Bipolar disorder occurs in all age groups, young and old. Until ... these children are suffering from a mental disorder. Yet, only a very few of those will have ... Facts for Families: Bipolar Disorder in Children and Teens Excerpts on Bipolar Disord
with the Serial Color-Word Test (Smith and Klein 1953). The latter consists in the analysis of reading times during the repeated confrontation with the Stroop task, i.e., the interference between reading the names and naming the colors of incongruent
16 What can your loved ones do? ... nia). Depression is ... Bipolar disorder in adults 7 more about this in the brochure about depression in adults)
systematic reviews and randomized controlled trials ... Evidence from meta-analysis of RCTs Ib: ... recommendations are not strictly based on systematic evidence at
corpora in Section 3, the common audiovisual baseline feature sets in Section 4 and the developed baseline recognition systems with the obtained results in Section 5, before concluding in Section 6
DHT Dihydrotestosterone DISC1 Disrupted in schizophrenia 1 DSM Diagnostic and statistical manual of mental disorders (versions follow in roman numerals, except for version 5) DUDIT Drug use disorders identification test ECT Electroconvulsive therapy
Predictors of Employment in Bipolar Disorder: a systematic ... (Meta-analysis Of Observational ... experience of conducting literature reviews on
From the DEPARTMENT OF MEDICAL EPIDEMIOLOGY AND BIOSTATISTICS Karolinska Institutet, Stockholm, Sweden
BIPOLAR DISORDER AND LITHIUM TREATMENT: ETIOLOGIES AND CONSEQUENCES Jie Song 宋洁
BIPOLAR DISORDER AND LITHIUM TREATMENT: ETIOLOGIES AND CONSEQUENCES THESIS FOR DOCTORAL DEGREE (Ph.D.) By
Jie Song Principal Supervisor:
Professor Paul Lichtenstein Karolinska Institutet Department of Medical Epidemiology and Biostatistics
Professor Thomas G. Schulze Institute of Psychiatric Phenomics and Genomics Medical Center of the University of Munich
Professor Mikael Landén Gothenburg University Institute of Neuroscience and Physiology
Professor Yvonne Forsell Karolinska Institutet Department of Public Health Sciences
Associate professor Sarah E. Bergen Karolinska Institutet Department of Medical Epidemiology and Biostatistics
Associate professor Sara Hägg Karolinska Institutet Department of Medical Epidemiology and Biostatistics
Associate professor Arvid Sjölander Karolinska Institutet Department of Medical Epidemiology and Biostatistics
Associate professor Lars Feuk Uppsala Universitet Department of Immunology, Genetics and Pathology
To my beloved family 致我最亲爱的爸爸妈妈
ABSTRACT Bipolar disorder is a chronic disabling psychiatric disorder marked by episodic disruptive mood swings, accompanied by disturbances in cognition and behavior. Without effective treatment, bipolar disorder can have devastating consequences, including suicide. Lithium has a unique and predominant role in the treatment of bipolar disorder, although the response varies considerably across individuals. Both the causes of bipolar disorder and the mechanism of lithium's therapeutic action remain only partially understood. The present thesis applied genetic epidemiological and pharmacoepidemiological designs to explore the etiologies of bipolar disorder and lithium response, the correlation between bipolar disorder and other mental disorders, and the effect of lithium medication for suicidal behavior in patients. In Study I, by using a population-based cohort, we extensively quantified the familial aggregation and heritability of bipolar disorder. The precise estimates of familial risks, which increased with genetic relatedness, in conjunction with the high heritability, added to the accumulative evidence for the importance of genetic predisposition in the etiology of bipolar disorder. Furthermore, to understand where bipolar disorder falls among the psychiatric disorders, we examined the familial co-aggregations and phenotypic correlations between them. The shared genetic determinants indicated pleiotropy for genes contributing to bipolar disorder and other psychiatric disorders. In Study II, we shifted focus to the two main subtypes: bipolar I disorder (BDI) and bipolar II disorder (BDII). Using a similar design for genetically informative data, we found a higher familial clustering within each subtype vs. across the subtypes. Risks for comorbid psychiatric conditions in each subgroup also yielded different patterns. Together with the estimated substantial genetic correlation, the results revealed a distinct but overlapped etiology between the two subtypes of bipolar disorder. In Study III, we examined the rates of suicide-related events during treatment with lithium and valproate among individuals with bipolar disorder. Using a within-individual study design, the observed rates of suicide-related events were reduced during lithium but not valproate treated periods. The suggestive difference between the two drugs supported the superiority of lithium against suicidal behavior. In Study IV, we performed genome-wide association studies (GWAS) for lithium response. Subsequent GWAS also tested the subgroup of individuals responding to lithium as cases and compared them with controls without bipolar disorder. No significantly associated genetic loci were identified or validated in both meta-analyses of GWAS. Nevertheless, for the latter, the estimated proportion of variance explained by common genetic variants (“SNP heritability”) is considerable. In conclusion, our work demonstrates the importance of genetic factors for the origins of bipolar disorder and lithium’s effects. Lithium may serve as a key in both treating and informing the underlying biology of bipolar disorder in future research.
LIST OF SCIENTIFIC PAPERS I. Song J, Bergen SE, Kuja-Halkola R, Larsson H, Landén M, Lichtenstein P. Bipolar disorder and its relation to major psychiatric disorders: a family-based study in the Swedish population. Bipolar disorders. 2015; 17:184-93. II. Song J, Kuja-Halkola R, Sjölander A, Bergen SE, Larsson H, Landén M, Lichtenstein P. A family-based study of bipolar I and bipolar II disorder: coaggregation and comorbidities (manuscript) III. Song J, Sjölander A, Joas E, Bergen SE, Runeson B, Larsson H, Landén M, Lichtenstein P. Suicidal behavior during lithium and valproate medication: a within-individual eight year prospective study of 50 000 patients with bipolar disorder (submitted) IV. Song J*, Bergen SE*, Di Florio A, Karlsson R, Charney A, Ruderfer DM, Stahl EA, The International Cohort Collection for Bipolar Disorder, Chambert KD, Moran JL, Gordon-Smith K, Forty L, Green EK, Jones I, Jones L, Scolnick EM, Sklar P, Smoller JW, Lichtenstein P, Hultman C, Craddock N, Landén M. Genome-wide association study identifies SESTD1 as a novel risk gene for lithium-responsive bipolar disorder. Mol Psychiatry. 2016 Sep;21(9):1290-7. (Corrigendum in press) * These authors contributed equally to this work
Introduction ..................................................................................................................... 1 1.1 Bipolar disorder ..................................................................................................... 2 1.1.1 Characteristics ........................................................................................... 2 1.1.2 Diagnosis ................................................................................................... 2 1.1.3 Genetics ..................................................................................................... 3 1.1.4 Etiological relationships with other psychiatric disorders ....................... 5 1.1.5 Etiological boundaries of subtypes ........................................................... 6 1.1.6 Suicidal behavior in bipolar disorder........................................................ 6 1.1.7 Pharmacological treatment ....................................................................... 6 1.2 Conundrum of lithium therapy.............................................................................. 7 1.2.1 A brief history of lithium: a drug of paramount importance ................... 7 1.2.2 Anti-suicidal effects of lithium ................................................................. 8 1.2.3 Lithium response ....................................................................................... 9 1.3 A bit of epidemiology for bipolar disorder ......................................................... 12 1.3.1 Genetic epidemiology: fusion of genetics and epidemiology................ 12 1.3.2 Pharmacoepidemiology: strengths and limitations ................................ 13 Aims............................................................................................................................... 14 Materials and methods .................................................................................................. 15 3.1 Data sources ......................................................................................................... 15 3.1.1 Swedish national registers....................................................................... 15 3.1.2 Swedish National Quality Registry for Bipolar Disorder ...................... 16 3.1.3 The Swedish Bipolar Cohort Collection ................................................ 16 3.1.4 St Göran Bipolar Project ......................................................................... 16 3.1.5 Bipolar Disorder Research Network ...................................................... 17 3.2 Measures .............................................................................................................. 17 3.2.1 Bipolar Disorder ...................................................................................... 17 3.2.2 Subtypes of bipolar disorder: BDI and BDII ......................................... 17 3.2.3 Other main psychiatric disorders ............................................................ 17 3.2.4 Suicide-related events ............................................................................. 18 3.2.5 Definition of treatment for bipolar disorder ........................................... 18 3.2.6 Definition of response to lithium treatment ........................................... 19 3.2.7 DNA Genotyping, quality control and imputation................................. 20 3.3 Observational study designs................................................................................ 21 3.3.1 Case-control studies ................................................................................ 21 3.3.2 Nested case-control studies ..................................................................... 21 3.3.3 Cohort studies .......................................................................................... 22 3.4 Research Approaches .......................................................................................... 22 3.4.1 Family studies ......................................................................................... 22 3.4.2 Methodology to quantify genetic contribution ....................................... 25 3.4.3 Within-individual design ........................................................................ 26 3.4.4 Genome-wide association studies ........................................................... 26
Statistical analysis................................................................................................27 3.5.1 Logistic regression and conditional logistic regression .........................27 3.5.2 Cox regression and stratified Cox regression .........................................27 3.5.3 Variance component model ....................................................................28 3.6 Approaches in Study I-IV ...................................................................................28 Results............................................................................................................................35 4.1 Study I ..................................................................................................................35 4.1.1 Familial aggregation of bipolar disorder ................................................35 4.1.2 Bipolar disorder and other psychiatric disorders....................................36 4.2 Study II ................................................................................................................37 4.2.1 Familial aggregation and co-aggregation for BDI and BDII .................37 4.3 Study III ...............................................................................................................39 4.3.1 Suicide-related events – analysis of repeated events .............................39 4.3.2 Completed suicides – analysis of non-repeated events ..........................40 4.4 Study IV ...............................................................................................................41 Discussion ......................................................................................................................44 5.1 Findings and interpretations ................................................................................44 5.1.1 Bipolar disorder runs in families ............................................................44 5.1.2 Bipolar disorder is correlated to other psychiatric disorders .................44 5.1.3 Differences exist in the etiology between BDI and BDII ......................45 5.1.4 Reduced suicidal behavior during lithium but not valproate use...........45 5.1.5 Lithium responders represent a homogenous group within bipolar disorder ....................................................................................................46 5.2 General considerations ........................................................................................46 5.2.1 Challenges using register-based data for family studies ........................46 5.2.2 Threats to validity in observational studies ............................................48 5.2.3 Generalizability .......................................................................................52 5.2.4 Ethical considerations .............................................................................53 Conclusions ...................................................................................................................54 Future perspectives ........................................................................................................55
Chinese summary（后记） .................................................................................................56 Acknowledgements （致谢） .............................................................................................58 Reference ..............................................................................................................................62
International Classification of Diseases and related health problems
Diagnostic and Statistical Manual of Mental Disorders
Genome-wide association study
Single nucleotide polymorphism 2
Psychiatric Genomics Consortium
The International Consortium on Lithium Genetics
Randomized controlled trial
Genome-wide complex trait analysis
Total Population Register
St Göran Bipolar Project
Anatomical Therapeutic Chemical
Cause of Death Register
National Patient Register
Prescribed Drug Register
Directed acyclic graph
Structural equation model
Minor allele frequency
1 INTRODUCTION Everyone has common mood fluctuations in daily life. However, for those with abnormally elevated and depressed states, there could be an underlying bipolar disorder. Also known as manic-depressive illness, bipolar disorder is a common and disruptive psychiatric disorder with symptoms documented across cultures throughout history.1 It is characterized by recurrent episodes of mania or hypomania, alternating with depression.2 As a lifelong episodic disorder, it affects more than 1% of worldwide population, with lifetime prevalence varying depending on different definitions.3-5 Bipolar disorder has been ranked as the second leading cause accounting for days out of role in society, in particular among young adults, connoting huge costs.5 Development of effective treatments is highly sought after but hampered by the scant understanding of the pathophysiology of bipolar disorder. Among the existing pharmacological treatments, lithium is the oldest mood stabilizer and remains the first line treatment. For over sixty years, despite the introduction of new pharmaceutical agents and that lithium is not promoted by any major pharmaceutical company, it continues to be the mainstay medication for bipolar disorder.6 How lithium acts is complicated and not well understood. It remains a mystery why a drug with multiple therapeutics targets presents the most specific clinical effects. Epidemiology can be defined as describing how health and disease are distributed in a group of people and what factors influence the distribution. When genetic determinants are among the factors, the term of genetic epidemiology emerges. For studies of drug effects, pharmacoepidemiology arises. The present thesis captures investigations into the etiology of bipolar disorder, lithium’s effects for suicide-related events, and genetic variants for both bipolar disorder and lithium response.
(by Yu Liu)
1.1.1 Characteristics Bipolar disorder, according to the current diagnostic system, refers to a recurrent mood disturbance that ranges from episodes of: (1) mania, characterized by elation/irritation in mood and related symptoms (e.g., increase in energy and reduce in need for sleep) or (2) hypomania, in which the symptoms are milder in severity or duration than in mania, and (3) depression, in which individuals experience low mood and related symptoms (e.g., reduced energy and loss of pleasure).7 In spite of the emphasis on mood elevation in nosology, episodes of depression occur more frequently than episodes of mania or hypomania. Figure 1.1 presents a classical life chart recording the mood changes in bipolar disorder.
Figure 1.1. Life chart showing progression of bipolar disorder. (Reprinted from Grande I, Berk M, Birmaher B, Vieta E. Bipolar disorder. Lancet. 2016; 387:1561-72. with permission from Elsevier)8. Depending on severity, symptoms of mania and hypomania are registered above the state of euthymia (normal mood state), while symptoms of subthreshold depression and major depression are depicted below euthymia. A mixed state involves the coexistence of symptoms of depression and mania. One feature not shown in the figure is rapid cycling, defined as four or more episodes (either mania or depression) in 12 months.
1.1.2 Diagnosis Like other psychiatric disorders, bipolar disorder is mainly diagnosed by assessments of behavior in combination with subjective reports of abnormal experiences, and it lacks of definitive and objective biomarkers.9 The most widely used diagnostic classification systems for psychiatric disorders are: the International Classification of Diseases (ICD) which is the official worldwide classification and the Diagnostic and Statistical Manual of Mental Disorders (DSM) which is the official classification in the United States (US). The current versions for each system are the tenth revision of the ICD (ICD-10) released in 199310 and the fifth edition of the DSM (DSM-5) released in 2013.11 This thesis defines bipolar disorder mainly based on the ICD-10 from the Swedish National Patient Register and identifies subtypes from the Bipolar Disorder Quality Register based on the fourth edition of the DSM, Text Revision (DSM-IV-TR). 2
According to DSM-IV-TR, bipolar disorder can be grouped into four main subtypes: bipolar I disorder (BDI), bipolar II disorder (BDII), cyclothymic disorder and bipolar disorder not otherwise specified. Among these four subgroups, BDI and BDII capture the majority of patients with bipolar disorder. The core difference of diagnoses between BDI and BDII is the presence of a manic episode: for BDI, it is required to experience at least one episode of mania, while for BDII it is required to have no episodes with mania but at least one episode of hypomania. The difference between DSM-IV and ICD-10 for diagnosis of bipolar disorder is summarized in Table 1.1, modified from a recent review.7 Table 1.1 Bipolar disorder and subtypes (modified from Phillips ML, Kupfer DJ. Bipolar disorder diagnosis: challenges and future directions. Lancet. 2013;381(9878):1663-71. with permission from Elsevier)7 DSM-IV
Diagnostic criteria: one episode of mania (or mixed mood), or one episode of hypomania plus a major depressive episode
Diagnostic criteria: requires two discrete mood episodes, at least one of which must be manic
BDI: at least one episode of full-blown mania or mixed episode (manic and depressive symptoms). Usually has at least one depressive episode
BDII: several protracted depressive episodes and at least one hypomanic episode, but no manic episodes
Cyclothymic disorder: several periods of hypomanic and depressive symptoms. Depressive symptoms do not meet criteria for depressive episodes
No discrimination between BDI and BDII
Bipolar disorder not otherwise specified: depressive and hypomanic-like symptoms and episodes that might alternate rapidly, but do not meet the full diagnostic criteria for any of the above disorders
1.1.3 Genetics Bipolar disorder is thought to result from multiple factors including genes and environments interacting together.12 It is one of the most heritable psychiatric disorders, with a pattern of polygenic inheritance.13 Classical genetic epidemiological studies using family, twin and adoption data since the 1960s (when bipolar disorder and major depression disorder were considered separately)14 have suggested a genetic basis for bipolar disorder. Caution is needed when interpreting the results due to the variations in study populations, designs and diagnostic definitions. Indicative figures are summarized in Table 1.2. The recurrence risk for first-degree relatives of individuals with bipolar disorder is approximately 5-10%,15 corresponding to a relative risk (risk in first-degree relatives of patients with bipolar disorder compared with risk in the general population) of 6-10.16, 17 Estimates of heritability (i.e., the degree to which the 3
variation in a trait in a population is due to genetic variation) using family data range between 59-93%.18-22 The advent of molecular genetics and concomitant technological revolution have provided compelling insights into the genetics of bipolar disorder. Early studies investigating biologically plausible candidate genes were not very successful, probably owing to the inadequate understanding of the underlying pathogenesis of bipolar disorder.23 Genome-wide association studies (GWAS), which are hypothesis-free, have opened a new era with more robust and replicable findings about the etiology of bipolar disorder. The genetic architecture of bipolar disorder comprises many common variants of small effects (in conjunction with multiple rare variants of bigger effects),13 which could be part of the reason that early GWAS were usually underpowered to identify genome-wide associations. This sparked collaborations for meta-analyses and the formation of consortia that produced more fruitful findings. For example, the Psychiatric Genomics Consortium (PGC) has performed the largest GWAS to date and found genome-wide associated loci in the genes CACNA1C and ODZ4.24 Significant results from other GWAS are also listed in Table 1.2. Furthermore, the information from genome-wide single nucleotide polymorphisms (SNPs) enables the estimation of the so-called SNP heritability (SNP-h2) which could be considered as a lower bound of heritability by using common variants. The SNP-h2 for bipolar disorder is estimated as 25-53% on a liability scale.25, 26 Table 1.2. Summary of the genetics of bipolar disorder (from independent studies or meta-analyses) Definition
Risk in general population (reference)
0.5-1.5% for a narrow definition of bipolar disorder (equivalent to BDI in DSM-IV)
Smith and Weissman (1992)27
0.6% for BDI, 0.4% for BDII, 1.4% for subthreshold bipolar disorder, 2·4% for the bipolar disorder spectrum
Merikangas et al. (2011)4
5-10% for a narrow definition of bipolar disorder (equivalent to BDI in DSM-IV)
Craddock and Jones (1999)15
8.7% for BDI and BDII
Smoller and Finn (2003)28
Relative risk for first-degree relatives
Barnett and Smoller (2009)16, Lichtenstein et al. (2009)17
Relative risk in adoptive relatives
4.3 (95% CI: 2.0-9.5) in adopted away offspring whose biological parents have bipolar disorder
Lichtenstein et al. (2009)17
59%-93% in twin studies
Bertelsen et al. (1977)18, Kendler et al. (1993, 1995)19, 20, McGuffin et al.
Risk for firstdegree relative
(2003)21, Kieseppa et al. (2004)22
SNP heritability (on liability scale)
Leading genetic variants from GWAS
59% using parent-offspring data
Lichtenstein et al. (2009)17
0.25 ( Linear mixed model with GCTA software29)
Cross-Disorder Group of the Psychiatric Genomics Consortium (2013)25
0.53 (LD-score regression)
Bulik-Sullivan BK et al. (2015)26
Ferreira et al. (2008)30, PGC-BD (2011)31, Green et al. (2013)32
Ferreira et al. (2008)30, Muhleisen et al. (2014)33,
PGC-BD (2011)24, Green et al. (2013)32, Muhleisen et al. (2014)33
Green et al. (2013)32
Green et al. (2013)32
Chen et al. (2013)34, Muhleisen et al. (2014)33
Muhleisen et al. (2014)33, Hou et al. (2016)35
Cichon et al. (2011)36
Abbreviations: SE, standard error, PGC-BD, The Psychiatric Genomics Consortium Bipolar Disorder Working Group, GCTA, Genome-wide complex trait analysis
1.1.4 Etiological relationships with other psychiatric disorders The current classification of psychiatric disorders, or psychiatric nosology, is defined by clinical syndromes with etiologies only partially understood. In light of recent findings of genetic overlap between psychiatric disorders, the etiologic boundaries between them have been a major topic.37 Both studies using family and genetic data have suggested shared genetic factors for bipolar disorder and other psychiatric disorders, including schizophrenia,17, 37, 38 major depressive disorder,28, 37 attention-deficit/hyperactivity disorder (ADHD),37, 39, 40 and autism spectrum disorders.41, 42 Moreover, bipolar disorder is also observed to be comorbid with anxiety disorders,43 personality disorders,44 substance use disorders45, 46 and eating disorders.47 These findings are likely to inform the future psychiatric nosology. Owing to the considerable familial and genetic overlap, the diagnostic classification of psychiatric disorders has been argued to favor a dimensional system. Moreover, studies are warranted to explore the shared pathophysiology across psychiatric disorders as well as common therapeutic mechanisms.
1.1.5 Etiological boundaries of subtypes Bipolar disorder has heterogeneous phenotypes. As stated in 1.1.2, BDII, for which diagnosis only requires hypomanic episodes, is differentiated from BDI for decades. Nevertheless, controversy exists about whether BDII should be treated as the same disorder of BDI, only with milder symptoms. In contrast to this point of view, BDII is observed to have more unfavorable illness features compared to BDI, including higher proportion of comorbidities and longer illness duration.48 Studies focusing on each subtype or comparisons between them are relatively sparse. Early family studies suggest an overlap of BDI and BDII in etiology, although ambiguities remain.49-52 In addition, a recent GWAS provides evidence for a partial difference in the genetic basis of BDI and BDII.53 As the diagnostic criteria change over time, updated and more solid evidences to support the etiological heterogeneity of BDI and BDII, possibly by quantification of familial co-aggregation, are warranted. 1.1.6 Suicidal behavior in bipolar disorder Individuals with bipolar disorder are at hugely elevated risk of suicide, higher than both the general population and most other psychiatric disorders.54, 55 Epidemiological studies have reported that about a quarter to a half of people with bipolar disorder attempt suicide in their lifetime at least once, and that roughly 8–20% of attempts result in death (completed suicides).56-59 The risk of death by suicide can be up to 20 times compared with the general population.54, 60 Factors significantly associated with both attempted and completed suicide, reported in metaanalyses, include sex (females are more likely to attempt suicide, whereas males have higher rates of death by suicide) and a first-degree family history of suicide.59 Specifically, factors associated with suicide attempt involve early age at onset, depressive polarity of first, current or most recent episode, and comorbidities of anxiety, substance misuse and borderline personality disorders.59, 61 In contrast, specific bipolar subtypes (BDI and BDII) and history of psychosis are less likely to correlate with suicide risk.10 In addition, little research regarding mixed symptoms in relation to suicidal behavior has been conducted; nevertheless, the existing evidence suggests a highly elevated risk for suicide attempts at mixed states.62-64 1.1.7 Pharmacological treatment Therapeutic approaches for bipolar disorder differ between mood states, which also adds complexity because the treatment that alleviates one symptom might cause a mood swing to the opposite direction (i.e., from depression to mania/hypomania or vice versa). Other factors influencing strategies of pharmacotherapy and psychotherapy pertain to medical and psychiatric comorbidities, treatment response, adherence and adverse effects. These factors should be considered in initiation of treatment to optimize efficacy while minimizing adverse effects and non-adherence.
Treatment of bipolar disorder conventionally focuses on two phases: acute and maintenance management. The main aim for acute management is to get a patient in state of mania or depression to a stable euthymic mood, with clinical and functional recovery. The primary goal for long-term management is to prevent relapse, reduce subthreshold symptoms, and enhance functioning while optimizing treatment. The first-line pharmacological treatment for acute mania is the use of antipsychotics with or without combination of mood stabilizers.65 Such drugs, except for a few atypical antipsychotics, have limited efficacy in treating acute bipolar depression.66, 67 Alternatively, antidepressants are commonly used for bipolar depression, although the evidence of efficacy is scarce.68, 69 Moreover, controversy exists for its role because it is known that antidepressants may induce a switch from depression to mania, and an adjuvant therapy with mood stabilizers is recommended.8, 70, 71 Long-term maintenance treatment is essential given the recurrent chronic nature of bipolar disorder. Pharmacotherapy could differ according to the patients’ predominant polarity of illness.72 Lithium is believed to be the best established long-term prophylaxis.73 Because the benefits of lithium are tempered by adverse effects and potential toxicity,74 anticonvulsants (also known as antiepileptic drugs), most commonly valproate and lamotrigine, are increasingly used as alternatives.75 However, equivalent efficacy is less often observed,76, 77 and combined therapy with lithium is recommended.78 True improvement in treatment of bipolar disorder has been hampered by our scant knowledge of the underlying biological mechanisms. Consequently, validated pharmacological targets are lacking. In fact, many newly introduced drugs are based on an expansion of use from other psychiatric disorders.79 One particular cornerstone is lithium. The unique therapeutic use in bipolar disorder makes it an important starting point. Research surrounding the mechanisms of lithium’s action is crucially valuable in the development of new targets and the consequent clarification of the illness biology. 1.2
CONUNDRUM OF LITHIUM THERAPY
1.2.1 A brief history of lithium: a drug of paramount importance The history of lithium goes back to its discovery as a new mineral (petalite) on the island of Utö in the Stockholm archipelago in 1800.80 It was later detected as a new element in 1817, by August Arfwedson during his working in the laboratory of Jacob Berzelius, a professor at Karolinska Institutet. Interestingly, lithium is believed to have existed for 13.8 billion years, as one of the few elements created by the Big Bang.81 The use of lithium in medicine (including psychiatry) can be traced back to the nineteenth century,82 and it was reintroduced to modern psychiatric therapy in 1949 by John Cade in treating patients with mania.83 Now after almost 70 years, despite its varying degree of use across countries (e.g., gradually decreased in the US), lithium remains the first-line treatment for bipolar disorder. 7
Lithium has multiple clinical effects, including the mood stabilizing action to prevent manic and depressive episode recurrence (prophylaxis of bipolar disorder),84 treating acute mania85 as well as reducing suicide risk in patients with bipolar disorder.86 It has also proven to be useful in the treatment of bipolar depression in combination with other drugs.87 1.2.2 Anti-suicidal effects of lithium Lithium has a unique role in suicide prevention. Accumulated evidence suggests that lithium reduces risk of suicidal behavior in people with bipolar disorder compared with placebo and other pharmacological drugs,88 even in individuals with partial response to the treatment.89-91 A few of the landmark studies or reviews are listed in Table 1.3. Notably, caution is required in the interpretation of lithium’s effects from the literature. The main source of evidence supporting lithium’s anti-suicidal effect consists of observational studies and meta-analyses rather than randomized controlled trials (RCTs).89 Adequately powered RCTs are commonly believed to provide the most reliable scientific evidence, in which a pronounced outcome can be assumed to be caused by the intervention.92, 93 However, owing to ethical and logistical issues (suicide is a rare and severe event), RCTs on effect of treatment on suicide are more difficult, and the results may be applicable only to selective groups. Therefore, observational studies with a wide range of people with treatment variability and unselected for suicidal risk are useful and powerful supplements. Because the non-random setting makes observational studies vulnerable to complex underlying factors that may lead to biased or invalid conclusion, approaches to handle the confounding are needed. More details regarding this issue are discussed in section 1.3. Given the increasing use of other mood stabilizers, particularly valproate,94, 95 examining its efficacy in reducing morbidity and mortality is important but currently not well known. Moreover, investigations between comparator medication groups of valproate and lithium could address the ethical concerns in RCTs as well as some confounding in observational studies (e.g., illness severity).96 Table 1.3 also lists several studies comparing lithium and valproate in relation to suicidal behaviors. Table 1.3 Summary of important studies for lithium (or compared with valproate) and suicidal behavior Literature
Type of study
Cipriani et al. (2013)88
Systematic review and metaanalysis of randomized controlled trials
Lithium is effective in reducing the number of suicides and all-cause mortality compared with placebo in patients with bipolar depression.
Observational study with Danish nationwide registers
Continued lithium treatment, irrespective of diagnosis of bipolar disorder, was associated with reduced suicide risk.
Baldessarini (2006, 2008)98, 99
Meta-analyses of observational studies and randomized trails
Risks of completed and attempted suicide were consistently lower, by roughly 80%, during treatment with lithium for patients with all major affective disorders (for an average of 18 months).
Lithium vs. valproate (or anticonvulsants) Oquendo (2011)96
A randomized controlled trial
No difference between lithium and valproate use in time to suicide attempt or suicide events (power to detect relative risk ≥ 5).
Observational study with retrospective cohort in US
Risk of suicide attempts and suicide deaths is lower during treatment with lithium than with valproate in patients with bipolar disorder.
Observational study with Danish nationwide registers
Similar reductions in rate of suicide are observed in lithium and anticonvulsants (including valproate), while switch or augmentation to lithium seems superior.
Rate of suicides or attempts during treatment with lithium are significantly lower versus anticonvulsants (carbamazepine, divalproex, or lamotrigine).
The mechanism behind lithium’s potential superior anti-suicidal effect remains unclear. One possibility has been proposed that lithium reduces the risk of suicide by reducing depressive recurrences. Another alternative (or supplemental) mechanism is that lithium’s anti-suicidal effect may be associated with its serotonin-mediated effects that result in reduced aggressive behavior and impulsivity.89, 103 Study designs comparing between pharmacological agents or with psychotherapeutic treatments, could help examine the candidate hypotheses.103 1.2.3 Lithium response Despite lithium’s superior beneficial effect on both acute stabilization and maintenance treatment of bipolar disorder, a number of patients show only partial or no response.104, 105 Research is required to better delineate who could benefit from lithium medication and who cannot alleviate suffering or even be at high risk of adverse effects. 220.127.116.11 A distinct subtype? Responders to lithium may represent a valid diagnostic subgroup. These individuals tend to have specific clinical features, including a typical course of complete remissions between distinctive affective episodes,106 a family history of bipolar disorder107, 108 and low rates of comorbid psychiatric conditions.109 Their response to lithium seems familial and stable over time.110, 111 Furthermore, clinical observational and neurobiological studies suggested that lithium responders distinctly differ from responders to other mood stabilizers in symptoms 9
(classical form vs. atypical form of illness).112, 113 Based on these characteristics, the response to long-term lithium treatment is speculated to define a distinct subtype of bipolar disorder – the “lithium-responsive bipolar disorder” (lithium-responsive BD). 18.104.22.168 Predictors and potential mechanisms Given the remarkable value of prescribing lithium to patients with good response, and to tailor alternative effective treatments promptly to those without favorable response, a reliable predictor of lithium response is a research priority. Unfortunately, to date there are no clinical tools available to predict the prophylactic effect of lithium. The current method of prediction relies mainly on clinical features described above (i.e., an episodic clinical course, full remission, family history and low psychiatric comorbidity), and plausible biomarkers are absent. The difficulty to identify biological predictors is attributed to the complex set of lithium’s actions, involving cell membrane regulation, transportation and ion distribution, neurotransmission and intracellular signaling.114, 115 How lithium acts in preventing episode recurrences remains only partially understood after decades of research. The current known mechanisms are summarized in Figure 1.2 (conceptualized at multiple levels) by Malhi et al.114 One of the oldest and the strongest candidates for mechanism of lithium’s action is “Inositol depletion hypothesis”.116, 117 It is suggested that lithium’s therapeutic effect is mediated through the inhibition of two important enzymes in the process of recycling inositol in the phosphoinositide pathway, which indirectly influences the actions of downstream neurotransmitter receptors.118 The other most investigated targets of lithium include the G proteins, glycogen synthase kinase 3β, and adenylyl cyclase.
Figure 1.2 Lithium actions. (Reprinted from Malhi GS, Tanious M, Das P, Coulston CM, Berk M. Potential mechanisms of action of lithium in bipolar disorder. Current understanding. CNS drugs 2013; 27(2): 135-53. with permission from Springer)114. AC, adenyl cyclase; bcl-2, B-cell lymphoma 2; BDNF, brain-derived neurotrophic factor; GSK, glycogen synthase kinase; MARCKS, myristoylated alanine-rich c kinase substrate; PKC, protein kinase C; indicates increased; indicates decreased.
22.214.171.124 Pharmacogenetics of lithium’s action: a field in its infancy As outlined previously, the features of familial clustering and multiple clinical effects suggest a polygenic basis of lithium response. The variation in response in population may be mediated by genetic factors, and a number of linkage and association studies have been performed to assess this possibility. Many are candidate gene studies, of which only a minority have reported findings observed in more than one study.115, 119 Accounting for the limited understanding of not only lithium’s pharmacodynamics but also the pathophysiology of bipolar disorder itself, the approach of GWAS that is hypothesis-free offers particular promise. The use of GWAS for lithium response is brand new, with only a few small investigations in the past few years. The International Consortium on Lithium Genetics (ConLiGen), using a large sample of more than 2500 patients for GWAS, has recently identified a responseassociated genetic region coding two long non-coding RNAs.120 Future GWAS will continue recruiting large cohorts of patients with long-term treatment assessment to boost power.121 Another impetus to investigate the genetic factors of lithium response is to decipher the complex genetic variability underlying bipolar disorder. Lithium responders may suffer from a subtype of bipolar disorder that is more homogeneous in etiology, possibly caused by abnormal pathways involving direct or indirect targets of lithium.122 Therefore, GWAS on a less heterogeneous group, for instance lithium responders, may be more fruitful than the classic case-control studies of the illness’ genetics. 126.96.36.199 Assessment of response to lithium: a challenge The response to lithium typically follows a bimodal distribution which has been suggested from studies evaluating long-term lithium treatment. By defining an “excellent” lithium response as a total prevention of future episodes on lithium monotherapy, approximately one third of patients belong to this “full responder” group.106, 123 A smaller group of partial responders will sometimes result in a distribution of lithium response with three peaks.124 It is unclear whether this group is similar to the full responders but varies in response because of non-optimized therapeutic level or inadequate adherence.84 Meanwhile, it should be acknowledged that, as stated above, patients with a typical form of bipolar disorder are more likely to be responders. In this respect, the reported loss of lithium efficacy in consecutive decades (which could result in decreased use) might be attributable to the expansion in diagnostic criteria for bipolar disorder.125-127 A conceivable explanation for the paucity of convincing genetic findings of lithium response is the variety of response definitions across studies. In fact, a comprehensive measurement of response to long-term lithium treatment is complicated by the inherent factors related to the irregular clinical course of bipolar disorder. Another obvious factor is the varying treatment adherence. Indeed, treatment with lithium requires close monitoring.128, 129 The most validated assessment of lithium response used in the largest GWAS is the “Retrospective 11
Criteria of Long-Term Treatment Response in Research Subjects with Bipolar Disorder” (the so-called “Alda Scale”).110 This rating scale scores lithium’s prophylactic effect by quantifying the degree of prevention against episode recurrences. In addition, it takes into account the illness severity before and after use of lithium, the length and compliance of treatment and concurrent medications. 1.3
A BIT OF EPIDEMIOLOGY FOR BIPOLAR DISORDER
The present thesis consists of epidemiological studies surrounding the etiologies and consequences of bipolar disorder and lithium treatment. By far, observational studies play the most predominant role in practical epidemiology, while experimental studies are only a minority. Observational studies have advantages in ethical acceptability, cost efficiency and real-life applicability.130 However, as is well known, it is important to acknowledge that an observed exposure-outcome association does not imply a causal relationship. Epidemiology increasingly uses designs to test what causal hypotheses would best explain the observed correlation.131 1.3.1 Genetic epidemiology: fusion of genetics and epidemiology In contrast to traditional epidemiology examining mostly environmental exposures, genetic epidemiology deals with “etiology, distribution and control of disease in groups of relatives and with inherited causes of disease in populations”.132 Research on psychiatric disorders follows a chain of genetic epidemiology that consists of progressive questions: (1) is the disorder a familial trait? (2) If so, how can we explain the inheritance pattern and measure the magnitude of familial aggregation? (3) In what mode is the disorder transmitted across generations? (4) Where are the genes mediating transmission located? (5) To elucidate the mechanism of the disorder, what are the genetic variants conferring the risks?133 Classical designs in genetic epidemiology using family, twin and adoption data are powerful tools to unravel the series of questions from the start in the chain. These studies examine whether a genetic component contributes to the disorder’s etiology, which lays the fundamental basis for subsequent questions. Molecular genetic studies emerge when the epidemiological exposures are measured as genetic variants. Owing to the rapid advance in technology, interest has shifted from candidate gene analysis to GWAS, because the latter is remarkably attractive in investigating disorders with unclear pathology. For Question (3) and (4) listed above, it is well indicated from segregation and linkage studies that bipolar disorder, like most other psychiatric disorders, follows a multifactorial polygenic model, and no single gene has a major effect on its development.134 In this thesis, we examined question (1), (2) and (5) using population- and register-based family and genetic data.
1.3.2 Pharmacoepidemiology: strengths and limitations Pharmacoepidemiology, a marriage of pharmacology and epidemiology, applies methodology from the latter to study the use and effects (both intended and side effects) of medicines.135 The main difference from clinical pharmacology is that pharmacoepidemiology has an emphasis on adverse effects and generally studies a larger number of people, for instance, participants from surveys or registers.136 Pharmacoepidemiology in an observational setting cannot provide as strong evidence for causal effects as the “gold standard” RCTs. However, its naturalistic property has several strengths to inform the efficacy and safe use of medications. In comparison to RCTs, well designed observational studies with large scale data are usually more powerful, efficient and cost-effective, particularly in studying long-term treatment or rare or chronic outcomes.137 Furthermore, study subjects can cover individuals that would not be recruited in RCTs due to ethical barriers (e.g., special situations like severe outcomes and pregnancy) or logistical reasons (e.g., real-word patterns like complicated simultaneous treatments, trends of drug prescription and compensation). In view of this, pharmacoepidemiology, with a broad generalizability, may provide unique value for research on special groups of people.137, 138 Nevertheless, observational studies are vulnerable to bias; in particular due to confounding factors. In pharmacoepidemiology, one of the most susceptible factors is the “confounding by indication”, where the confounder is the indication for a prescription.139 This situation refers to the fact that the reason why a medication is introduced is associated with (“indicated by”) the outcomes of interest. For instance, how bipolar disorder is treated would be most probably based on past and current symptoms. The rate of suicidal behavior also relates to the illness course and severity. In this respect, it is likely to observe an increasing rate of suicide in lithium-treated individuals, partly or solely because patients deemed at high risk for suicide might be more often prescribed lithium for its reported anti-suicidal property. In contrast, clinicians might also hesitate to prescribe lithium to high-risk patients considering the toxic effects in case of overdose, which could conversely result in a reduced rate of suicide in lithium users than non-users.96 In both situations, the association between lithium treatment and suicide risk is confounded and no accurate conclusions regarding causality can be drawn. To address the methodological limitations, various study designs are derived from traditional cohort or case-control studies to eliminate or at lease reduce the effect of confounding factors. Promisingly and interestingly, new designs to handle the unrecognized or unmeasurable confounding (e.g., genetic predisposition, prenatal and childhood environment) can be addressed by genetic epidemiology. Genetically informative designs, for instance sibling comparisons or within-individual comparisons, are powerful strategies to study the environmental exposures (i.e., drug use) without or with minimized residual familial or genetic confounding.137 Notwithstanding the control for cluster-invariant factors, these designs are unable to account for unmeasured cluster-varying confounding. Moreover, the applications are limited by underlying assumptions (e.g., no carry-over effect in sibling comparison)140 and generalizability.137 13
2 AIMS The overall aim of this thesis was to increase the understanding of the etiologies of bipolar disorder and lithium response as well as the effect of lithium medication on suicidal behavior. Specifically, in the different studies we aimed: Study I: (1) to provide the estimates of the familial aggregation and heritability of bipolar disorder; (2) to investigate the relationship between bipolar disorder and other psychiatric disorders from the perspectives of co-aggregation as well as relative contribution of shared genetic and environmental factors. Study II: to explore the familial aggregation for bipolar I disorder (BDI), bipolar II disorder (BDII), the co-aggregation and genetic correlation between them. Study III: to examine the risk of suicide-related events during use of lithium and valproate medication for bipolar disorder. Study IV: (1) to investigate the associations between common genetic variants and response to lithium medication; (2) to identify common genetic variants associated with the risk of a “subtype” of bipolar disorder that responds to lithium medication.
3 MATERIALS AND METHODS 3.1
3.1.1 Swedish national registers Sweden has assigned every resident a personal identification number since 1947, which is used for unique identification for public administration.141 This also enables accurate linkages between Swedish national registers and forms the basis of population-based research.142 A brief description of the Swedish national registers used in our studies is presented in Table 3.1. Table 3.1. Swedish national registers included in Study I-IV Register
Date of information
National Patient Register (NPR)143
National Board of Health and Welfare
Inpatient care of somatic diseases since 1964, psychiatric disorders since 1973, complete coverage of inpatient care since 1987;
Discharge date, a primary diagnosis and up to eight secondary diagnoses assigned by the attending physician according to the ICD system
I, II, III
Outpatient care since 2001 and coverage approximately 80% Total Population Register (TPR)
Established since 1968
Sex, date and place of birth, civil status, date of death
I, II, III
MultiGeneration Register 144
Based on TPR, Individuals born since 1932, or living in Sweden since 1961
Linkage between individuals and their biological and adoptive parents (which enables the identification of relatives)
Prescribed Drug Register (PDR)145
National Board of Health and Welfare
Since July 1, 2005
Drugs prescribed and dispensed in ambulatory care, classified according to the Anatomical Therapeutic Chemical (ATC) classification system;
Date, item (substance, brand name, formulation and package), amount prescribed and dispensed Cause of Death Register
National Board of Health and Welfare
Since 1952; Complete coverage since
Primary and contributing causes of deaths, coded in accordance with the ICD system on all
I, II, III
diseased individuals registered in Sweden at time of death
Based on TPR
Date of immigration and emigration, grounds for residence and the reason to immigrate
I, II, III
Swedish Twin Registry 147
Department of Medical Epidemiology and Biostatistics, Karolinska Institutet
Twins contacted via multiple wave of questionnaires depending on birth-year cohort
Twins born 1886-2008 (triplets and higher order multiples not included);
Medical information from population registers, in addition to questionnaire/interviews from both parents and children covering a broad range of exposures and behavior148
3.1.2 Swedish National Quality Registry for Bipolar Disorder The Swedish Healthcare Quality Registries have been increasingly used in the past two decades, aiming to collect individualized disease-specific data which is lacking from the government-administered registries stated above.149 One of these registries is the Swedish National Quality Register for Bipolar Disorder (BipoläR), established in 2004. Patients are followed with annual follow-ups. BipoläR contains data for patients diagnosed with BDI, BDII, bipolar disorder not otherwise specified, or schizoaffective disorder of bipolar type according to the DSM-IV-TR and ICD-10. The diagnosis was assigned by professionals treating the patients with full access to patients’ clinical data (including longitudinal data on the natural history and course of illness). Participation in the registry is voluntary for the clinicians as well as the patients. There have, however, been incentives from healthcare providers to increase the clinicians’ rate of participation. BipoläR contains much more detailed phenotypic information and was used in Study II and III for identification of BDI and BDII and in Study IV for assessment of objective lithium response. 3.1.3 The Swedish Bipolar Cohort Collection The Swedish Bipolar Cohort Collection (SWEBIC) research study is funded by the Stanley Medical Research Institute and the National Institute of Mental Health. The SWEBIC study includes approximately 7,300 patients who were mainly recruited from BipoläR (and partially from the NPR). Patients were contacted through an introduction letter, followed by a telephone interview. Written or oral informed consents were obtained from those who agreed to donate a blood sample. 3.1.4 St Göran Bipolar Project The St Göran Bipolar Project (SBP), based on a clinical population of bipolar disorder in two defined catchment areas in Stockholm and Gothenburg, is a structured work-up program with annual follow-ups.150 It collects patients’ phenotypic data (diagnosis and neuropsychological 16
testing), biological samples (blood and cerebrospinal fluid) as well as magnetic resonance imaging scans of the brain. Diagnosis of bipolar disorder was determined by a Swedish version of the Affective Disorder Evaluation and the M.I.N.I.-international neuropsychiatric interview. SBP also recruits population-based controls matched by age and sex through Statistics Sweden. 3.1.5 Bipolar Disorder Research Network The Bipolar Disorder Research Network (BDRN) in United Kingdom (UK) is a large ongoing program of research on the genetic and environmental determinants of bipolar disorder and related mood disorders (http://bdrn.org/). Participants were recruited via both systematic (UK National Health Service Community Mental Health Teams or lithium clinics) and non-systematic methods (website and patients support groups). 3.2
3.2.1 Bipolar Disorder The determination of diagnosis for bipolar disorder is performed by extracting information from the NPR for discharge codes and by applying an algorithm validated by our colleagues.151 Briefly, this algorithm requires at least two inpatient/outpatient admissions for a core diagnosis of bipolar disorder (Swedish version ICD-8: 296.0-296.3, 296.8, 296.9; ICD9: 296A-296E, 296W, 296X; ICD-10: F30, F31). It further excludes patients with sole diagnosis of depression (ICD-8: 296.2; ICD-9: 296B) and patients with more than one diagnosis of schizophrenia (ICD-8 & ICD-9: 295; ICD-10: F25). 3.2.2 Subtypes of bipolar disorder: BDI and BDII The ascertainment of BDI and BDII is from BipoläR (stated in 3.1.2), where diagnoses were made by the treating professionals according to DSM-IV-TR. These patients are further filtered by excluding those ever diagnosed with both BDI and BDII at different time of follow-up and those not meet the diagnosis of bipolar disorder by the above validated algorithm. 3.2.3 Other main psychiatric disorders Diagnoses of other psychiatric disorders are summarized in Table 3.2. To reduce potential misclassification, we required schizophrenia and major depressive disorder to have at least two diagnoses. Notably, in Study I we used a non-hierarchical diagnostic structure in the main analysis to enable the estimation of genetic correlation between bipolar disorder and schizophrenia. We relaxed the definition of bipolar disorder to not exclude patients with more than one schizophrenia diagnosis. These diagnoses were used in Study I-III.
Table 3.2. Diagnoses of main psychiatric disorders Psychiatric disorders
295.0-295.4, 295.6, 295.8, 295.9
295A-295E, 295G, 295W, 295X
296B, 300E, 311
Autism spectrum disorders
303, 304, 305A, 305X
F10-F19 except x.5
300 except 300.4
300 except 300E
F40-F42, F44-F45, F48
Schizoaffective disorder Major depressive disorder ADHD
Substance use disorders Anxiety disorders Personality disorders
3.2.4 Suicide-related events In Study II and III we defined suicide-related events as attempted suicides (nonfatal) or completed suicides (fatal). We included events with both determined intents (ICD-8 & ICD9: E950-E959; ICD-10: X60-X84) and undetermined intents (ICD-8 & ICD-9: E980-E989; ICD-10: Y10-Y34). Patient diagnoses and the event dates were retrieved from unplanned hospital visits in NPR and death dates in CDR. 3.2.5 Definition of treatment for bipolar disorder In Study III, where we investigated the effect of lithium’s suicide preventive properties, we defined medication status as a time-varying exposure (Figure 3.1). In Sweden, oral medications for psychiatric disorders are routinely dispensed no more than three months at a time. Therefore we defined an on-medication period as a time interval covered by a sequence of dispensed prescriptions (with the start at date of first dispensation and end at date of last dispensation), with no more than three months (92 days) between any two consecutive dispensations. If the gap between two dispensations was longer than three months, this period was defined as off medication. Consequently, the entire follow-up (1 October 2005 to 31 December 2013) was divided into on and off medication time intervals. Since the PDR has a coverage started on 1 July 2005, we set the start of follow-up to 1 October 2005 to define if the individual was on/off medication initially. This design emphasizes the concomitant effect of medication. 18
Figure 3.1: Medication status as a time-varying exposure (follow-up between 2005-10-01 and 2013-12-31)
The Anatomical Therapeutic Chemical (ATC) classification codes for drugs examined in Study III were: lithium sulphate N05AN01, sodium valproate/valproic acid N03AG01, lamotrigine N03AX09, antipsychotics N05A (excluding N05AN01), antidepressants N06A, anticonvulsants N03A (excluding N03AX09 and N03AG01) and thyroid hormones H03. 3.2.6 Definition of response to lithium treatment The measurement of lithium response was conducted differently in Sweden and UK. Both have available subjective and objective assessments: information in Sweden was gathered though both interviews and register records; information in UK was collected retrospectively by interviews and clinical-notes reviews. In Study IV, we treated lithium response as a binary trait based on both subjective and objective measurements (details in Page 33 Figure 3.9). The aim is to gain a sufficient sample size and to attain a higher certainty of lithium response.
Fig 3.2 Questions regarding lithium response from SWEBIC interview questionnaire (drug names in Swedish) 19
3.2.7 DNA Genotyping, quality control and imputation 188.8.131.52 DNA Genotyping in GWAS For the GWAS in Study IV, DNA was extracted from whole blood samples and SNP genotyping was conducted using genome-wide SNP chips. The genome-wide SNP chip (or SNP microarrays) is a small chip where hundreds of thousands of short primer DNA sequences (probes) are synthesized and arrayed on specific locations. Each probe is designed to hybridize to a fragmented sample DNA (target DNA) based on complementary binding of nucleotide bases. The generated signal will be measured with its intensity depending on the amount of target DNA and the hybridization affinity. These intensity measures are then processed into SNP genotype inferences. Details regarding the underlying technology can be found in comprehensive reviews.152, 153 The Affymetrix Genechips and Illumina’s Infinium Beadchips are the two most popular technologies used in GWAS currently. The former uses hybridization of allele specific matching and the latter relies on hybridization followed by base extension. Both platforms have high genotyping accuracy and were used in our study. 184.108.40.206 Quality control for GWAS Rigorous quality control (QC) is highly important for GWAS data to produce reliable findings. Researchers have established standard protocols for pre- and post-genotyping QC which are summarized in Figure 3.3.154 Important steps involve the check of genotyping efficiency for markers and samples, sample relatedness, population stratification and HardyWeinberg Equilibrium (HWE). After association analysis, the result should be evaluated for potential systematic bias. One way is to calculate the genomic inflation factor (also known as lambda λ).155 GWAS data are often re-utilized for pooled analyses, including mega-analyses which combine all individual data and meta-analyses which collect summary statistics. Consequently, the QC procedure should be adapted to specific research questions. 220.127.116.11 Genotype imputation Genotype imputation refers to prediction or imputation of genetic variants that are not directly genotyped with good accuracy in a sample of individuals.156 This “in silico” (using computer software or simulation) method uses the structure of linkage disequilibrium (LD) among a dense set of SNPs (reference panel) to infer the missing markers in a study sample genotyped at a subset of the SNPs. Genotype imputation is particularly useful in GWAS. It increases the power of GWAS (more markers available for association scans), helps to evaluate the evidence for associations, and enables meta- and mega-analyses with markers genotyped on different platforms.157 In Study IV, we attempted to confirm the imputed genotypes with genome-wide associations by direct genotyping with TaqMan assay.153
Fig 3.3 A flowchart overview of the quality control process of GWAS. Abbreviations: QC, quality control; HWE, Hardy -Weinberg Equilibrium, PCA, Principal component analysis
OBSERVATIONAL STUDY DESIGNS
3.3.1 Case-control studies A case-control study samples subject by the status of outcome rather than the exposures.158 Cases refer to a group of individuals having the outcomes, while controls are selected from the same source of population (representing the population at risk of the outcome) but without the outcomes. Case-control designs are well suited for studies of outcomes that are rare or with long latency. Because the information of exposures is collected retrospectively, it allows for examination of multiple exposures. However, at the same time it is more susceptible to recall or information bias. The GWAS in Study IV is a case-control study with genetic variants as the exposures, thus GWAS are less vulnerable to these types of bias. 3.3.2 Nested case-control studies A nested case-control design, defined as a case-control study nested within a specified cohort, selects controls for each case at the time of case occurrence. From then, the controls are at risk of becoming cases. Therefore, this design is sometimes called “risk set sampling”. Another name for this type of sampling is “incidence density sampling”, because it enables
the estimation of incidence rates (also called incidence densities). Compared to an ordinary case-control study that estimates the ratio of odds of prevalence, a nested case-control study estimates the ratio of odds of incidence. When the incidence rate is low, the incidence odds ratio (OR) will be a good approximation of the hazard ratio (HR). In Study I, we used nested case-control design and interpret the ORs as relative risk (incidence rate ratio). 3.3.3 Cohort studies Cohort studies refer to epidemiological investigations that follow a group of subjects with defined characteristics to determine the incidence of specific outcomes. Depending on the direction of investigation in time, cohort studies can be either prospective or retrospective. A prospective cohort study collects information on exposures before the start of follow-up, and this approach has several advantages. For instance, by safely inferring the temporal relationship between exposures and outcomes, a prospective cohort design avoids reverse causation. Also, it has no recall bias for which the quality of exposure could be affected by the status of outcome. In contrast, a retrospective cohort ascertains the outcome status at the start of study and looks backwards to examine the potential related exposures. It encounters more sources of bias and confounding (e.g., selection bias). The general shortcomings of cohort studies include inefficiency for studying outcomes either rare or with long latent periods. Notably, these limitations do not apply to the population-based cohorts established in the present thesis. Using register-based data, there is no clear distinction between retrospective or prospective cohorts. 3.4
3.4.1 Family studies Familial clustering arises from the phenomenon that the occurrence of a disorder among family members exceeds that in the general population. Studies demonstrating whether a disorder clusters in close family members (familial aggregation) are often undertaken as the first step to explore whether or not there is a genetic basis for the disorder, leading to “the chain of genetic epidemiological research” stated in section 18.104.22.168 Although familial aggregation may be due to both genetic and environmental factors, the former are often presumed when the risk of disorder follows the pattern of genetic relatedness within a family.159 Figure 3.4 describes the structure of biological relatedness used in family studies. Familial aggregation studies can be extended to examine whether two disorders cluster together within families (familial co-aggregation) to investigate potential genetic or environmental factors shared between them.
Figure 3.4. Structure of biological relatedness within family The solid diamond represents the individuals initially selected in a family (probands). Other symbols with dashed box represent different biological relationships to the proband. Figures in brackets are the percentage of shared genetic makeup between the relative and the proband on average at population level (For example, fullsiblings share in average 50% of their segregating alleles makeup). Abbreviations: MZ, monozygotic twin; DZ, dizygotic twin
22.214.171.124 Familial aggregation of one disorder The most widely used approach is to evaluate the presence of a disorder in the relatives of individuals (initially selected in a family) sampled with and without the disorder.160 In practice, the case-control design is usually applied with OR as the measure of association by viewing the disorder status as a binary variable and using logistic regression models.161 Compared with typical case-control studies, this design does not distinguish between exposure and outcomes, given that we are interested in the same disorder for probands and relatives. Moreover, this design can be used when the magnitude of association is the main interest regardless of the diagnosis order between the relative and the proband. From an etiological perspective, the case-control sampling design is rational if one believes that individuals’ genetic susceptibility to a disorder is set at birth but diagnosis could come later in life. 126.96.36.199 Familial co-aggregation of two disorders Similar approaches (i.e., case-control sampling and logistic regression) were also used to evaluate the familial co-aggregation of two disorders. We assessed the strength of the relationship between the presence of one disorder in the relatives of probands conditioned on the other disorder. To explore whether two disorders share familial factors we use a directed acyclic graph (DAG) for illustration. In a DAG that consists of variables and arrows between them, A is a cause of B if there is a directional arrow from A to B. 162 In Figure 3.5 (modified from VanderWeele et. al.163), A and B represent disorder A and disorder B, and 1 and 2 represent two individuals within the same family. The presence of A in individual 1 and 2 is 23
denoted by A1 and A2. Similarly, B1 and B2 denote the presence of B in individual 1 and 2. Let FA and FB represent familial causes unique for disorder A and B, respectively. Let Ei (i=1, 2) represents certain factor which is the cause of both disorder A and B but is unique to individual i. Finally we let FA-B denote the (set of) factors common to both individuals and are common causes of both A and B. Our test of familial co-aggregation is a test for the presence of factor FA-B.
Figure 3.5. A causal directed acyclic graph with familial co-aggregation
According to Figure 3.5, three open paths: A1 FA A2 B2 , A1 FA-B A2 B2 and A1 FA-B B2 may contribute to a statistical association between disorder A in individual 1 (A1) and disorder B in individual 2 (B2). In the absence of direct effect between A and B, the first two paths will disappear and an observed positive association between A1 and B2 would be attributed to the third path, which provides the evidence of the presence of etiological overlap between disorder A and B. If there exists a direct effect between A and B, additional strategies are required to examine the presence of FA-B. For example, we could adjust for A2 to close the first path when testing the association between A1 and B2. Moreover, we need to presume that factors unique to the disorder (FA and FB) and unique to the individual (E1 and E2) affect them in the same direction.31 The required assumptions are plausible in some settings according to the known biological mechanism or the main interest of research.161 In Studies I and II we used Swedish national registers to conduct family studies on bipolar disorder and its subtypes. Compared to the family history method (that only interviews one or several family members per family), family studies using population-based data identify the disorder status in every family member. It provides a more reliable and detailed estimate of familial aggregation. Although the status of a disorder is based only on clinical diagnoses, it is often considered a satisfactory substitute for best-estimate diagnosis.133, 164 However, family studies are not designed for, or capable of assessing the relative contribution of genes and environment. Instead, researchers turn to data from twins and extended family types to estimate the genetic and non-genetic influences on disorders as well as between disorders.
3.4.2 Methodology to quantify genetic contribution Several approaches have been developed to disentangle the relative contributions of genetic and environmental factors to the development of human complex traits or disorders. The degree to which the variation in a trait in a population is explained by genetic variation is measured as heritability.165 The estimated genetic correlation between disorders is one way to quantify their genetic overlap. The data used range from twins,166 extended family relatives (e.g., siblings with different degree of genetic relatedness) to unrelated populations.29 188.8.131.52 Univariate model for one disorder Estimation of heritability is based on the assumption that multiple genetic loci have a combined effect on a trait, and variation at these loci contributes to the variation of the trait in a population. For traits defined as binary variables, like a disorder diagnosed as present or absent, a liability-threshold model is regularly used, which has the assumption that the disorder has a normally distributed underlying liability. Correlations (tetrachoric correlations for binary variables) between relatives are the basis in estimation of the attributed genetic and non-genetic components. To decompose the variance in the liability of traits, researchers often use the structural equation model (SEM) framework to model the covariance matrix.167 The ACE model is commonly used to decompose the variance of a trait into three components: A, additive genetic factors which represents linear effects of all alleles; C, shared environmental factors (between family members); and E, non-shared environmental factors. The narrow sense heritability is defined as the proportion of phenotypic variance explained by additive genetic effects. This model was used in Study I and II. 184.108.40.206 Bivariate associations across disorders We can explore the genetic overlap between two disorders by applying bivariate quantitative genetic models. The genetic correlation is often estimated to examine the extent to which genetic influences on one trait correlates with genetic influences on another trait (i.e., pleiotropy of genes). Notably, genetic correlation evaluates the overlap between genetic influences rather than their magnitude, which means that it is independent of heritability. Two disorders can have low heritability but still be highly genetically correlated. In bivariate models, the tetrachoric correlations between two disorders and the extent to which the genetic and environment factors contribute to the correlations can be estimated. These types of analyses were carried out in Study I and II. 220.127.116.11 Genome-wide complex trait analysis (GCTA) With incorporation of molecular genetic data, researchers have developed the genome-wide complex trait analysis (GCTA) framework that takes advantage of information on unrelated individuals rather than relying on family members with genetic relatedness. GCTA was developed to address the “missing heritability” phenomenon,168 that is, the gap between variance explained by SNPs identified in GWAS and the much higher additive genetic variance estimated from classic twin and family designs. The basic concept of GCTA is to
construct a genetic relationship matrix derived from all genome-wide SNPs that are represented on the microarrays to estimate the genetic relationships between distantly-related individuals.169 Subsequently, GCTA implements the genomic-relatedness based restricted maximum-likelihood (GREML) approach, and it estimates the proportion of phenotypic variation that can be explained by genome-wide SNPs (the so-called SNP heritability, “SNPh2”).29 In Study IV, we used GCTA to estimate the SNP- h2 of bipolar disorder that responds to lithium treatment (“lithium-responsive BD”). 3.4.3 Within-individual design A within-individual design can be used to control for unmeasured confounding in observational studies. For exposures that vary over time, a within-individual design compares the risk of outcome between exposed and unexposed periods within the same individual. Consequently, all confounders constant within each individual during the study period, such as genetic predisposition and baseline severity of illness, are automatically controlled for. In cases where RCTs have limited possibilities to study rare adverse outcomes, observational studies could constitute the main source of examining drugs’ effects. Consequently, controlling for confounding becomes crucially important and within-individual design is an attractive approach. In Study III where we explored the rate of suicide-related events during medication periods with lithium and valproate, we used within-individual comparisons as our main analyses. 3.4.4 Genome-wide association studies Genome-wide association studies (GWAS) are hypothesis-free studies that test for the associations between a genome-wide set of common genetic variants and a trait.170 For complex traits where the location of causal genes are unknown, GWAS might be advantageous as an unbiased yet comprehensive approach.170 GWAS are driven by some known basic genetic concepts and assumptions. First, common diseases are assumed to be caused by small effects accumulated by genetic variations which are common in the population (“common disease-common variant hypothesis”);171 the most common form is SNP. Second, based on linkage disequilibrium (LD, the non-random association of alleles at adjacent loci), the identified genetic variants associated with a trait might be among the functional variants or, in most cases, indirectly correlated with the functional variants. Last, the International HapMap project172 and SNP databases (e.g., dbSNP) provide dense genomic marker maps, which allow researchers to capture the genetic variation in SNPs and their locations across the human genome. The design of GWAS considers several aspects, including but not limited to precise characterization of phenotypes (case-control or quantitative definition), standardization of phenotype for data harmonization, population stratification and corrections for multiple testing.173 Some aspects are described in section 3.2.7 and discussed in section 5.2.2.
GWAS have evolved explosively since the first successful one was published 12 years ago.174 Meanwhile, the knowledge of genetic architectures of psychiatric disorders has also advanced considerably. This is largely due to the rapid technological changes in genotyping, universally shared big data and newly established statistical methodologies. Following these advances, new hypotheses have emerged regarding the important contribution from rare variants, and these have also been tested.175 3.5
3.5.1 Logistic regression and conditional logistic regression Logistic regression is commonly used to estimate the relation between a binary outcome and exposure(s).176 This model assumes that a transformation of the risk of the outcome, the log of odds of the outcome, has a linear relationship with exposure variables. The estimated regression coefficients for exposures are originally on a log scale, that is, the log odds ratio of the outcome for per unit increase in the exposure. In practice, the exponentials of the logistic regression coefficients, the odds ratios (ORs) are most commonly presented. They refer to the odds of being a case, given the presence of the exposure, compared with the odds of being a case, given the absence of the exposure. Logistic regression is often applied in case-control studies, and was used in Study II and IV. Conditional logistic regression is often used in individually matched case-control studies. In this model, the estimation of exposure effects is conditional on the number of exposed and unexposed observations within each matched set.176 The matching results in independence between the outcome and matching variables, with ORs estimated to be constant across all strata of the matched variables. Additional confounders can be included in the conditional logistic regression to further reduce bias. We used this model in Study I. 3.5.2 Cox regression and stratified Cox regression Cox regression, or proportional hazards model, is most widely used in survival analysis.176 This model estimates the hazard ratio (HR), i.e., the ratio of hazard (or incidence rate) in the exposed group versus unexposed group. Specifically, each time an event occurs, it compares the values of exposures for subjects who experience the event to that for all others who have not experienced the event and are still being followed (the “risk sets”). Cox regression is well applied when the research of interest is the ratio of incidence rate rather than the baseline shapes. This model assumes that the ratio of the hazards between groups is constant (“proportional”) throughout follow-up. In Study III, where we aimed to perform a within-individual analysis for the time-to-event data, a variant of Cox regression, the stratified Cox regression is used. The relation between stratified Cox regression and ordinary Cox regression is comparable to the relation between conditional logistic regression and ordinary logistic regression. The estimated coefficients of exposures in stratified Cox regression are within-cluster HRs (within-individual HRs in Study III). 27
It is noteworthy that in large data sets, like the population-based data in Study III, any minor violation of the proportional hazard assumption will be tested as statistically significant. Instead, in practice, we could use graphical residual plot to assess whether a proportional assumption is severely violated. This issue arises in general with all assumptions in all statistical models used to fit large scale samples. 3.5.3 Variance component model Quantitative genetics is a branch of genetics analyzing complex traits; it is based on the assumption that the effects from multiple genes, where no allele is a necessary, nor sufficient cause, lead to a quantitative trait (in combination with environmental contributions).177 When we refer to traits with a dichotomous definition (e.g., psychiatric disorders), additional assumptions are applied. One is liability-threshold, which assumes that the risk for a disorder, i.e., the liability, is normally distributed but a disorder is not observed until above a certain threshold. The other assumption is that the disorders are indeed dimensional spectrums but are artificially classified into a dichotomy.14 The variance components model is widely used in quantitative genetic analysis aiming at identifying the causes of phenotypic differences in a population.14 This model decomposes the variance of a trait in a population into sources of variation (genes and environments). Classically, no direct information on the genetic variants for each person is used; rather a sample of genetically related individuals is required. The methods help assess the strength of genetic factors for a trait. This method has also been expanded to molecular genetics (i.e., GCTA, see 18.104.22.168).178 3.6
APPROACHES IN STUDY I-IV
Table 3.3 Summary of study approaches in Study I-IV
Study designs and Statistical analyses
Bipolar disorder in relative I
Bipolar disorder in proband
Other psychiatric disorder in proband Other psychiatric disorder in relative
Cohort study Cox regression Stratified Cox regression
Lithium response Lithium responsive-BD
Case-control study (GWAS) Logistic regression
22.214.171.124 Study I For the study of comorbid psychiatric conditions for individuals with bipolar disorder, we established a population-based cohort by linking Swedish national registers. We used “incidence density sampling”. Each time an individual was diagnosed with bipolar disorder, we selected controls from the general population who were alive and not diagnosed with bipolar disorder at that time (but could become cases later during follow-up). For the study of familial risks, we linked the cases and controls with their first-, second- and third-degree relatives, mating partners and adoptive relatives. The study design is illustrated in Figure 3.6.
Figure 3.6. Study designs for Study I a. By linking Swedish national registers, individuals diagnosed with bipolar disorder and the diagnosis dates were identified (Individual A, E and G are diagnosed with bipolar disorder, while B, C, D, F are not diagnosed with bipolar disorder); b. All possible combinations of relative pairs (with different relationships) are constructed (e.g., individual A and B are siblings and contributed two observations, one with sibling A as the case and B as the relative, the other with sibling B as the control and A as the relative; c. At the time an individual was first diagnosed with bipolar disorder (cases’ occurrence; solid dots), controls were sampled from the rest of cohort (“risk set”; open dots) who were alive and not diagnosed with bipolar disorder (e.g., when A was diagnosed with bipolar disorder, E and G have not been diagnosed with bipolar disorder, they belong to the control group and become cases later); d. Each proband-relative pair was matched to up to ten randomly selected control-relative pair. We matched on the proband’s year of birth, sex and counties of residence at the date the case was first diagnosed. We additionally matched the relatives of controls to the relative of the case by sex, birth year and biological/adoptive relationship.
Conditional logistic regressions were fitted for the clustered binary data above. We estimated the ORs of comorbid psychiatric conditions in cases with bipolar disorder compared with controls not affected with bipolar disorder. Furthermore, we estimated the ORs of comorbid psychiatric conditions among their relatives. The 95% confidence intervals (CIs) were
calculated by using robust standard errors which accounted for the non-independence of data due to familial clustering. To study the heritability we established a cohort with siblings born 1958-1985 with follow-up time between 1973 and 2009. We included five types of siblings with different genetic relatedness: monozygotic (MZ) twins, dizygotic (DZ) twins, full siblings, paternal halfsiblings and maternal half-siblings. We applied ACE models (stated in 126.96.36.199). The amount of shared genetic factors was fixed to 1 for MZ, 0.5 for DZ and full-siblings and 0.25 for halfsiblings. The amount of shared environmental factors was assumed to be 1 for MZ, DZ, fullsiblings, and maternal half-siblings and was assumed to be 0 for paternal half-siblings. We first fitted a univariate model with adjustment for sex to estimate the heritabilities for each psychiatric disorder. Then we fitted bivariate models to estimate the phenotypic correlation between bipolar disorder and the other psychiatric disorders. 188.8.131.52 Study II Unlike Study I, in the established population-based cohort, we defined controls as individuals not diagnosed with the disorder through the end of follow-up, and we adjusted for sex and birth year for cases and controls instead of matching them. These strategies are to compensate for the relatively small sample size of bipolar subtypes and model simplicity. For the examination of familial aggregation and co-aggregation, we focused more on the comparison between BDI and BDII. Figure 3.7 illustrates our research question.
Figure 3.7. A causal directed acyclic graph (DAG) comparing familial aggregation between BDI and BDII Subscript 1 and 2 represent two individuals within one family. The presence of BDI in individual 1 and 2 is denoted by BDI1 and BDI2. Similarly, BDII1 and BDII2 denote the presence of BDII in individual 1 and 2. FBDI and FBDII represent familial causes for BDI and BDII, respectively. Ei (i=1, 2) represents certain factor contributes to both BDI and BDII but unique to individual i. Our test of familial aggregation is to test for the presence of path from FBDI to BDIi (the red solid arrow) and from FBDII to BDIIi (the blue solid arrow). Our test of familial co-aggregation is to test for the presence of path from FBDI to BDIIi (the red dash arrow) and from FBDII to BDIi (the blue dash arrow). Moreover, we want to compare the strength between the familial aggregation and co-aggregation (i.e., to compare the strength between solid path and dash path).
We conducted pairwise comparisons of related individuals. For each type of relatives we used logistic regression to estimate the ORs and robust standard errors accounting for familial 30
clustering. Notably, the analyses were complicated by the fact that cases of BDI and BDII could only be identified from a subgroup of the patient population. Owing to this, we excluded the majority of patients with bipolar disorder who lacked information on status of subtype or with subtypes other than BDI or BDII. The exclusion resulted in the distortion of prevalence for cases but does not bias the estimated ORs. We estimated the following four ORs in each cohort of relatives: (1) the odds of BDI in a relative, given BDI in the proband (i.e. pair mate), compared to the odds of BDI in a relative, given no bipolar disorder in the control; (2) the odds of BDI in a relative, given BDII in the proband, compared to the odds of BDI in a relative, given no bipolar disorder in the control; (3) the odds of BDII in a relative, given BDII in the proband, compared to the odds of BDII in a relative, given no bipolar disorder in the control; (4) the odds of BDII in a relative, given BDI in the proband, compared to the odds of BDII in a relative, given no bipolar disorder in the control. Subsequently, we tested the difference between the two ORs of BDI as well as between the two ORs of BDII in post-estimation. The heritabilities of BDI and BDII were estimated using a sibling cohort and univariate model as in Study I. To explore the genetic correlation between these two subtypes, we modified the bivariate model. Since no individual can have both diagnoses concurrently, we excluded the contribution from the within-individual cross-disorder correlation to the likelihood and assumed no contribution from unshared environmental factors between disorders. 184.108.40.206 Study III We followed an open cohort of individuals diagnosed with bipolar disorder. Subjects entered the cohort on October 2005, or on age 15, or at the date of diagnosis of bipolar disorder, whichever occurred first. They were followed until emigration, death or December 2013. Stratified Cox regression was fitted to control for time-invariant covariates within each individual. Follow-up time was split into consecutive periods (see Figure 3.8): a new period started after a medication switch or a suicide attempt occurred; the follow-up period was reset to baseline after a suicide attempt to set the underlying time-scale as the time since last event. HRs and 95% CIs for suicide-related events were calculated between on- and off-medication periods with adjustment for categorical age and previous number of suicide attempts. We further tested for the difference between HRs for lithium and valproate medication. We estimated the population-attributable fraction (PAF) to measure the proportion of suiciderelated events that could be eliminated if all individuals in the cohort would be medicated with lithium during the entire follow-up. We carried out a number of additional analyses to test the robustness of association between lithium medication and suicide-related events. We also performed ordinary Cox regression to estimate the HRs at the population level (“between-individual comparison”).
Figure 3.8. Data preparation for individual A (follow-up split by treatment status and suicide-related events)
To test if our analyses were biased by unmeasured confounding, we performed additional analyses that included a comparison with “placebo medication”. We examined the risks of suicide-related events and completed suicides during use of thyroid medications which is not believed to have any effect on bipolar disorder or suicide. 220.127.116.11 Study IV A summary of sample ascertainment is shown in Figure 3.9 as a flowchart. The detailed quality control for Swedish and UK samples are described in Figure 3.10.
Figure 3.9. Sample ascertainment flow chart. (a) Subjectively defined lithium assessment. (b) Objectively defined lithium assessment
Swedish Wave 1
Swedish Wave 2
Per-sample QC Genotype call-rate ≥ 95%
Per-sample QC Genotype call-rate ≥ 98%
Subject heterozygosity rate ≤ 90%
Subject heterozygosity rate ≤ 85%
Ambiguous genetic sex exclusion
Ambiguous genetic sex exclusion
Random selecting one of subject pairs related with pi-hat ≥ 0.20
Random selecting one of subject from pairs related with pi-hat ≥ 0.10
Excluding outliers by multidimensional scaling (MDS)
Outlier excluded by MDS
Per-SNP QC Genotype call-rate ≥ 95%
Per-SNP QC Genotype call-rate ≥ 98%
P-HWE ≥ 1× 10-6
P-HWE ≥ 5× 10-5
MAF ≥ 0.01
MAF ≥ 0.01
differential missingness based on affection status with P ≥ 1× 10-6
differential missingness based on affection status with P ≥ 1× 10-3
differential missingness based on haplotype with P ≥ 1× 10-10
differential missingness based on haplotype with P ≥ 1× 10-10
Swedish Wave 1 898 cases with bipolar disorder, 2215 controls, 744,932 SNPs
Swedish Wave 2 1415 cases with bipolar disorder, 1271 controls, 598,894 SNPs
UK BDRN 2577 cases with bipolar disorder, 5413 controls, 393,635 SNPs
Genotype imputation (1000 Genomes Project, Phase 1 integrated data version 3)
Post-imputation QC Info score ≥ 0.6, MAF ≥ 0.01 More than 8 million SNPs shared in three waves Figure 3.10. Flowchart of genotyping QC Abbreviations: QC, quality control; HWE, Hardy -Weinberg Equilibrium, MAF, minor allele frequency
4 RESULTS 4.1
4.1.1 Familial aggregation of bipolar disorder We identified 54,723 individuals with bipolar disorder. The familial risks measured by ORs for different biological and adoptive relatives are shown in Figure 4.1 (the OR is interpreted as relative risk (RR) in Paper I, as stated in 3.3.2). Consistent with previous studies, we demonstrated that bipolar disorder runs in families, and the risks for bipolar disorder in relatives decreased with the genetic relatedness between the relatives and the proband affected with bipolar disorder (ORfirst-degree> ORsecond-degree> ORthird-degree). We observed a higher OR for maternal half-siblings compared to that for paternal half-siblings as well as other class of second-degree relatives. Using the sibling cohort, we estimated the heritability for bipolar disorder at 58%, with remaining variance attributed to non-shared environment.
Figure 4.1. Familial risks for bipolar disorder
In addition, the risks for bipolar disorder were also increased in probands’ mating partners, biological relatives in an adopted-away relationship as well as adoptive relatives without biological relationship. 35
Bipolar disorder and other psychiatric disorders
We observed increased risks of lifetime comorbid psychiatric conditions among individuals with bipolar disorder (ORs=9.7-22.9 as shown in Figure 4.2) and their first-degree relatives (ORs for full- and half-siblings are shown in Figure 4.2; results for other types of relationships can be found in online Supplementary Table 2-8 in Paper I). Throughout the comorbid psychiatric conditions, we observed a pattern that full-siblings experienced higher ORs than half-siblings, and maternal half-siblings had slightly higher ORs than paternal halfsiblings (except for schizophrenia). We also found increased risks for several comorbid psychiatric conditions among the adoptive siblings (see Table 3 in Paper I).
Figure 4.2. ORs for comorbid psychiatric conditions in individuals with bipolar disorder and their siblings (Risk for depression for individuals with bipolar disorder was not estimated regarding that depression is part of bipolar disorder and not a comorbid condition) Abbreviations: ADHD, attention-deficit/hyperactivity disorder; ASD: autism spectrum disorders
The phenotypic correlations between bipolar disorder and other psychiatric disorders are summarized in Figure 4.3. The correlations ranged 0.37-0.62, among which the highest was estimated to be between bipolar disorder and depression. Meanwhile, approximate one-half of the correlations were attributable to genetic factors.
Figure 4.3. Phenotypic correlations between bipolar disorder and other psychiatric disorders
4.2.1 Familial aggregation and co-aggregation for BDI and BDII We observed both familial aggregation for BDI and BDII and co-aggregation across these two subtypes. Figure 4.4 shows the ORs for BDI/BDII in first-degree relatives of probands affected with BDI/BDII compared to relatives of healthy controls. In general, the strength of familial aggregation for each subtype was higher than that of co-aggregation between them. The sex-adjusted heritability is estimated at 0.60 (95%CI 0.33-0.82) for BDI and 0.41 (95%CI 0.13-0.65) for BDII. The genetic correlation between BDI and BDII was estimated at 0.73 (0.60-1.00).
Figure 4.4. Familial aggregation of BDI and BDII and co-aggregation between them The first column represents the relation to probands (for parents, offspring and full-siblings, respectively, and for first-degree relatives by combing them together). The “BDI” and “BDII” in second column represent the risk for the subtype of bipolar disorder in probands (exposure). Four ORs were estimated for each type of relative: (a) the odds of BDI in a relative, given BDI in the proband, compared to the odds of BDI in a relative, given no bipolar disorder in the control (blue in the left); (b) the odds of BDI in a relative, given BDII in the proband, compared to the odds of BDI in a relative, given no bipolar disorder in the control (red in the left); (c) the odds of BDII in a
relative, given BDI in the proband, compared to the odds of BDII in a relative, given no bipolar disorder in the control (blue in the right); (d) the odds of BDII in a relative, given BDII in the proband, compared to the odds of BDII in a relative, given no bipolar disorder in the control (red in the right). The P-value is calculated by testing the difference between two ORs in each group in post-estimation.
Figure 4.5 shows the ORs for lifetime comorbid psychiatric conditions among individuals with BDI and BDII and the ORs for schizophrenia and major depressive disorders in their siblings, compared to the general population. Individuals with BDII showed significantly higher ORs for comorbid psychiatric conditions (except for autism spectrum disorders) and suicide-related events compared with individuals with BDI. Their full-siblings, unlike the full-siblings of BDI probands, did not show increased risk of schizophrenia.
Figure 4.5. ORs for comorbid psychiatric conditions in individuals with BDI and BDII *Due to the hierarchical structure of psychiatric diagnoses (i.e., a patient diagnosed with schizophrenia is usually not diagnosed with bipolar disorder, and a patient diagnosed with bipolar disorder is usually not diagnosed with major depressive disorder), we estimated the ORs of schizophrenia and major depressive disorders in the patients’ full-siblings instead.
4.3.1 Suicide-related events – analysis of repeated events A total of 51,535 patients entered the cohort from October 2005 to December 2013, among which 4,643 (9.0%) experienced suicide-related events (N=10,648) during 273,140 person years of follow-up. In the within-individual analyses, only individuals who change their exposure status and have suicide-related events during the follow-up contribute information; thus, 4,405 subjects were eligible for the within-individual analysis. Results are shown in Figure 4.6. The rate of suicide-related events decreased by 14% for periods on lithium treatment compared to off lithium treatment (HR 0.86, 95% CI 0.78-0.95). A similar result was also observed when we performed between-individual analysis among all the 51,535 patients. We further estimated the PAFs by using the estimated HRs from within- and between- individual analyses. Based on the exposure rate of lithium in the total population, 12% (95%CI 4%20%) of the suicide-related events could have been prevented if all the patients had taken lithium during the entire follow-up. Sensitivity analyses testing the robustness of the association between lithium treatment and suicide-related events showed no substantive difference from the main results (selected sensitivity analyses are shown in Figure 4.7). In contrast, the within-individual analysis found no significant difference in rates of suiciderelated events during periods on compared to off valproate treatment (HR 1.02, 95% CI 0.891.15; Figure 4.6). The test of difference in HRs had P-value of 0.038 between periods treated with lithium and valproate.
Figure 4.6. Rate of suicide-related events from 2005-10-01 to 2013-12-31 during medication periods compared to non-medication periods among patients with bipolar disorder in Sweden Stratified Cox regression was applied for within-individual analyses, with adjustment for time-varying covariates including categorical age, and previous number of suicide attempts. Ordinary Cox regression was applied for between-individual analyses, with adjustment for the same covariates as in stratified Cox regression and, additionally, with adjustment on time-fixed covariates including sex, length of baseline hospitalization periods due to psychiatric admissions and history of suicide-related events before entering follow-up.
Figure 4.7. Sensitivity within-individual analyses for association between lithium and suicide-related events Stratified Cox regression was applied for all analyses with adjustment for valproate medication (for medication with thyroid hormones, both lithium and valproate medication were adjusted), categorical age, and previous number of suicide attempts. Concurrent medications of other drugs (lamotrigine, antipsychotics, antidepressants, anticonvulsants) were defined using the same method as lithium and valproate.
4.3.2 Completed suicides – analysis of non-repeated events This part was previously included in our Study III. However, when we performed additional analyses to address one comment raised by a reviewer, we decided to delete the separate analyses for completed suicide in the revised manuscript. Here we elucidate the approaches, results and the encountered problems. During the follow-up, 590 completed suicides occurred (5.5% of the total number of suiciderelated events). When we restricted the outcome to these non-repeated events, we used conditional logistic regression for within-individual comparison. Since prescriptions cannot be dispensed after death, we defined the end of each medication period with a prolongation with 30 days (different cut-offs, 14 and 90 days, were also tested in sensitivity analyses). We used the “case-time-control method”,179 in which the exposure and outcome are “swapped” in the conditional logistic regression model. Due to the symmetry of the OR, this procedure does not change the interpretation of the exposure-outcome OR, but it does enable adjustment for monotonically increasing time-varying covariates (i.e., categorical age and previous number of medication switches). We further split follow-up time into months (30 days) for adjustment. PAFs were also estimated for completed suicides. 40
As shown in Figure 4.8, we found substantially lower risks of completed suicides during both lithium (OR 0.45, 95% CI 0.30-0.65) and valproate medication (OR 0.31, 95% CI 0.16-0.61). The between-individual analyses showed similar results. Based on the ORs and the proportions of suicides occurring during medication, the PAFs were estimated as 50% (95%CI 16%-83%) and 67% (95%CI 44%-90%) for lithium and valproate, respectively. These suggest that 50% and 67% of completed suicides would be prevented if all patients had taken lithium and valproate, respectively. However, when we used thyroid medication as a comparison, we also observed significantly decreased risk of completed suicides during the medication periods (Figure 4.8), whereas this drug did not seem to have any effect on all suicide-related events (Figure 4.7). It was not clear whether the “protective effect” for completed suicides from the “placebo” comparison is due to methodological limitation (non-repeated events using case-time control design) or residual confounding. In either situation, the magnitude of decreased risks for completed suicide for lithium and valproate requires additional investigation.
Figure 4.8. Risk of completed suicide from 2005-10-01 to 2013-12-31 during medication periods compared to non-medication periods among patients with bipolar disorder in Sweden A conditional logistic regression model, in which the exposure and outcome are “swapped”, was used in withinindividual analyses, with adjustment for time-varying covariates including categorical age, previous number of treatment switches and months. Ordinary Cox regression was used in between-individual analyses, with adjustment for the same covariates as for suicide-related events (Figure 4.6).
We observed little evidence for inflation in each GWAS and in meta-analyses (genomic inflation factor λ =0.97-1.05). Table 4.1 shows the top regions of genetic associations for each meta-analysis of GWAS for: (1) lithium responders vs. non-responders by subjective assessment; (2) lithium responders vs. non-responders by objective assessment; (3) lithium responders vs. controls by subjective assessment; (4) lithium responders vs. controls by objective assessment (N.B., a database error discovered after publication led to mismatched phenotypes for the objective lithium response measures in the Swedish data; therefore, Table 4.1 shows the updated results that are not the same as shown in the published Paper IV).
For analyses comparing lithium responders and non-responders, no SNPs achieved genomewide significance in each sample (except for one in GWAS for Sweden wave 2 with subjective assessment) and in the meta-analyses combining them. For analyses comparing lithium responders and healthy controls, the meta-analyses combining each GWAS together did not show supportive evidence for the genome-wide significant SNP found in one GWAS (in Sweden wave 1 with subjective assessment). Notably, it yielded another imputed variant with genome-wide significance (rs146727601, P=1.22×10-9, OR=4.12, for objective assessment of lithium response). However, validation genotyping attempts failed for this variant. The estimates of SNP heritability by using GCTA were 0.29 (95% CI 0.23–0.36, P < 0.0001 for the null hypothesis of being non-heritable) and 0.41 (95% CI 0.19–0.63, P < 0.0001) for lithium-responsive BD with subjective and objective assessment, respectively. Table 4.1. Top regions of genetic association for each meta-analysis Chr
Responders vs non-responders, subjective assessments
Responders vs non-responders, objective assessments
LD clumping was used to define regions of association. Positions are given in hg19 coordinates (UCSC website). Abbreviations: Chr, chromosome; Index SNP, markers with the strongest association in the genomic region; A1/A2, alternate and reference alleles; Freq, frequency of alternate alleles; OR, odds ratio; N, number of SNPs in the reported region
5 DISCUSSION 5.1
FINDINGS AND INTERPRETATIONS
5.1.1 Bipolar disorder runs in families With the largest cohort for bipolar disorder ever published, we confirmed that, as suggested by previous studies,17, 28 first-degree relatives of probands affected with bipolar disorder also had excess risk for bipolar disorder (ORs 6.4-6.8). Moreover, our data provide estimates for bipolar disorder among essentially all classes of relatives as well as mating partners and adoptive relationships. Genetic factors are suggested to play an important role in the etiology of bipolar disorder, which is supported from both the pattern of familial aggregation and the heritability estimate. This is also in line with molecular genetic studies of bipolar disorder.25 On the one hand, evidences from family aggregation (e.g., higher risk in maternal vs. paternal half-sibling, aggregation in adoptive families) supported the contribution from shared environmental factors to the etiology of bipolar disorder. On the other hand, analyses of variance decomposition suggest that the remaining variance of phenotype is mainly attributed to non-shared environments. Previous twin studies have found no significant contribution from the shared environment,20, 180 while later a Swedish study using parent-offspring data reported a small but significant contribution of this component to the etiology of bipolar disorder.17 In conclusion, if shared environmental effects exist, they are likely very modest. 5.1.2 Bipolar disorder is correlated to other psychiatric disorders In addition to the increased risks of other psychiatric disorders observed among individuals with bipolar disorder, we also found that these disorders were more prevalent in their family members. Shared genetic factors were estimated to contribute half or more to the considerable correlations between bipolar disorder and the other psychiatric disorders (Figure 4.3). These convergent lines of evidence support the pleiotropic effects of genes contributing to the etiology of psychiatric disorders. Our findings, along with reports from molecular genetic studies, support treating psychiatric disorders as syndromes of “inter-related clinical phenotypes” from the perspectives of etiology. Moreover, our data suggest that bipolar disorder is probably closer to mood disorders rather than psychotic disorders based on the stronger correlation between bipolar disorder and depression. Nevertheless, these findings should be interpreted with caution. First, our results are based on the diagnoses of psychiatric disorders with a non-hierarchical structure. For example, we allow an individual to be diagnosed with both schizophrenia and bipolar disorder. This is to enable our evaluation of comorbid psychiatric conditions and tetrachoric correlations within individuals. Second, the tetrachoric correlation implies phenotypic correlation, which is 44
higher between bipolar disorder and depression. This does not mean that the genetic correlation between them is also high. In fact, studies have suggested that bipolar disorder has higher genetic correlation with schizophrenia than with major depressive disorders.25, 181 5.1.3 Differences exist in the etiology between BDI and BDII Similar to earlier family studies,49-52 we consistently found familial aggregation of BDI and BDII. More importantly, the strength of aggregation for each subtype was higher than coaggregation across subtypes within families. Furthermore, the comorbid psychiatric conditions in patients with BDI and BDII revealed distinct patterns. These findings cumulatively offer evidence that BDI and BDII do not merely represent different levels of severity but also different (albeit overlapping) entities etiologically. Although our estimate of the genetic correlation between BDI and BDII using sibling data did not reject a complete correlation (the 95%CI included 1), a recent paper using genetic data reported the observation of genetic heterogeneity between BDI and BDII.53 Nevertheless, we should be cautious about the extent to which we can address the heterogeneity between BDI and BDII based on our findings and our exclusive definitions of BDI and BDII (i.e., individuals with diagnosis of both BDI and BDII at different time point were excluded). 5.1.4 Reduced suicidal behavior during lithium but not valproate use With the largest sample ever published and up to eight years of follow-up, we found decreased rates of suicide-related events during periods on lithium but not on valproate medication. The suggestive between-drug differences imply potentially unequal effects between lithium and valproate against suicidal behavior. Since within-individual comparisons exclusively draw information from those who ever attempted suicide during follow-up, the association we observed demonstrated that, even among high-risk patients (who were unlikely to be studied in RCTs), suicide-related events were reduced while on compared to off lithium medication. In addition, we estimated that, in the absence of potential confounding, more than 10% of suicide-related events could have been prevented if all patients had been treated with lithium during the entire follow-up. Note however that this estimation is based on the exposure rate in the study population and will vary depending on prescription rates and medication adherence. Although observational studies can never prove causality, we attempted to exclude a number of alternative explanations for the association between lithium and reduced suicide-related events. Moreover, by comparing treated and untreated periods within the same individual, we automatically controlled for all time-invariant confounders. Consequently, the likelihood of confounding by indication — the big threat in observational studies — is reduced. Therefore, the observed association supports the hypothesis that lithium is protective against suicidal behavior.
Additional issues should be considered when interpreting the results. First, compared to previous studies which performed analyses on a population level (between-individual comparisons), our findings, by design, emphasize the short-term effect of drugs. Second, we did not test whether the mechanism of lithium’s potential anti-suicidal effect is through reducing impulsive aggression. Moreover, alternative explanations for the associations can never be ruled out. For example, we cannot exclude a potential lithium-specific effect due to the required routine blood tests during lithium treatment to monitor for rare side-effects. It is also possible that those who quit medication may have done so because they face concurrent stressor events that lead to an increase risk of suicide events beyond the contributions from medication. Finally, the analyses for completed suicides were inconclusive regarding the magnitude of association. In fact, few studies have focused exclusively on completed suicide, largely because of its rarity and inherent difficulty, which makes it less amenable to firm conclusions. 5.1.5 Lithium responders represent a homogenous group within bipolar disorder Our GWAS failed to identify (or validate) any common genetic variants associated with lithium response and lithium-responsive bipolar disorder. One concern is the heterogeneous assessments of lithium response in Sweden and the UK. Despite of careful harmonization of measures across sites, potential phenotypic misclassification cannot be ruled out, and is a possible reason why loci with significant association in separate sample failed to hold in the following meta-analyses. As a result, a more elaborate assessment in lithium response might yield positive results. The other concern is the limited sample size by GWAS standards. To some extent, the different results between the analyses of responders vs. controls and responders vs. non-responders could reflect reduced power beyond the underlying biology. Nevertheless, when we defined a subtype of bipolar disorder by lithium response, the extent of genetic contribution (from common genetic variants) to the subtype’s variance was estimated at 0.29-0.41 (using subjective and objective assessments, respectively). This suggests that the subtype of bipolar disorder defined by lithium response is modestly heritable. In addition to increasing sample size, to define a homogeneous subgroup of complex traits by treatment response may serve as a productive approach to identify susceptible alleles. It could provide further understanding of the pathophysiology of the disorder and the lithium mechanism of action. 5.2
5.2.1 Challenges using register-based data for family studies Unlike typical family studies where probands are clearly delineated (e.g., consecutive admissions to an inpatient facility), the assignment of proband in a register is arbitrary because the ascertainment is fundamentally complete. Moreover, the family pedigrees derived from these data are complicated, with number of affected individuals, family sizes and relationship types varying across each family. To accommodate these analytical 46
challenges, we designated each individual in turn as the selected person (proband) from each family and constructed all possible relative pairs. As a consequence, non-independence between data appeared due to family clustering and would affect the standard error. We took account for this by using a robust sandwich estimator. In fact, the calculation of standard errors with and without the robust sandwich estimator yielded similar results owing to the very large sample size. One should bear in mind that individuals may have varying possibility to be diagnosed with one disorder in our study population. The reason is related to: (1) left truncation at the start time (i.e., the natural time origin of event is before individuals come under observation), (2) right censoring in the end of follow-up (i.e., individuals leave the study before the events occur), and (3) the changes in diagnostic criteria for psychiatric disorders. For example, for bipolar disorder with onset in late adolescence and adulthood, people with dissimilar birth year will be at different risk of diagnosis from 1973 through the end of follow-up. In Study I and II, we focused on the magnitude of familial liability for one disorder or between two disorders rather than the causal relationship between them. For this reason, we simplified our analysis by defining the disorder status as a binary variable with lifetime diagnosis. However, due to the left truncation and right censoring in register-based data, we lack the information for lifetime occurrence for all subjects. We used different approaches in Study I and II to account for this situation as well as to emphasize the main research interest. In Study I, where we aimed to estimate the familial recurrence risk for bipolar disorder, we used the nested case-control design and estimated the ratio of incidence risks (so-called “relative risk” in Study I). This method accommodated the structure of the register-based data. Without regard for when the cohort was drawn, it reduced bias caused by left truncation due to the different entry age of individuals into the population registers. It would have been possible to perform a time-to-event analysis by estimating the incidence rate ratios for the relatives of probands compared to relatives of controls. We used an alternative approach; we estimated the relatives’ total time-at-risk by matching on the year of birth, sex and living counties at the year of the case’s diagnosis. These variables could affect the probability of diagnoses of disorders, but by matching on them, we ensured approximately equivalent timeat-risk for the cases and controls. The advantage of matching is to preserve most of the informative data (all observations for disorder occurrence) and considerably reduce the analytic sample size (only a sample of control observations). This only slightly reduced the precision of estimates but substantially decreased the computing time. In Study II, we aimed to compare the degree of familial aggregations of BDI and BDII. Considering the limited sample size, we adjusted for sex and birth year instead of matching on them to keep all information from the entire cohort. One may argue that such design does not consider the left truncation and right censoring of data structure; however, factors caused by such data structure are unlikely to confound our comparison between ORs for BDI and BDII for the relatives. As is shown in Figure 3.7, factors unique to one individual (e.g., sex and birth year) do not correlate with the case status in his/her relative. 47
5.2.2 Threats to validity in observational studies 18.104.22.168 Selection bias caused by left-truncation and right-censoring In general, the register-based data largely eliminate the potential selection bias. However, the left-truncation before the start of register and the right-censoring at the end of follow-up (or due to migration) could possibly introduce selection bias. We discussed in the previous section how we account for this source of bias in family studies. When using sibling designs to estimate the heritability, the observed sibling correlation in psychiatric disorders could be affected by different birth years that introduce different timeat-risk for the identification of cases from registers. To compensate for this, we defined a sibling cohort born within a time period. For example, in Study I, we defined the sibling cohort with individuals born during 1958-1985 and followed them during 1973-2009, during which time the psychiatric diagnoses can be captured by the register. These individuals were older than 24 years old at the end of follow-up —a period approximately covers the mean age at onset. 22.214.171.124 Misclassification of exposure and outcome Misclassification can be induced from the use of sub-optimal measurement of exposures and outcomes using register data. Non-differential (random) misclassification usually dilutes the true association. Differential (non-random) misclassification, whose degree varies depending on the status of the exposure and/or the outcome, can either drive up or pull down the association. Quality of psychiatric diagnosis For ascertainment of psychiatric disorders, numerous factors can introduce the misclassification and thus affect the observed association in different directions. (1) The validated algorithm we used to define bipolar disorder decreases the false-positive rates but increases the probability that individuals with bipolar disorder may be misclassified as unaffected. (2) Although medical records identified from the entire Swedish population substantially reduced the risk of recall and reporting biases, it raised potential misclassification by the weaker sensitivity of disorder ascertainment. Compared to surveybased studies, register-based data capture only treatment-seeking individuals. (3) Physicians may be influenced by family history in assessing diagnosis, which could induce the diagnostic bias for relatives of probands. If so, the familial risks estimated in our studies could be biased upwards. (4) In a similar vein, if the physician’s assessment for one disorder is affected by the appearance of another one, the evaluation of comorbid psychiatric conditions also may not reflect the true magnitude of associations. Ascertainment of suicide-related events A universal standard definition of suicidal behavior is difficult, and measures vary across studies. The definition of suicide-related events in Study III includes cases with both 48
determined and undetermined intent. This helps to reduce the underestimation of cases and generalize our findings for all suicides. Meanwhile, it is argued that events without suicidal intents may have a different etiology. Therefore we also present the results for events with determined and undetermined intents separately. We included both fatal (completed suicide) and non-fatal deaths (suicide attempt) in our analyses. On the one hand, as previously discussed, more difficulties arise in withinindividual comparisons for non-repeated events. On the other hand, attempted suicides and completed suicides are believed to represent the same behavior (with different consequences depending on severity and situations), and we obtained similar results when restricting the outcomes to suicidal attempts only. Measurement of treatment status Potential misclassification of exposure to medication in Study III is unavoidable since we never know the actual consumption of drugs. We carefully addressed the possible bias resulting from the misclassification of lithium medication from the following aspects. (1) Individuals may not take the drug as prescribed and dispensed. Nevertheless, the nonadherence can be considered in a similar manner as intention-to-treat analyses in RCTs that allow for non-compliance after protocol allocation. If anything, the reported association would be a conservative estimate. (2) The reduced HR could be underestimated for several reasons. First, we conservatively classified individuals as untreated since the date when the last prescription was dispensed, while they could be truly on medication for some time. Second, we treated individuals with only one dispense date as off medication throughout the follow-up, although some may actually had taken the drugs. To test to what extent the association was biased downwards, we used alternative definitions of exposure time and observed no material differences in association. Assessment of lithium response One of the most crucial issues for pharmacogenetic studies is a reliable assessment of treatment response. However, the irregular clinical course of bipolar disorder, and difficulty in accurately measuring treatment adherence are only two of the reasons that make the assessment of lithium response complicated and ambiguous. In Study IV, we thoroughly utilized all available information from both self-assessment and clinical records. For subjective self-assessment, we distinguished patients with a clear good response to lithium as responders from those with partial or no response. This measurement corresponded approximately to the dichotomous definition used by ConLiGen consortium and has been proven to have moderate to substantial inter-rater reliability.124 For objective assessment, we measured the efficacy of lithium treatment in assessing affective episodes annually by using the clinical notes from the Swedish Bipolar Quality Register. In this way, we defined a more homogenous subgroup as full responders to lithium with follow-up time over two years. This is comparable to the UK-BDRN subgroup with excellent and beneficial response to lithium for at least three years. 49
In addition to measuring the clinical improvement level, a refined evaluation should also consider relevant factors to determine whether the reduction of illness activity is actually due to the treatment. We have tried to take into account complementary evaluations including the number and frequency of episodes, the length and severity of symptoms before and during lithium medication, the length and compliance of lithium use and additional treatment. For those defined as lithium responders and with available information, over 80% have used lithium as the sole mood stabilizer, among whom two thirds have not taken any antidepressant or antipsychotic; more than 85% have been assessed with stable serum lithium level within the intended therapeutic range. However, due to the considerable missing rate of data, we did not include these criteria in the assessment because of the trade-off between precise certainty of response and sufficient sample size. Moreover, the absence of detailed retrospective data which hampers the contrast between pre- and post-treatment course further affected the accuracy of response assessment. Therefore, the negative findings for common genetic markers for lithium response here may be due to the potential misclassification of lithium response. Meanwhile, the evaluation of heterogeneity is crucial for meta-analyses. Our study only consists of two samples from Sweden and the UK, which presents strengths compared with studies of multicenter collaborations. To assure the proper use of fixed-effects models, we also checked the Q-test of heterogeneity for the highest associated markers and none has shown significant evidence of heterogeneity. However, a major limitation is the degree of correspondence between the responsiveness phenotype in these two samples. It also suggests caution in interpretation of the negative findings in meta-analyses. 126.96.36.199 Confounding The observed association between two factors can be that one causes the other, or they share a common cause or both. Confounding can be viewed as the presence of common causes (directly or indirectly) to the exposure and the outcome. For observational studies where randomization is not available, control for confounding is extremely important for study design and analysis, followed by the consideration of residual confounding in the interpretation of association. In Study I and II, we examined whether shared familial factors contribute to the association between the occurrence of one disorder in the individual and recurrence of the same disorder (or co-occurrence of another disorder) in the family members. With the implementation of causal diagrams, it is easier to distinguish the potential confounders related to our research question. Interestingly, age and sex as the common confounders that are adjusted in many studies actually do not confound our analyses. The reasons that we adjusted for them are due to concerns regarding the structure of register-based data. Nevertheless, the structure of a causal diagram is derived depending on subjective knowledge. For example, considering that bipolar disorder may affect the development (or ascertainment) of another disorder, we
adjusted for the status of bipolar disorder in the siblings when estimating their risks for other psychiatric disorders. In Study III, many efforts were taken to minimize bias from different source of confounding. First, to rule out a reverse causation that the medication is introduced because of the occurrence of suicidal behavior, we used a prospective longitudinal design and set the underlying time scale as time since last event. Second, we used multivariable regression analysis to adjust for potential confounders (both time-invariant and time-varying variables); nevertheless, regression models can only control for measured confounders, given that they are measured correctly. Most importantly, to reduce one of the biggest threats in pharmacoepidemiological studies, the “confounding by indication”, we employed a time-series within-individual comparison that controls for unobservable time-invariant confounding for each person during the followup. However, within-individual designs cannot exclude the potential existence of unmeasured time-varying confounding within each person, such as varying severity of symptoms of bipolar disorder (e.g., rapid cycling and mixed states), frequency of hospital visits, treatment site and other types of concomitant treatment. Alternatively, one can use negative controls to detect the absence of confounding or to estimate the magnitude of residual confounding.182 Analyses are repeated to examine the associations by using a negative control, either an exposure control presumed not to be causally associated with the outcome, or an outcome control that is not presumed to be causally associated with the exposure. If an association is observed in the primary analysis but not in negative controls, the causal interpretation of the association between the exposure and the outcome in the initial analysis is strengthened. Conversely, if a similar association is also observed in the analysis using negative controls, the initial analysis is likely to be biased. In Study III, we repeated our analyses by using thyroid hormone therapy as a negative control and found it was also associated with decreased rate of completed suicides. Therefore the residual confounding is not ignorable, and it threatened the validity for analyses of completed suicide. 188.8.131.52 Multiple testing and false positives When numerous statistical tests are performed, the possibility to observe a rare event just by chance increases, which will increase the likelihood to reject the null hypothesis incorrectly (“Type I error”). In Study IV, where we investigated the statistical associations for millions of genetic variates in parallel, it is crucial to guard against spurious discoveries due to false positives. The most widely used method is to correct the significance threshold by the number of tests performed— the so-called Bonferroni correction. Although Bonferroni correction has been criticized as being strict, researchers have adopted this adjustment in GWAS to reduce false positive findings by accounting for the number of relatively independent haplotype blocks in the genome. Type I error is particularly problematic in “omics” studies. Even after correction, it is still commonly seen that many significantly
associated variants reported fail to replicate. Meanwhile, we should bear in mind that the failure to replicate findings across populations may also be due to different LD patterns between SNPs. 5.2.3 Generalizability Generalizability, also called external validity, refers to the extent to which the findings of a study can be generalized to subjects other than the study population or in other situations. In Study I and II, the familial aggregations we tested are based on the whole Swedish population with register data. Considering the diagnostic criteria and social and genetic backgrounds varying across countries and districts, our estimates might be more representative to North Europeans but less transferable to other places distant from Sweden. In Study III, several issues should be kept in mind regarding the generalizability. (1) The estimated association of lithium and valproate medication could be dissimilar in Nordic countries and the United States. As discussed in the introduction, lithium is less prescribed in the US. Compared to Sweden, physicians in the US are more likely to be concerned about lithium’s toxicity and possible lethal outcomes in overdose when prescribing mood stabilizers to patients at high risk of suicide, which makes the confounding by indication more complicated. (2) The estimate of PAF only applies to our study population during the eight year period, since PAF estimates are based on exposure proportion and thus will vary depending on prescription rates and medication adherence. (3) The within-individual comparison also alters the generalizability. On the one hand, the estimation was based on individuals discontinuing their medication and having suicidal behavior during the follow-up. We don’t know how representative they are to the rest of patient group. On the other hand, logistic (or probit) regression for dichotomous outcomes, conditioning (stratifying or adjusting) on covariates associated with the outcome but not with the exposure, will probably increase the regression coefficient. This property is called “noncollapsibility”. It is not a bias but implies different interpretations for the population average estimates (marginal effects) and cluster-specific estimates (conditional effects). For GWAS, most of the overwhelming number of findings so far have been discovered in populations of European ancestry. The associations found in one European population are often more replicable in other European populations but are less successful when replication is attempted in subjects of non-European descent. This can be explained by many factors, such as low MAF markers, reduced portability of tag SNP and lower accuracy of genotype imputation in non-European populations.
5.2.4 Ethical considerations Studies included in this thesis mainly raise ethical considerations regarding the balance between the protection of personal integrity and the public health benefit. Study I-III used the national register data that do not require informed consent for research purposes in accordance to the current Swedish law. To study such data, researchers must submit ethical applications clarifying the specific research purpose.183 Only data related to the research questions will be derived from the registers and anonymized before access to the researcher. The linkage for encryption is destroyed afterwards and no direct contact is allowed between the researchers and participants.141 In the GWAS in Study IV, all projects involved obtained written or oral informed consent from the participants before the donation of blood samples and access to medical records. Only by gaining public confidence through safeguarding personal data and vulnerable participants can we keep the national register data as a sustainable goldmine for medical research.
(by Yu Liu)
“ see a universe consists of facts, logic and rationality, glowing in the darkness of unknown and underneath the disorder of sensitivity chaos, with the endless power of curiosity. ” Yu Liu’s illustration for the front cover, A Bipolar Universe 53
Bipolar disorder aggregates in families, with the strength increased with increasing genetic relatedness between relatives and the proband. For the etiology of bipolar disorder, both the pattern of aggregation and heritability estimation confirmed the important role of genetic factors. Several psychiatric conditions co-occur with bipolar disorder within individuals, and co-aggregate with bipolar disorder in their close relatives. Together with the estimated correlations, our results suggest pleiotropic effects of shared genetic determinants across different psychiatric disorders.
Bipolar disorder is in both the mood and psychotic spectra, but our results suggest a closer relationship to mood disorders.
Familial co-aggregation and genetic correlation confirmed the overlap in etiology between BDI and BDII. More importantly, each subtype was more likely to run in families than to cross subtypes within a family, which demonstrated that BDI and BDII exhibit differences in etiology but not merely in severity.
Medication with lithium but not valproate is associated with reduced rates of suiciderelated events. By controlling for all confounders constant within each individual during the study period, the association suggested a special anti-suicidal effect of lithium. No common genetic variants have been identified to be associated with response to lithium through our GWAS.
When we used lithium response to define a subtype of bipolar disorder (“lithiumresponsive BD”), one imputed variant reached genome-wide significance but failed in genotyping validation. Nevertheless, the considerable extent of variance explained by common genetic variants (SNP-h2) showed that the lithium-responsive BD is modestly heritable.
The precise feature of familial risks encourages the initiation of preventive family-based screening in primary care. It is increasingly acknowledged that genetic overlap is present between psychiatric diagnoses. The magnitude of shared genetic factors might contribute to establishing a more informative diagnostic system; thus yielding better treatments and medications for patients. Lithium should be considered for individuals with bipolar disorder with suspected suicidal intentions, although risk for suicide is only one of many concerns when providing clinical care. It may provide new insights into the pathophysiology of bipolar disorder by studying the specific response to lithium.
7 FUTURE PERSPECTIVES The complexity of bipolar disorder offers numerous puzzles in psychiatry. We only looked into a few of these and explored from specific directions with specialized approaches. Every time we move a step forward, we encounter more questions and are inspired with more ideas. During the course of this thesis, several questions are raised and hopefully could be continued in the future, with combinations of register-based data and genetic data. To provide a more complete understanding for each subtypes of bipolar disorder, I want to further examine the genetic correlations between each subtype and other psychiatric disorders (especially schizophrenia and major depressive disorder). This will, like Study I, facilitate understanding of the correlation between genetic overlap and overlap in clinical phenomenology. BDII is more likely to be misdiagnosed as major depressive disorder. Hopefully, future work could help refine this distinction and potentially improve the treatment strategies. The putative effect of lithium against suicidal behavior remains to be thoroughly investigated. It would be interesting to compare lithium with other drugs based on their specific efficacy. This could help understand the underlying mechanisms for lithium’s potential anti-suicidal effect. Moreover, GWAS on suicide in individuals with bipolar disorder will provide new insights. GWAS for lithium response invite follow-up. First, we aim to develop a better assessment of lithium response, possibly with dimensional measures combining the quality register, selfreport surveys and more detailed dispense records. Second, in the future information collection, how to harmonize our phenotype definition with others to promote collaborations is an important concern. Finally, individually associated alleles for lithium response may confer only modest effect, but the combined effect of all associated variants could be substantial. SNP profiling using a polygenic risk score may become clinically useful in prediction for therapeutic responses. Incorporating information on family history and environmental risk factors could further help population risk stratification and screening programs.
也许很多人都知道锂电池，却不知道锂也是一种药物。作为在宇宙大爆炸之初就合成 的三大元素之一，锂直到 1800 年才在斯德哥尔摩附近的群岛上以新矿石（透锂长石） 的身份被发现。1817 年，卡罗琳斯卡医学院教授，“现代化学之父”Jöns Jacob Berzelius 的私人实验室里，化学家 Arfwedson 从矿石中检测到了这个新元素的存在。 碳酸锂是治疗双向情感障碍的最古老也最主流药物。1949 年澳洲医生 John Cade 偶然 发现锂能使具有攻击行为的豚鼠变得安静，而将其带入现代精神治疗药物的视野。70 年过去，其他类别的精神治疗药剂大都有了长足的发展，而对于双向情感障碍的治疗 却进步甚微。不同于其他精神药物的多适用性，锂专用于双相情感障碍的维持治疗。 这一介导多种信号传导、具有复杂药理作用的药物，却在精神病学中有着最特异的临 床应用，这像一个悖论，又像能解开双向情感障碍复杂生理机制的一把钥匙。 锂似乎具有独特的“抗自杀效果”，我们的第三个研究结果支持了这一观点。值得一 提的是，观察性研究不同于随机对照实验，事物的相关性并不能代表因果性 ("Association is not causation")。例如，双向情感障碍患者中，接受锂治疗的病人往往自 杀率也很高，但这可能是因为医生们对有自杀倾向的病人会更多地采用锂治疗。在流 行病学中，我们往往需要发挥无限的想象力，假设是否有外部因素的存在，掩盖甚至 歪曲了研究因素与事件之间的真实关系。我们更需要多样化的实验设计，对可能的假 设进行检验，对潜在的混杂因素进行控制。流行病学对严密逻辑思维的需求使它被喻 为“医学中的哲学”。 课题四是全基因组关联分析（genome-wide association studies，GWAS），旨在探索人 类基因组中可能影响到锂治疗效果的遗传变异。准确来讲，这是一个失败的研究。原 始数据库的一个错误在文章发表后的一年被发现，曾经的候选基因与锂治疗效果不再 具有统计学意义上的关联。这一打击使我对自己的科研素质产生怀疑，对今后的研究 结果往往保持更谨慎的结论。局限性（limitation）已成为写论文稿最喜欢的部分，写 及此处，恨不得把整个分析过程中累积的问题和不确定性一股脑“跃然纸上”。 这本论文浓缩了我自 2012 年秋来到斯德哥尔摩后，在卡罗琳斯卡医学院（Karolinska Institute, KI）开展的工作。KI 建校于 1810 年，自 1901 年起担任诺贝尔生理医学奖的 评审工作。我所在的医学流行病学与生物统计系（MEB）是 1997 年成立的新科系， 已从最早的 40 人发展成为 300 人的宏大科研队伍，并负责采集和维护瑞典的生物样 本库和双胞胎信息，后者是世界上最大的双胞胎数据库。
ACKNOWLEDGEMENTS （致谢） Congratulations that you finally get into (well, skip into) this section — the most important part that I want to write the first and the most, long before this thesis started. It’s an incredible privilege that I have my PhD studies at the terrific MEB and meet a lot of brilliant and fantastic people shining during this journey. Finally I am given the chance to express my most sincere gratitude to all of you, despite that I’m tongue-tied, vocabulary-limited, and the words below are far from sufficient to express all my feelings. Paul Lichtenstein, my main supervisor, I am extremely delighted and privileged to study in the psychiatric epidemiology group. You are such a cool-appearance but warm-hearted gentleman (sometimes I cannot get your humor, but I know that’s my problem). You give me enough freedom in research, but generously provide me wise suggestions when I was stuck and encouraged me to move forward. I learned a lot from you, your professional efficiency, your speedy replies, your brainstorm in discussion, your thorough comments to my manuscripts and the kappa, your knowledgeable views and gift in targeting the core of problems. What I appreciate the most is that you train me to think about what I want to pursue in work and outside work. I feel amazingly lucky that I can never find a better supervisor than you. Mikael Landén, my fantastic co-supervisor. I am enormously honored to be your student and to work on your project on bipolar disorder. Intelligent as you always comes up with a lot of new research questions. Thank you for giving me insightful advices, showing your trust in me, and offering me opportunities to be part of different research projects. It’s a lot of fun in Gothenburg, the workshop and the group hanging-out! Sarah E Bergen, my friend-like co-supervisor. Thank you for bringing me to the world of genetics! I enjoy our chatting about everything. You are a language master. You patiently and wisely teach me how to respond to editors, to reviewers, to collaborate with researchers and to communicate science in writing. You never know how invaluable guidance and warmth you have given to me. Arvid Sjölander, my strong-mind co-supervisor. You are one of the best teachers showing us the beauty of epidemiology through the course of causal inference. Every question meets a solution in front of you. My energy and confidence are fully charged after discussion with you, in conjunction with a lot of knowledge to be digested. Fang Fang, my dear and wise mentor. Your optimistic mood is the best medicine pulling me out of sometimes negative thinking and back on track. Your feasible advices promote me to stop impractical wandering but develop mature thoughts. Not only your success in career, but also your positive outlook on work and life makes you the role model for many of us.
Robert Karlsson and Ralf Kuja-Halkola, you are like my external co-supervisors! You are always nice, supportive and knowledgeable to help everyone stuck in methodological and modelling problems. You are wonderful teachers, researchers and friends. Lu Yi, thanks for sharing your knowledge both in and outside work with me. It’s a treasure that you came to our group. It’s always hard to stop once I start chatting with you:) Many thanks to my study collaborators and co-authors, Henrik Larsson, Catarina Almqvist Malmros, Arianna Di Florio, Nick Craddock, Bo Runeson, Erik Joas, Christina Hultman, Helga Westerlind, Ali Manouchehrinia, Jan Hillert, Virginija Danylaité Karrenbauer, the leaders and members from The International Cohort Collection for Bipolar Disorder (ICCBD) Thank you for invaluable input and support. Thanks to Chang Zheng and Chen Qi, for your friendly contacts before my coming to Sweden. Qi, my thesis preparation will be much harder without your helps. Thanks to my past and current office mates. Erik Pettersson, I never forget the first day I came to MEB and how terrible the food I felt when you brought me to the Chinese restaurant. Thanks for sharing with me your research ideas, beautiful photos and candies. Chen Ruoqing, we came to MEB the same day, went through all the procedures together (even now) and share with each other so many joys. Like your singing, wish we travel together in the future! Lv Donghao, How can you know anything! If not having an office mate as know-all as you, I would not even find a good place to live. Yao Shuyang, still remember you invited us to your master graduation ceremony, you took me to the midsummer celebration, the cake you made for us is so yummy that finally every one asked for the recipe. Other past, present and external members of the psychiatric epidemiology “family” at MEB: Patrik Magnusson and Niklas Långström, without you we’ll miss a lot of fun; Amir Sariaslan, Martin Cederlöf, Alexander Viktorin, Agnieszka Butwicka, Agnes Ohlsson Gotby, Joanna Martin, Mark Taylor, Isabell Brikell, Laura Ghirardi, Andreas Jangmo, Carolyn Cesta, Mina Rydell, Tong Gong, Xu Chen, thank you for the rewarding group meetings and journal clubs; Christina Norrby and Marcus Boman, thanks for your guidance when I came to MEB knowing nothing about SAS and database; Johan Zetterqvist, thank you for sharing me with your experience and codes in medication studies; Carl Sellgren, Simon Kyaga, Yasmina Molero Samuelson, Emma Frans, Therese Ljung, Vilhelmina Ullemar, Barbro Sandin, Isabelle Kizling, Camilla Palm, Gunilla Hedlund, Monica Sagerstål, Charlotte Skoglund, Linda Halldner, Sebastian Lundstrom, Sven Sandin, Antti Latvala, Rozita Broumandi, it has been a great pleasure meeting you smart and kind people. Special thanks to Gunilla Sonnebring, for taking care of us all! Past and present members of Mikael Landén’s group in MEB and Gothenburg, Anders Juréus, Erik Pålsson, Erik Joas, Jessica Holmen-Larsson, Anne Snellman, Birgitta Ohlander, Joel Jakobsson, Annika Blom, Mathias Kardell, Christoph Abé, Benita
Gezelius, Timea Sparding, Milka Krestelica, Jessica Pege, Bozenna Iliadou, Oscar Svantesson. Special thanks to Jessica for taking care of me in Mölndal hospital. Patrick Sullivan and Cynthia Bulik, I deeply appreciate it that you let me join your group training. Thanks to all the wonderful teachers for their efforts. Your talks let me see an incredible future on psychiatric genetics. Yudi Pawitan, Marie Reilly, Shen Xia and Ning Zheng, your knowledge in biostatistics (and puzzles) are sooo cool, despite that I may not understand all of them. MEB is a wonderful working place. This is owing to a lot of people: The past and current head of department, Hans-Olov Adami, Nancy Pedersen, Henrik Grönberg and Paul Lichtenstein, and the vice dean Kamila Czene and Per Hall; Thanks to Amelie Plymoth, Camilla Ahlqvist, Gunilla Nilsson Roos for taking care of our PhD students; the TA and IT/Data group, including Rikard Öberg, Lennart Martinsson, Marie Jansson and Johan Rosén; Vivekananda Lanka the SAS master; Anna Johansson, wonderful teaching and beautiful Lucia songs; the PubMEB members who offer cozy pubs on Thursdays; the Christmas party and MEB-Day organizing committees; the Praesepe editorial board; …and thanks to “sleepless”( inexhausted) Jonas Ludvigsson for “feeding” us with limitless social activities. Thanks to Vilhelmina Ullemar, Johanna Holm, Carolyn Cesta, Isabell Brikell, Emilio Ugalde, Qing Shen who contribute a lot to our PhD group. The Thursday student seminars are always stimulating owing to Hannah Bower and Elisabeth Dahlqwist. Other past and present friends and colleagues at MEB, Kathleen Bokenberger, Alexander Ploner, Daniela Mariosa, Ida Karlsson, Bénédicte Delcoigne, Frida Lundberg, Mariam Lashkariani, Donal Barrett, Gabriel Isheden, Anna Plym, Elisa Longinetti, Favelle Lamb, Anna Plym, Gustaf Brander, Cecilia Radkiewicz, Alireza Salehi. I am grateful to have met you and got to know you. My excellent Chinese friends at MEB: Song Huan, Liu Bojing, Gong Tong, Wang Jiangrong, Chen Xu, He Wei, Zhu Janwei, Huang Tingting, Lei Jiayao, Shen Qing, Mei Wang, Yang Haomin, Zhan Yiqiang, Yang Fei, Liu Xingrong, Wang Chen, Huang Jiaqi, Liu Zhiwei, Suo Chen, Song Ci, Jiang Ruijingfang, Zhang Ruyue, I enjoy our lunch time chatting and after work gathering. Life becomes more colorful with you. Your companionship during my PhD journey makes it a joyful experience. I always feel grateful that how blessed I am to meet so many wonderful friends in Sweden. Liu Xuejin, thank you for the company, the skating, the travelling, the badminton playing, the cruise, the swimming and the chatting! May all the best to Chang Peiliang and you, together with xiao Yingzi; Xu Lidi and Zhang Peng, you are an amazing couple (now with Maimai, an amazing triple). Thank you for inviting me to witness your happiness by playing somewhat special characters
Zhang Xinhai, Gao Jiangning, Yin Gaoyu, Bai Haitong, Liu Hailong, Zha Yingying, I enjoyed all the happy time we have spent together. Ma Limin, Miao Xinyan, Jiang Xintong, great time hanging out with you in early days when I came to Sweden. Badminton family members: Sun Chengjun, Zhang Qiang, Han Hongya, Zou Honyan, Huang Genping, Hu Wei, Zhang Xiaonan, Guo Jing, Ma Yuanjun, Xuan Yang, Zhang Qiong, Wang Tongmei, Sun Tianyang, Yao Liqun, Li Xin, Luo Jiangnan, Li Xichen. Joining you change my life outside work. All weekends became: playing badminton + parties! Wang Rui, Ren Xiaoyuan, Tang Xiao, Liu Chang, Shi Tiansheng, Wang Yiqiao, Yang Dong, Yan Qinzi, Li Tian, Li Xiaofei, Wang Zhao, Meng Liesu, Zhu Lin, Long Mengni, Fu Jing, Hua Kai, wonderful memories with you great guys. Yuan Yuan and Ma Jun, you are a fantastic couple and artists. Friends outside Sweden: Huang Fei, we know each other since teenagers and you are my lifetime soulmate. Ma Yuqing and Zhao Jing, I expect to hang out with you in Xi’an and cannot wait to witness your happiness. Zhang Xin and Zhang Weiyi, chatting (texting) with you throughout these years enrich my mind. Liu Yu, your quest for perfection makes the front cover an artwork. Wang Quanli, you are my best teacher and friend at XJTU university. My gratitude to all the patients, volunteers, nurses and clinicians who contribute to projects of this thesis. Thanks to CSC, SLS, the Swedish Research Council for the financial supports. Yanpeng, once I was told I’m tender in soul, you are the safeguard in the entrance. 感谢你出 现在我的生命中，认识你之后时常感慨，你早干嘛去了:P~ miao~ HH forever. Last but most, to my dearest families for their unconditional love and never ending support.
Goodwin FK, Jamison KR. Clinical description. In: Manic-depressive illness. New York: Oxford University Press. 1990:15-55.
Oswald P, Souery D, Kasper S, et al. Current issues in bipolar disorder: a critical review. Eur Neuropsychopharmacol. 2007;17(11):687-95. Epub 2007/05/22.
Ketter TA. Diagnostic features, prevalence, and impact of bipolar disorder. J Clin Psychiatry. 2010;71(6):e14. Epub 2010/06/25.
Merikangas KR, Jin R, He JP, et al. Prevalence and correlates of bipolar spectrum disorder in the world mental health survey initiative. Arch Gen Psychiatry. 2011;68(3):241-51.
Alonso J, Petukhova M, Vilagut G, et al. Days out of role due to common physical and mental conditions: results from the WHO World Mental Health surveys. Molecular Psychiatry. 2011;16(12):1234-46.
Rybakowski JK. Lithium in neuropsychiatry: a 2010 update. The World Journal of Biological Psychiatry. 2011;12(5):340-48.
Grande I, Berk M, Birmaher B, Vieta E. Bipolar disorder. Lancet. 2016;387(10027):1561-72.
Vieta E, Phillips ML. Deconstructing Bipolar Disorder: A Critical Review of its Diagnostic Validity and a Proposal for DSM-V and ICD-11. Schizophrenia Bulletin. 2007;33(4):886-92.
WHO. The ICD-10 classification of mental and behavioural disorders: diagnostic criteria for research. . Geneva: World Health Organization. 1993.
American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders (5th ed.; DSM–5). Washington, DC: American Psychiatric Association. 2013.
Craddock N, Sklar P. Genetics of bipolar disorder. The Lancet. 2013;381(9878):165462.
Sullivan PF, Daly MJ, O'Donovan M. Genetic architectures of psychiatric disorders: the emerging picture and its implications. Nat Rev Genet. 2012;13(8):537-51.
Plomin R, Defries JC, Knopik VS, Neiderhiser JM. Behavioral genetics: 6th edition. New York: Worth. 2013:243.
Craddock N, Jones I. Genetics of bipolar disorder. J Med Genet. 1999;36(8):585-94. Epub 1999/08/28.
Barnett JH, Smoller JW. The genetics of bipolar disorder. Neuroscience. 2009;164(1):331-43. Epub 2009/04/11.
Lichtenstein P, Yip BH, Bjork C, et al. Common genetic determinants of schizophrenia and bipolar disorder in Swedish families: a population-based study. Lancet. 2009;373(9659):234-9. Epub 2009/01/20.
Bertelsen A, Harvald B, Hauge M. A Danish twin study of manic-depressive disorders. The British Journal of Psychiatry. 1977;130(4):330.
Kendler KS. Twin studies of psychiatric illness. Current status and future directions. Arch Gen Psychiatry. 1993;50(11):905-15. Epub 1993/11/01.
Kendler KS, Pedersen NL, Neale MC, Mathe AA. A pilot Swedish twin study of affective illness including hospital- and population-ascertained subsamples: results of model fitting. Behav Genet. 1995;25(3):217-32. Epub 1995/05/01.
McGuffin P, Rijsdijk F, Andrew M, Sham P, Katz R, Cardno A. The heritability of bipolar affective disorder and the genetic relationship to unipolar depression. Arch Gen Psychiatry. 2003;60(5):497-502. Epub 2003/05/14.
Kieseppa T, Partonen T, Haukka J, Kaprio J, Lonnqvist J. High concordance of bipolar I disorder in a nationwide sample of twins. Am J Psychiatry. 2004;161(10):1814-21. Epub 2004/10/07.
Seifuddin F, Mahon PB, Judy J, et al. Meta-analysis of genetic association studies on bipolar disorder. Am J Med Genet B Neuropsychiatr Genet. 2012;159b(5):508-18. Epub 2012/05/11.
Psychiatric GCBDWG. Large-scale genome-wide association analysis of bipolar disorder identifies a new susceptibility locus near ODZ4. Nat Genet. 2011;43(10):977-83.
Cross-Disorder Group of the Psychiatric Genomics Consortium. Genetic relationship between five psychiatric disorders estimated from genome-wide SNPs. Nat Genet. 2013;45(9):984-94.
Bulik-Sullivan BK, Loh P-R, Finucane HK, et al. LD Score regression distinguishes confounding from polygenicity in genome-wide association studies. Nat Genet. 2015;47(3):291-95.
Smith AL, Weissman MM. Epidemiology. In: Paykel ES, ed. Handbook of affective disorders. Edinburgh: Churchill Livingstone. 1992:111-29.
Smoller JW, Finn CT. Family, twin, and adoption studies of bipolar disorder. Am J Med Genet C Semin Med Genet. 2003;123C(1):48-58.
Yang J, Lee SH, Goddard ME, Visscher PM. GCTA: a tool for genome-wide complex trait analysis. Am J Hum Genet. 2011;88(1):76-82.
Ferreira MA, O'Donovan MC, Meng YA, et al. Collaborative genome-wide association analysis supports a role for ANK3 and CACNA1C in bipolar disorder. Nat Genet. 2008;40(9):1056-8. Epub 2008/08/20.
Yao S, Kuja-Halkola R, Thornton LM, et al. Familial liability for eating disorders and suicide attempts: Evidence from a population registry in sweden. JAMA Psychiatry. 2016;73(3):284-91.
Green EK, Hamshere M, Forty L, et al. Replication of bipolar disorder susceptibility alleles and identification of two novel genome-wide significant associations in a new bipolar disorder case-control sample. Mol Psychiatry. 2013;18(12):1302-07.
Muhleisen TW, Leber M, Schulze TG, et al. Genome-wide association study reveals two new risk loci for bipolar disorder. Nat Commun. 2014;5:3339.
Chen DT, Jiang X, Akula N, et al. Genome-wide association study meta-analysis of European and Asian-ancestry samples identifies three novel loci associated with bipolar disorder. Mol Psychiatry. 2013;18(2):195-205. Epub 2011/12/21.
Hou L, Bergen SE, Akula N, et al. Genome-wide association study of 40,000 individuals identifies two novel loci associated with bipolar disorder. Hum Mol Genet. 2016.
Cichon S, Mühleisen TW, Degenhardt FA, et al. Genome-wide Association Study Identifies Genetic Variation in Neurocan as a Susceptibility Factor for Bipolar Disorder. The American Journal of Human Genetics. 2011;88(3):372-81.
Cross-Disorder Group of the Psychiatric Genomics C, Genetic Risk Outcome of Psychosis C. Identification of risk loci with shared effects on five major psychiatric disorders: a genome-wide analysis. Lancet. 2013;381(9875):1371-9. Epub 2013/03/05.
Badner JA, Gershon ES. Meta-analysis of whole-genome linkage scans of bipolar disorder and schizophrenia. Mol Psychiatry. 2002;7(4):405-11. Epub 2002/05/03.
Faraone SV, Biederman J, Wozniak J. Examining the Comorbidity Between Attention Deficit Hyperactivity Disorder and Bipolar I Disorder: A Meta-Analysis of Family Genetic Studies. American Journal of Psychiatry. 2012;169(12):1256-66.
Larsson H, Rydén E, Boman M, Långström N, Lichtenstein P, Landén M. Risk of bipolar disorder and schizophrenia in relatives of people with attention-deficit hyperactivity disorder. The British Journal of Psychiatry. 2013;203(2):103.
Bulik-Sullivan B, Finucane HK, Anttila V, et al. An atlas of genetic correlations across human diseases and traits. Nat Genet. 2015;47(11):1236-41.
Sullivan PF, Magnusson C, Reichenberg A, et al. Family history of schizophrenia and bipolar disorder as risk factors for autism. Arch Gen Psychiatry. 2012;69(11):1099103. Epub 2012/07/04.
Simon NM, Otto MW, Wisniewski SR, et al. Anxiety disorder comorbidity in bipolar disorder patients: data from the first 500 participants in the Systematic Treatment Enhancement Program for Bipolar Disorder (STEP-BD). Am J Psychiatry. 2004;161(12):2222-9. Epub 2004/12/01.
Garno JL, Goldberg JF, Ramirez PM, Ritzler BA. Bipolar disorder with comorbid cluster B personality disorder features: impact on suicidality. J Clin Psychiatry. 2005;66(3):339-45. Epub 2005/03/16.
Regier DA, Farmer ME, Rae DS, et al. Comorbidity of mental disorders with alcohol and other drug abuse. Results from the Epidemiologic Catchment Area (ECA) Study. Jama. 1990;264(19):2511-8. Epub 1990/11/21.
McElroy SL, Kotwal R, Keck PE, Jr., Akiskal HS. Comorbidity of bipolar and eating disorders: distinct or related disorders with shared dysregulations? J Affect Disord. 2005;86(2-3):107-27. Epub 2005/06/07.
Dell'Osso B, Holtzman JN, Goffin KC, et al. American tertiary clinic-referred bipolar II disorder compared to bipolar I disorder: More severe in multiple ways, but less severe in a few other ways. J Affect Disord. 2015;188:257-62.
Coryell W, Endicott J, Reich T, Andreasen N, Keller M. A family study of bipolar II disorder. Br J Psychiatry. 1984;145:49-54.
Heun R, Maier W. The distinction of bipolar II disorder from bipolar I and recurrent unipolar depression: results of a controlled family study. Acta Psychiatr Scand. 1993;87(4):279-84.
Endicott J, Nee J, Andreasen N, Clayton P, Keller M, Coryell W. Bipolar II. Combine or keep separate? J Affect Disord. 1985;8(1):17-28.
Gershon ES, Hamovit J, Guroff JJ, et al. A family study of schizoaffective, bipolar I, bipolar II, unipolar, and normal control probands. Arch Gen Psychiatry. 1982;39(10):1157-67.
Charney AW, Ruderfer DM, Stahl EA, et al. Evidence for genetic heterogeneity between clinical subtypes of bipolar disorder. Transl Psychiatry (in press). 2016;6.
Webb RT, Lichtenstein P, Larsson H, Geddes JR, Fazel S. Suicide, hospitalpresenting suicide attempts, and criminality in bipolar disorder: examination of risk for multiple adverse outcomes. J Clin Psychiatry. 2014;75(8):e809-16.
Nordentoft M, Mortensen PB, Pedersen CB. Absolute risk of suicide after first hospital contact in mental disorder. Arch Gen Psychiatry. 2011;68(10):1058-64.
Schaffer A, Isometsa ET, Tondo L, et al. Epidemiology, neurobiology and pharmacological interventions related to suicide deaths and suicide attempts in bipolar disorder: Part I of a report of the International Society for Bipolar Disorders Task Force on Suicide in Bipolar Disorder. Aust N Z J Psychiatry. 2015;49(9):785-802.
Dennehy EB, Marangell LB, Allen MH, Chessick C, Wisniewski SR, Thase ME. Suicide and Suicide Attempts in the Systematic Treatment Enhancement Program for Bipolar Disorder. Journal of affective disorders. 2011;133(3):423-27.
Marangell LB, Bauer MS, Dennehy EB, et al. Prospective predictors of suicide and suicide attempts in 1,556 patients with bipolar disorders followed for up to 2 years. Bipolar Disord. 2006;8(5 Pt 2):566-75.
Schaffer A, Isometsa ET, Tondo L, et al. International Society for Bipolar Disorders Task Force on Suicide: meta-analyses and meta-regression of correlates of suicide attempts and suicide deaths in bipolar disorder. Bipolar Disord. 2015;17(1):1-16.
Pompili M, Gonda X, Serafini G, et al. Epidemiology of suicide in bipolar disorders: a systematic review of the literature. Bipolar Disord. 2013;15(5):457-90. Epub 2013/06/13.
Oquendo MA, Currier D, Liu SM, Hasin DS, Grant BF, Blanco C. Increased risk for suicidal behavior in comorbid bipolar disorder and alcohol use disorders: results from the National Epidemiologic Survey on Alcohol and Related Conditions (NESARC). J Clin Psychiatry. 2010;71(7):902-9. Epub 2010/07/30.
Saunders KE, Hawton K. Clinical assessment and crisis intervention for the suicidal bipolar disorder patient. Bipolar Disord. 2013;15(5):575-83. Epub 2013/04/10.
Goldberg JF, Garno JL, Leon AC, Kocsis JH, Portera L. Association of Recurrent Suicidal Ideation With Nonremission From Acute Mixed Mania. American Journal of Psychiatry. 1998;155(12):1753-55.
Holma KM, Haukka J, Suominen K, et al. Differences in incidence of suicide attempts between bipolar I and II disorders and major depressive disorder. Bipolar Disord. 2014;16(6):652-61. Epub 2014/03/19.
Grande I, Vieta E. Pharmacotherapy of Acute Mania: Monotherapy or Combination Therapy with Mood Stabilizers and Antipsychotics? CNS Drugs. 2015;29(3):221-27.
De Fruyt J, Deschepper E, Audenaert K, et al. Second generation antipsychotics in the treatment of bipolar depression: a systematic review and meta-analysis. J Psychopharmacol. 2012;26(5):603-17. Epub 2011/09/24.
Tohen M, Vieta E, Calabrese J, et al. Efficacy of olanzapine and olanzapinefluoxetine combination in the treatment of bipolar I depression. Arch Gen Psychiatry. 2003;60(11):1079-88. Epub 2003/11/12.
Sidor MM, MacQueen GM. An update on antidepressant use in bipolar depression. Curr Psychiatry Rep. 2012;14(6):696-704. Epub 2012/10/16.
Cipriani A, Furukawa TA, Salanti G, et al. Comparative efficacy and acceptability of 12 new-generation antidepressants: a multiple-treatments meta-analysis. Lancet. 2009;373(9665):746-58. Epub 2009/02/03.
Viktorin A, Lichtenstein P, Thase ME, et al. The risk of switch to mania in patients with bipolar disorder during treatment with an antidepressant alone and in combination with a mood stabilizer. Am J Psychiatry. 2014;171(10):1067-73.
Vieta E. Antidepressants in bipolar I disorder: never as monotherapy. Am J Psychiatry. 2014;171(10):1023-6. Epub 2014/10/02.
Colom F, Vieta E, Daban C, Pacchiarotti I, Sanchez-Moreno J. Clinical and therapeutic implications of predominant polarity in bipolar disorder. J Affect Disord. 2006;93(1-3):13-7. Epub 2006/05/03.
Geddes JR, Burgess S, Hawton K, Jamison K, Goodwin GM. Long-term lithium therapy for bipolar disorder: systematic review and meta-analysis of randomized controlled trials. Am J Psychiatry. 2004;161(2):217-22. Epub 2004/02/03.
McKnight RF, Adida M, Budge K, Stockton S, Goodwin GM, Geddes JR. Lithium toxicity profile: a systematic review and meta-analysis. Lancet. 2012;379(9817):7218. Epub 2012/01/24.
Hayes J, Prah P, Nazareth I, et al. Prescribing Trends in Bipolar Disorder: Cohort Study in the United Kingdom THIN Primary Care Database 1995–2009. PLOS ONE. 2011;6(12):e28725.
Cipriani A, Reid K, Young AH, Macritchie K, Geddes J. Valproic acid, valproate and divalproex in the maintenance treatment of bipolar disorder. Cochrane Database Syst Rev. 2013(10):Cd003196. Epub 2013/10/18.
Hayes JF, Marston L, Walters K, Geddes JR, King M, Osborn DP. Lithium vs. valproate vs. olanzapine vs. quetiapine as maintenance monotherapy for bipolar
disorder: a population-based UK cohort study using electronic health records. World Psychiatry. 2016;15(1):53-8. 78.
investigators B, collaborators, Geddes JR, et al. Lithium plus valproate combination therapy versus monotherapy for relapse prevention in bipolar I disorder (BALANCE): a randomised open-label trial. Lancet. 2010;375(9712):385-95.
Geddes JR, Miklowitz DJ. Treatment of bipolar disorder. The Lancet.381(9878):1672-82.
Johnson FN. The History of Lithium Therapy. London: Macmillan. 1984:1-4.
Boesgaard AM, Steigman G. Big Bang Nucleosynthesis: Theories and Observations. Annual Review of Astronomy and Astrophysics. 1985;23(1):319-78.
Shorter E. The history of lithium therapy. Bipolar disorders. 2009;11(0 2):4-9.
Cade JF. Lithium salts in the treatment of psychotic excitement. Med J Aust. 1949;2(10):349-52. Epub 1949/09/03.
Alda M. Lithium in the treatment of bipolar disorder: pharmacology and pharmacogenetics. Mol Psychiatry. 2015;20(6):661-70.
Poolsup N, Li Wan Po A, de Oliveira IR. Systematic overview of lithium treatment in acute mania. J Clin Pharm Ther. 2000;25(2):139-56. Epub 2000/06/10.
Girardi P, Brugnoli R, Manfredi G, Sani G. Lithium in Bipolar Disorder: Optimizing Therapy Using Prolonged-Release Formulations. Drugs R D. 2016.
Nivoli AMA, Colom F, Murru A, et al. New treatment guidelines for acute bipolar depression: A systematic review. Journal of Affective Disorders. 2011;129(1–3):1426.
Cipriani A, Hawton K, Stockton S, Geddes JR. Lithium in the prevention of suicide in mood disorders: updated systematic review and meta-analysis. Bmj. 2013;346:f3646. Epub 2013/07/03.
Tondo L, Isacsson G, Baldessarini R. Suicidal behaviour in bipolar disorder: risk and prevention. CNS Drugs. 2003;17(7):491-511. Epub 2003/05/20.
Baldessarini RJ, Tondo L, Hennen J. Treating the suicidal patient with bipolar disorder. Reducing suicide risk with lithium. Ann N Y Acad Sci. 2001;932:24-38; discussion 39-43. Epub 2001/06/20.
Lewitzka U, Severus E, Bauer R, Ritter P, Müller-Oerlinghausen B, Bauer M. The suicide prevention effect of lithium: more than 20 years of evidence—a narrative review. International Journal of Bipolar Disorders. 2015;3:15.
Guyatt GH, Oxman AD, Kunz R, Vist GE, Falck-Ytter Y, Schünemann HJ. What is “quality of evidence” and why is it important to clinicians? BMJ : British Medical Journal. 2008;336(7651):995-98.
Schulz KF, Altman DG, Moher D. CONSORT 2010 Statement: updated guidelines for reporting parallel group randomised trials. BMC Medicine. 2010;8(1):18.
Blanco C, Laje G, Olfson M, Marcus SC, Pincus HA. Trends in the treatment of bipolar disorder by outpatient psychiatrists. Am J Psychiatry. 2002;159(6):1005-10. Epub 2002/06/04.
Kessing LV, Vradi E, Andersen PK. Nationwide and population-based prescription patterns in bipolar disorder. Bipolar Disord. 2016;18(2):174-82.
Oquendo MA, Galfalvy HC, Currier D, et al. Treatment of suicide attempters with bipolar disorder: a randomized clinical trial comparing lithium and valproate in the prevention of suicidal behavior. Am J Psychiatry. 2011;168(10):1050-6.
Kessing LV, Sondergard L, Kvist K, Andersen PK. Suicide risk in patients treated with lithium. Arch Gen Psychiatry. 2005;62(8):860-6.
Baldessarini RJ, Tondo L, Davis P, Pompili M, Goodwin FK, Hennen J. Decreased risk of suicides and attempts during long-term lithium treatment: a meta-analytic review. Bipolar Disord. 2006;8(5 Pt 2):625-39.
Baldessarini RJ, Tondo L. Lithium and suicidal risk. Bipolar Disord. 2008;10(1):1145. Epub 2008/01/18.
Goodwin FK, Fireman B, Simon GE, Hunkeler EM, Lee J, Revicki D. Suicide risk in bipolar disorder during treatment with lithium and divalproex. JAMA. 2003;290(11):1467-73. Epub 2003/09/18.
Sondergard L, Lopez AG, Andersen PK, Kessing LV. Mood-stabilizing pharmacological treatment in bipolar disorders and risk of suicide. Bipolar Disord. 2008;10(1):87-94.
Baldessarini RJ, Tondo L. Suicidal risks during treatment of bipolar disorder patients with lithium versus anticonvulsants. Pharmacopsychiatry. 2009;42(2):72-5.
Hayes JF, Pitman A, Marston L, et al. Self-harm, Unintentional Injury, and Suicide in Bipolar Disorder During Maintenance Mood Stabilizer Treatment: A UK PopulationBased Electronic Health Records Study. JAMA Psychiatry. 2016;73(6):630-7.
Garnham J, Munro A, Slaney C, et al. Prophylactic treatment response in bipolar disorder: results of a naturalistic observation study. J Affect Disord. 2007;104(13):185-90. Epub 2007/04/20.
Rybakowski JK. Lithium in neuropsychiatry: a 2010 update. World J Biol Psychiatry. 2011;12(5):340-8. Epub 2011/03/03.
Grof P. Sixty years of lithium responders. Neuropsychobiology. 2010;62(1):8-16. Epub 2010/05/11.
Alda M, Grof P, Rouleau GA, Turecki G, Young LT. Investigating responders to lithium prophylaxis as a strategy for mapping susceptibility genes for bipolar disorder. Prog Neuropsychopharmacol Biol Psychiatry. 2005;29(6):1038-45. Epub 2005/06/11.
Grof P, Alda M, Grof E, Zvolsky P, Walsh M. Lithium response and genetics of affective disorders. J Affect Disord. 1994;32(2):85-95. Epub 1994/10/01.
Tighe SK, Mahon PB, Potash JB. Predictors of Lithium Response in Bipolar Disorder. Therapeutic Advances in Chronic Disease. 2011;2(3):209-26.
Grof P, Duffy A, Cavazzoni P, et al. Is response to prophylactic lithium a familial trait? J Clin Psychiatry. 2002;63(10):942-7. Epub 2002/11/06.
Berghofer A, Alda M, Adli M, et al. Long-term effectiveness of lithium in bipolar disorder: a multicenter investigation of patients with typical and atypical features. J Clin Psychiatry. 2008;69(12):1860-8. Epub 2008/11/26.
Passmore MJ, Garnham J, Duffy A, et al. Phenotypic spectra of bipolar disorder in responders to lithium versus lamotrigine. Bipolar Disord. 2003;5(2):110-4. Epub 2003/04/12.
Kruger S, Alda M, Young LT, Goldapple K, Parikh S, Mayberg HS. Risk and resilience markers in bipolar disorder: brain responses to emotional challenge in bipolar patients and their healthy siblings. Am J Psychiatry. 2006;163(2):257-64. Epub 2006/02/02.
Malhi GS, Tanious M, Das P, Coulston CM, Berk M. Potential mechanisms of action of lithium in bipolar disorder. Current understanding. CNS Drugs. 2013;27(2):135-53.
Pisanu C, Melis C, Squassina A. Lithium Pharmacogenetics: Where Do We Stand? Drug Dev Res. 2016.
Berridge MJ, Downes CP, Hanley MR. Lithium amplifies agonist-dependent phosphatidylinositol responses in brain and salivary glands. Biochem J. 1982;206(3):587-95. Epub 1982/09/15.
Hallcher LM, Sherman WR. The effects of lithium ion and other agents on the activity of myo-inositol-1-phosphatase from bovine brain. J Biol Chem. 1980;255(22):10896-901. Epub 1980/11/25.
Berridge MJ. Inositol trisphosphate and calcium signalling mechanisms. Biochim Biophys Acta. 2009;1793(6):933-40. Epub 2008/11/18.
Severino G, Squassina A, Costa M, et al. Pharmacogenomics of bipolar disorder. Pharmacogenomics. 2013;14(6):655-74. Epub 2013/04/11.
Hou L, Heilbronner U, Degenhardt F, et al. Genetic variants associated with response to lithium treatment in bipolar disorder: a genome-wide association study. Lancet. 2016.
Oedegaard KJ, Alda M, Anand A, et al. The Pharmacogenomics of Bipolar Disorder study (PGBD): identification of genes for lithium response in a prospective sample. BMC Psychiatry. 2016;16:129. Epub 2016/05/07.
Cruceanu C, Alda M, Turecki G. Lithium: a key to the genetics of bipolar disorder. Genome Med. 2009;1(8):79.
Rybakowski JK, Chlopocka-Wozniak M, Suwalska A. The prophylactic effect of long-term lithium administration in bipolar patients entering treatment in the 1970s and 1980s. Bipolar Disord. 2001;3(2):63-7. Epub 2001/05/03.
Manchia M, Adli M, Akula N, et al. Assessment of Response to Lithium Maintenance Treatment in Bipolar Disorder: A Consortium on Lithium Genetics (ConLiGen) Report. PLoS One. 2013;8(6):e65636. Epub 2013/07/11.
Grof P, Alda M. Discrepancies in the efficacy of lithium. Archives of General Psychiatry. 2000;57(2):191-91.
Baldessarini RJ, Tondo L. Does lithium treatment still work? Evidence of stable responses over three decades. Arch Gen Psychiatry. 2000;57(2):187-90. Epub 2000/02/09.
Kleindienst N, Greil W. Lithium in the long-term treatment of bipolar disorders. Eur Arch Psychiatry Clin Neurosci. 2003;253(3):120-5.
Kallner G, Lindelius R, Petterson U, Stockman O, Tham A. Mortality in 497 patients with affective disorders attending a lithium clinic or after having left it. Pharmacopsychiatry. 2000;33(1):8-13. Epub 2000/03/18.
Kessing LV, Hansen HV, Hvenegaard A, et al. Treatment in a specialised out-patient mood disorder clinic v. standard out-patient treatment in the early course of bipolar disorder: randomised clinical trial. Br J Psychiatry. 2013;202(3):212-9. Epub 2013/01/26.
Grossman J, Mackenzie FJ. The randomized controlled trial: gold standard, or merely standard? Perspect Biol Med. 2005;48(4):516-34. Epub 2005/10/18.
Lahey BB, D'Onofrio BM, Waldman ID. Using epidemiologic methods to test hypotheses regarding causal influences on child and adolescent mental disorders. J Child Psychol Psychiatry. 2009;50(1-2):53-62. Epub 2009/02/18.
Lao KSJ, Chui CSL, Man KKC, Lau WCY, Chan EW, Wong ICK. Medication safety research by observational study design. International Journal of Clinical Pharmacy. 2016;38(3):676-84.
Black N. Why we need observational studies to evaluate the effectiveness of health care. BMJ. 1996;312(7040):1215.
Strom BL. Pharmacoepidemiology, 3rd ed.: Chichester ; New York : Wiley; 2000.
Sjolander A, Frisell T, Kuja-Halkola R, Oberg S, Zetterqvist J. Carryover Effects in Sibling Comparison Designs. Epidemiology. 2016;27(6):852-8.
Ludvigsson JF, Otterblad-Olausson P, Pettersson BU, Ekbom A. The Swedish personal identity number: possibilities and pitfalls in healthcare and medical research. Eur J Epidemiol. 2009;24(11):659-67.
Ludvigsson JF, Almqvist C, Bonamy AK, et al. Registers of the Swedish total population and their use in medical research. Eur J Epidemiol. 2016;31(2):125-36.
Ludvigsson JF, Andersson E, Ekbom A, et al. External review and validation of the Swedish national inpatient register. BMC Public Health. 2011;11:450.
Ekbom A. The Swedish Multi-generation Register. Methods Mol Biol. 2011;675:21520.
Wettermark B, Hammar N, Fored CM, et al. The new Swedish Prescribed Drug Register--opportunities for pharmacoepidemiological research and experience from the first six months. Pharmacoepidemiol Drug Saf. 2007;16(7):726-35.
The National Board of Health and Welfare. Statistics on Causes of Death 2015. Sweden: The National Board of Health and Welfare. 2016;2016-8-4.
Magnusson PK, Almqvist C, Rahman I, et al. The Swedish Twin Registry: establishment of a biobank and other recent developments. Twin Res Hum Genet. 2013;16(1):317-29.
Lichtenstein P, De Faire U, Floderus B, Svartengren M, Svedberg P, Pedersen NL. The Swedish Twin Registry: a unique resource for clinical, epidemiological and genetic studies. J Intern Med. 2002;252(3):184-205.
Emilsson L, Lindahl B, Koster M, Lambe M, Ludvigsson JF. Review of 103 Swedish Healthcare Quality Registries. J Intern Med. 2015;277(1):94-136.
Haydeh O, Wennberg M, CarlswärdKjellin A, Gezelius B, Blom A, Landén M. The St. Göran bipolar project: a prospective longitudinal study on bipolar disorder. International Clinical Psychopharmacology. 2012;28:e63-e64.
Sellgren C, Landen M, Lichtenstein P, Hultman CM, Langstrom N. Validity of bipolar disorder hospital discharge diagnoses: file review and multiple register linkage in Sweden. Acta Psychiatr Scand. 2011;124(6):447-53.
LaFramboise T. Single nucleotide polymorphism arrays: a decade of biological, computational and technological advances. Nucleic Acids Res. 2009;37(13):4181-93.
Ragoussis J. Genotyping technologies for genetic research. Annu Rev Genomics Hum Genet. 2009;10:117-33.
Turner S, Armstrong LL, Bradford Y, et al. Quality control procedures for genomewide association studies. Curr Protoc Hum Genet. 2011;Chapter 1:Unit1 19.
Devlin B, Roeder K. Genomic control for association studies. Biometrics. 1999;55(4):997-1004. Epub 2001/04/21.
Marchini J, Howie B. Genotype imputation for genome-wide association studies. Nat Rev Genet. 2010;11(7):499-511.
Porcu E, Sanna S, Fuchsberger C, Fritsche LG. Genotype imputation in genome-wide association studies. Curr Protoc Hum Genet. 2013;Chapter 1:Unit 1 25.
Song JW, Chung KC. Observational Studies: Cohort and Case-Control Studies. Plastic and reconstructive surgery. 2010;126(6):2234-42.
Zimmerman R, Pal DK, Tin A, Ahsan H, Greenberg DA. Methods for assessing familial aggregation: family history measures and confounding in the standard cohort, reconstructed cohort and case-control designs. Hum Hered. 2009;68(3):201-8.
Hudson JI, Laird NM, Betensky RA, Keck PE, Jr., Pope HG, Jr. Multivariate logistic regression for familial aggregation of two disorders. II. Analysis of studies of eating and mood disorders. Am J Epidemiol. 2001;153(5):506-14.
Hudson JI, Javaras KN, Laird NM, VanderWeele TJ, Pope HG, Jr., Hernan MA. A structural approach to the familial coaggregation of disorders. Epidemiology. 2008;19(3):431-9.
Greenland S, Pearl J, Robins JM. Causal diagrams for epidemiologic research. Epidemiology. 1999;10(1):37-48.
VanderWeele TJ, Robins JM. Directed acyclic graphs, sufficient causes, and the properties of conditioning on a common effect. Am J Epidemiol. 2007;166(9):1096104.
Gershon ES, Guroff JJ. Information from relatives. Diagnosis of affective disorders. Arch Gen Psychiatry. 1984;41(2):173-80.
Stoltenberg SF. Coming to terms with heritability. Genetica. 1997;99(2-3):89-96.
Rijsdijk FV, Sham PC. Analytic approaches to twin data using structural equation models. Brief Bioinform. 2002;3(2):119-33.
Bollen KA. Structural equations with latent variables. New York: Wiley. 1989.
Maher B. Personal genomes: The case of the missing heritability. Nature. 2008;456(7218):18-21.
Visscher PM, Yang J, Goddard ME. A commentary on 'common SNPs explain a large proportion of the heritability for human height' by Yang et al. (2010). Twin Res Hum Genet. 2010;13(6):517-24. Epub 2010/12/15.
Hirschhorn JN, Daly MJ. Genome-wide association studies for common diseases and complex traits. Nat Rev Genet. 2005;6(2):95-108.
Reich DE, Lander ES. On the allelic spectrum of human disease. Trends Genet. 2001;17(9):502-10.
Sachidanandam R, Weissman D, Schmidt SC, et al. A map of human genome sequence variation containing 1.42 million single nucleotide polymorphisms. Nature. 2001;409(6822):928-33.
Bush WS, Moore JH. Chapter 11: Genome-Wide Association Studies. PLOS Computational Biology. 2012;8(12):e1002822.
Klein RJ, Zeiss C, Chew EY, et al. Complement factor H polymorphism in agerelated macular degeneration. Science. 2005;308(5720):385-9.
Veltman JA, Brunner HG. De novo mutations in human genetic disease. Nat Rev Genet. 2012;13(8):565-75.