Familial traits of bipolar disorder: A systematic review and meta‐analysis

Genetic studies of bipolar disorder (BD) have shown varied results, which is in part because of the heterogeneity of the disorder. Identifying clinical phenotypes of BD could reduce variability and benefit research. Since BD has a robust genetic component, studies can investigate clinical traits that cluster in families to identify phenotypes with a probable genetic basis.


| INTRODUCTION
Bipolar disorder (BD) is a severe psychiatric disorder characterized by depressive and manic or hypomanic episodes with periods of euthymic mood. It is among the most prevalent causes of disability worldwide. 1 Furthermore, it is estimated up to 20% of those affected with BD will end their life by suicide. 2 The heterogeneity of BD needs to be investigated in order to determine differences in etiology and treatment response. BD has a strong genetic basis, and a positive family history is its most robust risk factor. 3 Genome-wide analyses suggest that BD is a complex polygenic disorder, and they have identified many associated genes; though, results vary from study to study. 4 Defining more homogenous groups of BD could reduce the clinical variability and allow for more practical results from studies.
The identification of potential subtypes of BD through phenotypic classification has been suggested to reduce the heterogeneity. 5,6 Along these lines, multiple subtypes of BD have been indicated defined by characteristics such as psychotic features, 7 age of onset, 8 lithium treatment response, 9 and comorbid conditions, such as panic disorder. 10 As proposed by Robins and Guze, 11 the element of familiality is essential to validate subtypes in the absence of known etiology. The familial nature of a trait is a starting point for studies to establish whether there is a genetic origin. Subgroups can be formed based on familiality, reducing heterogeneity within groups, and the genetic contribution can be quantified. Because of this, familial clinical traits of BD may define phenotypic profiles that could be utilized in research areas such as genetics and treatment response.

| Aims of the study
In this paper, we present the results from a systematic review of the current literature on familial traits of BD in relation to potential subtypes. In addition, a metaanalysis of the familial effects was performed for the traits identified.

| Eligibility
Eligibility criteria included studies with adult participants (aged 18+) diagnosed with BD (type I, II, not-otherwisespecified) and/or schizoaffective disorder (SA). If participants diagnosed with major depressive disorder (MDD) were included, the analysis must have reported separate results from subjects with BD and SA. A measure of familiality was required for the trait of interest (e.g., prevalence of the feature in affected relatives, similarity of characteristic between relatives). We included traits of BD that could easily be clinically measured; therefore, biological markers and neuropsychological traits were excluded. Where relatives with a trait (such as comorbid condition) were used, the relatives must have had an eligible diagnosis. Additionally, we included only studies published in English.

| Sources
We searched the PubMed, PsycINFO, and Embase databases for literature dating from inception until March 7, 2022. Additional studies were identified through handsearching references.

| Search
The following search was used in all databases: (Bipolar disorder) AND (familial* OR family history OR pedigree OR genetic) AND (aggregate* OR heritable OR heritability OR cluster* OR co-aggregat* OR coaggregat* OR

Summations
• Fourteen traits were identified, and seven were eligible for meta-analysis.

| Study selection
The screening was conducted independently by two reviewers. Each reviewer determined eligibility through screening titles and abstracts, after which, the decisions of each reviewer were compared. Consensus for conflicts was reached through discussion. Each reviewer then screened the full texts of the remaining studies, and again, consensus was reached for any conflicts. For studies with overlapping samples, the study with a larger sample was selected.

| Data extraction and analysis
Reviewers individually extracted relevant data from the remaining studies and compared results. Any discrepancies were discussed and resolved. Extracted data included author details, year of publication, sample size, groups compared, trait examined, definition for trait, effect sizes, and their variance or confidence intervals. Study quality was assessed by each reviewer independently, using the Newcastle-Ottawa Scale. 12 Most studies reported their findings in one of three ways: odds ratios (OR), raw data, or intraclass correlations (ICC). We chose to present our results in ORs since they are common in this type of research. ORs were calculated from raw data or statistical methods such as mixed effects logistic regression. In contrast, the ICCs in the studies included here were calculated using a one-way random effect model. ORs can be interpreted as an association between exposure and outcome 13 ; exposure in this case would be having a relative with the trait of interest. An OR >1.00 suggests the outcome (having the trait of interest or not) is associated with the presence or absence of the trait of interest in the relative. However, ICCs are a continuous outcome that measures the similarity between two groups (ie., the similarity between affected relatives). 14 Because of the difference in statistical methods between studies, a formula was derived to convert ICCs to ORs (see Supporting Information).
Random effect meta-analysis with inverse-variance weighting was conducted using the R meta package 15 for each identified trait where two or more studies were available. The heterogeneity of studies was measured by the I 2 statistic and corresponding p-value, and publication bias was assessed with funnel plots.

| RESULTS
The search resulted in 1117 citations after removing duplicates, then 1040 articles were excluded based on irrelevance of titles and abstracts through the screening process. The remaining 77 articles were screened in full text, leaving 16 articles for our analyses (Figure 1). The assessment of study quality indicated half the articles were good quality and half were of poor quality, according to the Agency for Healthcare Research and Quality Methods standards (Table S1). 16 The studies were published from 1984 to 2021, with a median publication year of 2005. There were just two articles from the last decade. The included articles had sample sizes ranging from 27 to 5046, with a median sample of 421 individuals. Most studies used participants from North America or the United Kingdom. A few smaller samples were from Austria, Germany, Italy, and New Zealand, and the largest sample (N = 5046) was from Taiwan.

| Age of onset
The effect of familiality for the age of illness onset (2 studies, N = 810) was significant (OR: 4.50; 95% CI: [3.25, 6.22]; p < 0.001), indicating that persons with BD are much more likely to have a similar age of onset as their affected relative (I 2 = 0%, p = 0.95; Figure 2A).

| Bipolar type
Type of BD (2 studies, N = 5536) showed a significant positive effect of familiality (OR: 2.05 [1.50, 2.79], p < 0.001). The finding suggested that affected relatives of individuals with BD are twice as likely to have the same type (I vs. II) than unrelated individuals. However, there was evidence of significant heterogeneity between studies (I 2 = 92%, p < 0.01; Figure 2B).

| Lithium response
One study (N = 64) of lithium treatment response found a familial effect (OR: 3.71 [1.28, 10.82]; p = 0.016), indicating that the relatives of good lithium responders were over 3 times more likely to respond well to lithium themselves, compared with an unrelated individual.

| Polarity at onset
Polarity of the mood episode at illness onset had one study (N = 971), which showed a significant result (OR: 1.17 [1.03, 1.34]; p = 0.019). The study found individuals with BD were slightly more likely to have the same polarity of mood episode (ie., depressive, manic, or mixed) at illness onset as their affected relatives.

| Psychosis
The familiality of features of psychosis (4 studies, N = 1995) was significant (OR: 2.20 [1.51, 3.20]; p < 0.001). This suggested individuals with BD are more than twice as likely to experience features of psychosis if their affected relative has had features of psychosis, compared with unrelated individuals. There was evidence of heterogeneity (I 2 = 54%, p = 0.09; Figure 2C).

| Mood-incongruent psychosis
There was one study (N = 784) of the familiality of mood-incongruent psychosis which reported a significant result (OR: 2.52 [1.66, 3.83]; p < 0.001). The finding asserts that relatives of individuals with BD who experienced mood-incongruent psychosis are 2.5 times as likely to experience mood-incongruent psychosis as unrelated individuals.

| Puerperal psychosis
A study of puerperal psychosis (N = 152) reported a significant effect of familiality (OR: 6.54 [2.55, 16.77]; p < 0.001), meaning the affected female relatives of a woman with BD who developed puerperal psychosis were 6.5 times more likely to experience puerperal psychosis, compared with someone without the family history. Weighted by inverse-variance method. OR, odds ratio; CI, confidence interval. Significant heterogeneity defined as p < 0.10.

| Obsessive compulsive disorder
One study (N = 850) found affected relatives of individuals with BD and comorbid OCD were significantly more likely to have comorbid OCD as well (OR: 3.10 [1.31; 7.09]; p = 0.009), compared with unrelated individuals.
Of the seven traits that were meta-analyzed, five had significant heterogeneity between studies. Age of onset and alcoholism were the only two traits with minimal heterogeneity (Figure 2A, F). This is even though most studies individually had significant results demonstrating effects in the positive direction, supporting familiality. In addition, the examination of publication bias showed bias is likely present for all seven traits except suicide attempt ( Figures S1-S8).

| DISCUSSION
We identified 16 studies that described 14 traits that could define potential subtypes of BD. Half of these traits (lithium response, polarity at onset, mood-incongruent psychosis, puerperal psychosis, and comorbid OCD, social anxiety disorder, and specific phobia) were reported only in a single study. Therefore, the random effects meta-analysis was conducted on the remaining seven traits. The features with the largest effects were puerperal psychosis (OR: 6.54), age of onset (OR: 4.50), and lithium response (OR: 3.71). Puerperal psychosis and lithium response were represented by one study each and had wide confidence intervals, 95% CI: [2.55, 16.77] and [1.28, 10.82], respectively. In contrast, age of onset had two studies with minimal heterogeneity, and the 95% confidence interval was narrower, [3.25, 6.22]. The metaanalysis results for rapid cycling (OR: 4.95 [0.96, 25.40]) and suicide attempt (OR: 1.01 [0.65, 1.67]) did not reach significance, p = 0.055 and p = 0.879, respectively. However, there were limitations that should be considered when interpreting our results. Two significant limitations were the overall lack of literature and the heterogeneity of studies. The inclusion criteria were not particularly strict, but many papers were excluded for reasons including not reporting statistics for BD samples separate from those with MDD. Our aim was to investigate familial traits of BD, and while the current diagnostic classification systems are not perfect, the presence of mania or hypomania is a defining feature of BD. Because of this, we did not include subjects diagnosed with MDD. Furthermore, various studies measured familiality based on a positive family history of the trait but did not state the diagnoses of the relatives at all. Other studies were excluded because they used the same data for their analysis, most often this was from the NIMH database. Unfortunately, there were previously suggested subtypes of BD that were not included in the review because they did not meet the inclusion criteria (ie., Not enough information reported, no familial measure) such as irritable-type mania 30 and predominant polarity of episodes. 31 Lastly, 13 of the 16 studies were from 2008 or older, meaning only three eligible studies were published in the following span of 14 years. This gap appears to coincide with the development of rapid, efficient, and inexpensive genotyping technology. 32 The shift of focus from family studies to genetic studies likely contributed to the lack of published literature.
In addition to the quality of the studies, the heterogeneity between studies made the interpretation of the results difficult. There were many combinations of BDI, BDII, BD not-otherwise-specified, and SA for eligible diagnoses, so for each trait, the studies varied in their definition of BD. Similarly, the definition of the trait of interest varied in some cases, the most extreme example being rapid cycling. Rapid cycling was defined as four distinct mood episodes in a 12-month period by Saunders et al 7 In contrast, Pritz and Mitterauer 24 defined rapid cycling as clear-cut mood shifts between manic and depressive phases, lasting several hours to 2 weeks. The inconsistent definition may be a factor in the non-significant finding from the meta-analysis.
There was one particularly unexpected result. We anticipated seeing a significant finding regarding suicide attempts because suicide behavior is well-established as familial, 33 but we did not. This is despite suicide attempt being the single trait that appeared free of publication bias in this case. Most studies of familial suicide behavior used a positive family history as their measure, 34 but the diagnoses of the relatives either included MDD or were not used for analysis. It is possible that suicide behavior is familial in affective disorders, but BD does not contribute to the familiality, specifically. On the other hand, the meta-analysis for suicide attempt showed significant heterogeneity, which may have affected the results.
It is clear from our results that more studies investigating familial traits of BD are needed. The shift back to more comprehensive phenotypic analyses including family studies could be the answer for improving the utility of results from future genetic studies. Other areas of medicine have had success using family studies in this manner. For example, inflammatory bowel disease (IBD) includes Crohn's disease and ulcerative colitis; two separate conditions with very similar symptoms. 35 A higher prevalence of both diseases was noted among the relatives of IBD patients, and population studies later supported this observation. 36 Further, it was determined that ulcerative colitis and Crohn's disease each cluster in families at different rates, pointing to genetic and/or environmental differences. 37 Indeed, a genetic distinction between ulcerative colitis and Crohn's disease has since been reported, and notably, the genetic differences suggest Crohn's disease can be further divided into two subtypes, Ideal and colonic. 38 We suggest a similar process should be used to explore BD.
The phenotypes we identified should be followed up on, and their validity as subtypes should be tested in areas such as treatment response and genetic associations. The issue of variability between the studies could be helped by developing a standard or set of guidelines for research into phenotypic classification. The guidelines could give recommended methods of analysis, encourage inclusion of different types of BD, and indicate the results and other information to be included in published papers. An effort to include various ethnicities and minorities should be made as well to increase generalizability. The heterogeneity of BD is a challenge for researchers, but the identification of valid subtypes can help future genetic studies in the identification of genes related to BD as well as be applied clinically where treatment response is shown to differ.