- Email: [email protected]
Predictors and outcomes of somatization in bipolar I disorder: A latent class mixture modeling approach
Author’s Accepted Manuscript Predictors and outcomes of somatization in bipolar I disorder: A latent class mixture modeling approach Juliet Beni Edgcomb, Berit Kerner www.elsevier.com/locate/jad
PII: DOI: Reference:
S0165-0327(17)30750-4 https://doi.org/10.1016/j.jad.2017.11.083 JAD9397
To appear in: Journal of Affective Disorders Received date: 27 April 2017 Revised date: 3 October 2017 Accepted date: 13 November 2017 Cite this article as: Juliet Beni Edgcomb and Berit Kerner, Predictors and outcomes of somatization in bipolar I disorder: A latent class mixture modeling a p p r o a c h , Journal of Affective Disorders, https://doi.org/10.1016/j.jad.2017.11.083 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting galley proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Title: Predictors and outcomes of somatization in bipolar I disorder: A latent class mixture modeling approach Juliet Beni Edgcomb, M.D., Ph.D.1 Semel Institute for Neuroscience and Human Behavior, University of California, Los Angeles, CA, USA Berit Kerner, M.D. Semel Institute for Neuroscience and Human Behavior, University of California, Los Angeles, CA, USA Witten/Herdecke University, Witten, Germany Abstract Background: Mood disorders are often associated with somatic symptoms. The role of somatic symptoms on disease progression in unipolar depression is substantially better characterized than that role in bipolar disorder. Moreover, the contribution of comorbid anxiety disorders and medical illness is not well understood. Method: We investigated 527 patients with bipolar I disorder clustered within 102 families using a latent class approach. Predictors were added stepwise into the model. Anxiety and commonly associated medical illnesses were added as covariates. Results: The rate of somatic symptoms in this sample was 73% (mean 1.7 symptoms), and 27.3% had a comorbid anxiety disorder. A two-class model, with a subgroup at high-risk for somatization, gave the best fit to the data. Multilevel mixture modeling accounted for family clusters. Somatic symptoms were independently associated with disease severity, defined as earlier age of first seeking psychiatric help (x=21.7 vs x=24.7, p=0.005) and first psychiatric hospitalization (x=25.7 vs x=28.2, p=0.03), greater probability of attempting suicide (x=0.41 vs x=0.32, p=0.047), and rapid-cycling disease course (x=0.57 vs x=0.36, p<0.001). Persons with few or no somatic symptoms were more likely to be hospitalized for severe mania (x=0.63 vs x=0.51; p=0.013), but did not significantly differ in hospitalization for severe depression. Limitations: The study is correlational. Information on pharmacologic interventions and comorbid diseases was limited. Conclusions: Somatic symptoms in bipolar disorder could be an independent indicator for disease severity, suicidality, and rapid-cycling disease course. In severe mental illness, somatic and psychological symptoms must be jointly addressed. Keywords bipolar disorder, somatic, somatization, medical, physical symptoms, anxiety bipolar disorder, somatic, somatization, medical, physical symptoms, anxiety 1
Corresponding author: Juliet Beni Edgcomb, M.D., Ph.D. UCLA Semel Institute for Neuroscience and Human Behavior 760 Westwood Plaza Los Angeles, CA, 90024 [email protected] Telephone: (310)-794-2053 Fax: (310)-825-0340
Background Bipolar disorder is a mental illness defined by episodic mood shifts between mania and depression (American Psychiatric Association (APA), 2013). Comorbid medical conditions and physical symptoms influence the symptom profile and disease severity of bipolar disorder (Perugi et al., 2015; Forty et al., 2014). Individuals with bipolar disorder suffer from somatic symptoms at a rate nearly double that of the general population (OR 1.82), a rate like that observed in unipolar depression (Edgcomb et al., 2016). Moreover, other psychiatric disorders often co-occur with bipolar disorder and among these, anxiety disorders are highly prevalent. More than half of individuals with bipolar disorder (55.8%) have a comorbid anxiety diagnosis and nearly a third (31.8%) have two or more diagnoses related to anxiety (Boylan et al., 2004). Despite the high frequency of anxiety and somatic symptoms in persons with bipolar disorder, it is poorly understood how these constructs are related. Somatic symptoms are defined as physical dysfunctions (e.g. of appetite, digestion, sleep) or bodily sensations that are perceived as unpleasant or worrisome (Kapfhammer, 2006). Symptoms are not feigned or intentionally produced, and may or may not occur in the presence of organic disease. Somatic symptoms may be conceptualized on a spectrum, ranging from a single physical dysfunction or sensation, to Somatic Symptom Disorder, defined in the Diagnostic and Statistical Manual of Mental Disorders (5th ed.; DSM–5; American Psychiatric Association, 2013) as consisting of one or more somatic symptoms, excessive thoughts, feelings, and behaviors, and with symptomatology present for six months or more. Numerous studies have investigated the complex association of anxiety and physical symptoms in unipolar depressive disorders (Henningsen et al., 2003), but there is a paucity of data on the relationship between anxiety, somatization, and bipolar spectrum disorders. Prior studies on somatization have had relatively small sample sizes of bipolar patients (Mykletun et al., 2010), were limited to anxiety subscale scores rather than discrete anxiety diagnoses (Perlis et al., 2006) and have focused predominantly on softer domains of bipolarity (e.g. bipolar II, bipolar III, cyclothymic disorder) (Coplan et al., 2015). Large epidemiological studies have established that anxiety disorders occur more frequently in persons with bipolar disorder compared to the general population (Bauer et al., 2005; Pavlova et al., 2015). Over the last two decades, research has demonstrated high prevalence rates of discrete anxiety disorders amongst persons with bipolar disorder (Nabavi et al., 2015): 20.8-36.8% have panic disorder (Chen and Dilsaver, 1996; Pini et al., 1997), 20% have agoraphobia (Freeman et al., 2002), 10-35% have obsessive-compulsive disorder (Chen and Dilsaver, 1995), 47.2% have social phobia (Kessler et al., 2005), and 14-32% have generalized anxiety disorder (Young et al., 1993). Some studies have suggested that bipolar disorder may have stronger association with anxiety disorder diagnoses compared unipolar depression (Chen and Dilsaver, 1995). Genetic data is now emerging to add to the extensive epidemiological and clinical data demonstrating this association (Chang et al., 2013). Prior conceptual models (Freeman et al., 2002) have suggested that anxiety and bipolar disorders may co-occur at high frequency (1) due to the high prevalence of both disease conditions, (2) because anxiety and bipolar disorders are separate disorders with a pathophysiological association, and (3) because anxiety and bipolar disorders are manifestations of the same underlying abnormality. This paradigm may in turn be applied to comorbidity with somatic symptoms, albeit with a distinct pathophysiological association and/or underlying abnormality. Somatization refers to the conversion of a mental state into somatic symptoms. When controlling for the presence of comorbid medical illness, we hypothesize that the extent to which an individual experiences somatic symptoms may be used as a proxy measure for somatization. Although the relationship between anxiety disorders, somatic symptoms, and markers of disease severity is well-established in unipolar depression, no studies have thus far investigated the relationship between anxiety disorders and somatic symptoms in bipolar I disorder (BP1D). We hypothesize that:
1. There is a subset of individuals with BP1D who experience a high level of somatic symptoms. This group can be clearly differentiated from the subsets of individuals with BP1D who experience either no or a low level of somatic symptoms. 2. Individuals who exhibit a high level of somatic symptoms experience higher bipolar disease severity as measured by disease indicators. 3. The propensity to experience higher disease severity persists when accounting for the presence of comorbid anxiety disorders and commonly associated physical illnesses. Method Study design A stepwise approach was used to incorporate predictors, account for clustering, and investigate distal outcomes. First, we identified a sample of persons with bipolar I disorder. Second, we extracted dichotomous responses on the presence or absence of eight somatic symptoms. Third, we ran a latent class analysis using a stepwise approach. In brief, the model was run in six steps as follows: (1) somatization variables as indicators, (2) somatic symptom variables and demographic variables (age, gender) as indicators, (3) two-level hierarchical modeling to investigate the effect of familial clustering, (4) somatic symptom variables and demographics as indicators with continuous (DCON) and categorical (DCAT) predictors, (5) somatic symptom variables, demographics, and medical conditions as indicators with continuous (DCON) and categorical (DCAT) predictors, (6) somatic symptom variables, demographics medical conditions, and anxiety disorders as indicators with continuous (DCON) and categorical (DCAT) predictors. Sample Data were provided by the National Institute of Mental Health (NIMH) Repository and Genomic Resources Center for Collaborative Genomics Research on Mental Disorders (NIMH Repository and Genomics Resource). The sample consisted of 527 individuals from 102 families, age 16 and older. All individuals had been interviewed by trained health care providers and had been assessed with the Diagnostic Interview for Genetic Studies (DIGS, Version 3.0) (Nurnberger et al., 1994). Consensus clinical diagnoses according to DSM-IV (4th ed.; DSM–IV, American Psychiatric Association (APA), 2013) were reached by at least three independent mental health experts using Best Estimate procedures (Leckman et al., 1982). The interviewers also collected information on demographics, family history, somatic symptoms, the longitudinal disease course of psychiatric disturbances, and medical conditions. Variable selection In this analysis, responses to questions about the perception of pain (DIGS 3.0, I1660-I1760), were used as indicators in the latent class analysis (LCA) as a measure of somatic symptoms, : “Have you ever been bothered a lot by problems with pains in your...” (1) abdomen or stomach (other than during menstruation), (2) back, (3) joints, (4) arms or legs (other than in the joints), (5) chest, (6) painful sexual intercourse (other than after childbirth), (7) genitals or rectum (other than during intercourse), and (8) other pain. The presence or absence of medical conditions was coded (0 absent, 1 present) and included in the model as auxiliary variables. Diseases included (Variable identification codes included in parenthesis) were thyroid disorder (I124), overactive thyroid (I125), underactive thyroid (I126), enlarged thyroid (I127), Cushing’s disease (I136), migraines (I139), ulcers or other bowel diseases (I142), peptic ulcers (I145), Crohn’s disease (I148), ulcerative colitis (I151), lupus (I154), learning disability (I157), vitamin deficiency (I940), meningitis or other brain disorders (I160), Parkinson’s disease or movement disorder (I49), multiple sclerosis (I52), Huntington’s disease (I55), stroke (I58), epilepsy, convulsions, or seizure (I61), and traumatic brain injury (I68). The presence of comorbid anxiety disorders was determined from corresponding International Classification of Diseases (ICD-9) codes established by diagnostic criteria from at least three independent raters, as described above. Comorbid anxiety disorders investigated included: anxiety disorders (300.20, 300.3, 300.21, 300.22, 300.01, 300.29, 300.23), obsessive-compulsive disorder (300.3), agoraphobia with panic disorder (300.21),
agoraphobia without panic disorder (300.22), panic disorder without agoraphobia (300.01), simple phobia (300.29), and social phobia (300.23). The following continuous predictors were extracted from the Diagnostic Interview for Genetic Studies (DIGS Version 3.0: age at initial interview, age of onset to seek psychiatric help (I3130), years of schooling (I360), number of psychiatric hospitalizations (I3650), number of admissions for drug use or alcohol detoxification (I3760), age of first psychiatric hospitalization (I3680), and number of suicide attempts (I4680). Categorical predictors included: sex (I10), lifetime history of psychiatric hospitalization (I3650), hospitalization for most severe mania (I5810), hospitalization for most severe depression (I4470), lifetime history of suicide attempt (I4670), rapid-cycling disease course (I6330), and lifetime use of cocaine (I8890), stimulants (I8900), sedatives (I8910), opiates (I8920), phencyclidine (I8930), hallucinogens (I8940), solvents (I8950), and other substances (I8960). Latent class mixture modeling We performed the Latent Class Analysis (LCA) using Mplus Version 7 (Muthén and Muthén, 1998-2010). We estimated the Latent Class (LC) membership for each individual using the estimation maximization (EM) algorithm based on the probability of endorsing a profile of variables (Muthén and Shedden, 1999). The starting values were from 200 random sets, so as to avoid local maxima. We selected the optimal number of classes when the Bayesian Information Criterion (BIC) reached a minimum (Nylund et al., 2007). In addition, the LC solution was checked by using the Akaike Information Criterion (AIC) (Akaike, 2011), the BIC and sample size adjusted BIC (Schwarz, 1978), the Lo-Mendel-Rubin (LMR) test (Lo et al., 2001), and the Bootstrapped Likelihood Ratio Test (BLRT) (McLachlan, 1987). We deleted the observations with missing data on covariates in the analyses that included covariates. In this mixture modeling framework, U stands for the binary, categorical, or count observed indicator variables and C refers to a latent categorical variable with K classes, C = (c ,c ,…,c ), where c =1, if individual i belongs to class k and zero if otherwise. Given c , conditional independence is assumed for U, i
i1
i2
iK
ik
i
P(u ,u ,…,u |c ) = P(u | c )P(u | c )P(u | c ) i1
i2
iτ
i
i1
i
i2
i
iτ
i
The data sample consisted of individuals nested within families. A two-level mixture model with maximum likelihood estimation was run with somatic symptom and demographic variables (age, gender) as indicators. The ‘cluster’ option was then specified in Mplus (Muthén and Muthén, 1998-2010). Family clusters were identified by a randomly assigned numerical code. The multilevel extension of the modeling framework allowed for random slopes and random intercepts that varied across clusters in hierarchical data (Heck and Thomas, 2015). The random effects across-cluster variation in slopes and intercepts was combined with individual-level indicators (somatic symptoms, demographics) to determine model fit when accounting for familial clustering. Model fit statistics for a two-level model accounting for familial clustering was compared with a two-level model that treated each participant as unrelated (i.e. independent). Model fit statistics suggested an insignificant effect of familial clustering. Using the pseudo-class method (Clark and Muthén, 2009; Asparouhov, 2010) we included auxiliary variables in the model to evaluate additional potential class predictors. Lanza et al’s method (2013) is based on draws from the posterior probabilities, from which we tested the equality of means across the latent classes. We interpreted the magnitude of the mean differences between the classes as an indicator of the strength of the prediction that the auxiliary variables influenced the class membership (Asparouhov and Muthén, 2007; Asparouhov and Muthén, 2014). We used multiple imputation to derive the latent class membership variables from the posterior distribution obtained by the LCA model estimation, after estimating the latent class model without including covariates. We then analyzed the imputed class variables together with the auxiliary variables, using the multiple imputation technique developed by Rubin (1987). The Lanza et al (2013) method was used as, unlike other mixture model auxiliary options, both categorical and continuous outcomes could be incorporated simultaneously in the model. For each auxiliary variable, we evaluated conditional class specific means, based on the estimated latent class model (option DCAT and DCON in Mplus).
Results The sample consisted of 57% female and 43% male participants who belonged to 102 families (Table 1). 75.1% of respondents were Caucasian. The sample’s mean age was 42.1 years (SD 14.3), while the average age of onset of bipolar disorder was 23.9 years (SD 11.4). Thus, a large number of individuals had been symptomatic for at least 18 years, had been hospitalized for bipolar disorder several times, and, in many cases, had attempted suicide, all of which indicated a severe form of disease. Comorbid anxiety disorders were prevalent (27%). Panic disorder was the most common comorbid anxiety disorder (13%). Medical disorders were frequently reported, the three most prevalent being migraine (22%), gastrointestinal disease (21%) and thyroid disorders (19%). All individuals in this sample had answered questions about frequency and perception of pain as indicators of somatic symptoms and the responses to these questions were used as indicators for the latent class approach. To explore the structure of this sample, we utilized a latent class mixture modeling approach. As our sample consisted of individuals grouped in families, we tested if it would be more appropriate to use a clustering approach within a two-level analysis. The comparison of the latent class models using a two-level clustering approach and a simple one-level analysis indicated that there was no difference in the model fit between the two approaches. To reduce the computational burden, we therefore ran our models without the two-level approach. The model fit indices suggested a division of the sample of patients with bipolar I disorder into two subclasses (Fig. 1, Table 3 and 4), based on the presence of comorbid somatic symptoms and their correlations. The first Latent Class (LCA1) is characterized by high probability of somatic symptoms (20-79%, Table 4). This class accounted for approximately one-quarter (27%) of the sample, and was 72% female. The second Latent Class (LCA2) is characterized by low probability of somatic symptoms (2-29%). This class represented approximately three-quarters (73%) of the sample and 52% of patients in this latent class were female. Compared with LCA2 (low somatization), individuals in LCA1 (high somatization) had a significantly greater probability of being diagnosed with thyroid disorder (33% vs 13%), migraine (35% vs 17%), peptic ulcers or bowel disease (44% vs 9%), and epilepsy, convulsions, or seizure (17% vs 5%) (Table 3). As these disorders disproportionately affected the high somatization group, we then added medical conditions as covariates into the model. Using the Lanza et al method (2013) to determine conditional class specific categorical variable probabilities and continuous variable means, severity predictors were compared between classes. When accounting for the aforementioned comorbid medical conditions, LCA1 (high somatization) was characterized by an older age to first seek psychiatric help (25.2 vs 23.1, p=0.03), fewer years of schooling (13.4 vs 14.2, p=0.003), higher number of psychiatric hospitalizations (5.17 vs 3.12, p=0.003), higher age of first hospitalization (30.9 vs 25.7, p<0.001), higher number of suicide attempts (1.64 vs 0.58, p=0.003), greater likelihood of attempting suicide (45% vs 29%, p =0.01), and greater probability of a rapid-cycling disease course (51% vs 36%). As expected, LCA1 was also characterized by higher probability of comorbid anxiety disorders (42% vs 18%, p<0.001), including obsessive-compulsive disorder (7% vs 1%, p=0.03), agoraphobia (11% vs 3%, p=0.02), panic disorder (19% vs 9%, p=0.02), simple phobia (12% vs 3%, p=0.009), and social phobia (8% vs 3%, p=0.05). In addition, individuals in LCA1 were more likely to have used tobacco (71% vs 59%, p=0.03), stimulants (34% vs 22%, p=0.04), opiates (23% vs 12%, p=0.03) and phencyclidine (19% vs 8%, p=0.02). There was no significant difference between classes for rates of cocaine, hallucinogen, solvents, or other drug use. We then repeated the analysis adding common medical conditions and anxiety disorders as predictors into the model. When accounting for the presence of comorbid medical conditions and anxiety disorders, compared with LCA2, individuals in LCA1 had a earlier age to first seek psychiatric help (21.7 vs 24.7, p=0.005) and earlier age of first psychiatric hospitalization (25.7 vs 28.2, p=0.03). Moreover, those in LCA1 tended to have more
suicide attempts compared with those in LCA2 (1.46 vs 0.72, p=0.06). Age of interview, years of schooling, total number of hospitalizations, and number of drug/alcohol admissions did not differ between classes. Individuals in LCA1 were less likely to be hospitalized for severe mania (51% vs 63%, p=0.01), more likely to have attempted suicide (41% vs 32%, p=0.04), and more likely to have a rapid cycling course (57% vs 36%, p<0.001). Discussion In this study, our goal has been to determine if persons with bipolar disorder and comorbid somatic symptoms experience increased markers of disease severity and poorer outcomes compared with those who report low or no somatic symptoms. In order to establish this, we have used a latent class mixture modeling approach to determine if a population of persons with bipolar disorder could be meaningfully subdivided into groups with differing degrees of somatization. Our results demonstrate a two-class model, wherein a group of individuals is prone to high levels of somatization and can be distinguished from those with low or no somatization. In our study, approximately one-quarter of the study sample was identified as high somatizers and these individuals experienced greater bipolar disease severity as measured by various indicators. Our findings are consistent with prior meta-analytic studies, which have found a rate of somatic symptoms amongst persons with bipolar disorder of close to 50% (Edgcomb et al., 2016; Stubbs et al., 2015). However, much of the prior research has been sparse and highly heterogeneous in methodology. Because of this, it has been difficult to meaningfully conceptualize groups within the bipolar population with different probabilities of somatization. We note that although our sample consisted of individuals nested within family clusters, when we accounted for clustering using two-level hierarchical modeling, there was no significant difference in model fit indices. As expected, the high somatization class had a greater probability of reporting comorbid medical illness, including thyroid disorders, migraine, peptic ulcers or bowel disease, and neurologic disorders. The high somatization class was also at increased risk of comorbid anxiety disorders, including obsessive-compulsive disorder, agoraphobia, panic disorder, simple phobia, and social phobia. When accounting for the presence of these anxiety disorders in addition to the comorbid medical illness, those who are classified as high somatizers were more likely to be female, were older, had a younger age of first seeking psychiatric help and first psychiatric hospitalization, were more likely to have lifetime suicide attempt, tended to have more suicide attempts, were less likely to be hospitalized for severe mania, and were more likely to have a rapid-cycling disease course. Age of interview, years of schooling, total number of hospitalizations, and number of drug/alcohol admissions did not differ between classes. Prior research has demonstrated that persons with severe mental illness, including bipolar disorder, experience higher levels of bodily pain, poorer physical functioning, and lower self-rated general health compared with the general population, and this in turn contributes to treatment nonadherence and poorer quality of life. To our knowledge, our study is the first to identify a subgroup of persons with bipolar disorder who have a high probability of somatization, and describe this group as high risk for a variety of markers of disease severity when accounting for illnesses that may otherwise explain this relationship (i.e. comorbid anxiety disorders and medical illnesses). Our finding that the high somatization group is characterized by younger age of onset of psychiatric illness and first psychiatric hospitalization is in keeping with prior research suggesting that age may distinguish subtypes of bipolar disorder (Leboyer et al., 2005). Moreover, the high somatization group was characterized by a longer total disease period, in that these individuals were first hospitalized at a younger age and were older at time of interview. Numerous studies have established the relationship between pain-related diagnoses and suicide risk (Hooley, 2014; Leverich, 2003), with the greatest risk for suicide found amongst individuals with psychogenic pain (Ilgen et al., 2008). This is consistent with our finding that the high somatization group was characterized by a higher probability of ever attempting suicide and having more individual attempts.
In our study, the high somatization group additionally had a higher probability of experiencing a rapid-cycling disease course of bipolar disorder. Prior research has demonstrated that persons with rapid-cycling bipolar disorder have a greater number of medical comorbidities (Kemp et al., 2010) and are more likely to have migraine headaches (Gordon-Smith, 2015) compared with the general population. Moreover, persons with a rapid-cycling disease course experience more functional impairment, likely due to an increased number of mood episodes and shorter periods of euthymia (McIntyre et al., 2006). A growing body of research has suggested that the cumulative neurobiological insults of multiple acute mood episodes, including increased oxidative stress, increases the presence of pro-inflammatory markers and deficits in neuroprotection (Berk et al, 2011). Our finding of an association of somatization with rapid-cycling disease course may suggest the role of incorporating physical symptoms in neuroprogression and clinical staging models of bipolar disorder. Whereas the likelihood of hospitalization for most severe depression did not differ between the high and low somatizing groups, those who had a high level of somatization were less likely to be hospitalized for severe mania. These findings suggest that the pattern of mood episodes may vary between individuals who experience a greater degree of somatic symptoms. Limitations As our measures of somatic symptoms utilized self-report, under or over reporting of symptoms could have occurred. By using probability-based multivariate statistical methods, the degree of uncertainty introduced by individual classification was minimized. In addition, the latent class mixture model was run recursively until a convergent solution was determined. Detailed information on pharmacologic and other therapeutic interventions was not available. Moreover, our operationalized definition of somatization as quantity of somatic symptoms, in the context of controlling for comorbid medical illnesses, is circumscribed by the limited available information on comorbid disease conditions. Our study is correlational, and any conclusions about cause and effect must be limited. Of note, although we suggest that greater somatic symptom burden is associated with markers of bipolar disease severity, this relationship is likely not unidirectional. More severe bipolar illness may be characterized by greater somatic symptom burden through a variety of mechanisms, including vulnerabilities acquired via neuroprogression. In order to maximize internal validity, multiple sources of information, including medical records, self-rated scales, and interview-based scales were referenced; however, retrospective recall of the experience of somatic symptoms, hospitalizations, and disease onset, may have introduced some uncertainty. These limitations are common to secondary analyses of existing data, and are mitigated, in part, by the use of a relatively large sample size. Future research Future research should investigate a broader range of somatic symptoms (e.g. functional syndromes, neurologic symptoms, other pain syndromes) and bipolar spectrum disorders (i.e. bipolar II, cyclothymia) beyond that which was available for our analyses. In our study, we controlled for the presence of commonly associated disease conditions; however, our list was limited by the data available and we could not account for all significant disease conditions. In a more precise characterization, it would be important to establish which somatic symptoms were, in fact, medically explained, and which remained medically unexplained. Although the DSM-V removes this distinction in a clinical context, it nevertheless remains important to developing an understanding of somatic symptoms and the implications of these symptoms on treatment. Our results suggest the importance of more precisely characterizing patterns of mood symptoms in bipolar disorder as they relate to somatic symptoms, especially the role of rapid-cycling bipolar disorder. Conclusions Nearly one-third of persons with severe and enduring mental illness are managed entirely by primary care (Rielly et al., 2012). In these circumstances, patients often present with a combination of mood, anxiety, and somatic symptoms (Simms et al., 2012). Significant comorbidity among somatic symptoms and mood disorders has resulted in a demand for conceptual models to explain this association. Our study puts forward a model that defines a group of individuals with bipolar I disorder who are at increased risk for somatic symptoms and, in turn, other markers of disease pathology, beyond that which would be explained solely by the presence of
comorbid medical illnesses and anxiety disorders. These results may inform models of clinical practice and emphasize the importance of somatic symptoms in severe mental illness. Acknowledgements We would like to thank the UCLA Statistical Consulting Group for their help throughout the course of this work. We are indebted to the investigators of the National Institute of Mental Health (NIMH) Bipolar Genetics Initiative. Funding sources The sample was provided by the National Institute of Mental Health (NIMH) Repository and Genomic Resources Center for Collaborative Genomics Research on Mental Disorders. This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Contributors Juliet Edgcomb and Berit Kerner designed the study, conducted the analyses, and wrote the first draft of the manuscript. All authors have contributed to and have approved the final manuscript.
Declaration of interest Conflicts of interest: none
References Akaike, H., 2011. Akaike’s information criterion, in: Lovric, M. (Ed.), International Encyclopedia of Statistical Science. Springer, Berlin Heidelberg, pp. 25-25 American Psychiatric Association (APA), 2013. Diagnostic and Statistical Manual of Mental Disorders, third (DSM-III), fourth (DSM-IV, DSM-IV-TR), and fifth edition (DSM-5). American Psychiatric Press, Arlington, VA. Asparouhouv, T., 2007. Wald test of mean equality for potential latent class predictors in mixture modeling. Technical appendix. Los Angeles: Muthén and Muthén, Los Angeles https://www.statmodel.com/download/MeanTest1.pdf (accessed 25 September 2017) Asparouhov, T., 2010. Bayesian analysis using Mplus. Technical appendix. Los Angeles: Muthén and Muthén, Los Angeles http://statmodel2.com/download/Bayes3.pdf (accessed 25 September 2017) Asparouhov, T., Muthén, B., 2014. Auxiliary variables in mixture modeling: Using the BCH method in Mplus to estimate a distal outcome model and an arbitrary secondary model. Mplus Web Notes, 21(2). https://www.statmodel.com/MixtureModeling.shtml (accessed 25 September 2017) Bauer, M.S., Altshuler, L., Evans, D.R., Beresford, T., Williford, W.O., Hauger, R., VA Cooperative Study#430 Team, 2005. Prevalence and distinct correlates of anxiety, substance, and combined comorbidity in a multi-site public sector sample with bipolar disorder. J. Affect. Disord. 85 (3), 301-315. Berk, M., Kapczinski, A., Andreazza, A.C., Dean, O.M.., Giorlando, F., Maes, M., Yücel, M., Gama, C.S., Dodd, S., Dean, B., Magalhäes, P.V., Amminger, P., McGorry, P., Malhi, G.S., 2011. Pathways underlying neuroprogression in bipolar disorder: Focus on inflammation, oxidative stress and neurotrophic factors. Neurosci. Biobehav. Rev. 35 (3), 804-817. Boylan, K.R., Bieling, P.J., Marriott, M., Begin, H., Young, L.T., MacQueen, G.M., 2004. Impact of comorbid anxiety disorders on outcome in a cohort of patients with bipolar disorder. J. Clin. Psychiatry. 65 (8), 11061113. Chang, Y.H., Lee, S.Y., Chen, S.L., Tzeng, N.S., Wang, T.Y., Lee, I.H., Chen, P.S., Huang, S.Y., Yang, Y.K., Ko, H.C., Lu, R B. 2013. Genetic variants of the BDNF and DRD3 genes in bipolar disorder comorbid with anxiety disorder. J. Affect. Disord. 151 (3), 967-972. Chen, Y. W., Dilsaver, S. C. 1995. Comorbidity for obsessive-compulsive disorder in bipolar and unipolar disorders. Psychiatry Res. 59 (1-2), 57-64. Chen, Y.W., Dilsaver, S.C., 1996. Lifetime rates of suicide attempts among subjects with bipolar and unipolar disorders relative to subjects with other Axis I disorders. Biol. Psychiatry. 39 (10), 896-899. Clark, S.L., Muthén, B., 2009. Relating latent class analysis results to variables not included in the analysis. https://www.statmodel.com/download/relatinglca.pdf (accessed 25 September 2017) Coplan, J., Singh, D., Gopinath, S., Mathew, S.J., Bulbena, A., 2015. A novel anxiety and affective spectrum disorder of mind and body—the ALPIM (Anxiety-Laxity-Pain-Immune-Mood) syndrome: A preliminary report. J. Neuropsychiatry Clin. Neurosci. 27 (2), 93-103.
Edgcomb, J.B., Tseng, C.H., Kerner, B., 2016. Medically unexplained somatic symptoms and bipolar spectrum disorders: A systematic review and meta-analysis. J. Affect. Disord. 204, 205-213. Forty, L., Ulanova, A., Jones, L., Jones, I., Gordon-Smith, K., Fraser, C., Farmer, A., McGuffin, P., Lewis, C.M., Hosang, G.M., Rivera, M., Craddock, N., 2014. Comorbid medical illness in bipolar disorder. Br. J. Psychiatry. 205 (6), 465-472. Freeman, M.P., Freeman, S.A., McElroy, S.L., 2002. The comorbidity of bipolar and anxiety disorders: prevalence, psychobiology, and treatment issues. J. Affect. Disord. 68 (1), 1-23. Gordon-Smith, K., Forty, L., Chan, C., Knott, S., Jones, I., Craddock, N., Jones, L.A., 2015. Rapid cycling as a feature of bipolar disorder and comorbid migraine. J. Affect. Disord. 175, 320-324. Heck, R.H., Thomas, S.L., 2015. An introduction to multilevel modeling techniques: MLM and SEM approaches using Mplus. Quantitative Methodology Series, third ed. Routledge, New York. Henningsen, P., Zimmermann, T., Sattel, H., 2003. Medically unexplained physical symptoms, anxiety, and depression: A meta‐analytic review. Psychosom. Med. 65 (4), 528-533. Hooley, J.M., Franklin, J.C., Nock, M.K., 2014. Chronic pain and suicide: Understanding the association. Curr. Pain Headache Rep. 18 (8), 435. Ilgen, M.A., Zivin, K., McCammon, R.J., Valenstein, M., 2008. Pain and suicidal thoughts, plans and attempts in the United States. Gen. Hosp. Psychiatry. 30 (6), 521-527. Kapfhammer, H., 2006. Somatic symptoms of depression. Dialogues Clin. Neurosc. 8 (2), 227-239. Kemp, D.E., Gao, K., Chan, P.K., Ganocy, S.J., Findling, R.L., Calabrese, J.R., 2010. Medical comorbidity in bipolar disorder: relationship between illnesses of the endocrine/metabolic system and treatment outcome. Bipolar Disord. 12 (4), 404-413. Kessler, R.C., Chiu, W.T., Demler, O., Merikangas, K.R., Walters, E.E., 2005. Prevalence, severity, and comorbidity of 12-month DSM-IV disorders in the National Comorbidity Survey Replication. Arch. Gen. Psychiatry. 62 (6), 617-627. Lanza, S.T., Tan, X., Bray, B.C., 2013. Latent class analysis with distal outcomes: A flexible model-based approach. Struct. Equ Modeling. 20 (1), 1-26. Leboyer, M., Henry, C., Paillere‐Martinot, M.L., Bellivier, F., 2005. Age at onset in bipolar affective disorders: A review. Bipolar Disord. 7 (2), 111-118. Leckman, J.F., Sholomskas, D., Thompson, D., Belanger, A., Weissman, M.M., 1982. Best estimate of lifetime psychiatric diagnosis: a methodological study. Arch. Gen. Psychiatry. 39 (8), 879-883. Leverich, G.S., Altshuler, L.L., Frye, M.A., Suppes, T., Keck P.E. Jr., McElroy, S.L., Denicoff, K.D., Obrocea, G., Nolen, W.A., Kupka, R., Walden, J., Gunze, H., Perez, S., Luckenbaugh, D.A., Post, R.M., 2003. Factors associated with suicide attempts in 648 patients with bipolar disorder in the Stanley Foundation Bipolar Network. J. Clin. Psychiatry. 64 (5), 506-515. Lo, Y., Mendell, N.R., Rubin, D.B., 2001. Testing the number of components in a normal mixture. Biometrika. 88 (3), 767-778.
McIntyre, R.S., Soczynska, J.K., Bottas, A., Bordbar, K., Konarski, J.Z., Kennedy, S.H., 2006. Anxiety disorders and bipolar disorder: A review. Bipolar Disord. 8 (6), 665-676. McLachlan, G.J., 1987. On bootstrapping the likelihood ratio test statistic for the number of components in a normal mixture. Applied Statistics. 36, 318-324. Muthén, B., Shedden, K., 1999. Finite mixture modeling with mixture outcomes using the EM algorithm. Biometrics. 55 (2), 463-469. Muthén, L.K., Muthén, B.O., 1998-2010. Mplus User's Guide, sixth ed. Muthen and Muthen, Los Angeles. https://www.statmodel.com/ (accessed 25 September 2017). Mykletun, A., Jacka, F., Williams, L., Pasco, J., Henry, M., Nicholson, G.C., Kotowicz, M.A., Berk, M., 2010. Prevalence of mood and anxiety disorder in self reported irritable bowel syndrome (IBS). An epidemiological population based study of women. BMC Gastroenterol. 10, 88. Nabavi B., Mitchell AJ, Nutt D., 2015. A lifetime prevalence of comorbidity between bipolar affective disorder and anxiety disorders: A meta-analysis of 52 interview-based studies of psychiatric population. EBioMedicine. 2 (10), 1405-1419. NIMH Repository and Genomics Resource. NIMH Center for Collaborative Genomics Research on Mental Disorders. https://www.nimhgenetics.org/available_data/bipolar_disorder/ (accessed 25 September 2017). Nurnberger, J.I. Jr., Blehar, M.C., Kaufmann, C.A., York-Cooler, C., Simpson, S.G., Harkavy-Friedman, J., Severe, J.B., Malaspina, D., Reich, T., 1994. Diagnostic interview for genetic studies: Rationale, unique features, and training. NIMH Genetics Initiative. Arch. Gen. Psychiatry. 51 (11), 849-859. Nylund, K.L., Asparouhov, T., Muthén, B.O., 2007. Deciding on the number of classes in latent class analysis and growth mixture modeling: A Monte Carlo simulation study. Structural Equation Modeling. 14 (4), 535-569. Pavlova, B., Perlis, R.H., Alda, M., Uher, R., 2015. Lifetime prevalence of anxiety disorders in people with bipolar disorder: a systematic review and meta-analysis. Lancet Psychiatry. 2 (8), 710-717. Perlis, R.H., Ostacher, M.J., Patel, J.K., Marangell, L.B., Zhang, H., Wisniewski, S.R., Ketter, T.A., Miklowitz, D.J., Otto, M.W., Gyulai, L., Reilly-Harrington, N.A., Nierenberg, A.A., Sachs, G.S., Thase, M.E., 2006. Predictors of recurrence in bipolar disorder: Primary outcomes from the Systematic Treatment Enhancement Program for Bipolar Disorder (STEP-BD). Am. J. Psychiatry. 163 (2), 217-224. Perugi, G., Quaranta, G., Belletti, S., Casalini, F., Mosti, N., Toni, C., Dell’Osso, L., 2015. General medical conditions in 347 bipolar disorder patients: Clinical correlates of metabolic and autoimmune-allergic diseases. J. Affect. Disord. 170, 95-103. Pini, S., Cassano, G.B., Simonini, E., Savino, M., Russo, A., Montgomery, S.A., 1997. Prevalence of anxiety disorders comorbidity in bipolar depression, unipolar depression and dysthymia. J. Affect. Disord. 42 (2-3), 145-153. Reilly, S., Planner, C., Hann, M., Reeves, D., Nazareth, I., Lester, H., 2012. The role of primary care in service provision for people with severe mental illness in the United Kingdom. PloS One. 7 (5), e36468. Rubin, D.B., 1987. Comment, Journal of the American Statistical Association. 82 (398), 543-546.
Schwarz, G., 1978. Estimating the dimension of a model. Ann. Statist. 6 (2), 461-464. Simms, L.J., Prisciandaro, J.J., Krueger, R.F., Goldberg, D.P., 2012. The structure of depression, anxiety and somatic symptoms in primary care. Psychol. Med. 42 (1), 15-28. Stubbs, B., Eggermont, L., Mitchell, A.J., De Hert, M., Correll, C.U., Soundy, A., Rosenbaum, S., Vancampfort, D., 2015. The prevalence of pain in bipolar disorder: A systematic review and large‐scale meta‐ analysis. Acta Psychiatr. Scand. 131 (2), 75-88. Young, L.T., Cooke, R.G., Robb, J.C., Levitt, A.J., Joffe, R.T., 1993. Anxious and non-anxious bipolar disorder. J. Affect. Disord. 29 (1), 49-52.
Figure 1 Latent class solution for reporting somatic symptoms in bipolar I disorder. The likelihood to report somatic symptoms indicated heterogeneity in the Bipolar I Disorder (BID) sample. The figure shows the probability (y-axis) of reporting somatic symptoms (x-axis) for the two latent classes. Latent class 1 (red line) (27% of the sample) had a high probability of reporting symptoms. Latent class 2 (blue line) (73% of the sample) had a low probability of reporting symptoms.
Table 1. Demographic characteristics of the Bipolar I Disorder (BID) sample Number of individuals, N (%) 526 (100) Gender, female, N (%) 300 (57) Ethnicity African American 23 (4.4) Asian American 3 (0.6) Caucasian 395 (75.1) Hispanic or Latino/a 2 (0.4) Native American 3 (0.6) Mixed ethnicity 33 (6.3) Unknown 68 (12.9) Age at interview, mean (S.D.) 42.1 (14.3) Age of onset to seek psychiatric help, mean (S.D.) 23.9 (11.4) Number of hospitalizations, mean (S.D.) 3.85 (6.28) Years of education, mean (S.D.) 13.9 (3.08) Number of suicide attempts, mean (S.D.) 0.92 (2.84) Anxiety disordersa All anxiety disordersb, N (%) 143 (27.3) Obsessive-compulsive disorder, N (%) 20 (0.04) Agoraphobia, N (%) 33 (0.06) Panic disorder, N (%) 68 (0.13) Simple phobia, N (%) 35 (0.07) Social phobia, N (%) 28 (0.05) Medical conditions Thyroid disease, N (%) 99 (0.19) Hyperthyroidism, N (%) 13 (0.02) Hypothyroidism, N (%) 62 (0.13) Migraine, N (%) 115 (0.22) Gastrointestinal disease, N (%) 112 (0.21) Vitamin deficiency, N (%) 21 (0.04) Meningitis, N (%) 5 (0.01) Movement disorder, N (%) 16 (0.03) Stroke, N (%) 5 (0.01) Seizure, N (%) 32 (0.06) Traumatic brain injury, N (%) 43 (0.08) Loss of consciousness, N (%) 29 (0.05) Other neurologic problems, N (%) 42 (0.08) a. According to DSM-IV. b. “All anxiety disorders” refers to any anxiety disorder diagnoses, including OCD, agoraphobia, panic disorder, simple phobia, social phobia, and unspecified anxiety states. Of note, some individuals had more than one anxiety disorder diagnosis. Table 2. Comparison model fit indices accounting for clustering in two-level model vs no clustering AIC
BIC
SABIC
Entropy
LMR adjusted LRT
BLRT
No cluster
8825.262
8927.629
8851.447
0.66
P<0.0001
P<0.0001
Cluster
8825.264
8927.631
8851.449
0.66
P=0.0001
P<0.001
AIC, Akaike Information Criterion; BIC, Bayesian Information Criterion; SABIC, sample size adjusted BIC; LMR, Lo-Mendel-Rubin, LRT, likelihood ratio test, BLRT, Bootstrapped Likelihood Ratio Test.
Table 3. Distribution of comorbid medical conditions Latent Class 1
Latent Class 2
Chi-square
P-value
Thyroid disorder 0.33 (SE = 0.05) 0.13 (SE = 0.03) 12.35** P<0.001 Hyperthyroidism 0.01 (SE = 0.01) 0.03 (SE = 0.01) 1.83 P=0.176 Hypothyroidism 0.24 (SE = 0.04) 0.07 (SE = 0.02) 12.43** P<0.001 Enlarged thyroid 0.03 (SE = 0.02) 0.02 (SE = 0.01) 0.116 P=0.73 Migraine 0.35 (SE = 0.05) 0.17 (SE = 0.02) 10.26** P=0.001 Ulcers or bowel disease 0.44 (SE = 0.05) 0.09 (SE = 0.04) 33.24** P<0.001 Peptic ulcer disease 0.26 (SE = 0.10) 0.46 (SE = 0.04) 2.57 P=0.109 Ulcerative Colitis 0.24 (SE = 0.12) 0.03 (SE = 0.03) 2.65 P=0.103 Vitamin deficiency 0.09 (SE = 0.03) 0.02 (SE = 0.01) 3.16 P=0.076 Learning disability 0.10 (SE = 0.03) 0.11 (SE = 0.02) 0.12 P=0.732 Meningitis or brain disorder 0.03 (SE = 0.02) 0.01 (SE = 0.004) 1.60 P=0.206 Stroke 0.04 (SE = 0.02) 0.01 (SE = 0.004) 3.14 P=0.076 Epilepsy, convulsions, or seizure 0.17 (SE = 0.01) 0.05 (SE = 0.01) 7.52** P=0.006 Traumatic brain injury 0.16 (SE = 0.05) 0.25 (SE = 0.05) 1.19 P=0.276 Lost consciousness 0.58 (SE = 0.11) 0.71 (SE = 0.06) 0.67 P=0.413 The probability of comorbid medical disease differed between classes. Individuals in Latent Class I had a higher probability of being affected by hypothyroidism, migraine, peptic ulcer or bowel disease, movement disorders, and other neurologic problems, compared with those in Latent Class 2. Insufficient data was available for the following medical conditions: Cushing’s disease, Lupus, Parkinson’s disease, Huntington’s disease. * p<0.05 **p<0.01 Table 4. Results of auxiliary variable analyses using the Lanza et al method (2013) A. Continuous variables (Mean) Class 1 (SE) Class 2 (SE) Chi-square P-value Age 45.6 (1.1) 40.5 (1.2) Age of onset to seek psychiatric help 21.7 (0.86) 24.7 (0.60) 7.82** 0.005 Years of schooling 13.7 (0.26) 13.9 (0.15) 0.35 0.553 Number of psych hospitalizations 4.7 (0.76) 3.5 (0.24) 2.40 0.121 Number of drug/alcohol admission 0.46 (0.14) 0.21 (0.04) 2.81 0.094 Age of first hospital admission 25.7 (0.96) 28.2 (0.65) 4.59* 0.032 Number of suicide attempts 1.46 (0.39) 0.72 (0.09) 0.062 3.48+ B. Categorical variables (Probability) Class 1 (SE) Class 2 (SE) Chi-square P-value Sex 0.72 (0.03) 0.52 (0.03) Ever hospitalized 0.80 (0.03) 0.84 (0.02) 1.08 0.297 Hospitalized for most severe mania 0.51 (0.05) 0.63 (0.03) 6.21* 0.013 Hospitalized for most severe depression 0.55 (0.05) 0.47 (0.03) 2.52 0.112 Ever suicide attempt 0.41 (0.04) 0.32 (0.03) 3.93* 0.047 Rapid Cycling 0.57 (0.04) 0.36 (0.02) 17.5* <0.001 Results for continuous (A) and categorical (B) variables. When accounting for comorbid anxiety disorders and commonly associated organic diseases, high and low somatization classes significantly differed in age of onset, age of first psychiatric hospitalization, probability of lifetime hospitalization for severe mania, probability ever attempting suicide, and probability of experiencing a rapidcycling disease course. + p<0.10 *p<0.05 **p<0.01
Highlights We investigated somatic symptoms in patients with bipolar disorder. Latent class analysis distinguished high and low somatization classes. We controlled for comorbid anxiety disorders and organic medical disease. High somatization was associated with age of onset, suicidality, and rapid-cycling.
Figure 1 Latent class solution for reporting somatic symptoms in bipolar I disorder. The likelihood to report somatic symptoms indicated heterogeneity in the Bipolar I Disorder (BID) sample. The figure shows the probability (y-axis) of reporting somatic symptoms (xaxis) for the two latent classes. Latent class 1 (red line) (27% of the sample) had a high probability of reporting symptoms. Latent class 2 (blue line) (73% of the sample) had a low probability of reporting symptoms.
PII: DOI: Reference:
S0165-0327(17)30750-4 https://doi.org/10.1016/j.jad.2017.11.083 JAD9397
To appear in: Journal of Affective Disorders Received date: 27 April 2017 Revised date: 3 October 2017 Accepted date: 13 November 2017 Cite this article as: Juliet Beni Edgcomb and Berit Kerner, Predictors and outcomes of somatization in bipolar I disorder: A latent class mixture modeling a p p r o a c h , Journal of Affective Disorders, https://doi.org/10.1016/j.jad.2017.11.083 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting galley proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Title: Predictors and outcomes of somatization in bipolar I disorder: A latent class mixture modeling approach Juliet Beni Edgcomb, M.D., Ph.D.1 Semel Institute for Neuroscience and Human Behavior, University of California, Los Angeles, CA, USA Berit Kerner, M.D. Semel Institute for Neuroscience and Human Behavior, University of California, Los Angeles, CA, USA Witten/Herdecke University, Witten, Germany Abstract Background: Mood disorders are often associated with somatic symptoms. The role of somatic symptoms on disease progression in unipolar depression is substantially better characterized than that role in bipolar disorder. Moreover, the contribution of comorbid anxiety disorders and medical illness is not well understood. Method: We investigated 527 patients with bipolar I disorder clustered within 102 families using a latent class approach. Predictors were added stepwise into the model. Anxiety and commonly associated medical illnesses were added as covariates. Results: The rate of somatic symptoms in this sample was 73% (mean 1.7 symptoms), and 27.3% had a comorbid anxiety disorder. A two-class model, with a subgroup at high-risk for somatization, gave the best fit to the data. Multilevel mixture modeling accounted for family clusters. Somatic symptoms were independently associated with disease severity, defined as earlier age of first seeking psychiatric help (x=21.7 vs x=24.7, p=0.005) and first psychiatric hospitalization (x=25.7 vs x=28.2, p=0.03), greater probability of attempting suicide (x=0.41 vs x=0.32, p=0.047), and rapid-cycling disease course (x=0.57 vs x=0.36, p<0.001). Persons with few or no somatic symptoms were more likely to be hospitalized for severe mania (x=0.63 vs x=0.51; p=0.013), but did not significantly differ in hospitalization for severe depression. Limitations: The study is correlational. Information on pharmacologic interventions and comorbid diseases was limited. Conclusions: Somatic symptoms in bipolar disorder could be an independent indicator for disease severity, suicidality, and rapid-cycling disease course. In severe mental illness, somatic and psychological symptoms must be jointly addressed. Keywords bipolar disorder, somatic, somatization, medical, physical symptoms, anxiety bipolar disorder, somatic, somatization, medical, physical symptoms, anxiety 1
Corresponding author: Juliet Beni Edgcomb, M.D., Ph.D. UCLA Semel Institute for Neuroscience and Human Behavior 760 Westwood Plaza Los Angeles, CA, 90024 [email protected] Telephone: (310)-794-2053 Fax: (310)-825-0340
Background Bipolar disorder is a mental illness defined by episodic mood shifts between mania and depression (American Psychiatric Association (APA), 2013). Comorbid medical conditions and physical symptoms influence the symptom profile and disease severity of bipolar disorder (Perugi et al., 2015; Forty et al., 2014). Individuals with bipolar disorder suffer from somatic symptoms at a rate nearly double that of the general population (OR 1.82), a rate like that observed in unipolar depression (Edgcomb et al., 2016). Moreover, other psychiatric disorders often co-occur with bipolar disorder and among these, anxiety disorders are highly prevalent. More than half of individuals with bipolar disorder (55.8%) have a comorbid anxiety diagnosis and nearly a third (31.8%) have two or more diagnoses related to anxiety (Boylan et al., 2004). Despite the high frequency of anxiety and somatic symptoms in persons with bipolar disorder, it is poorly understood how these constructs are related. Somatic symptoms are defined as physical dysfunctions (e.g. of appetite, digestion, sleep) or bodily sensations that are perceived as unpleasant or worrisome (Kapfhammer, 2006). Symptoms are not feigned or intentionally produced, and may or may not occur in the presence of organic disease. Somatic symptoms may be conceptualized on a spectrum, ranging from a single physical dysfunction or sensation, to Somatic Symptom Disorder, defined in the Diagnostic and Statistical Manual of Mental Disorders (5th ed.; DSM–5; American Psychiatric Association, 2013) as consisting of one or more somatic symptoms, excessive thoughts, feelings, and behaviors, and with symptomatology present for six months or more. Numerous studies have investigated the complex association of anxiety and physical symptoms in unipolar depressive disorders (Henningsen et al., 2003), but there is a paucity of data on the relationship between anxiety, somatization, and bipolar spectrum disorders. Prior studies on somatization have had relatively small sample sizes of bipolar patients (Mykletun et al., 2010), were limited to anxiety subscale scores rather than discrete anxiety diagnoses (Perlis et al., 2006) and have focused predominantly on softer domains of bipolarity (e.g. bipolar II, bipolar III, cyclothymic disorder) (Coplan et al., 2015). Large epidemiological studies have established that anxiety disorders occur more frequently in persons with bipolar disorder compared to the general population (Bauer et al., 2005; Pavlova et al., 2015). Over the last two decades, research has demonstrated high prevalence rates of discrete anxiety disorders amongst persons with bipolar disorder (Nabavi et al., 2015): 20.8-36.8% have panic disorder (Chen and Dilsaver, 1996; Pini et al., 1997), 20% have agoraphobia (Freeman et al., 2002), 10-35% have obsessive-compulsive disorder (Chen and Dilsaver, 1995), 47.2% have social phobia (Kessler et al., 2005), and 14-32% have generalized anxiety disorder (Young et al., 1993). Some studies have suggested that bipolar disorder may have stronger association with anxiety disorder diagnoses compared unipolar depression (Chen and Dilsaver, 1995). Genetic data is now emerging to add to the extensive epidemiological and clinical data demonstrating this association (Chang et al., 2013). Prior conceptual models (Freeman et al., 2002) have suggested that anxiety and bipolar disorders may co-occur at high frequency (1) due to the high prevalence of both disease conditions, (2) because anxiety and bipolar disorders are separate disorders with a pathophysiological association, and (3) because anxiety and bipolar disorders are manifestations of the same underlying abnormality. This paradigm may in turn be applied to comorbidity with somatic symptoms, albeit with a distinct pathophysiological association and/or underlying abnormality. Somatization refers to the conversion of a mental state into somatic symptoms. When controlling for the presence of comorbid medical illness, we hypothesize that the extent to which an individual experiences somatic symptoms may be used as a proxy measure for somatization. Although the relationship between anxiety disorders, somatic symptoms, and markers of disease severity is well-established in unipolar depression, no studies have thus far investigated the relationship between anxiety disorders and somatic symptoms in bipolar I disorder (BP1D). We hypothesize that:
1. There is a subset of individuals with BP1D who experience a high level of somatic symptoms. This group can be clearly differentiated from the subsets of individuals with BP1D who experience either no or a low level of somatic symptoms. 2. Individuals who exhibit a high level of somatic symptoms experience higher bipolar disease severity as measured by disease indicators. 3. The propensity to experience higher disease severity persists when accounting for the presence of comorbid anxiety disorders and commonly associated physical illnesses. Method Study design A stepwise approach was used to incorporate predictors, account for clustering, and investigate distal outcomes. First, we identified a sample of persons with bipolar I disorder. Second, we extracted dichotomous responses on the presence or absence of eight somatic symptoms. Third, we ran a latent class analysis using a stepwise approach. In brief, the model was run in six steps as follows: (1) somatization variables as indicators, (2) somatic symptom variables and demographic variables (age, gender) as indicators, (3) two-level hierarchical modeling to investigate the effect of familial clustering, (4) somatic symptom variables and demographics as indicators with continuous (DCON) and categorical (DCAT) predictors, (5) somatic symptom variables, demographics, and medical conditions as indicators with continuous (DCON) and categorical (DCAT) predictors, (6) somatic symptom variables, demographics medical conditions, and anxiety disorders as indicators with continuous (DCON) and categorical (DCAT) predictors. Sample Data were provided by the National Institute of Mental Health (NIMH) Repository and Genomic Resources Center for Collaborative Genomics Research on Mental Disorders (NIMH Repository and Genomics Resource). The sample consisted of 527 individuals from 102 families, age 16 and older. All individuals had been interviewed by trained health care providers and had been assessed with the Diagnostic Interview for Genetic Studies (DIGS, Version 3.0) (Nurnberger et al., 1994). Consensus clinical diagnoses according to DSM-IV (4th ed.; DSM–IV, American Psychiatric Association (APA), 2013) were reached by at least three independent mental health experts using Best Estimate procedures (Leckman et al., 1982). The interviewers also collected information on demographics, family history, somatic symptoms, the longitudinal disease course of psychiatric disturbances, and medical conditions. Variable selection In this analysis, responses to questions about the perception of pain (DIGS 3.0, I1660-I1760), were used as indicators in the latent class analysis (LCA) as a measure of somatic symptoms, : “Have you ever been bothered a lot by problems with pains in your...” (1) abdomen or stomach (other than during menstruation), (2) back, (3) joints, (4) arms or legs (other than in the joints), (5) chest, (6) painful sexual intercourse (other than after childbirth), (7) genitals or rectum (other than during intercourse), and (8) other pain. The presence or absence of medical conditions was coded (0 absent, 1 present) and included in the model as auxiliary variables. Diseases included (Variable identification codes included in parenthesis) were thyroid disorder (I124), overactive thyroid (I125), underactive thyroid (I126), enlarged thyroid (I127), Cushing’s disease (I136), migraines (I139), ulcers or other bowel diseases (I142), peptic ulcers (I145), Crohn’s disease (I148), ulcerative colitis (I151), lupus (I154), learning disability (I157), vitamin deficiency (I940), meningitis or other brain disorders (I160), Parkinson’s disease or movement disorder (I49), multiple sclerosis (I52), Huntington’s disease (I55), stroke (I58), epilepsy, convulsions, or seizure (I61), and traumatic brain injury (I68). The presence of comorbid anxiety disorders was determined from corresponding International Classification of Diseases (ICD-9) codes established by diagnostic criteria from at least three independent raters, as described above. Comorbid anxiety disorders investigated included: anxiety disorders (300.20, 300.3, 300.21, 300.22, 300.01, 300.29, 300.23), obsessive-compulsive disorder (300.3), agoraphobia with panic disorder (300.21),
agoraphobia without panic disorder (300.22), panic disorder without agoraphobia (300.01), simple phobia (300.29), and social phobia (300.23). The following continuous predictors were extracted from the Diagnostic Interview for Genetic Studies (DIGS Version 3.0: age at initial interview, age of onset to seek psychiatric help (I3130), years of schooling (I360), number of psychiatric hospitalizations (I3650), number of admissions for drug use or alcohol detoxification (I3760), age of first psychiatric hospitalization (I3680), and number of suicide attempts (I4680). Categorical predictors included: sex (I10), lifetime history of psychiatric hospitalization (I3650), hospitalization for most severe mania (I5810), hospitalization for most severe depression (I4470), lifetime history of suicide attempt (I4670), rapid-cycling disease course (I6330), and lifetime use of cocaine (I8890), stimulants (I8900), sedatives (I8910), opiates (I8920), phencyclidine (I8930), hallucinogens (I8940), solvents (I8950), and other substances (I8960). Latent class mixture modeling We performed the Latent Class Analysis (LCA) using Mplus Version 7 (Muthén and Muthén, 1998-2010). We estimated the Latent Class (LC) membership for each individual using the estimation maximization (EM) algorithm based on the probability of endorsing a profile of variables (Muthén and Shedden, 1999). The starting values were from 200 random sets, so as to avoid local maxima. We selected the optimal number of classes when the Bayesian Information Criterion (BIC) reached a minimum (Nylund et al., 2007). In addition, the LC solution was checked by using the Akaike Information Criterion (AIC) (Akaike, 2011), the BIC and sample size adjusted BIC (Schwarz, 1978), the Lo-Mendel-Rubin (LMR) test (Lo et al., 2001), and the Bootstrapped Likelihood Ratio Test (BLRT) (McLachlan, 1987). We deleted the observations with missing data on covariates in the analyses that included covariates. In this mixture modeling framework, U stands for the binary, categorical, or count observed indicator variables and C refers to a latent categorical variable with K classes, C = (c ,c ,…,c ), where c =1, if individual i belongs to class k and zero if otherwise. Given c , conditional independence is assumed for U, i
i1
i2
iK
ik
i
P(u ,u ,…,u |c ) = P(u | c )P(u | c )P(u | c ) i1
i2
iτ
i
i1
i
i2
i
iτ
i
The data sample consisted of individuals nested within families. A two-level mixture model with maximum likelihood estimation was run with somatic symptom and demographic variables (age, gender) as indicators. The ‘cluster’ option was then specified in Mplus (Muthén and Muthén, 1998-2010). Family clusters were identified by a randomly assigned numerical code. The multilevel extension of the modeling framework allowed for random slopes and random intercepts that varied across clusters in hierarchical data (Heck and Thomas, 2015). The random effects across-cluster variation in slopes and intercepts was combined with individual-level indicators (somatic symptoms, demographics) to determine model fit when accounting for familial clustering. Model fit statistics for a two-level model accounting for familial clustering was compared with a two-level model that treated each participant as unrelated (i.e. independent). Model fit statistics suggested an insignificant effect of familial clustering. Using the pseudo-class method (Clark and Muthén, 2009; Asparouhov, 2010) we included auxiliary variables in the model to evaluate additional potential class predictors. Lanza et al’s method (2013) is based on draws from the posterior probabilities, from which we tested the equality of means across the latent classes. We interpreted the magnitude of the mean differences between the classes as an indicator of the strength of the prediction that the auxiliary variables influenced the class membership (Asparouhov and Muthén, 2007; Asparouhov and Muthén, 2014). We used multiple imputation to derive the latent class membership variables from the posterior distribution obtained by the LCA model estimation, after estimating the latent class model without including covariates. We then analyzed the imputed class variables together with the auxiliary variables, using the multiple imputation technique developed by Rubin (1987). The Lanza et al (2013) method was used as, unlike other mixture model auxiliary options, both categorical and continuous outcomes could be incorporated simultaneously in the model. For each auxiliary variable, we evaluated conditional class specific means, based on the estimated latent class model (option DCAT and DCON in Mplus).
Results The sample consisted of 57% female and 43% male participants who belonged to 102 families (Table 1). 75.1% of respondents were Caucasian. The sample’s mean age was 42.1 years (SD 14.3), while the average age of onset of bipolar disorder was 23.9 years (SD 11.4). Thus, a large number of individuals had been symptomatic for at least 18 years, had been hospitalized for bipolar disorder several times, and, in many cases, had attempted suicide, all of which indicated a severe form of disease. Comorbid anxiety disorders were prevalent (27%). Panic disorder was the most common comorbid anxiety disorder (13%). Medical disorders were frequently reported, the three most prevalent being migraine (22%), gastrointestinal disease (21%) and thyroid disorders (19%). All individuals in this sample had answered questions about frequency and perception of pain as indicators of somatic symptoms and the responses to these questions were used as indicators for the latent class approach. To explore the structure of this sample, we utilized a latent class mixture modeling approach. As our sample consisted of individuals grouped in families, we tested if it would be more appropriate to use a clustering approach within a two-level analysis. The comparison of the latent class models using a two-level clustering approach and a simple one-level analysis indicated that there was no difference in the model fit between the two approaches. To reduce the computational burden, we therefore ran our models without the two-level approach. The model fit indices suggested a division of the sample of patients with bipolar I disorder into two subclasses (Fig. 1, Table 3 and 4), based on the presence of comorbid somatic symptoms and their correlations. The first Latent Class (LCA1) is characterized by high probability of somatic symptoms (20-79%, Table 4). This class accounted for approximately one-quarter (27%) of the sample, and was 72% female. The second Latent Class (LCA2) is characterized by low probability of somatic symptoms (2-29%). This class represented approximately three-quarters (73%) of the sample and 52% of patients in this latent class were female. Compared with LCA2 (low somatization), individuals in LCA1 (high somatization) had a significantly greater probability of being diagnosed with thyroid disorder (33% vs 13%), migraine (35% vs 17%), peptic ulcers or bowel disease (44% vs 9%), and epilepsy, convulsions, or seizure (17% vs 5%) (Table 3). As these disorders disproportionately affected the high somatization group, we then added medical conditions as covariates into the model. Using the Lanza et al method (2013) to determine conditional class specific categorical variable probabilities and continuous variable means, severity predictors were compared between classes. When accounting for the aforementioned comorbid medical conditions, LCA1 (high somatization) was characterized by an older age to first seek psychiatric help (25.2 vs 23.1, p=0.03), fewer years of schooling (13.4 vs 14.2, p=0.003), higher number of psychiatric hospitalizations (5.17 vs 3.12, p=0.003), higher age of first hospitalization (30.9 vs 25.7, p<0.001), higher number of suicide attempts (1.64 vs 0.58, p=0.003), greater likelihood of attempting suicide (45% vs 29%, p =0.01), and greater probability of a rapid-cycling disease course (51% vs 36%). As expected, LCA1 was also characterized by higher probability of comorbid anxiety disorders (42% vs 18%, p<0.001), including obsessive-compulsive disorder (7% vs 1%, p=0.03), agoraphobia (11% vs 3%, p=0.02), panic disorder (19% vs 9%, p=0.02), simple phobia (12% vs 3%, p=0.009), and social phobia (8% vs 3%, p=0.05). In addition, individuals in LCA1 were more likely to have used tobacco (71% vs 59%, p=0.03), stimulants (34% vs 22%, p=0.04), opiates (23% vs 12%, p=0.03) and phencyclidine (19% vs 8%, p=0.02). There was no significant difference between classes for rates of cocaine, hallucinogen, solvents, or other drug use. We then repeated the analysis adding common medical conditions and anxiety disorders as predictors into the model. When accounting for the presence of comorbid medical conditions and anxiety disorders, compared with LCA2, individuals in LCA1 had a earlier age to first seek psychiatric help (21.7 vs 24.7, p=0.005) and earlier age of first psychiatric hospitalization (25.7 vs 28.2, p=0.03). Moreover, those in LCA1 tended to have more
suicide attempts compared with those in LCA2 (1.46 vs 0.72, p=0.06). Age of interview, years of schooling, total number of hospitalizations, and number of drug/alcohol admissions did not differ between classes. Individuals in LCA1 were less likely to be hospitalized for severe mania (51% vs 63%, p=0.01), more likely to have attempted suicide (41% vs 32%, p=0.04), and more likely to have a rapid cycling course (57% vs 36%, p<0.001). Discussion In this study, our goal has been to determine if persons with bipolar disorder and comorbid somatic symptoms experience increased markers of disease severity and poorer outcomes compared with those who report low or no somatic symptoms. In order to establish this, we have used a latent class mixture modeling approach to determine if a population of persons with bipolar disorder could be meaningfully subdivided into groups with differing degrees of somatization. Our results demonstrate a two-class model, wherein a group of individuals is prone to high levels of somatization and can be distinguished from those with low or no somatization. In our study, approximately one-quarter of the study sample was identified as high somatizers and these individuals experienced greater bipolar disease severity as measured by various indicators. Our findings are consistent with prior meta-analytic studies, which have found a rate of somatic symptoms amongst persons with bipolar disorder of close to 50% (Edgcomb et al., 2016; Stubbs et al., 2015). However, much of the prior research has been sparse and highly heterogeneous in methodology. Because of this, it has been difficult to meaningfully conceptualize groups within the bipolar population with different probabilities of somatization. We note that although our sample consisted of individuals nested within family clusters, when we accounted for clustering using two-level hierarchical modeling, there was no significant difference in model fit indices. As expected, the high somatization class had a greater probability of reporting comorbid medical illness, including thyroid disorders, migraine, peptic ulcers or bowel disease, and neurologic disorders. The high somatization class was also at increased risk of comorbid anxiety disorders, including obsessive-compulsive disorder, agoraphobia, panic disorder, simple phobia, and social phobia. When accounting for the presence of these anxiety disorders in addition to the comorbid medical illness, those who are classified as high somatizers were more likely to be female, were older, had a younger age of first seeking psychiatric help and first psychiatric hospitalization, were more likely to have lifetime suicide attempt, tended to have more suicide attempts, were less likely to be hospitalized for severe mania, and were more likely to have a rapid-cycling disease course. Age of interview, years of schooling, total number of hospitalizations, and number of drug/alcohol admissions did not differ between classes. Prior research has demonstrated that persons with severe mental illness, including bipolar disorder, experience higher levels of bodily pain, poorer physical functioning, and lower self-rated general health compared with the general population, and this in turn contributes to treatment nonadherence and poorer quality of life. To our knowledge, our study is the first to identify a subgroup of persons with bipolar disorder who have a high probability of somatization, and describe this group as high risk for a variety of markers of disease severity when accounting for illnesses that may otherwise explain this relationship (i.e. comorbid anxiety disorders and medical illnesses). Our finding that the high somatization group is characterized by younger age of onset of psychiatric illness and first psychiatric hospitalization is in keeping with prior research suggesting that age may distinguish subtypes of bipolar disorder (Leboyer et al., 2005). Moreover, the high somatization group was characterized by a longer total disease period, in that these individuals were first hospitalized at a younger age and were older at time of interview. Numerous studies have established the relationship between pain-related diagnoses and suicide risk (Hooley, 2014; Leverich, 2003), with the greatest risk for suicide found amongst individuals with psychogenic pain (Ilgen et al., 2008). This is consistent with our finding that the high somatization group was characterized by a higher probability of ever attempting suicide and having more individual attempts.
In our study, the high somatization group additionally had a higher probability of experiencing a rapid-cycling disease course of bipolar disorder. Prior research has demonstrated that persons with rapid-cycling bipolar disorder have a greater number of medical comorbidities (Kemp et al., 2010) and are more likely to have migraine headaches (Gordon-Smith, 2015) compared with the general population. Moreover, persons with a rapid-cycling disease course experience more functional impairment, likely due to an increased number of mood episodes and shorter periods of euthymia (McIntyre et al., 2006). A growing body of research has suggested that the cumulative neurobiological insults of multiple acute mood episodes, including increased oxidative stress, increases the presence of pro-inflammatory markers and deficits in neuroprotection (Berk et al, 2011). Our finding of an association of somatization with rapid-cycling disease course may suggest the role of incorporating physical symptoms in neuroprogression and clinical staging models of bipolar disorder. Whereas the likelihood of hospitalization for most severe depression did not differ between the high and low somatizing groups, those who had a high level of somatization were less likely to be hospitalized for severe mania. These findings suggest that the pattern of mood episodes may vary between individuals who experience a greater degree of somatic symptoms. Limitations As our measures of somatic symptoms utilized self-report, under or over reporting of symptoms could have occurred. By using probability-based multivariate statistical methods, the degree of uncertainty introduced by individual classification was minimized. In addition, the latent class mixture model was run recursively until a convergent solution was determined. Detailed information on pharmacologic and other therapeutic interventions was not available. Moreover, our operationalized definition of somatization as quantity of somatic symptoms, in the context of controlling for comorbid medical illnesses, is circumscribed by the limited available information on comorbid disease conditions. Our study is correlational, and any conclusions about cause and effect must be limited. Of note, although we suggest that greater somatic symptom burden is associated with markers of bipolar disease severity, this relationship is likely not unidirectional. More severe bipolar illness may be characterized by greater somatic symptom burden through a variety of mechanisms, including vulnerabilities acquired via neuroprogression. In order to maximize internal validity, multiple sources of information, including medical records, self-rated scales, and interview-based scales were referenced; however, retrospective recall of the experience of somatic symptoms, hospitalizations, and disease onset, may have introduced some uncertainty. These limitations are common to secondary analyses of existing data, and are mitigated, in part, by the use of a relatively large sample size. Future research Future research should investigate a broader range of somatic symptoms (e.g. functional syndromes, neurologic symptoms, other pain syndromes) and bipolar spectrum disorders (i.e. bipolar II, cyclothymia) beyond that which was available for our analyses. In our study, we controlled for the presence of commonly associated disease conditions; however, our list was limited by the data available and we could not account for all significant disease conditions. In a more precise characterization, it would be important to establish which somatic symptoms were, in fact, medically explained, and which remained medically unexplained. Although the DSM-V removes this distinction in a clinical context, it nevertheless remains important to developing an understanding of somatic symptoms and the implications of these symptoms on treatment. Our results suggest the importance of more precisely characterizing patterns of mood symptoms in bipolar disorder as they relate to somatic symptoms, especially the role of rapid-cycling bipolar disorder. Conclusions Nearly one-third of persons with severe and enduring mental illness are managed entirely by primary care (Rielly et al., 2012). In these circumstances, patients often present with a combination of mood, anxiety, and somatic symptoms (Simms et al., 2012). Significant comorbidity among somatic symptoms and mood disorders has resulted in a demand for conceptual models to explain this association. Our study puts forward a model that defines a group of individuals with bipolar I disorder who are at increased risk for somatic symptoms and, in turn, other markers of disease pathology, beyond that which would be explained solely by the presence of
comorbid medical illnesses and anxiety disorders. These results may inform models of clinical practice and emphasize the importance of somatic symptoms in severe mental illness. Acknowledgements We would like to thank the UCLA Statistical Consulting Group for their help throughout the course of this work. We are indebted to the investigators of the National Institute of Mental Health (NIMH) Bipolar Genetics Initiative. Funding sources The sample was provided by the National Institute of Mental Health (NIMH) Repository and Genomic Resources Center for Collaborative Genomics Research on Mental Disorders. This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Contributors Juliet Edgcomb and Berit Kerner designed the study, conducted the analyses, and wrote the first draft of the manuscript. All authors have contributed to and have approved the final manuscript.
Declaration of interest Conflicts of interest: none
References Akaike, H., 2011. Akaike’s information criterion, in: Lovric, M. (Ed.), International Encyclopedia of Statistical Science. Springer, Berlin Heidelberg, pp. 25-25 American Psychiatric Association (APA), 2013. Diagnostic and Statistical Manual of Mental Disorders, third (DSM-III), fourth (DSM-IV, DSM-IV-TR), and fifth edition (DSM-5). American Psychiatric Press, Arlington, VA. Asparouhouv, T., 2007. Wald test of mean equality for potential latent class predictors in mixture modeling. Technical appendix. Los Angeles: Muthén and Muthén, Los Angeles https://www.statmodel.com/download/MeanTest1.pdf (accessed 25 September 2017) Asparouhov, T., 2010. Bayesian analysis using Mplus. Technical appendix. Los Angeles: Muthén and Muthén, Los Angeles http://statmodel2.com/download/Bayes3.pdf (accessed 25 September 2017) Asparouhov, T., Muthén, B., 2014. Auxiliary variables in mixture modeling: Using the BCH method in Mplus to estimate a distal outcome model and an arbitrary secondary model. Mplus Web Notes, 21(2). https://www.statmodel.com/MixtureModeling.shtml (accessed 25 September 2017) Bauer, M.S., Altshuler, L., Evans, D.R., Beresford, T., Williford, W.O., Hauger, R., VA Cooperative Study#430 Team, 2005. Prevalence and distinct correlates of anxiety, substance, and combined comorbidity in a multi-site public sector sample with bipolar disorder. J. Affect. Disord. 85 (3), 301-315. Berk, M., Kapczinski, A., Andreazza, A.C., Dean, O.M.., Giorlando, F., Maes, M., Yücel, M., Gama, C.S., Dodd, S., Dean, B., Magalhäes, P.V., Amminger, P., McGorry, P., Malhi, G.S., 2011. Pathways underlying neuroprogression in bipolar disorder: Focus on inflammation, oxidative stress and neurotrophic factors. Neurosci. Biobehav. Rev. 35 (3), 804-817. Boylan, K.R., Bieling, P.J., Marriott, M., Begin, H., Young, L.T., MacQueen, G.M., 2004. Impact of comorbid anxiety disorders on outcome in a cohort of patients with bipolar disorder. J. Clin. Psychiatry. 65 (8), 11061113. Chang, Y.H., Lee, S.Y., Chen, S.L., Tzeng, N.S., Wang, T.Y., Lee, I.H., Chen, P.S., Huang, S.Y., Yang, Y.K., Ko, H.C., Lu, R B. 2013. Genetic variants of the BDNF and DRD3 genes in bipolar disorder comorbid with anxiety disorder. J. Affect. Disord. 151 (3), 967-972. Chen, Y. W., Dilsaver, S. C. 1995. Comorbidity for obsessive-compulsive disorder in bipolar and unipolar disorders. Psychiatry Res. 59 (1-2), 57-64. Chen, Y.W., Dilsaver, S.C., 1996. Lifetime rates of suicide attempts among subjects with bipolar and unipolar disorders relative to subjects with other Axis I disorders. Biol. Psychiatry. 39 (10), 896-899. Clark, S.L., Muthén, B., 2009. Relating latent class analysis results to variables not included in the analysis. https://www.statmodel.com/download/relatinglca.pdf (accessed 25 September 2017) Coplan, J., Singh, D., Gopinath, S., Mathew, S.J., Bulbena, A., 2015. A novel anxiety and affective spectrum disorder of mind and body—the ALPIM (Anxiety-Laxity-Pain-Immune-Mood) syndrome: A preliminary report. J. Neuropsychiatry Clin. Neurosci. 27 (2), 93-103.
Edgcomb, J.B., Tseng, C.H., Kerner, B., 2016. Medically unexplained somatic symptoms and bipolar spectrum disorders: A systematic review and meta-analysis. J. Affect. Disord. 204, 205-213. Forty, L., Ulanova, A., Jones, L., Jones, I., Gordon-Smith, K., Fraser, C., Farmer, A., McGuffin, P., Lewis, C.M., Hosang, G.M., Rivera, M., Craddock, N., 2014. Comorbid medical illness in bipolar disorder. Br. J. Psychiatry. 205 (6), 465-472. Freeman, M.P., Freeman, S.A., McElroy, S.L., 2002. The comorbidity of bipolar and anxiety disorders: prevalence, psychobiology, and treatment issues. J. Affect. Disord. 68 (1), 1-23. Gordon-Smith, K., Forty, L., Chan, C., Knott, S., Jones, I., Craddock, N., Jones, L.A., 2015. Rapid cycling as a feature of bipolar disorder and comorbid migraine. J. Affect. Disord. 175, 320-324. Heck, R.H., Thomas, S.L., 2015. An introduction to multilevel modeling techniques: MLM and SEM approaches using Mplus. Quantitative Methodology Series, third ed. Routledge, New York. Henningsen, P., Zimmermann, T., Sattel, H., 2003. Medically unexplained physical symptoms, anxiety, and depression: A meta‐analytic review. Psychosom. Med. 65 (4), 528-533. Hooley, J.M., Franklin, J.C., Nock, M.K., 2014. Chronic pain and suicide: Understanding the association. Curr. Pain Headache Rep. 18 (8), 435. Ilgen, M.A., Zivin, K., McCammon, R.J., Valenstein, M., 2008. Pain and suicidal thoughts, plans and attempts in the United States. Gen. Hosp. Psychiatry. 30 (6), 521-527. Kapfhammer, H., 2006. Somatic symptoms of depression. Dialogues Clin. Neurosc. 8 (2), 227-239. Kemp, D.E., Gao, K., Chan, P.K., Ganocy, S.J., Findling, R.L., Calabrese, J.R., 2010. Medical comorbidity in bipolar disorder: relationship between illnesses of the endocrine/metabolic system and treatment outcome. Bipolar Disord. 12 (4), 404-413. Kessler, R.C., Chiu, W.T., Demler, O., Merikangas, K.R., Walters, E.E., 2005. Prevalence, severity, and comorbidity of 12-month DSM-IV disorders in the National Comorbidity Survey Replication. Arch. Gen. Psychiatry. 62 (6), 617-627. Lanza, S.T., Tan, X., Bray, B.C., 2013. Latent class analysis with distal outcomes: A flexible model-based approach. Struct. Equ Modeling. 20 (1), 1-26. Leboyer, M., Henry, C., Paillere‐Martinot, M.L., Bellivier, F., 2005. Age at onset in bipolar affective disorders: A review. Bipolar Disord. 7 (2), 111-118. Leckman, J.F., Sholomskas, D., Thompson, D., Belanger, A., Weissman, M.M., 1982. Best estimate of lifetime psychiatric diagnosis: a methodological study. Arch. Gen. Psychiatry. 39 (8), 879-883. Leverich, G.S., Altshuler, L.L., Frye, M.A., Suppes, T., Keck P.E. Jr., McElroy, S.L., Denicoff, K.D., Obrocea, G., Nolen, W.A., Kupka, R., Walden, J., Gunze, H., Perez, S., Luckenbaugh, D.A., Post, R.M., 2003. Factors associated with suicide attempts in 648 patients with bipolar disorder in the Stanley Foundation Bipolar Network. J. Clin. Psychiatry. 64 (5), 506-515. Lo, Y., Mendell, N.R., Rubin, D.B., 2001. Testing the number of components in a normal mixture. Biometrika. 88 (3), 767-778.
McIntyre, R.S., Soczynska, J.K., Bottas, A., Bordbar, K., Konarski, J.Z., Kennedy, S.H., 2006. Anxiety disorders and bipolar disorder: A review. Bipolar Disord. 8 (6), 665-676. McLachlan, G.J., 1987. On bootstrapping the likelihood ratio test statistic for the number of components in a normal mixture. Applied Statistics. 36, 318-324. Muthén, B., Shedden, K., 1999. Finite mixture modeling with mixture outcomes using the EM algorithm. Biometrics. 55 (2), 463-469. Muthén, L.K., Muthén, B.O., 1998-2010. Mplus User's Guide, sixth ed. Muthen and Muthen, Los Angeles. https://www.statmodel.com/ (accessed 25 September 2017). Mykletun, A., Jacka, F., Williams, L., Pasco, J., Henry, M., Nicholson, G.C., Kotowicz, M.A., Berk, M., 2010. Prevalence of mood and anxiety disorder in self reported irritable bowel syndrome (IBS). An epidemiological population based study of women. BMC Gastroenterol. 10, 88. Nabavi B., Mitchell AJ, Nutt D., 2015. A lifetime prevalence of comorbidity between bipolar affective disorder and anxiety disorders: A meta-analysis of 52 interview-based studies of psychiatric population. EBioMedicine. 2 (10), 1405-1419. NIMH Repository and Genomics Resource. NIMH Center for Collaborative Genomics Research on Mental Disorders. https://www.nimhgenetics.org/available_data/bipolar_disorder/ (accessed 25 September 2017). Nurnberger, J.I. Jr., Blehar, M.C., Kaufmann, C.A., York-Cooler, C., Simpson, S.G., Harkavy-Friedman, J., Severe, J.B., Malaspina, D., Reich, T., 1994. Diagnostic interview for genetic studies: Rationale, unique features, and training. NIMH Genetics Initiative. Arch. Gen. Psychiatry. 51 (11), 849-859. Nylund, K.L., Asparouhov, T., Muthén, B.O., 2007. Deciding on the number of classes in latent class analysis and growth mixture modeling: A Monte Carlo simulation study. Structural Equation Modeling. 14 (4), 535-569. Pavlova, B., Perlis, R.H., Alda, M., Uher, R., 2015. Lifetime prevalence of anxiety disorders in people with bipolar disorder: a systematic review and meta-analysis. Lancet Psychiatry. 2 (8), 710-717. Perlis, R.H., Ostacher, M.J., Patel, J.K., Marangell, L.B., Zhang, H., Wisniewski, S.R., Ketter, T.A., Miklowitz, D.J., Otto, M.W., Gyulai, L., Reilly-Harrington, N.A., Nierenberg, A.A., Sachs, G.S., Thase, M.E., 2006. Predictors of recurrence in bipolar disorder: Primary outcomes from the Systematic Treatment Enhancement Program for Bipolar Disorder (STEP-BD). Am. J. Psychiatry. 163 (2), 217-224. Perugi, G., Quaranta, G., Belletti, S., Casalini, F., Mosti, N., Toni, C., Dell’Osso, L., 2015. General medical conditions in 347 bipolar disorder patients: Clinical correlates of metabolic and autoimmune-allergic diseases. J. Affect. Disord. 170, 95-103. Pini, S., Cassano, G.B., Simonini, E., Savino, M., Russo, A., Montgomery, S.A., 1997. Prevalence of anxiety disorders comorbidity in bipolar depression, unipolar depression and dysthymia. J. Affect. Disord. 42 (2-3), 145-153. Reilly, S., Planner, C., Hann, M., Reeves, D., Nazareth, I., Lester, H., 2012. The role of primary care in service provision for people with severe mental illness in the United Kingdom. PloS One. 7 (5), e36468. Rubin, D.B., 1987. Comment, Journal of the American Statistical Association. 82 (398), 543-546.
Schwarz, G., 1978. Estimating the dimension of a model. Ann. Statist. 6 (2), 461-464. Simms, L.J., Prisciandaro, J.J., Krueger, R.F., Goldberg, D.P., 2012. The structure of depression, anxiety and somatic symptoms in primary care. Psychol. Med. 42 (1), 15-28. Stubbs, B., Eggermont, L., Mitchell, A.J., De Hert, M., Correll, C.U., Soundy, A., Rosenbaum, S., Vancampfort, D., 2015. The prevalence of pain in bipolar disorder: A systematic review and large‐scale meta‐ analysis. Acta Psychiatr. Scand. 131 (2), 75-88. Young, L.T., Cooke, R.G., Robb, J.C., Levitt, A.J., Joffe, R.T., 1993. Anxious and non-anxious bipolar disorder. J. Affect. Disord. 29 (1), 49-52.
Figure 1 Latent class solution for reporting somatic symptoms in bipolar I disorder. The likelihood to report somatic symptoms indicated heterogeneity in the Bipolar I Disorder (BID) sample. The figure shows the probability (y-axis) of reporting somatic symptoms (x-axis) for the two latent classes. Latent class 1 (red line) (27% of the sample) had a high probability of reporting symptoms. Latent class 2 (blue line) (73% of the sample) had a low probability of reporting symptoms.
Table 1. Demographic characteristics of the Bipolar I Disorder (BID) sample Number of individuals, N (%) 526 (100) Gender, female, N (%) 300 (57) Ethnicity African American 23 (4.4) Asian American 3 (0.6) Caucasian 395 (75.1) Hispanic or Latino/a 2 (0.4) Native American 3 (0.6) Mixed ethnicity 33 (6.3) Unknown 68 (12.9) Age at interview, mean (S.D.) 42.1 (14.3) Age of onset to seek psychiatric help, mean (S.D.) 23.9 (11.4) Number of hospitalizations, mean (S.D.) 3.85 (6.28) Years of education, mean (S.D.) 13.9 (3.08) Number of suicide attempts, mean (S.D.) 0.92 (2.84) Anxiety disordersa All anxiety disordersb, N (%) 143 (27.3) Obsessive-compulsive disorder, N (%) 20 (0.04) Agoraphobia, N (%) 33 (0.06) Panic disorder, N (%) 68 (0.13) Simple phobia, N (%) 35 (0.07) Social phobia, N (%) 28 (0.05) Medical conditions Thyroid disease, N (%) 99 (0.19) Hyperthyroidism, N (%) 13 (0.02) Hypothyroidism, N (%) 62 (0.13) Migraine, N (%) 115 (0.22) Gastrointestinal disease, N (%) 112 (0.21) Vitamin deficiency, N (%) 21 (0.04) Meningitis, N (%) 5 (0.01) Movement disorder, N (%) 16 (0.03) Stroke, N (%) 5 (0.01) Seizure, N (%) 32 (0.06) Traumatic brain injury, N (%) 43 (0.08) Loss of consciousness, N (%) 29 (0.05) Other neurologic problems, N (%) 42 (0.08) a. According to DSM-IV. b. “All anxiety disorders” refers to any anxiety disorder diagnoses, including OCD, agoraphobia, panic disorder, simple phobia, social phobia, and unspecified anxiety states. Of note, some individuals had more than one anxiety disorder diagnosis. Table 2. Comparison model fit indices accounting for clustering in two-level model vs no clustering AIC
BIC
SABIC
Entropy
LMR adjusted LRT
BLRT
No cluster
8825.262
8927.629
8851.447
0.66
P<0.0001
P<0.0001
Cluster
8825.264
8927.631
8851.449
0.66
P=0.0001
P<0.001
AIC, Akaike Information Criterion; BIC, Bayesian Information Criterion; SABIC, sample size adjusted BIC; LMR, Lo-Mendel-Rubin, LRT, likelihood ratio test, BLRT, Bootstrapped Likelihood Ratio Test.
Table 3. Distribution of comorbid medical conditions Latent Class 1
Latent Class 2
Chi-square
P-value
Thyroid disorder 0.33 (SE = 0.05) 0.13 (SE = 0.03) 12.35** P<0.001 Hyperthyroidism 0.01 (SE = 0.01) 0.03 (SE = 0.01) 1.83 P=0.176 Hypothyroidism 0.24 (SE = 0.04) 0.07 (SE = 0.02) 12.43** P<0.001 Enlarged thyroid 0.03 (SE = 0.02) 0.02 (SE = 0.01) 0.116 P=0.73 Migraine 0.35 (SE = 0.05) 0.17 (SE = 0.02) 10.26** P=0.001 Ulcers or bowel disease 0.44 (SE = 0.05) 0.09 (SE = 0.04) 33.24** P<0.001 Peptic ulcer disease 0.26 (SE = 0.10) 0.46 (SE = 0.04) 2.57 P=0.109 Ulcerative Colitis 0.24 (SE = 0.12) 0.03 (SE = 0.03) 2.65 P=0.103 Vitamin deficiency 0.09 (SE = 0.03) 0.02 (SE = 0.01) 3.16 P=0.076 Learning disability 0.10 (SE = 0.03) 0.11 (SE = 0.02) 0.12 P=0.732 Meningitis or brain disorder 0.03 (SE = 0.02) 0.01 (SE = 0.004) 1.60 P=0.206 Stroke 0.04 (SE = 0.02) 0.01 (SE = 0.004) 3.14 P=0.076 Epilepsy, convulsions, or seizure 0.17 (SE = 0.01) 0.05 (SE = 0.01) 7.52** P=0.006 Traumatic brain injury 0.16 (SE = 0.05) 0.25 (SE = 0.05) 1.19 P=0.276 Lost consciousness 0.58 (SE = 0.11) 0.71 (SE = 0.06) 0.67 P=0.413 The probability of comorbid medical disease differed between classes. Individuals in Latent Class I had a higher probability of being affected by hypothyroidism, migraine, peptic ulcer or bowel disease, movement disorders, and other neurologic problems, compared with those in Latent Class 2. Insufficient data was available for the following medical conditions: Cushing’s disease, Lupus, Parkinson’s disease, Huntington’s disease. * p<0.05 **p<0.01 Table 4. Results of auxiliary variable analyses using the Lanza et al method (2013) A. Continuous variables (Mean) Class 1 (SE) Class 2 (SE) Chi-square P-value Age 45.6 (1.1) 40.5 (1.2) Age of onset to seek psychiatric help 21.7 (0.86) 24.7 (0.60) 7.82** 0.005 Years of schooling 13.7 (0.26) 13.9 (0.15) 0.35 0.553 Number of psych hospitalizations 4.7 (0.76) 3.5 (0.24) 2.40 0.121 Number of drug/alcohol admission 0.46 (0.14) 0.21 (0.04) 2.81 0.094 Age of first hospital admission 25.7 (0.96) 28.2 (0.65) 4.59* 0.032 Number of suicide attempts 1.46 (0.39) 0.72 (0.09) 0.062 3.48+ B. Categorical variables (Probability) Class 1 (SE) Class 2 (SE) Chi-square P-value Sex 0.72 (0.03) 0.52 (0.03) Ever hospitalized 0.80 (0.03) 0.84 (0.02) 1.08 0.297 Hospitalized for most severe mania 0.51 (0.05) 0.63 (0.03) 6.21* 0.013 Hospitalized for most severe depression 0.55 (0.05) 0.47 (0.03) 2.52 0.112 Ever suicide attempt 0.41 (0.04) 0.32 (0.03) 3.93* 0.047 Rapid Cycling 0.57 (0.04) 0.36 (0.02) 17.5* <0.001 Results for continuous (A) and categorical (B) variables. When accounting for comorbid anxiety disorders and commonly associated organic diseases, high and low somatization classes significantly differed in age of onset, age of first psychiatric hospitalization, probability of lifetime hospitalization for severe mania, probability ever attempting suicide, and probability of experiencing a rapidcycling disease course. + p<0.10 *p<0.05 **p<0.01
Highlights We investigated somatic symptoms in patients with bipolar disorder. Latent class analysis distinguished high and low somatization classes. We controlled for comorbid anxiety disorders and organic medical disease. High somatization was associated with age of onset, suicidality, and rapid-cycling.
Figure 1 Latent class solution for reporting somatic symptoms in bipolar I disorder. The likelihood to report somatic symptoms indicated heterogeneity in the Bipolar I Disorder (BID) sample. The figure shows the probability (y-axis) of reporting somatic symptoms (xaxis) for the two latent classes. Latent class 1 (red line) (27% of the sample) had a high probability of reporting symptoms. Latent class 2 (blue line) (73% of the sample) had a low probability of reporting symptoms.