Background
Bipolar disorders (BDs) are a group of chronic psychiatric illnesses with diverse clinical courses composed of combinations of (hypo)manic and depressive episodes (Sadock et al.
2014). Although a number of guidelines have been established for the pharmacotherapy of BDs (Yatham et al.
2013; Jeong et al.
2015; Goodwin
2009), long-term drug response and its clinical correlates in BD patients receiving standard clinical care have not been well explored. Considering the wide variation between individuals in the manifestation of illness, including biphasic and recurring courses, difficulties in defining outcomes should be expected. Another obstacle to the investigation of long-term response is variability in drug options and changes in the drugs that are taken during the course of treatment (Ghaemi et al.
2006; Baek et al.
2014).
The criteria for long-term response have been defined in various ways. One or a few isolated variables, such as time to recurrence, reduction of episode frequency, length, or severity, and reduction of the time spent in the hospital, have frequently been used in previous studies (Kleindienst et al.
2005). These simple variables, however, do not reflect the diversity of treatment courses observed in natural clinical settings. Global assessment of functioning (GAF) (American Psychiatric Association
2000) and the clinical global impressions scale for use in bipolar illness (CGI-BP) (Spearing et al.
1997) have also been used to evaluate long-term efficacy (Kusalic and Engelsmann
1998; Post et al.
2010). However, these scales are generally only appropriate for assessing a patient’s state at a single cross-sectional time point. A scale specifically tailored to the retrospective assessment of prophylactic lithium response in BDs was introduced by Alda and colleagues (Grof et al.
2002) (the retrospective criteria of lithium response in research subjects; the Alda scale), and has been widely used in genetic studies on lithium response (Grof et al.
2002; Passmore et al.
2003; Hou et al.
2016; Chen et al.
2014; Squassina et al.
2011). As the scale considers confounding factors, such as polypharmacy, compliance, and disease course before administration of medication, it can be adapted to various clinical cases and settings involving long-term treatment.
A number of previous studies investigated potential predictors of prophylactic lithium treatment in BDs. According to a systematic review by Kleindienst and colleagues (
2005), depression-mania episode sequence, earlier onset of illness, a high number of previous hospitalizations, and continuous cycling seem to be associated with poor outcomes. In later studies, psychotic symptoms, inter-episodic residual symptoms, mixed episodes, and rapid cycling (Backlund et al.
2009; Pfennig et al.
2010; Silva et al.
2016) were identified as predictors of poor long-term response to lithium. A positive correlation of long-term response with hyperthymic temperament (Rybakowski et al.
2013) and typically episodic course of illness with earlier onset (Garnham et al.
2007) were also reported. Comorbid anxiety disorders and alcohol abuse/dependence were associated with poor prophylactic efficacy in both early and recent studies (O’Connell et al.
1991; Young et al.
1993; Kliwicki
2014). Controversial results were generated regarding bipolar I disorder (BD-I) vs. bipolar II disorder (BD-II) (Kleindienst et al.
2005; Garnham et al.
2007; Rybakowski
2014) and a family history of mood disorders (Kliwicki
2014; Mendlewicz et al.
1973; Maj et al.
1985; Misra and Burns
1977; Coryell et al.
2000).
In contrast to the lithium response, studies on long-term response to the broader category of mood stabilizers are limited. A recent study evaluating the prophylactic efficacy of lithium, valproate, and carbamazepine reported an association between the likelihood of relapse and a mixed episode and the total number of manic or depressive symptoms prior to the observational period (Peselow et al.
2015). In naturalistic observational studies of the Stanley Foundation Treatment Outcome Network that assessed the prospective outcomes of BD patients receiving various combinations of pharmacologic treatments, family history of drug abuse, history of childhood abuse, a greater number of prior episodes, and comorbid substance abuse were associated with poor outcome (Post et al.
2010; Nolen et al.
2004). To the best of our knowledge, long-term studies investigating specific predictors of treatment response to valproate are lacking (Carvalho and McIntyre
2015).
According to a recent multi-center investigation of prescription patterns in Korea, valproate is more commonly used than lithium, and polypharmacy is used in 80.86% of patients (Baek et al.
2014). Also, medication changes owing to adverse drug effects, lack of response, and phase changes frequently occur during the course of treatment for BDs (Yatham et al.
2013; Baek et al.
2014; Arvilommi et al.
2010). Given the complexity of pharmacotherapy, the long-term effect of an isolated mood stabilizer is quite difficult to delineate. In addition, the effects of a single medication given alone would be hard to generalize to real-world BD treatment. A more global view of treatment effects on the long-term outcome of BD is required.
This study was designed to estimate the clinical response of individuals with BDs to long-term treatment with mood stabilizers. In order to reflect the prescription patterns shown by a recent nationwide survey that includes data from our clinic (Baek et al.
2014), treatment with the most commonly used mood stabilizers, i.e., valproate and lithium, was selected as target mood stabilizer treatment. The overall response rate was retrospectively assessed using the Alda scale based on observational data collected over a period of more than 2 years. This study also aimed to identify the factors associated with treatment response among all of the comprehensive clinical variables investigated.
Discussion
In this retrospective investigation of the clinical response of patients with BDs to long-term (2 years or more) treatment with valproate and/or lithium, one-third of the patients were good responders, with a total Alda score of 5 or higher. When our analysis excluded patients receiving both valproate and lithium, we did not find any significant differences in the long-term clinical effects between the two drugs. Previous experience with mixed episodes (according to DSM-IV criteria), delusions during manic episodes, appetite increase during depressive episodes, and comorbid anxiety disorders were related to a worse response.
As the present study does not focus on the effects and associated factors of a single drug, direct comparison of clinical response between the current study and previous studies, most of which investigated the effects of single mood stabilizers in isolation, would be difficult. In order to explore long-term response in a naturalistic clinical setting, we felt it necessary to consider medication changes and combinations of different mood stabilizers during the course of treatment. Therefore, we selected valproate, lithium, and a combination of the two drugs as a single target treatment. In the current study, valproate and lithium seemed to be associated with similar long-term responses. In addition, the supplementary analysis showed that baseline demographic and clinical features did not affect selection of one drug over another, except that there was a preference for valproate in BD-II. Valproate and lithium are the most frequently used medications in our clinic and in other mood disorder clinics in tertiary-care hospitals in Korea (Baek et al.
2014). According to a report from the Stanley Foundation Treatment Outcome Network (Post et al.
2010), these two drugs are the most frequently prescribed medications at the time of improvement and have high overall success rates in outpatients treated for BDs.
Although several studies have investigated the response to mood stabilizers using the Alda scale, they used different response criteria. In a previous report by Garnham and colleagues (
2007), which defined the ‘full-responder’ group as those with a total score of 7 or higher, the rate of full response was 30% in lithium users and 13% in valproate users. In the ConLiGen study (Manchia et al.
2013), which used the same criteria, 33% of lithium-treated patients were classified as ‘full responders.’ When adopting the ConLiGen criteria, the full-responder rate was just 15.0% in our study (12 out of 80 patients). The mean total score was also lower in the current study (3.4 ± 2.5) compared to in the ConLiGen studies, i.e., 4.4 ± 3.1 in their initial clinical report (Manchia et al.
2013), and 4.3 ± 3.3 and 3.9 ± 3.0 in their genome-wide association study (Hou et al.
2016). However, the mean A score of the current subjects (6.7 ± 1.9) was higher than that of the ConLiGen study subjects, which ranged from 6.0 to 6.4 (Hou et al.
2016; Manchia et al.
2013). This indicates a higher B score in the current subjects. The B score reflects baseline clinical characteristics that could affect the true causal relationship between treatment and outcome, including previous mood episodes, treatment duration, compliance, and additional medication. One prominent feature of our sample related to B score is a high rate (76.3%) of psychotic features that could result in the usage of additional antipsychotic medications.
To identify the clinical factors associated with long-term mood stabilizer treatment, we investigated a variety of baseline characteristics, including disease onset and course, symptoms of episodes, and comorbidities. Worse treatment response in patients with more previous mixed episodes, delusions during manic episodes, and appetite increase during depressive episodes was observed not only in the main analysis of all subjects but also in the supplementary analysis of BD-I patients alone. Analysis of comorbid psychiatric conditions could be performed only in the overall sample because of the small sample size of the BD-I group, considering the low rates of comorbidities. Anxiety disorders were only observed in the moderate/poor responders. Although study designs and target drugs differ between studies, the current results are roughly in agreement with previous reports of predictors of treatment response to mood stabilizers. Mixed episodes predicted poor long-term response in a study on lithium (Backlund et al.
2009) and a study on multiple mood stabilizers (Peselow et al.
2015). Psychotic features were also reported as a predictive factor of a poor response to lithium (Kleindienst et al.
2005; Backlund et al.
2009; Pfennig et al.
2010; Silva et al.
2016). According to the current results, among psychotic symptoms, delusions during manic episodes were specifically associated with worse responses. Increase in appetite is a major symptom of atypical depression, and atypical features of depression were reported to be associated with a greater rate of psychiatric comorbidities, increased distress, suicidal ideation, and disability, all of which might lead to negative treatment outcomes (Sadock et al.
2014; Matza et al.
2003). To the best of our knowledge, this is the first study to report on the atypical symptoms in depressive episodes as a predictor of poor long-term mood-stabilizer response in BDs. Various comorbid disorders were expected to occur and are known to be associated with poor response in BD patients (Kleindienst et al.
2005; Kliwicki
2014; Rybakowski
2014), and a concordant finding was detected only for anxiety disorders in the current study. In the case of other comorbid conditions, including eating disorders and alcohol/substance-related disorders, the small sample size might have limited our ability to detect associations. Other reported predictors of response, i.e., age at onset, a high number of previous hospitalizations, rapid cycling, and hyperthymic temperament (Kleindienst et al.
2005; Rybakowski et al.
2013), did not show a significant association in the current study, and need to be analyzed in future studies with larger sample sizes. Regarding the controversial results of previous studies on BD-I vs. BD-II (Kleindienst et al.
2005; Garnham et al.
2007; Rybakowski
2014), the current data could not generate any conclusions owing to the extremely skewed use of valproate in BD-II.
This study has several limitations. First, because of the relatively small number of subjects, false negative results are to be expected. The statistical power may be particularly limited to detect differences in patients taking lithium vs. valproate. A large number of patients in the clinic were excluded because they had not taken the index medications for more than 2 years. Second, as this is a retrospective study, a conclusive causal relationship could not be determined between clinical factors and poor response rates. Third, in this naturalistic observational study, many uncontrolled confounding variables might decrease the assay sensitivity. Choice of medication (valproate vs. lithium) and diagnosis of BD-I vs. BD-II were considered as possible confounding factors in the supplementary analyses, whereas the effects of other potential confounding variables, such as medication dose or plasma level, adjunctive medications, and non-pharmacologic treatment, were not excluded. In the comparison of long-term response between drugs, there might be additional confounding variables affecting the selection of a specific mood stabilizer. Fourth, lack of use of classical longitudinal illness metrics, e.g., time to recurrence/recovery is also a limitation of this study considering difficulties in direct comparison of the current results with previous studies. Finally, as all of the subjects were Korean patients, the current results may have limited generalizability to other populations.
This study also has the following strength. It is a naturalistic observational study that reflects a real-world clinical setting. Although retrospective evaluation was performed, reliable assessment of treatment response is expected considering the long follow-up period (at least 2 years) at a single institute and the involvement of the treating clinicians in outcome assessment. In addition, baseline disease characteristics and clinical variables were independently assessed in previous studies by the authors before the current assessment of treatment response.
Authors’ contributions
SWA, JHB, YC, SYY, KL, and KSH have contributed to study design, data collection, and clinical rating. SWA and KSH wrote the manuscript. YC contributed to data collection and management, and chart review of the patients. YKK and TP undertook the statistical analysis. All authors read and approved the final manuscript.