Customize your JAMA Network experience by selecting one or more topics from the list below.
Schulze TG, Hedeker D, Zandi P, Rietschel M, McMahon FJ. What Is Familial About Familial Bipolar Disorder?Resemblance Among Relatives Across a Broad Spectrum of Phenotypic Characteristics. Arch Gen Psychiatry. 2006;63(12):1368–1376. doi:10.1001/archpsyc.63.12.1368
Copyright 2006 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2006
Current diagnostic criteria for bipolar affective disorder define a phenotype that is highly heritable, yet clinically variable. A more homogeneous definition might facilitate genetic and other studies, but the best approach is unclear. Familial features of bipolar disorder should help define more homogeneous subtypes, but there are few data indicating which clinical features of bipolar disorder are the most familial.
To study the familiality of phenotypic features in families ascertained through individuals with bipolar affective disorder.
The study comprises 1246 individuals in 172 multiplex families ascertained for genetic linkage studies of bipolar disorder. The familiality of 40 diverse phenotypic features was studied using mixed-effects regression analysis.
Substance abuse, alcoholism, psychosis, history of suicide attempt, and the level of social functioning were all strongly familial in this sample. Several other traits, including clinical subtype, earliest age at onset, and comorbid panic disorder, showed a suggestion of familiality that did not hold up to conservative correction for multiple testing.
This is the largest and most comprehensive study to assess the familiality of phenotypic features in bipolar disorder. Our results suggest that comorbid conditions and social functioning should be considered along with other familial clinical features in formulating subtypes of bipolar disorder suitable for further studies. Familial variables may help reduce diagnostic heterogeneity in genetic and other biological studies.
The phenotype has become an increasing focus of genetic research in bipolar affective disorder (BPAD) and other genetically complex conditions. Current diagnostic criteria define a BPAD phenotype that is highly heritable, yet clinically variable. For example, there is a variable occurrence of psychosis, suicidality, and rapid cycling among people with the bipolar I subtype (BPI). It is widely suspected that this clinical variability reflects underlying heterogeneity at the biological and genetic levels.
Such heterogeneity, if present, could account for the slower than expected progress in mapping and identifying genes that contribute to BPAD. There have been recent successes in identifying consistent linkage signals,1,2 some at genome-wide levels of significance.3,4 Promising association findings are also emerging for G72 (DAOA)5-7 and a handful of other genes. Yet these successes have been hard won and may not point to an efficient strategy for continued progress.
Ongoing gene-mapping efforts could benefit from an approach that seeks to decrease clinical variability among cases while maintaining the high heritability that makes BPAD a good target for gene mapping efforts in the first place. This is what is meant by “refining the phenotype.”8,9
Phenotype refinement currently encompasses approaches based on individual clinical features,10 quantitative traits,11 or (suspected) endophenotypes. However, that field lacks a clear conceptual framework to guide the selection of clinical variables, their analysis, and the interpretation of results.8,12,13 There is a large amount of phenotypic data collected over the years and stored in databases, only waiting to be used.14,15 The question that arises is which phenotypic data are most useful in the formulation of alternative phenotypes. One approach to identify the most promising phenotypic variables to use in gene mapping is to assess which traits are most familial, then use these to define subtypes among people affected with the broader disorder.16 In such a strategy, familial resemblance is used as a proxy for heritability, which would be an ideal criterion for data reduction but is often difficult to assess in human pedigree data. While everything familial is not genetic, variables that are not familial are unlikely to be genetically informative.
Herein we present such an approach that uses one of the larger BPAD family samples currently available. We show that many traditional clinical features are familial, along with comorbid conditions, and, unexpectedly, a measure of social functioning. Our results suggest that genetic and other biological studies of BPAD might usefully consider comorbid conditions and social functioning along with selected other clinical features in formulating BPAD subtypes most suitable for genetic and other biological studies.
This analysis is based on multiplex families recruited within the framework of a bipolar disorder genetics study by the Department of Psychiatry at The Johns Hopkins University, Baltimore, Md. These families have been used for a variety of molecular and formal genetic studies on BPAD. The family ascertainment is detailed elsewhere.17,18 All families were ascertained through a proband with a history of BPI, and contained at least 2 additional siblings, or at least 1 sibling and only 1 parent with a major affective disorder. Families also had evidence of major affective disorder in no more than 1 parental lineage by family history.
After complete description of the study to the subjects, written informed consent was obtained. Subjects were interviewed by a psychiatrist using the Schedule for Affective Disorders and Schizophrenia–Lifetime Version (SADS-L).19 Two additional psychiatrists reviewed the interview, family informant data, and any medical records before assigning a best-estimate diagnosis under Research Diagnostic Criteria.20 The diagnosis of bipolar II disorder (BPII) required recurrent major depression as well as hypomanias. Diagnostic reliability was excellent.21
The present study is based on a total of 172 families comprising 1246 individuals. All families included in this study fulfilled the ascertainment criteria (previously outlined17,18) at the time of identification, although not all relatives could be recruited. All 1246 subjects were interviewed with SADS-L. Two hundred fifty-three individuals had a diagnosis of BPI, 215 had BPII, 150 had recurrent unipolar disorder, 12 had schizoaffective-bipolar disorder, 354 individuals could not be unequivocally assigned to a major mood disorder diagnosis (this uncertain category was kept broad by design to encourage the best-estimate clinicians to assign only confident diagnoses), and 262 individuals were considered unaffected by a major mood disorder. For this analysis, we considered 2 affection status models. The narrow model included BPI, BPII, and schizoaffective-bipolar, resulting in a total of 480 affected individuals. The broad model added recurrent unipolar disorder, resulting in a total of 630 affected individuals.
The full sample of 1246 individuals was used in the analysis. Individuals without a diagnosis of major mood disorder were studied only with those variables that are not nested within a major mood disorder diagnosis (Table 1). (A PDF version of the SADS-L score sheet used can be provided on request.)
The individual variables are all directly taken from the SADS-L interview (Table 1). Some variables were nested in a diagnosis because data on these variables were only gathered from individuals who endorsed the screening items for that diagnosis. Data on other variables were collected from everyone regardless of diagnosis. Up to 1246 individuals have contributed data on these variables. κ values 0.81 or greater22 have been established for the phenotypic characteristics associated with affective episodes assessed by the SADS-L. For diagnosis-independent traits, such as the variables describing social functioning, κ values between 0.41 and 0.8122 have been established.23-25
Familiality was assessed using a mixed-effects regression model framework, as described and advocated in several recent articles.26-28
For the estimation of familiality, we used mixed-effects logistic regression as implemented in MIXOR29 for dichotomous and ordinal outcomes, mixed-effects continuous regression using MIXREG30 for continuous outcomes, and mixed-effects nominal regression via MIXNO31 for nominal outcomes. For each variable, mixed model analyses were performed, with age and sex treated as fixed effects and family membership treated as a random effect. For clustered data, the mixed-effects model assumes that data within clusters (ie, families) are dependent. The degree of dependency is jointly estimated along with the regression coefficients of the fixed effects, thus adjusting for dependency resulting from clustering of the data. The degree of dependency attributable to families is characterized by the between-families variance, which is estimated in the mixed model. This estimated variance represents the population variance of family effects, and therefore our results pertain to the population of families of which this sample is representative.
Two models were run for each variable: one that included only the fixed effects and no random effect and another model that added family membership as a random effect. Log likelihoods for each model were compared using the likelihood ratio test with 1 df. Because a finding of significant nonfamiliality would be uninterpretable, and variances are nonnegative by definition, a 1-tailed P value was applied, as is typical for this test.32 All P values were corrected for multiple comparisons by using the Bonferroni-Holm method.33 We expect this to be conservative here because we did not account for any correlation between the dependent variables tested. The strength of each familial effect was measured by calculating the intraclass correlation coefficient (ICC) and its confidence intervals.34 The ICC, which theoretically varies from 0 to 1, indicates the proportion of unexplained variance attributable to family membership.
We tested 40 variables in this data set (Table 1). Variables in the category “Diagnostic criteria” had been rated for the most severe episode. The variables substance abuse and alcoholism are nonoverlapping Research Diagnostic Criteria lifetime diagnoses. The SADS-L assesses symptoms of mania independently from those of hypomania. The majority of variables were dichotomous or ordinal and thus were analyzed using logistic models. Age at onset and number of manic/hypomanic/depressive symptoms were analyzed under continuous models, and subtype was analyzed under a nominal model. The variables that were nested within the major affective diagnosis were analyzed under a narrow and a broad affection status model (Table 1), except for items nested within mania or hypomania. These were analyzed only under the narrow model since the broad model, by definition, does not contain any additional cases of mania and hypomania.
Given the strong evidence for the heritability of BPI, it is possible that phenotypic characteristics are also more familial in BPI individuals than in individuals with other major mood disorders. We tested this by adding BPI diagnosis as an additional fixed effect with 2 levels (BPI vs non-BPI cases of major mood disorder).
Tools for performing heritability analyses on dichotomous variables in family samples are not yet available, to our knowledge. In quantitative data, however, the QTDT35 software performs such an analysis under a variance components framework. We used QTDT to assess heritability in those quantitative variables that proved to be familial. Z transformation was applied where appropriate.
The overall sample of 1246 individuals comprised 529 males (42%) and 717 females (58%); the mean ± SD age at evaluation was 49 ± 17 years. The 480 individuals affected under the narrow affection status model comprised 191 males (40%) and 289 females (60%); the mean ± SD age at evaluation was 43 ± 14 years; the mean ± SD age at onset was 23 ± 10 years. The 630 individuals affected under the broad affection status model comprised 226 males (36%) and 404 females (64%); the mean ± SD age at evaluation was 43 ± 18 years, and the mean ± SD age at onset was 23 ± 10 years.
The results of the mixed-effects regression model analyses are presented in Table 2. Results are sorted in ascending order of the P value of the likelihood ratio test. The variables social relations, substance abuse, alcoholism, psychosis (both broad- and narrow-affection model), and suicide attempt were significantly familial in this sample, after correction for multiple testing. The ICC values ranged from 0.304 (substance abuse) to 0.130 (social relations).
An additional 15 variables showed a nominally significant familial effect but did not withstand correction for multiple testing. These included age at onset, diagnostic subtype, healthiest level of functioning, suicidal ideation in depression, poor judgment in hypomania, and several comorbid conditions (Table 2).
No significant familial effect was detected for most of the individual symptoms used in the diagnosis of mania, depression, and hypomania.
The 5 variables that were significantly familial (after correction) were further assessed in a model that included BPI diagnosis as a fixed effect. This resulted in changes in the ICC, but they tended to be small. Values increased for psychosis, alcoholism, and suicide attempt, and decreased for substance abuse and social relations (Table 3).
Social relations was the only nondichotomous variable that proved to be significantly familial after correction. Therefore this was the only variable that was subjected to heritability analysis. When compared with a null model that included only individual-specific variance, social relations (Z transformed) proved to be significantly heritable (P<.001). An estimated 21% of the total variance was attributable to genes.
Current definitions of BPAD define a clinically variable disorder. Onset varies from early childhood to late adulthood, although about 80% of cases begin between the ages of 18 and 24 years.36 The clinical course can be characterized by infrequent episodes followed by long periods of remission, or by rapid, unpredictable mood swings with little recovery between episodes.37 Some people suffer mainly from depressions, while others manifest only manias. Psychotic features or suicidal behavior may complicate the picture for some patients, and some suffer from comorbid anxiety disorders or substance abuse. Perhaps most puzzling is that many patients maintain a high level of social and occupational functioning throughout their illness, while others pursue a chronic, deteriorating course.38,39 Do these clinical differences reflect differences in etiology, or are they better explained by differences among people with the disease?
For these reasons, phenotype definition remains a key problem in genetic studies of BPAD. Overly narrow definitions restrict sample sizes and may inadvertently end up emphasizing atypical forms of the illness attributable to nongenetic factors. Overly broad definitions, however, may encompass heterogeneous conditions and complicate gene-mapping efforts.
What is needed is some external standard by which to judge the value of a phenotype definition. In this article, we assume that familiality—resemblance among relatives—is a useful standard that can guide the selection of variables that will be most effective in defining genetically homogenous forms of the illness. We demonstrate that several traits associated with bipolar disorder, such as substance abuse, are indeed highly familial, while other phenotypic features not traditionally associated with bipolar disorder are also highly familial in this sample, especially social functioning.
The phenotypic features we have studied encompass different kinds of variables, each bearing a different relationship to BPAD. Most proximal are the characteristics that make up the same diagnostic criteria for BPAD (Table 1). These will be as familial as the disease itself, by definition, except where the same criteria can be met in different ways. For example, any 4 of 7 symptoms can satisfy the Research Diagnostic Criteria criterion B for mania. The next group of characteristics are best thought of as features of the illness, present in all cases to some degree (Table 1). Examples would be age at onset, episode frequency, and other variables associated with clinical course. Many so-called endophenotypes could also be thought of as in this category. Next are the characteristics that are present only in some people with the disease, and are not required for diagnosis, such as psychosis, rapid cycling, or a history of suicide (Table 1). One step further brings us to associated (or comorbid) disorders, which can also stand alone, such as panic or alcohol dependence (Table 1). While the comorbidity of these disorders with BPAD is established by epidemiological studies, people with BPAD may assortatively mate with people who have another disorder, leading to a co-occurrence of both disorders in the offspring by simple vertical transmission.40,41 Familiality estimates of such comorbid conditions may thus be inflated in families ascertained on the basis of multiple cases of BPAD, and should be viewed with caution.
Finally, we can consider traits that might occur in anyone, regardless of disease status, but which may nevertheless influence disease presentation, such as personality, social relations, employment status, or intelligence (Table 1). The familiality of tall stature in families of probands with Marfan syndrome is an example.42 Some of the clinical characteristics of Marfan syndrome are mimicked or exacerbated in people with very tall stature, and people with Marfan syndrome, who tend to be very tall, often marry other tall people without the disease. Thus normal variation can influence the clinical presentation of even a simple mendelian disease.
The difficulty in identifying alleles that can account for the high heritability of BPAD could reflect a situation where there are many alleles, each of small effect, with a threshold of liability to clinical disease.43,44 In this case, gene mapping would depend on the collection of samples with sufficient power for detecting alleles of small effect, along with the development of methods for detecting gene-gene interactions when main effects are undetectable. Under this framework, clinical characteristics of individual cases may not be of particular interest because they are unlikely to point to genetic differences that are distinguishable within the large set of contributory risk alleles. For example, hypertension is widely considered a polygenic trait, and the key issue is not what kind of hypertension one has but how severe it is, in other words, not which risk alleles have been inherited but how many.45
In this article, we work from an alternative conceptual framework, namely, that the clinical heterogeneity of BPAD may reflect genetic heterogeneity, with several different, but clinically similar forms of BPAD, each largely attributable to the effects of relatively few alleles. Under this framework, clinical characteristics of individual cases are potential manifestations of particular risk alleles, and thus of key importance. For example, there are many different forms of nonsyndromic deafness, most characterized by a particular constellation of symptoms, age at onset, and mode of inheritance.46 Most of these clinically distinct forms of deafness are attributable to distinct loci or alleles. Under this framework, familiality may be the best a priori criterion by which to choose clinical features that may reflect particular risk alleles.
The familiality of phenotypic features in mood disorders has only recently become the focus of research. Korszun et al47 reported on familiality of symptom dimensions in major depression using a large sibling-pair sample (n = 1034). They assessed the familiality of 26 individual symptom items from the Schedules for Clinical Assessment in Neuropsychiatry,48 along with 4 symptom dimensions identified from these 26 variables by factor analysis. A total of 13 variables showed nominally significant familiality, with ICCs ranging from 0.081 (preoccupation with death or catastrophe) to 0.307 (subjective restlessness). In another sibling-pair study of major depression, Farmer et al49 analyzed the familiality of the 7 scales of the Temperament and Character Inventory,50 using the Pearson correlation in sibling pairs. All 7 scales were significantly correlated among siblings, with correlation coefficients in the 0.14 to 0.28 range.
One of the first studies to assess familiality for the purpose of finding variables that could help define genetically more homogenous phenotypes of BPAD was performed by MacKinnon et al,51 using the first 57 (n = 528) families of the sample used in the present study. They found that first-degree relatives of BPAD probands with comorbid panic disorder have a significantly higher prevalence of panic disorder than relatives of probands without panic disorder. Similarly, Potash et al52 showed that psychotic symptoms aggregate in families, a finding not supported by earlier studies of schizoaffective disorder.53 Recently, we used a sample that overlaps with the current sample to show that the frequency of manic and depressive episodes54 and the polarity at onset4 are also strongly familial in BPAD. The current study examined many more variables, and thus provides a broader clinical context for the earlier findings.
One of the largest BPAD samples to be studied for familiality of clinical features so far (160 sibling pairs) was reported by O’Mahony et al.55 Using sibling-pair correlation analysis, they found that age at onset, a dimensional measure of psychosis, and the proportion of manic to depressive episodes showed significant within-pair correlation. Using smaller samples and diverse approaches, other authors have studied the intrafamilial resemblance of age at onset,56 puerperal psychosis,57 and suicidal ideation,58 all of which showed some intrafamilial correlation.
Our study supports some of these findings and extends the focus beyond clinical symptoms to a broader spectrum of phenotypic characteristics. The main strengths of our study are its large sample size, stringent phenotype characterization, and robust statistical methods.
This sample of 1246 subjects in 172 families is one of the largest of its kind and the largest to be studied to date for the familiality of phenotypic features. The phenotypic assessment, which has been described in detail elsewhere,17,18 is characterized by a conservative best-estimate procedure. Because the diagnosis of a major mood disorder was not assigned in uncertain cases, individuals who displayed lifetime psychopathology that did not meet diagnostic criteria for a major mood disorder were not considered affected. These individuals still contribute information on comorbid traits and social functioning to the familiality analysis.
We have pursued a different analytic approach than that used by some previous studies. For example, MacKinnon et al51 and Potash et al52 used a proband-predictive model implemented in the generalized estimation equation algorithm, which has several strengths.59 A major advantage of the mixed-effects regression model we used over traditional proband-predictive models is that it estimates the familial clustering of variables directly rather than treating clustering as a nuisance parameter. The mixed-effects regression model approach also allows for the incorporation of covariates into the model, effectively correcting for any familiality that is attributable to those covariates. We used age at interview and sex as covariates here, because the former correlates strongly within our families, and the latter may affect some of the phenotypic features we studied. Despite the methodological differences, our results support previous findings that psychosis and suicide attempts are familial.52,52 Likewise, suicidal ideation, comorbid panic disorder, and age at onset showed moderate levels of familiality in this study (although the P values did not withstand conservative correction).
On the other hand, the corrective influence of covariates may have led to results that differ from those of some previous studies. For example, Leboyer et al56 found strong familial correlation (ρ = 0.42; P<.001) for age at onset, based on sibling-pair ICC in a sample of 59 multiple-affected sibships. The familial resemblance of age at onset in our sample, which was based on the full family structure, was not significant after correction for multiple testing. Age at onset is a censored variable, so sets of people correlated in chronological age—such as siblings—will also tend to be correlated in age at onset. Leboyer et al56 tried to overcome this problem by analyzing a subsample of 21 sibships interviewed after the age of 35 years, which resulted in substantially reduced significance levels in their data but did not eliminate their findings. Our use of age at interview as a covariate is similar to the approaches by Wickham et al60,61 and Dikeos et al.58 While this approach will not completely compensate for all censoring present in these data, it should eliminate large biases due to age.
Our study also identified several other variables as strongly or moderately familial, including mood disorder subtype, substance abuse, alcoholism, and level of social functioning. This familiality was not due to the ascertainment of families multiply affected with BPI, because the measure of familiality (ICC) was virtually unchanged after controlling for BPI diagnosis.
Systematic family studies in mood disorders62-65 support a familial relationship among mood disorder subtypes, but familiality of subtype per se, has not, to our knowledge, been studied before.
Whereas we found strong familial resemblance for both substance abuse and alcoholism, our study cannot contribute much to the debate as to whether the observed familial aggregation of affective disorders and addictive behavior is due to common genetic causes or other mechanisms.41,66-68
Our study is the first to demonstrate that the quality of social relations—a trait not traditionally associated with mood disorders—is highly familial in families of probands with BPAD. The related variables healthiest level of functioning, sickest level of functioning, and employment also showed familial resemblance, albeit at significance levels that did not withstand correction for multiple testing. These results support the idea, exemplified by several recent studies on the interplay between genetic variation and life events,69-72 that variables characterizing psychosocial functioning may be of greater importance in psychiatric genetics than has been previously appreciated. A recent study by Hecht et al,73 comparing healthy first-degree relatives of affective disorder patients with control individuals without a personal or family history of psychiatric disorder, suggests that good social functioning is familial in BPI.
The motivation for performing this study was to assist genetic studies in BPAD by providing a means to select variables most useful in generating refined definitions of the affected phenotype. The hypothesis that this approach will help to reduce the genetic heterogeneity of complex phenotypes has some support in BPAD. In a sample overlapping with that of the present study, McMahon and colleagues74,75 showed that stratifying families on the basis of BPII disorder substantially increased evidence of linkage to chromosome 18q22. Based on their findings on the familiality of psychosis in BPAD,52 Potash et al10 demonstrated suggestive evidence of linkage to chromosomes 13q31 and 22q12 in families with psychotic BPAD. In another example, Faraone et al76 found that age at onset of mania was heritable and revealed suggestive evidence of linkage to several chromosomal regions. These early attempts provide proof of principle, but the ultimate validation of this approach will require the definitive identification of risk alleles.
There are some limitations to this study design. The sample used in this study was highly selected. Families were ascertained on the basis of multiple relatives affected with BPAD. Therefore, it may not be appropriate to generalize our findings to less-selected samples. The primary goal of this study was to test whether certain clinical features aggregate within families and therefore might help identify genetically more homogeneous subgroups of BPAD for genetic and other biological studies. Only multiply affected families are informative for such an analysis. Preliminary results from another family sample, the National Institute of Mental Health Genetics Initiative, support many of the findings presented herein, with significant familial effects observed for substance abuse, alcoholism, psychosis, and suicide attempt (T.G.S., unpublished data, 2006).
Second, familiality is not the same as heritability. Our estimates of familiality do not consider the degree of relatedness among subjects and cannot disentangle familial resemblance due to shared alleles from familial resemblance due to shared environment. As currently available tools do not allow formal assessments of heritability of dichotomous measures in family samples, we could not formally test to what degree the familiality estimates in these data may be affected by shared familial environment. However, our strongest finding in a nondichotomous variable—social relations—showed significant heritability in the QTDT,32 which is consistent with the results of the mixed regression. Thus, at least the most significant finding in this study cannot be explained by shared environment alone.
A third limitation of this study design is its dependency on the frequency of the features studied. For instance, variables closely associated with BPAD, such as those used in the diagnostic criteria, do not vary sufficiently among affected individuals to estimate variances accurately, complicating any assessment of familiality. The same holds for phenotypic characteristics that are rare in the sample.
Fourth, the phenotype characterization procedure used a lifetime-based assessment of symptoms, rated for the worst episode only, as required by SADS-L. This may not appropriately mirror the course of a disorder, so that considering presence or absence of symptoms on an “ever” or “during most episodes” basis may give different results. On the other hand, only a “worst episode”–based approach may identify certain subgroups of major affective disorder, as demonstrated for psychosis in depression by Coryell et al.77
Finally, we did not consider any potential correlations between phenotypic characteristics in this study. These correlations probably exist but would not undermine our findings. Indeed, sets of familial characteristics may define even more familial subtypes of BPAD that would be ideal for genetic studies. This will be the subject of future studies.
In summary, this is the first study of the familiality of a broad spectrum of phenotypic characteristics in BPAD. Taken together, our results suggest that definitions of the affected phenotype of BPAD might usefully consider comorbid conditions and social functioning along with traditional symptoms. People whose bipolar disorder is complicated by substance abuse, psychosis, and poor social relations may well differ genetically from those whose illness shows none of these features. The systematic assessment of familiality may help to resolve this genetically complex disorder into more homogeneous subtypes better suited for genetic and other biological studies.
Correspondence: Thomas G. Schulze, MD, Division of Genetic Epidemiology in Psychiatry, Central Institute of Mental Health, I5, 68159 Mannheim, Germany (firstname.lastname@example.org).
Submitted for Publication: October 18, 2005; final revision received February 23, 2006; accepted March 7, 2006.
Financial Disclosure: None reported.
Funding/Support: Supported by the National Institute of Mental Health Intramural Research Program and the Edward F. Mallinckrodt, Jr, Foundation. Dr Schulze was supported through a Young Investigator Award from the National Alliance for Research on Schizophrenia and Depression. Additional support for family recruitment was provided by National Institute of Health grants R01MH061613 and R01MH042243, the Charles A. Dana Foundation, and the Ted and Veda Stanley Foundation.
Previous Presentation: Data from this study were presented in part at the World Congress on Psychiatric Genetics, Dublin, Ireland, October 2004.
Acknowledgment: We thank Sylvia G. Simpson, Dean F. MacKinnon, and Melvin G. McInnis for contributing to the family evaluations; Jo Steele for data management; Brittany Nguyen for assistance with the statistical analyses; Kathleen Merikangas for advice on the issue of comorbidity; members of the Bipolar Disorder Phenome Project (Layla Kassem, Victor López-Soto, Dean F. MacKinnon, Evaristus Nwulia, James B. Potash, and Sylvia G. Simpson) for their critique of the manuscript; and all the family volunteers who make this work possible. We also thank J. Raymond DePaulo for sharing clinical data and DNA samples.
Create a personal account or sign in to: