizing all forms of abnormal perception, mood, cognition or other behavioral disturbance; imposing independent category status on disturbances that are fundamentally dimensional in nature and frequently inter-correlated (see common forms of anxiety and depression 142 143 ); creating diagnostic hierarchies that presume etiological or pathophysiological dominance of one form of disturbance over another (for example, psychotic compared with mood or cognitive phenomena); or using the classification system as the sole basis for allocating research funds, licensing treatments, providing access to health care, determining legal matters, reimbursing health care costs or supporting access to other personal entitlements system.

Contrary to prevailing wisdom, there is no urgent need to have one over-arching international diagnostic system or bible for all perceptual, mood, cognitive and other behavioral syndromes. Unfortunately, the recent attempts to include all things within the one descriptive system have underpinned a rather unhelpful debate about the inclusion or exclusion of fewer or more categories 11 12 144 . Another unintended side-effect of the forced international orthodoxy was the lack of support of other plausible models or active support for the development of alternative approaches 13 16 145 .

An emerging body of work, utilizing appropriate genetic modeling and brain imaging data provides provisional support for such alternative approaches 20 . These data favor a model in which mental disorders represent broad patterns of psychopathology. Within this model, various genetic and environmental factors determine the development of common brain circuits that underpin normal behavior and cognition. When perturbed, changes in the function of these circuits give rise to deviant behavior or cognitive function. This model has influenced the National Institutes of Mental Health, which now actively promotes investigation of brain-circuit-based approaches and proposes that they may underpin a new approach to the classification of common mental disorders 20 146 .

Within the circuitry models, changes from normal function (thereby giving rise to disorders) are by their nature dimensional rather than categorical. Generally, correlations exist between the degree of accumulated dysfunction of the underlying circuits (or related systems) and the extent of behavioral or cognitive change. This dimensional approach shifts the emphasis from making a specific diagnosis to determining the current level of cognitive or behavioral change in individual patients. Clinical and related neurobiological research then examines the extent to which common risk variables (for example, developmental delay, trait neurocognitive impairments, alcohol or other substance misuse) increase risk to measured behavioral or cognitive change, increase risk to markers of brain impairment (for example, poor neuropsychological function, excessive cortical thinning, disturbed neurophysiological markers) or increase risk to poor outcomes - independently of any specific relationship with classical ‘diagnostic’ entities 108 147 148 .

Core dimensions of cognition and behavior are hypothesized to be common to all persons and not viewed as a unique set of characteristics occurring only in those who present with mental health problems. Perturbations of brain circuits will typically result in ‘trans-diagnostic’ rather than pathognomonic symptom sets. An important prediction of such systems is that many objective measures of the structure (for example, brain imaging) or outputs of such systems (for example, cognition - including traditional neuropsychological and social cognition, circadian, hypothalamus-pituitary-adrenal axis function, neurophysiological or immune) will also show little diagnostic specificity. Over the last 30 years, this has been the overwhelming experience with proposed diagnostic markers, such as the dexamethasone suppression test for major depression 149 . Preliminary analyses of data from those in the early phases of a range of major psychotic or mood disorders also support this perspective 58 59 60 61 72 73 76 99 106 136 150 , 151 .

A range of diverse strategies can be advanced to support these new goals (see Table 2). These do not need to come together into one single bible for clinical or research purposes, as the initial aim is to develop, evaluate and refine them frequently as evidence emerges. A potential consequence of the key shift from many categories to the use of fewer underlying dimensions is that it is likely to lead to the need for an alternative term to mental disorders. A very broad descriptive concept that encapsulates a 21st century approach to disturbances of perception, mood, cognition and other behavioral dimensions may be required. The clear public, professional, clinical and research benefits likely to derive from abandoning the historical and cultural divisions between clinical neurology and psychiatry and clinical psychology have been highlighted by others 13 26 .

In doing so, such strategies move away from the very broad categories we currently use (for example, major depression) to describe very heterogeneous populations. Generally, more specific states (for example, depression in association with late-life vascular disease; first episode psychosis; cannabis-associated recurrence of psychosis; depressive episode following previous manic episode) provide a better fit with known risk factors, actual pathophysiology or known illness course and may be much more useful in research and some fields of clinical practice 83 .

In association with all disorders, more general age-of-onset and stage-of-illness concepts should be recorded. Those that have clear pathways to illness need to be emphasized (for example, childhood attention deficit hyperactivity disorder leading to early age of onset of substance misuse; childhood schizotypal or autistic behaviors leading to early onset psychotic disorder; adolescent social anxiety leading to alcohol misuse). Concurrently, this may help to reduce the overuse of the concept of comorbidity - limiting it in the future to the co-occurrence of genuinely independent conditions (for example, psychotic disorder and alcohol dependence).

It is clear that one of the most robust ways of differentiating diagnostic groups is on the basis of response (or non-response) to specific treatments. Much greater effort needs to be invested in the reverse translation agenda (that is, working back from the bedside to the bench). For example, we urgently need to unravel what neurobiological factors differentiate those with schizophrenia who respond to clozapine or those with bipolar disorder who respond to lithium. The goal is to develop better predictors of treatment-response before initiating therapies.

At this time, we argue that we may be better served by investing heavily in a much smaller number of more discrete syndromes that are partially validated by clinical course, objective markers or predictors of treatment response. That is, rather than prioritize reliability, we could vigorously pursue those more homogeneous groups that are identified in clinical practice or family studies 90 91 92 93 . Inevitably, this would advance the push for more personalized approaches to health care. Further, relevant groups could be preferentially recruited to specific preventive or treatment trials (for example, late-onset depression; childhood-onset obsessive-compulsive disorder; psychotic disorder preceded by childhood schizotypal behaviors; first-episode mania; depressive disorders characterized by disturbed hypothalamic-pituitary function, circadian disruption or immune activation).

Another clear way of approaching this desire for greater specificity is to focus on syndromes that are closely linked with discrete environmental exposures, social adversity or inter-current medical events 152 . These natural experiments can be used to facilitate more targeted pathophysiological studies (for example, post-infective or post-stroke neuropsychiatric syndromes; post-traumatic depressive states; depression or prolonged fatigue states following cancer therapies; alcohol or other substance misuse neuropsychiatric syndromes; cannabis-associated first episode of psychosis 89 153 154 155 156 ).

We would also promote an overt move away from checklist-driven criteria that rely on poorly validated symptom sets (for example, treating independent symptoms such as sleep disturbance and suicidal ideation as equivalent items for the diagnosis of major depressive disorders). Instead, we favor a move back towards clear syndromes that share key pathophysiological, symptom or illness course features (for example, psychotic depression; melancholic disorders that are associated with concurrent psychomotor change; first-episode psychosis; bipolar disorder diagnosed by a discrete manic episode). This process would also be assisted by greater use of objective markers of key features of disorders (for example, computer-generated measures of neurocognitive function; actigraphy-generated measures of sleep-wake cycle; self-report based measures of cognitive styles) and wider use of clinician or self-reported measures of key symptom dimensions.

One of the most important developments in recent years has been the importing of the concept of clinical staging from general medicine. In other clinical domains (for example, oncology, coronary heart disease, inflammatory join disease), it is totally inadequate to choose treatments, or plan health care, for persons who suffer from recurring or progressive conditions simply on the basis of a broad diagnostic category (for example, breast cancer). We suggest that it is equally meaningless in mental health to select specific treatments on the basis of broad categories such as schizophrenia, bipolar disorder or major depression. There is a wealth of evidence indicating that patients at different points along the illness continuum of all of these conditions show quite different patterns of response to various interventions 21 88 157 158 .

Consequently, we have proposed a general framework for clinical staging that can be applied to the more severe mood or psychotic disorders (Figure 1). This framework is readily applied to those who present for health care and clearly differentiates those in early phases (stages 1a ‘seeking help’ or 1b ‘attenuated syndromes’) from those who have reached a higher threshold for disorder (stage 2 and above - see Figure 1). Current clinical and related neurobiological studies (for example, magnetic resonance imaging; neuropsychological and sleep/circadian studies) of this framework provide provisional evidence in support of its reliability and validity 22 58 59 .

As compared with current DSM and ICD disorder thresholds, and particularly for the common anxiety and depressive disorders, this approach raises the bar for the initiation of more specific or intensive pharmacological or behavioral strategies. That is, although this approach does encourage more active health care for those at lower levels of illness, it also promotes the use of safe, easily-delivered and non-specific psychological intervention, health-care, suicide prevention and other secondary prevention strategies for those who have not yet reached the higher threshold for a stage 2 disorder 54 84 .

All of these strategies move the emphasis from the rather sterile academic debates about the virtues of competing symptom sets (or illness thresholds) 159 160 161 to the provision of relevant treatment at key points along an illness path. In dialogue with individual patients, this would result in the provision of far more accurate, multi-dimensional and context-relevant information. In our view, stating clearly that a 21-year-old man has a first-onset psychosis, with a strong family history of mania and associated five-year history of persistent cannabis use is more likely to guide treatment selection, health care planning and accurate prognostic statements than assigning any of the current psychosis disorder categories. Similarly, stating that a 17-year-old girl has the recent onset of a depressive disorder characterized by psychomotor slowing and preceded by prolonged fatigue, seasonal changes in mood and energy, with no evidence of childhood anxiety or concurrent alcohol or substance misuse is more useful than attaching any of the current depression categories. While these approaches are already common in everyday clinical practice they are not captured in our current diagnostic systems.

Further, we need systems that support innovative research paradigms such as those promoted by the youth mental health for common mental disorders and early intervention in psychosis movements 162 163 . Within these novel paradigms, it is far more important to recruit patients who share key demographic, illness stage, prior treatment characteristics or family history (for example, age, gender, duration of illness, lack of prior exposure to medical treatments, family history of psychosis or mania) rather than pre-selecting on the basis of poorly-validated syndromal constructs such as major depression.

Such novel research paradigms inevitably focus attention on those genetic or environmental risk factors that are common across disorders (and hence may be modified with resulting benefits to a large number of persons), distinct from the narrow search for unique risk factors that link specifically to each separate disorder. Key considerations such as the role of intra-uterine environments, early childhood infections, childhood sexual or emotional abuse, early adolescent alcohol or substance misuse, grossly distorted adolescent sleep patterns, and physical inactivity in the teenage years will then emerge as major foci for targeted research and potential public health programs 58 60 108 164 165 166 .

These alternative approaches to classification would give rise to a new wave of early intervention, biomarker, clinical intervention and other longitudinal studies. Specifically, they would also promote reverse translation initiatives - meaning those research programs that work back from clinical research based on well-characterized, more narrow and probably more homogeneous patient groups to elucidate more fundamental biological correlates. Additionally, the reverse translation agenda has other important attributes, including actively responding to topics of real significance to those living with the illness (for example, cognitive impairment in those with psychotic disorders) and capitalizing on phenomena that have been well-replicated in studies on humans (for example, lithium-responsive bipolar disorder or clozapine-responsive schizophrenia).

The combination of key clinical insights and patient priorities plays a crucial role in setting the reverse translation research agenda 2 . Categories deserving of detailed neurobiological research are those clinical situations in which a robust link with relevant pathophysiological, risk, illness-stage or treatment variables has already been demonstrated (for example, circadian-based depressive disorders - see Table 3; cannabis-associated psychosis; first-episode mania; or clozapine-responsive schizophrenia disorders).

On the basis of making new links in these smaller but clinically-defined cohorts with the best available markers of the active pathophysiology, we would then expect new insights that could provide a basis for working forwards again (that is, bench back to bedside). A new wave of more relevant animal models, molecular targeting or other rapid assay modalities could emerge. The goal then is to use that new knowledge to implement better targeted and more individualized prevention or active treatment strategies (that is, traditional forward translation programs 167 186 187 188 ). Such translational research programs (that is, those that incorporate both reverse and forward strategies) would then genuinely link 21st century neurobiology to clinical practice in an iterative and mutually-informative discourse.

The impact of the move away from investigating traditional schizophrenia to more focused first-episode psychosis (and then related prodromal or at-risk clinical research) in Australia, Europe and the United Kingdom demonstrates the extent to which these novel approaches can genuinely transform clinical practice 27 56 163 . By contrast, those working in the anxiety and mood disorders fields have been slow to comprehend the significance of these developments 28 85 . We believe the time is now right for a more general shift in direction in favor of those diagnostic practices that focus attention on key developmental, illness course, reverse translation and strategic intervention approaches. Although we can recognize the genuine progress that the DSM and ICD revisions from 1980 onwards supported, there is now no longer any good reason for the international mental health community to be constrained by the ongoing revisions of these systems.