Papers eligible for screening were written in English, published in a peer-reviewed journal and reported data on humans with mental disorders according to the DSM-III (13), -IIIR (14), -IV (15), 1994), -IVTR (16), or -5 (17), the Research Diagnostic Criteria (18) or the International Classification of Diseases (19). We used the same criteria as Cosci and Fava (11), meaning that the following papers were eligible for screening: category a) papers wherein a (full or partial) staging model was proposed including a motivation of its existence; category b) papers studying clinical features related to a staging model; category c) papers studying treatment interventions related to a staging model. Additional inclusion criteria for category b) were: inclusion of at least 10 patients; papers with participants with co-occurring mental or organic disorders were also included. Additional inclusion criteria for category c) were: inclusion of at least 10 patients, inclusion of a comparison group or a crossover design, at least a double-blind design in the case of pharmacological treatments, at least a single-blind design in the case of nonpharmacological treatments. Papers with a primary focus on neuroanatomy or biological markers were excluded.

The search strategy of Cosci and Fava was replicated, including the databases used by Cosci and Fava, which were Medline, psychINFO, EMBASE and Cochrane. The databases were systematically searched from May 2012 to June 2023. Reference lists of relevant systematic reviews on clinical staging and of all included papers were checked for additional papers. Search terms were ‘stage OR stages OR staging’, combined using the Boolean ‘AND’ operator with different categories of mental disorders. Abstracts, titles, and keywords were searched. We combined with the following search terms: ‘mental disorder’, ‘psychiatric disorder’, ‘mood disorder’, ‘anxiety disorder’, ‘substance abuse disorder’, ‘schizophrenia’, ‘eating disorder’, ‘conduct disorder’, and ‘personality disorder’. These search terms deviated in two ways from Cosci and Fava’s (11) original search terms. First, we chose to include additional categories of mental disorders for which a potentially progressive course may be assumed, i.e., personality and conduct disorders. Second, we wanted to account for changes in the transition to DSM-5 and therefore performed an additional search including the search terms ‘obsessive compulsive disorders’ and ‘posttraumatic stress disorder’. In the Supplementary Material we reported the search strategy.

The searches in the databases were conducted by one reviewer (S.Cl.) and references were exported to Endnote (20). After removal of duplicates in Endnote (21), the remaining papers were exported to Rayyan (22). The last duplicates were removed by hand in Rayyan. Papers were screened by four independent reviewers (S.Cl., N.S., S.Cr., L.P.). Given the high number of papers retrieved during this procedure, we chose to first assess interrater reliability of eligibility ratings. Two pairs were made (S.Cl./N.S. and S.Cr/L.P) and both pairs of reviewers independently rated 100 papers based upon title, abstract and key words. As interrater reliability was excellent for both pairs (k = 0.96 to 1.00) (23), reflecting (almost) perfect agreement between both screeners, all references were divided and screened independently by the four reviewers. In case of doubt, a reviewer could consult the other reviewer of the pair to decide upon eligibility. Disagreements were solved by consensus (24). If there still was doubt about inclusion of the paper for the full text round (round 2), inclusion was discussed by all authors to find a consensus. In round 2, all possible eligible papers in the full texts round were screened by two independent researchers. We used the same procedure in case of disagreement as in round 1.

The first data extraction was performed on 23/08/2022. Data were extracted from the papers in two different ways. Firstly, a summary of findings was reported in a pre-designed table including the headings ‘objective’, ‘sample’, ‘stages and N’, and ‘conclusions’. This was done by N.S. and S.Cl. and supervised by H.V. Secondly, preliminary drafts were written for the result section according to a pre-designed format, including ‘full or partial staging model’, ‘reason for inclusion (criterion a, b, or c)’ and ‘main findings’. The various disorder sections were divided between N.S. and S.Cl. They composed summaries for each disorder section with assistance from L.P. and B.B. Each summary was supervised by one of the senior researchers (H.V., A.V., S.v.A., J.H., L.D.). Further details (objective, sample, stages and N and conclusions) of the included studies were reported in Tables 1 – 9 .

The selection procedures and resulting outcomes are presented in a PRISMA flowchart ( Figure 1 ). In line with procedures described for data extraction, we used the extracted data to summarize findings for each category of mental disorders. Data are presented in a table that summarizes key findings from the selected studies. In addition, brief text summaries describe the number of studies, their reasons for inclusion (i.e. whether they met inclusion criterion a, b, or c), whether the studies related to an existing staging model, the study objective, and a summary of findings.

Our search revealed 12985 papers. After removing duplicates, 6804 papers were screened on titles, abstracts, and keywords. Of these, 217 papers were subject to full text screening, of which 45 papers were included. Inspection of the reference lists of these 45 papers revealed another 26 papers that met inclusion criteria. The PRISMA flowchart is shown in Figure 1 .

Twenty-one papers met our inclusion criteria for schizophrenia or related psychotic disorders. One paper was included because of a theoretical focus on staging (‘category a’ see full definition in method section), seventeen papers focused on clinical characteristics of different stages (‘category b’) and three papers comprised interventions for one or several stages of a staging model (‘category c’). An overview of papers with a summary of aims and conclusions can be found in Table 1 .

Ten papers referred to McGorry and colleagues’ (4–6) model or to an adaptation of this model. Armando and colleagues (25) proposed an extension of McGorry’s model, by including associated impairments in social functioning in the formulation of stages. In addition, they described several stage-appropriate adaptations, based upon mentalization-based treatment. Eight papers investigated clinical features related to McGorry’s staging model. Berendsen et al. (26) supported construct validity by demonstrating worse affective, catatonic, and negative clinical markers in later stages. Although their paper revealed no difference in treatment response between stages, they found that duration of untreated psychosis (DUP) was associated with more symptomatology both at baseline and at two-year follow-up. In a follow-up paper, they proposed subdividing stage 2 in less (2a) versus more (2b) than one year of DUP, which was supported by their finding that negative symptoms were more severe in stage 2b than in stage 2a (27). In a subsequent paper, working memory and processing speed as measures of cognitive performance were found to be lower in later stages at baseline, however this was not the case anymore at 3 and 6 years of follow-up (28). Godin et al. (29) studied stage stability and validated their model by using number of episodes, daily functioning, and total scores on the Positive and Negative Syndrome Scale (PANSS). A subdivision of stages 2 and 3 based upon mood and cognition was proposed by data-driven cluster analyses. Kommescher et al. (30) found more maladaptive coping styles in high-risk populations (stage 0-1) and in stage 4. Li et al. (31) focused on stability of Psychotic disorder Not Otherwise Specified (PNOS) and suggested a subdivision of stage 2 in first episode PNOS (2a) or schizophrenia (2b). Another validation paper (32) yielded significant differences in the factor structure of the PANSS between stages independent of age of onset and duration of the illness, indicating staging may serve as an appropriate model to deal with the clinical heterogeneity of schizophrenia. Peralta et al. (33) focused on stages from the first psychotic episode onwards and made alterations to the McGorry model based on stability and progression of non-remitting illness. Validation of this adjusted model followed mainly from differences between stage 2 and 3A and addressed the factors involved in full versus incomplete remission. Finally, one paper investigated stage-tailored interventions referring to McGorry’s model. Ruiz-Oriondo et al. (34) found that combining integrated psychological therapy (IPT) with emotional management therapy (EMT) improved clinical symptoms, cognitive performance, social outcome, and quality of life in stage 4.

The remaining eleven papers departed from other staging models than McGorry’s. Nine papers studied correlates of stages and two articles [i.e., Rapado-Castro et al. (35), and Falkai et al. (36)] studied a corresponding intervention. Duration of illness was used in a paper by Bagney et al. (37) to define treatment groups and to study executive functioning related to negative symptoms. Rapado-Castro et al. (35) found a significant interaction between duration of illness and response to N-acetyl cysteine (NAC) for positive symptoms and functional variables, but not for negative or general symptoms. Specifically, this mediator effect for DUP in response to treatment was more evident in subjects with 20 years or more DUP, suggesting a potential advantage of adjunctive NAC on positive symptoms in patients with chronic schizophrenia. Ortiz et al. (38) used the four-stage model proposed by Cosci and Fava (11) to study clinical and psychopathological differences. Sauvé et al. (39) also used a four-stage model to study negative symptoms by categorising data from 47 studies on point prevalence. Carrion et al. (40) described and clinically validated a partial four-stage model that focused on the prodromal phases of psychosis as well as on conversion rates between stages. Dragioti et al. (41) developed and described a clinical staging approach using the PANSS pyramidal model and arbitrarily chose stages based on age groups and gender, as duration of illness was not available. Fountoulakis et al. (42) used duration of illness as the primary factor to describe stages and plotted illness duration against PANSS, factors resulting in four major stages (43). Positive symptoms were dominant in the first three years, excitement and hostility in the period between 3 and 12 years, depression and anxiety after 12 years, and neurocognitive impairment after 25 years. Negative symptoms were found to be mostly stable during all stages, with a mild increase from after 18 years onwards. Fountoulakis et al. (44) also studied the role of gender, age at onset (four groups) and duration of illness (seven groups) on the course of schizophrenia. They found a later onset and more benign course of the disorder for females, a relation between early onset and slower progression of the disorder for both sexes, while they did not find an effect of the disorder duration. Falkai et al. (36) used an early (<5 years illness duration) and late (≥ 15 years) stage to study the use of cariprazine versus placebo. Cariprazine was found to be more effective than placebo in both stages with a larger treatment response in the early stage. Lastly, Fusar-poli et al. (45) conducted a qualitative paper, with input from patient experts, and themes were discussed between patients, family members and professionals to design enriched stages, improving their validity, based upon the lived experience of patients.

Twenty papers met inclusion criteria for bipolar and related disorders. Four papers presented theoretical proposals for a staging model, ten papers described clinical features of stages, five papers studied stage-related treatments and one paper (46) studied both clinical features and a stage-related intervention. Four different staging models, or adaptations, and associated interventions were mentioned, and eight papers used correlates of stages. See Table 2 .

Two studies focused upon clinical characteristics, based on a previously proposed clinical staging model of Duffy and colleagues (47, 48). The first study found that, regardless of bipolar parent substance abuse, offspring in stage 1, 2 and 4 were likely to develop substance use disorder compared to stage 0, while no further significant differences were found between stages 1 to 4 (49). The second study found that offspring of one parent diagnosed with bipolar disorder with no long-term response to lithium were more likely to develop psychotic and neurodevelopmental disorders compared to offspring with a lithium responding parent. Therefore, they advised including parents’ response to lithium in the staging model (50). A subsequent conceptual paper distinguished between classic episodic and spectrum bipolar disorder within the staging model and underlined the high heritability of the disorder and the need for a thorough family history assessment to add context to early (otherwise nonspecific) risk syndromes (51).

Berk et al. (52) described an adaptation of McGorry’s staging model of psychotic and severe mood disorders (4) for bipolar disorder, which was also used by Chanen et al. (53) in a combined model for borderline personality disorder (BPD), unipolar depression and bipolar disorder given their common underlying risk factors, age of onset, co-occurrence and overlapping core symptoms (see personality disorders, below). The same model was then slightly simplified, and interventions tailored to each stage were proposed (54). A third paper used the model of Berk et al. (52) to study clinical features. Illness progression was monitored during five years after onset of bipolar I or II disorder with slower progression through the earlier stages and faster transition in the later stages (55).

Van der Markt et al. (56) refined a combined model based on the staging models of Berk et al. (57), Kupka and Hillegers (58) and Duffy et al. (51). Retrospective life charts were used to study progression through stages in the five years after onset of bipolar disorder (stage 2) with the majority reaching stage 3 and a smaller amount reaching stage 4. Two papers referred to the staging model described by Kapczinski et al. (59). Rosa et al. (60) evaluated functional and neurocognitive performance across stages and found a progressive impairment in functioning along the stages with neurocognitive decline in the later stages. Goi et al. (61) studied patterns of pharmacological treatment with more functional impairment when more medication was used from the second stage onward. Two papers compared different staging models with each other. Van der Markt et al. (62) compared the models of Berk et al. (57) and Kapczinski et al. (59) and found a low association suggesting the models act complementary. Furthermore, they suggested using ‘dispersion over the stages’ to assess the clinical utility of a model. Macellaro et al. (63) also compared these models with the models of Kupka and Hillegers (58) and Duffy (51) and retrospectively found stage progression in all four models during a ten year period. Results showed that an increased number of mood episodes worsened severity of next episodes and reduced treatment response.

Finally, several papers discussed correlates of stages for bipolar disorder. Most studies used ‘number of episodes’ as a proxy to define stages. One study (64) studied functional outcomes in patients with only one versus more (depressive, manic, or mixed) episodes. First episode patients were younger and more autonomous, had less cognitive complaints and better work performance, and were better able to enjoy spare time and their companions. Magalhães et al. (46) observed more severe symptoms and a lesser quality of life after more episodes had passed. In a subset of (chronic) patients, antidepressant use was evaluated with no significant difference between subgroups.

Four papers studied differential effects of interventions for bipolar disorder. Peters et al. (65) found that intensive psychotherapy was associated with more likely and faster recovery compared to collaborative care in patients with less previous episodes and shorter duration of disease (i.e., earlier stage). Similarly, Morris et al. (66) noted more beneficial treatment effects for structured group psychoeducation compared to optimized unstructured group support in patients with fewer episodes. Murray et al. (67) defined the late stage as more than six episodes and found that online mindfulness therapy improved quality of life. This contrasts with findings of Kamali et al. (68) who found that self-reported number of lifetime and past-year mood episodes predicted neither the effects of lithium nor quetiapine. However, illness duration was on average more than 23 years and number of episodes high. Reinares and colleagues (69) proposed still another partial staging model of remitted patients and used latent class analysis to find a two-class model fitting the data best. Classification in ‘good’ or ‘poor’ outcome stages was based on functional outcome with episode density (number of episodes divided by disorder duration), residual level of depressive symptoms, estimated verbal intelligence and inhibitory control as strongest predictors. Grande et al. (70) used cluster analysis on several clinical variables (number of episodes, age of onset, time since first episode and functioning) to divide their study population in an early and late stage with the first functioning better, having a later age of onset and fewer episodes.

Our search revealed one newly proposed staging model including a study on clinical correlates, two papers on clinical correlates of existing staging models and one stage-related treatment paper. A more extensive description of findings can be found in Table 3 .

Rhee and colleagues (71) proposed and studied a staging model for Major Depressive Disorder (MDD) with three stages, namely stage 1 ‘new onset of MDD’ (past-year but not prior MDD), stage 2 ‘chronic or unrecovered MDD’ (past-year and prior MDD) and stage 3 ‘recovery from MDD’ (prior but no past-year MDD). Verduijn and colleagues (72) studied the construct and predictive validity of a clinical staging model for MDD, as proposed by McGorry and colleagues (4), consisting of different stages (0, 1A, 1B, 2, 3A, 3B, 3C, 4) based on the severity and duration of symptoms and the number of episodes. At 2-year follow-up, most of the clinical characteristics, such as severity and duration of symptoms and disability, worsened across the stages. Reneses and colleagues (73) modified Hetrick and colleagues’ model (74), by separating those with “recurrence from a previous depressive episode that was stabilized with a complete remission” into stage 3b1. They found that ‘treatment resistance’ – operationalized using the Maudsley Staging Model based upon a) disease episode duration, b) symptom severity and c) treatment failures – distinguished best between clinical stages and they argued that this variable differentiates best the stages in Hetrick and colleagues’ (74) model. Finally, a study by Dodd and colleagues (75) found no significant differences between the number of previous depressive episodes and treatment response.

The two papers identified refer to clinical correlates (see Table 4 ). Clarke and colleagues (76) used Hickie and colleagues’ (77) (see below) transdiagnostic model to classify participants according to the stage of their anxiety disorder. They found that clinical features, such as greater psychological distress, higher levels of depression, and more alcohol use distinguished participants in stages 2 and 3 from participants in stage 1b. There were no differences in reported levels of anxiety between stage 1b and the higher stages. Bokma and colleagues (78) adapted Hickie and colleagues’ (77) and McGorry and colleagues’ (4) models, differentiating between ‘subclinical’ stages (0, 1a, 1b), clinical stages with comorbidity (2b, 3b, 4b) and without (2a, 3a, 4a) comorbidity. Linear trends were found with severity of anxiety, depression and disability increasing in the later stages.

Our search found two conceptual proposals and one study of clinical correlates (see Table 5 ). Fineberg and colleagues (79) proposed a partial staging model, starting with subthreshold symptoms that may represent the first at-risk stage for obsessive compulsive disorder (OCD), for which they suggest psychoeducation for parents on not accommodating of the obsessive-compulsive symptoms. They additionally proposed an ultra-high-risk group, defined by at-risk symptoms and vulnerability because of environmental risk factors. The next stage includes patients who do meet the syndromal threshold for OCD, for which they recommend cognitive behavior therapy and medication (SSRI). Finally, they describe later stages as consisting of patients who appear to be resistant to treatment.

Fontenelle and Yücel (80) proposed a staging model for OCD, based upon a review, considering clinical markers, potential biomarkers, and outcome characteristics. They proposed stage 0, 0a, 0b, 0ab, 1, 2, 2a, 2b and 3. The authors identify different types of symptoms, levels of family accommodation, and comorbidity for these stages and they also propose different treatment strategies. This model was studied by Benatti and colleagues (81) in a follow-up study. About two-thirds remained unchanged. The worsened group displayed a higher prevalence of comorbid disorders, a higher prevalence of somatic obsessions and more unfavorable employment features. The improved group showed higher rates of magical thinking and violence/harm obsessions.

Three conceptual papers and one study of correlates (see Table 6 ) were identified. McFarlane and colleagues (82) distinguished between five stages of posttraumatic stress disorder (PTSD). Their model describes a range of possible neurobiological changes associated with different stages and they stress how these biological processes may also affect the immune system and underpin somatic comorbidities. A proof-of-concept study used machine-learning techniques to predict stage assignment, based upon McFarlane’s staging model (83). The authors found that number of symptoms, the Clinician-Administered PTSD Scale-DSM-5 edition (CAPS-5) total score, global severity score, and presence of current/previous trauma were most predictive of PTSD stage. Nijdam et al. (84) proposed to extend McFarlane’s model with information processing systems and psychophysiological stress and emotional reactivity, and consciousness. The authors proposed interventions for each stage of the disorder based upon their extended approach. The search also retrieved a hypothetical staging model for persistent complex bereavement disorder (PCBD) following the loss of a loved one in traumatic circumstances (85), to inform treatment decisions.

Three papers on staging models for feeding and eating disorders were identified, two conceptual papers and a study of clinical correlates (see Table 7 ). Treasure et al. (86) proposed a staging model, based upon a systematic review, distinguishing between four stages (high risk, early, syndrome, severe enduring). They focused on the progression of anorexia nervosa (AN), as few data were found to support staging models for bulimia nervosa and binge eating disorders. Stages for AN were largely based upon the presence of predisposing features, illness features and neuroprogressive features, and duration of illness. The authors propose different interventions with so-called ‘psychoprotective’ (cognitive dissonance strategies and social media literacy) and ‘neuro-protective’ (promoting healthy eating and physical activity) interventions for the high-risk stage, and family support and family-based therapy for the early stage. A slightly modified version of the same model was included in Treasure et al. (87), differentiating ‘high risk’ in a ‘high risk’ and an ‘ultra-high risk/prodrome stage’. The authors explore different interventions for the severe and enduring anorexia nervosa (SE-AN) stage, based upon a maintenance model including interpersonal consequences, behavioral consequences, and reactions to chronic stress. Their suggestions include caregiver skill training, habit reversal therapy, exposure techniques, neuromodulation, and pharmacotherapy. The clinical utility of a concept of SE-AN was further explored in an outpatient service (88). Results highlighted different courses and service use associated with different stages of AN.

No papers were found for staging models in conduct disorders.

No staging models were found for substance use disorders.

Two conceptual papers on staging models were found for borderline personality disorder (BPD) (see Table 8 ). Chanen et al. (53) (2016) presented a combined model for BPD and mood disorder, based upon their similar age of onset, common risk factors, frequent co-occurrence and overlapping clinical phenomenology. This model further outlines a mental health care response to each stage. Interventions are more generic and simpler at stage 0 and 1a and become more specialist and intensive at stages 1b and 2, and further on. A second staging model for BPD has been proposed by Hutsebaut et al. (89). Their model uses features of BPD, co-occurring psychopathology, and psychosocial disability as criteria to distinguish between five stages.

The thirteen included papers all studied clinical correlates and are outlined in Table 9 . Twelve papers used Hickie and colleagues’ (77) transdiagnostic staging model. In a study of young people seeking mental health care, assignment to stages was reliable and at follow-up 11%, 19%, and 33% of respective stages 1a, 1b, and 2 progressed to a later stage. In a large-scale longitudinal observational study, Iorfino and colleagues (90) investigated the transition rates from a help-seeking stage and an attenuated syndrome stage to a discrete disorder stage in patients with anxiety, mood, and psychotic disorders. Stage progression from the help-seeking stage was associated with older age, self-harm, manic-like experiences and lower social functioning, and engagement in education, employment, or training. Progression from the attenuated syndrome stage was associated with older age, psychotic-like experiences, previous use of psychiatric medication and a history of childhood psychiatric disorders. Addington and colleagues (91, 92) followed youth with emerging severe mental illness (SMI), assigned to similar stages. Of the help-seeking participants, 50% remained symptomatic, and 7.5% moved to the next stage or developed a SMI. Of the attenuated syndrome participants, 9% developed a SMI and one-third had symptom remission within 12 months. Hermens and colleagues (93) used the same staging model in a sample of young people seeking mental health care for psychotic and/or depressive symptoms. They found that the discrete disorder group displayed the most impaired neuropsychological profile (i.e., problems in memory and executive functioning), compared with the attenuated syndrome group, with a healthy control group being least impaired. To further inform empirical research in transdiagnostic staging, Ratheesh et al. (94) explored interrelationships and risk factors among mood, psychotic and anxiety 1b symptom stages. Anxiety, depressive and psychotic stages were found to be inter-related and associated with the following risk factors: sex at birth (female), more emotional and behavioral difficulties in early adolescence, and life events in late adolescence. Hypomania was not interrelated with these stages nor with these risk factors. Based on these findings, the authors suggested that anxiety, psychotic and depressive symptoms could form a combined transdiagnostic stage in their studied cohort.

Two papers investigated neuropsychological functioning across stages. Romanowska and colleagues (95) reported that participants in the help-seeking stage 1a performed worse than healthy controls on measures of speed of processing, working memory, reasoning, problem solving and overall neurocognitive functioning. Participants in the attenuated syndrome stage 1b showed lower processing speed and poorer working memory than participants in the asymptomatic stage 0. Participants in stage 0 performed worse than healthy controls in working memory, reasoning, and problem solving. Tickell and colleagues (96) found differences in verbal learning, verbal memory, visual memory and set shifting between stage 1b and 2+. Both groups showed similar improvement in neuropsychological functioning at follow-up.

In a study of disturbed sleep-wake cycle patterns, Scott and colleagues (97) found that help-seeking young people experienced more wake time after sleep onset, compared with healthy young people. Participants in the mild symptom and established disorder stages had lower sleep efficiency, compared with healthy young people. Delayed sleep increased across stages. Cross and colleagues (98) used the medical records of young people to assign them to one of four groups: stage 1a lacking a current (DSM-IV) diagnosis, stage 1a with a current diagnosis, stage 1b lacking a current diagnosis, or stage 1b with a current diagnosis. Age, gender, severity of illness, functioning and psychological distress had differential associations with both planned treatment intensity and attendance rates.

One paper used an online survey method to identify perceived utility of a wide range of attributes of mental health care for different disorders in the attenuated syndrome and discrete disorder stages (99). Hartmann et al. (100) introduced CHARMS (Clinical High At Risk Mental State), a pluripotent, at-risk mental state concept. CHARMS criteria are proposed as an extension of UHR (Ultra High Risk) criteria, extending these to psychosis (UHR), high risk for severe depression (HRD), high risk for mania (HRM) and high risk for BPD (HRB).

Finally, Wigman and colleagues (101) used experience sampling methods in a sample of women, 95% twins, to study the reciprocal impact of momentary mental states (positive affect, negative affect, and paranoia) over time across different stages of severity of psychopathology; staging was operationalized across four levels of increasing severity of psychopathology, based on the total score of the Symptom Checklist. They found that more severe stages were characterized by stronger connections and more variable connections between mental states. Moreover, severe stages indicated more individual-specific associations between mental states over time. These results suggest that, as individuals move through progressive stages, the dynamics between different mental states become increasingly stronger (referring to the nomothetic concept of staging), and the differences between individuals become progressively larger (referring to the idiographic concept of profiling).

Notable strengths include using the same eligibility criteria as Cosci and Fava (11), allowing for comparison and identification of potential trends in publications about clinical staging. We also performed an a priori interrater reliability test, showing very high to almost perfect levels of agreement, to avoid differences among raters in selection of papers for eligibility. Our systematic review was also preregistered.

Limitations include introducing extra search terms referring to disorders for which we assumed a potentially lifelong progressive course, like personality and conduct disorder (that may progress into antisocial personality disorder). However, we did not include neurodevelopmental disorders in our search, such as autism spectrum disorder, and we excluded papers with a primary focus on neuroanatomy or biological markers. Furthermore, we chose to include extra search terms, i.e., PTSD and OCD, in a follow-up search to include categories of mental disorders that have been re-allocated into new individual categories in DSM-5. Although this inclusion was justified given by the transition to DSM-5, which occurred after the publication of Cosci and Fava’s review, this deviated from our preregistered search protocol. Furthermore, we deviated from Cosci and Fava’s (11)) eligibility criteria by excluding papers referring to stages of change in substance abuse and eating disorders. Also, we did not include papers referring to treatment-resistant depression, as we decided that such papers did not reflect the criterion of a partial or full staging model in our eligibility criteria. Finally, we did not perform any quality assessment of the designs used in the included studies in this review, which could be seen as a shortcoming. Although recommended, we believe the usefulness of such a quality assessment is limited given the aim of our systematic review to provide a broad overview of the current state of clinical staging and given the type of papers retrieved by the search. The results of our review revealed very heterogeneous studies, varying from RCT’s, cohort studies, longitudinal studies, mixed methods studies, survey studies and qualitative studies. Widely used tools for Risk of Bias analyses, such as Cochrane’s tool or ROBINS-I, are not suitable to report on the quality of heterogeneous studies. Therefore, we opted to merely describe findings, as a reflection of academic and clinical interest in staging. Consequently, no conclusions can be drawn about the quality of evidence derived from these studies. Our results should be interpreted largely as a reflection of efforts to validate staging models, rather than as justification of these models in and of themselves. reviews could rigorously assess the quality of evidence.