Exploring the clinical utility of two staging models for bipolar disorder

Abstract Objective To assess the clinical utility of two staging models for bipolar disorder by examining distribution, correlation, and the relationship to external criteria. These are primarily defined by the recurrence of mood episodes (model A), or by intra‐episodic functioning (model B). Methods In the Dutch Bipolar Cohort, stages according to models A and B were assigned to all patients with bipolar‐I‐disorder (BD‐I; N = 1396). The dispersion of subjects over the stages was assessed and the association between the two models calculated. For both models, change in several clinical markers were concordant with the stage was investigated. Results Staging was possible in 87% of subjects for model A and 75% for model B. For model A, 1079 participants (93%) were assigned to stage 3c (recurrent episodes). Subdividing stage 3c with cut‐offs at 5 and 10 episodes resulted in subgroups containing 242, 510, and 327 subjects. For model B, most participants were assigned to stage II (intra‐episodic symptoms, N = 431 (41%)) or stage III (inability to work, N = 451 (43%)). A low association between models was found. For both models, the clinical markers “age at onset,” “treatment resistance,” and “episode acceleration” changed concordant with the stages. Conclusion The majority of patients with BD‐I clustered in recurrent stage 3 of Model A. Model B showed a larger dispersion. The stepwise change in several clinical markers supports the construct validity of both models. Combining the two staging models and sub‐differentiating the recurrent stage into categories with cut‐offs at 5 and 10 lifetime episodes improves the clinical utility of staging for individual patients.


| INTRODUC TI ON
Clinical staging models are widely used in medicine to classify disease progression in chronic conditions such as cancer, cardiac failure, autoimmune disorders, and dementia. 1 Only relatively recently has the concept of staging been introduced in the field of psychiatry, first by Fava and Kellner, 2 later by McGorry et al. 3 Bipolar disorder (BD) is characterized by recurrent (hypo)manic and depressive episodes alternating with euthymic intervals and has a highly heterogeneous longitudinal course and outcome. Diagnostic and prognostic precision would therefore benefit from a staging system that determines and predicts illness progression in individual patients that can guide treatment decisions as early as possible.
Given the lack of established biomarkers underpinning the pathophysiology, current staging models in psychiatry rely exclusively on clinical characteristics.
In recent years, numerous studies have aimed at identifying biomarkers for bipolar disorder, and some progress has been made. Teixeira et al 4 reviewed the current state of peripheral (blood-derived), genetic, neuroimaging, and neurophysiological candidates for biomarkers of bipolar disorders and found several promising candidates, including peripheral inflammation markers and Brain-Derived Neurotrophic Factor (BDNF). Several studies have found bipolar illness to be associated with chronic low-grade inflammation with exacerbations during mood episodes. [5][6][7][8] A decrease in BDNF was found during manic and depressive episodes. High BDNF levels were found to be associated with a good treatment response to lithium. 9,10 Other interesting candidates include oxidative stress markers such as lipid peroxidation, DNA/RNA damage and nitric oxide, neuronal markers such as S100B and Neuron Specific Enolase (NSE), and metabolic markers such as GLP-1, ghrelin, adiponectin, and GIP. 11 No specific genetic maker for BD has been identified, although BD has high heritability. 12,13 The current consensus implicates a role of multiple genetic variants that are dependent on environmental interactions and epigenetic mechanisms. When combined, they increase the chance of developing BD. 4 Most neuroimaging studies found alterations in cortical thickness similar to those in schizophrenia and major depressive disorder. Several fMRI studies found a different activity in brain regions responsible for emotional regulation and cognitive control. 4 Various candidates for neurophysiological biomarkers have been identified, using electroencephalography (EEG) or magnetoencephalography (MEG), such as lower delta inter-hemispheric coherence in the frontal region and greater parietal-temporal and central parietal region alpha hemispheric coherence, 14 suggesting dysfunctional longrange cortical connectivity in BD.
Two staging models prevail for BD. The model as proposed by Berk et al 15 (further called "model A" in our paper) is largely defined by the occurrence and recurrence of mood episodes. It starts with an at-risk stage (0), moving from a prodromal stage 1 to first episode stage 2, to recurrent stage 3 to chronic unremitting illness stage 4.
The model is fueled by the neuroprogression hypothesis that every mood episode is toxic to the brain. 16 We previously applied this model to patients in order to evaluate the stage progression over the course of the first 5 years after the diagnosis of BD. 17 Frank et al 18 applied this model to subjects with BD to assess genetic and neuroimaging markers for each stage, after which progressive changes were found.
Kapczinski et al 19 proposed an alternative staging model ('model B') based upon intra-episodic functional impairment. This   includes a latent stage and four clinical stages, defined by the absence of intra-episodic symptoms (stage I), intra-episodic symptoms (stage II), intra-episodic impairment with inability to work   (stage III) In this study, we have investigated both models next to each other using data from the Dutch Bipolar Cohort. For a staging model to have clinical utility, sufficient distribution over the stages is necessary, since this is a measure for the distinctiveness of the model and the ability to define illness progression. Our first aim was therefore to assess the dispersion of the subjects over the different stages for both models. We hypothesized to find a lower dispersion in Model A over Model B, as Model A will likely exhibit a ceiling effect as all nonchronic subjects with recurrent (ie, two or more) mood episodes are assigned to the same stage.
Second, the association between the two models was determined. Since both models are based on an underlying concept of illness progression, we expected to find a high association between both models. Alternatively, a low association would suggest that both models reflect different aspects of the illness, supporting the idea that a combination of models would be synergetic.
However, both models may represent different indicators of the underlying progress of BD. Given a lack of specific biological markers, the validity of both models may be compared using clinical markers of disease progression, such as age at onset, BD in parents, childhood trauma, treatment resistance, and longitudinal illness course. We expected to find alterations in these markers concordant with the stages in the two models.

| Participants and procedure
Data were acquired from the Dutch Bipolar Cohort (DBC), performed by the University Medical Center Utrecht (UMCU), the Netherlands, in collaboration with the University of California at Los Angeles (UCLA). Patients were recruited via clinicians (19.2%), the Dutch BD patient association (15.8%), pharmacies (33.6%), advertisements (6.9%), self-referral (5%), participated in previous studies of the UMCU (4.5%), or from miscellaneous undocumented resources (15.0%). 28 The diagnosis bipolar-I-disorder was confirmed using the Structured Clinical Interview for DSM-IV (SCID-I; 29 ).
Inclusion criteria for all participants were: a minimum age of 18 years at inclusion, at least three grandparents of Dutch ancestry, and a thorough understanding of the Dutch language. The study was approved by the medical ethical committee of the UMCU and all participants gave written informed consent. More information on this cohort is provided in the studies of Vreeker et al 33 and Van Bergen et al. 28 The sample characteristics are presented in Table 1.

| Application procedure for staging
All subjects were assigned to a stage from both model A and model B using a decision flowchart ( Figure 1).
For model A, (sub)stages were allocated using a set number of items originating from the Questionnaire for Bipolar Disorder (QBP; 30 ). Patients were divided into groups on the basis of the current mood episode. Those with a current mood episode were divided into two groups, euthymia in the previous year versus the absence of a period of recovery in the last year, the latter qualifying for stage 4. The total lifetime number of previous manic and depressive episodes was summed. In case of a current mood episode, one additional mood episode was added to the total. Patients were allocated to stage 2 (one mood episode), stage 3a (one mood episode with current residual symptoms), stage 3b (two mood episodes) and stage 3c (multiple recurrent mood episodes). Since the majority of subjects were allocated to group 3c, this stage was further subdivided, with cut-off points at 5 and 10 episodes, in accordance with cut-off points previously defined by Berk et al and Magalhães et al. 16,34 For model B, 19 subjects were assigned to a stage ranging from latent to stage IV, using a predetermined set of items from the Questionnaire for Bipolar Disorder (QBP; 30 ) ( Figure 1). Stages I to IV were assigned based on social, occupational, and psychological functioning (GAF > or <80), current mood episode (yes or no), employment over the last year (yes or no), work limitations (present or not present), and limitations in functioning (present or not present).

| External criteria
Several markers for clinical disease progression were assessed. These were selected based on earlier recommendations 24 including BD in parents, childhood physical abuse, childhood sexual abuse, age at onset, episode acceleration, increasing or decreasing episode severity and treatment resistance ( Table 2). Treatment resistance was opera-

| Statistics
Data were analyzed using SPSS24.0 35 . A Spearman's rank correlation was calculated as a measure of association between models.
The markers for clinical disease progression were tested for construct validity using analysis of variance (ANOVAs) and X 2 statistics.

| Clinical sample
The sample consisted of 1396 patients. All demographic and clinical variables and test statistics are listed in Table 1.

| Association between models
A Spearman's rank correlation was calculated between model A and B, with stage 3c of model A subdivided in cutoffs at 5 and 10 episodes (as shown in Table 3). The correlation for models A and B was 0.21 (P < .05), signifying a low association.

| D ISCUSS I ON
Using data from the Dutch Bipolar Cohort, we investigated the applicability of two different staging models for subjects with BD  36 The clinical utility for this subdivision has previously been found to align with episode dependent treatment resistance. 36,37 After 10 episodes, a decrease in moodstabilizing response was found for lithium by Swann et al. 37  Ideally, for a clinical useful staging model, clinical markers show an alteration concordant with the staging model. 24 In both models, the markers "age at onset," "episode acceleration," and "treatment resistance" show an increase concordant with the stages, supporting the construct validity of the staging model. We found "age at onset" to lower in higher stages in both models, underscoring onset at childhood or adolescence to be associated with worse outcome. 38,39 We found episode acceleration over progressive stages, indicating less time to recover or remain well. This is in line with the kindling model. 40 The increasing number of medication classes over the stages is in line with earlier studies on treatment resistance. 21,37 Future studies may focus on selecting those variables putting individuals at a higher risk of progressing to more advanced stages, adjusting clinical interventions accordingly.