Bipolar affective disorder is characterised by extreme depressive and manic affective states, which are often associated with adverse outcomes, including reduced functioning, impaired quality of life and increased risk of death by suicide. Reference Sanchez-Moreno, Martinez-Aran, Tabares-Seisdedos, Torrent, Vieta and Ayuso-Mateos1–Reference Harris and Barraclough3 Response to treatment is limited, with high rates of relapse. Reference Simhandl, Konig and Amann4 A better understanding of the risk factors for bipolar disorder is vital for refining detection and intervention strategies. Although research has typically focused on the biogenetic determinants of bipolar symptoms, environmental risk factors are increasingly being considered. Reference Etain, Henry, Bellivier, Mathieu and Leboyer5 This review and meta-analysis explore the association between bipolar disorder and childhood adversity.
Childhood adversity is associated with a variety of negative outcomes in the general population. Reference Arnow6 In individuals with bipolar affective disorder it has been linked to increased mood cycling, greater numbers of affective episodes and the presence of psychosis. Reference Etain, Aas, Andreassen, Lorentzen, Dieset and Gard7,Reference Upthegrove, Chard, Jones, Gordon-Smith, Forty and Jones8 However, the question of whether childhood adversity relates to the development of this disorder remains unresolved. Previous reviews have observed high rates of adversity in many, but not all, samples of people with bipolar disorder. Reference Daruy-Filho, Brietzke, Lafer and Grassi-Oliveira9–Reference Alloy, Abramson, Urosevic, Walshaw, Nusslock and Neeren14 To date, no research has attempted to integrate empirical findings using meta-analytic methods. To do so would provide a more rigorous method for testing the null hypothesis, and also allow for consideration of the size and consistency of the effects.
Authors have proposed that emotional abuse and neglect may convey greater risk of bipolar disorder than other forms of maltreatment such as sexual or physical abuse. Reference Etain, Henry, Bellivier, Mathieu and Leboyer5 Comparison of effect sizes for different forms of adversity might help to clarify whether specific adversity subtypes are more strongly related to bipolar symptoms. Meta-analytic approaches might also elucidate whether childhood adversity is associated with a particular form of bipolar disorder. Type 1 bipolar disorder is characterised by periods of mania (episodes of extremely elated mood, arousal and levels of activity, often in the presence of psychosis), whereas type 2 disorder presents only attenuated symptoms of mania with limited impact on functioning (hypomania). Given the evidence for an association between adversity and severe psychopathologic disorder, characterised by psychotic symptoms, Reference Varese, Smeets, Drukker, Lieverse, Lataster and Viechtbauer15 levels of childhood adversity may be elevated in patients with type 1 disorder. Lastly, diagnoses of major depression and schizophrenia appear more likely in individuals with a history of childhood adversity. Reference Varese, Smeets, Drukker, Lieverse, Lataster and Viechtbauer15,Reference Rubino, Nanni, Pozzi and Siracusano16 It is possible that childhood maltreatment is related to one particular form of psychiatric disorder. The final and exploratory aim of this review was therefore to compare rates of childhood adversity in individuals diagnosed with bipolar disorder with those in people diagnosed with schizophrenia or major depression.
We examined three a priori hypotheses: first, rates of childhood adversity would be elevated in samples with bipolar disorder compared with non-clinical controls; second, effect sizes for emotional abuse and neglect would be higher than for other forms of adversity; and third, rates of childhood adversity would be greater in individuals with type 1 disorder compared with type 2. We made no prediction regarding rates of childhood adversity in bipolar disorder compared with the other clinical samples.
The review was carried out in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) standards. A systematic search of four databases (Medline, EMBASE, PsycINFO and Web of Science) identified peer-reviewed articles published between January 1980 and October 2014. We used blocks of search terms pertaining to bipolar disorder (bipolar, mani*, cyclothymi*, manic-depressi* OR hypomani*) AND childhood adversity (child abuse, physical abuse, sexual abuse, psychological abuse, emotional abuse, neglect*, trauma*, advers*, maltreat*, bully*, bullied, victim* OR parental loss). The search terms were partly adapted from past reviews, Reference Maniglio10,Reference Maniglio11,Reference Varese, Smeets, Drukker, Lieverse, Lataster and Viechtbauer15 and where possible were ‘exploded’ in the field of Bipolar Disorder. The search in Web of Science was restricted to the areas of psychiatry and psychology by field. In addition to the systematic search, we screened the reference lists of the included manuscripts and previous reviews. Reference Etain, Henry, Bellivier, Mathieu and Leboyer5,Reference Daruy-Filho, Brietzke, Lafer and Grassi-Oliveira9–Reference Alloy, Abramson, Urosevic, Walshaw, Nusslock and Neeren14,Reference Fisher and Hosang17 We also examined journal articles citing at least one of the included studies. Where data were insufficient to assess eligibility or calculate an effect size, further information was requested from the study's corresponding author.
The review included case–control studies (comparing two existing groups distinguished by a defining outcome, i.e. bipolar status v. control) and epidemiological (prospective and cross-sectional) studies, where a quantitative measure of childhood adversity was administered to individuals with a formal diagnosis of bipolar disorder according to the DSM (DSM-III or later) or ICD (ICD-9 or -10). We defined childhood adversity as the experience of neglect, abuse, bullying or the loss of parents before the age of 19 years. Studies exploring loss through separation (e.g. divorce of parents), expressed emotion and/or stressful life events occurring in adulthood (after the age of 18 years) were not included. We excluded relatively common parenting practices (e.g. spanking, shouting), as these were assumed to be subject to cultural variability. Also excluded were case-note reviews that opportunistically assessed – rather than systematically measured – childhood adversity, owing to the increased likelihood of response bias. When both 12-month and lifetime diagnoses were provided, the latter were selected for effect size extraction. Reference Scott, Smith and Ellis18 Only articles published in peer-reviewed English-language journals were included in the analysis.
Only studies with at least one eligible control sample were included. These samples were defined a priori as comprising healthy individuals without an identified DSM or ICD diagnosis (in the epidemiological studies, this was defined as respondents known to be free of the outcome of interest, i.e. bipolar disorder), and individuals with a DSM or ICD diagnosis of major depression or non-affective psychosis (e.g. schizophrenia, schizoaffective disorder, schizophreniform disorder or delusional disorder).
Screening and data extraction
The lead author (J.P.C.) screened articles in three stages: at title level, abstract level and article level. A third of titles (1800) were double-rated separately by a postgraduate researcher, with adequate levels of agreement (95%, κ = 0.65). All of the abstracts (446) were double-rated with similarly high levels of agreement (87%, κ = 0.71). The majority of discrepancies were due to the primary coder (J.P.C.) being overly inclusive. Two authors extracted data and calculated effect sizes using a data spreadsheet. The intraclass correlation (ICC) between the two sets of effect sizes indicated high levels of agreement (ICC = 0.98, P<0.001). For the four cases where the primary authors were in disagreement, the wider team arbitrated. Extracted data included study and effect size descriptors. When possible the authors extracted binary (e.g. frequency tables, percentages), as opposed to d-family (e.g. means, standard deviations), effect sizes based on the use of odds ratio as the overall metric.
Methodological quality was explored using the Newcastle–Ottawa Quality Assessment Scale (NOQAS), Reference Wells, Shea, O'Connell, Peterson, Welch and Losos19 which assesses the selection and comparability of the samples and the suitability of the adversity exposures. Gender was selected as the most important covariate or matching criterion, given the studies showing greater levels and impact of childhood adversity in women compared with men. Reference Tolin and Foa20 Quality ratings were based on the effect sizes of interest, rather than other analyses reported in the papers. Independent, masked quality ratings by a postgraduate researcher demonstrated good interrater reliability with the lead author (ICC = 0.83, P<0.001).
We used Comprehensive Meta-Analysis (version two) to compute effect sizes and conduct the analyses. All effect sizes were converted to odds ratios to aid interpretation of the results. Effects were integrated using random effects meta-analysis. Visual inspection of funnel plots and regression tests of funnel plot asymmetry (Egger's test) established the presence of publication and selection bias. Where selection bias was deemed likely, Duval & Tweedie's trim and fill method was employed to identify and correct for hypothetically missing effects.
The analysis consisted of four stages. In stage one we considered the overall effects from studies comparing bipolar and non-clinical samples on measures of childhood adversity. This analysis focused on the association between childhood adversity and bipolar disorder regardless of adversity type, and considered both single (e.g. sexual abuse) and multiple (e.g. sexual and emotional abuse) exposures. When extracting data in the presence of more than one measure of adversity we used the most global or wide-reaching assessment (e.g. total levels of adversity). Where this information was unavailable, we contacted the corresponding author of the primary manuscript to request information regarding an aggregated effect. In the absence of this information, we calculated separate effect sizes for each type of adversity, which were then aggregated in the main analysis. The second stage of analysis examined independent associations between different types of exposures and bipolar disorder. In the third stage overall effects were extracted for studies that compared childhood adversity between samples with type 1 and type 2 bipolar disorder. Finally, we independently examined differences in childhood adversity between bipolar disorder and other psychiatric groups (major depression, schizophrenia).
Some manuscripts contained the results of both the unadjusted analyses and those adjusting for covariates. In order to increase comparability among the eligible studies we included the unadjusted results in the main analyses and then conducted a sensitivity analysis with the adjusted effects. In the presence of multiple levels of adjustment we included the analysis with the largest number of demographic and/or clinical covariates. The majority of the aforementioned analyses explored the impact of childhood adversity generally, rather than the specific effects of adversity subtypes over and above the other forms of adversity. Therefore, we did not include effects that examined the impact of exposures while controlling for other types of childhood adversity (e.g. Stikkelbroek et al). Reference Stikkelbroek, Prinzie, de Graaf, ten Have and Cuijpers21 The full review protocol (CRD42015017201) is available through the PROSPERO website (http://www.crd.york.ac.uk/PROSPERO).
The screening procedure is summarised in Fig. 1 and the characteristics of the included articles are given in online Table DS1. Reference Scott, Smith and Ellis18,Reference Stikkelbroek, Prinzie, de Graaf, ten Have and Cuijpers21–Reference Gilman, Ni, Dunn, Breslau, McLaughlin and Smoller26,Reference Agid, Shapira, Zislin, Ritsner, Hanin and Murad28–Reference Perna, Vanni, Di Chiaro, Cavedini and Caldirola48 Eleven authors provided clarification or further information from which to generate an effect size. Only 11 studies reported the exact prevalence of childhood adversity within bipolar samples, which ranged from 8% (Laursen et al) to 77% (Fowke et al), with a weighted average exposure of 10.5%. Reference Laursen, Munk-Olsen, Nordentoft and Bo22,Reference Fowke, Ross and Ashcroft23 This estimate includes parental loss (4 studies), sexual abuse (3 studies) and composite adversity measures (4 studies). Thirteen case–control and six epidemiological studies were included in the main analysis. The case–control studies included 1259 cases and 1118 controls, whereas the epidemiological studies surveyed over 2.1 million respondents. The epidemiological research included three population-based cross-sectional design studies, Reference Stikkelbroek, Prinzie, de Graaf, ten Have and Cuijpers21,Reference Afifi, MacMillan, Boyle, Taillieu, Cheung and Sareen24,Reference Molnar, Buka and Kessler25 two retrospective cohort design studies, Reference Scott, Smith and Ellis18,Reference Laursen, Munk-Olsen, Nordentoft and Bo22 and one quasi-prospective study. Reference Gilman, Ni, Dunn, Breslau, McLaughlin and Smoller26 The latter examined childhood adversity as a predictor of transition to psychosis over a 3-year period in adulthood. The cohort design studies linked data on current diagnosis to registers on parental loss and child protection status. Reference Scott, Smith and Ellis18,Reference Laursen, Munk-Olsen, Nordentoft and Bo22 The most commonly used assessment of adversity in the case–control studies (seven studies) was the Childhood Trauma Questionnaire, a 28-item self-report measure of emotional and physical abuse, emotional and physical neglect and sexual abuse. Reference Bernstein and Fink27 Measures of childhood adversity in the epidemiological studies were generally single items derived from validated measures.
Stage one investigated the overall association between childhood adversity and bipolar disorder. Figure 2 shows the ORs for each of the included studies, and the aggregated effects of childhood adversity on bipolar disorder. The analysis showed an overall effect of 2.63 (95% CI 2.00–3.47, P<0.001), suggesting that individuals with bipolar disorder are 2.6 times more likely to have experienced childhood adversity when compared with a non-clinical control group. Similar effect sizes were observed for the case–control (OR = 2.88, 95% CI 2.04–4.06, P<0.001) and epidemiological studies (OR = 2.24, 95% CI 1.40–3.57, P = 0.001). There was no significant difference (Q(1) = 0.74, P = 0.391) in the strength of the effect sizes between the two subgroups.
Heterogeneity was examined using the Q-test and I 2 statistics. Results showed that the strength of the relationship between childhood adversity and bipolar disorder varied considerably across studies (Q(18) = 79.53, P<0.001), with 77% of the observed dispersion attributable to true statistical heterogeneity. This level of heterogeneity is generally thought to be high and should be considered when interpreting the results.
Selection bias analyses
Regarding publication bias, funnel plots of standard error against log odds ratios indicated a roughly symmetrical distribution of studies around the mean effect sizes. When combining the case–control and epidemiological literature the result of Egger's test was non-significant (β = 0.12, s.e. = 1.08, P = 0.456) indicating no evidence of publication or selection bias. Duval & Tweedie's trim and fill found two hypothetical missing studies, which brought the imputed OR to 2.47 (95% CI 1.8–3.1).
One-study-removed analysis suggested that the withdrawal of any particular study would not greatly alter the results. Three of the epidemiological studies provided effect sizes adjusted for covariates in addition to unadjusted scores. Repeating the analysis using adjusted scores yielded similar results (OR = 2.58, 95% CI 1.96–3.36, P<0.001) with equivalent levels of statistical heterogeneity (Q(18) = 79.2, P<0.001, I 2 = 77.27). This was also true when including only the epidemiological studies in the analysis (OR = 2.14, 95% CI 1.36–3.39, P = 0.001).
In stage two we examined associations between specific adversity subtypes and bipolar disorder. Table 1 shows the results of the analyses exploring whether specific types of childhood adversity are elevated in bipolar disorder. Grandin et al and Neeren et al both reported analyses from the Longitudinal Investigation of Bipolar Spectrum Disorders project; Reference Grandin, Alloy and Abramson31,Reference Neeren, Alloy and Abramson49 we selected information from the paper by Neeren et al because it specifically reported effects pertaining to the impact of adversity subtypes. The results of these separate meta-analyses showed significant effects of all childhood adversity subtypes, with the exception of parental loss, on bipolar disorder. Emotional abuse showed the strongest effect (OR = 4.04, 95% CI 3.12–5.22, P<0.001).
|Odds ratios||Heterogeneity tests|
|k||OR (95% CI)||P||I 2||Q||P|
|Physical abuse||12||2.86 (2.22–3.69)||<0.001||70||36.55||<0.001|
|Sexual abuse||12||2.58 (2.08–3.20)||<0.001||35||16.94||0.109|
|Emotional abuse||9||4.04 (3.12–5.22)||<0.001||23||10.40||0.238|
|Physical neglect||7||2.26 (1.74–2.93)||<0.001||0||5.41||0.492|
|Emotional neglect||7||2.62 (2.03–3.38)||<0.001||0||5.94||0.430|
|Parental loss||5||1.16 (0.75–1.78)||0.514||51||8.23||0.084|
Differences between bipolar disorder subtypes were studied in stage three. Four identified studies provided data to compare rates of childhood adversity across subtypes. Reference Savitz, Van Der Merwe, Stein, Solms and Ramesar32,Reference Watson, Gallagher, Dougall, Porter, Moncrieff and Ferrier38,Reference Janiri, Sani, Danese, Simonetti, Ambrosi and Angeletti39,Reference Parker, Fletcher, McCraw, Futeran and Hong47 No significant difference in childhood adversity was observed between type 1 and type 2 bipolar disorder (OR = 0.93, 95% CI 0.48–1.81, P = 0.827; Q(3) = 6.91, P = 0.075, I 2 = 56.58).
In stage four we looked at differences between bipolar disorder and psychiatric controls, major depression and schizophrenia. Data from 11 studies were used to compare rates of childhood adversity in bipolar disorder and major depression (see online Fig. DS1). Childhood adversity was significantly greater in bipolar disorder (OR = 1.24, 95% CI 1.02–1.50, P = 0.031), with low levels of statistical heterogeneity (Q(10) = 12.83, P = 0.233, I 2 = 22.08). However, Egger's test approached significance (β = 0.75, s.e. = 0.43, P = 0.058), indicating the possibility of publication bias. After Duval & Tweedie's trim and fill adjusted for three hypothetical missing studies, the imputed OR fell to 1.09 (95% CI 0.88–1.36). Based on the post hoc hypothesis that the absence of an effect was due to the type of adversity considered, we repeated the analyses removing four studies that focused on parental loss. This elevated the effect size (OR = 1.54, 95% CI 1.19–2.00, P<0.001; Q(6) = 4.30, P<0.001, I 2 = 0) showing significantly higher rates of childhood adversity in bipolar disorder when compared with major depression. No hypothetically missing study was detected, with no indication of publication bias (β = −1.34, s.e. = 1.25, P = 0.166).
No significant difference in rates of childhood adversity was found when comparing bipolar disorder and schizophrenia in the analysis of five studies (OR = 0.89, 95% CI 0.79–1.01, P = 0.067; Q(4) = 2.32, P = 0.677, I 2 = 0; see online Fig. DS2). Egger's test was non-significant (β = −0.52, s.e. = 0.42, P = 0.152) and no hypothetically missing studies were estimated.
The NOQAS ratings for the case–control studies are given in online Table DS2. Generally, the quality of the studies in the main analysis was adequate, with eight studies employing an appropriately matched control group and/or controlling for covariates in the analysis. Only one study failed to substantiate participants' diagnoses through interview. Reference Fowke, Ross and Ashcroft23 There was a non-significant trend of better study quality producing larger effects (b = 0.22, s.e. = 0.12, 95% CI −0.01 to 0.45; Z = 1.82, P = 0.066) in the case–control studies. Quality ratings for the case–control studies included in the secondary analysis were lower than for those in the main analysis. This was largely due to studies not controlling for covariates or employing matching criteria. The majority of the studies included in the secondary analysis employed a rigorous method of ascertaining diagnoses. Epidemiological studies included nationally representative samples with data obtained through structured interviews or record linkage; these studies adequately controlled for a range of covariates in their analyses, including gender.
The results of the meta-analysis suggest that individuals with bipolar disorder are 2.6 times more likely to experience childhood adversity when compared with a non-clinical control group. This effect did not appear to be the result of study design or bias, and remained robust and significant even after controlling for hypothetically missing studies. The findings should be interpreted in the context of relatively few longitudinal studies and none with a prospective cohort design, limiting the ability to make causal inferences. Nevertheless, there appears to be a strong and significant association between childhood adversity and bipolar disorder.
We found some variances in this association when specific types of exposure were analysed separately. Emotional abuse was four times more likely to have occurred in bipolar disorder groups than in healthy controls, an effect seemingly larger than for other types of adversity. This is in contrast to a recent meta-analysis that observed roughly equivalent effect sizes for adversity subtypes on psychosis. Reference Varese, Smeets, Drukker, Lieverse, Lataster and Viechtbauer15 Interestingly, parental loss did not significantly differ between bipolar and non-clinical samples. One explanation is that the impact of losing a parent is highly dependent on the context and stage at which it occurs. Reference Etain, Henry, Bellivier, Mathieu and Leboyer5 Indeed, research has suggested that younger age at parental loss, maternal loss in particular and death by unnatural causes are more strongly associated with a bipolar disorder diagnosis. Reference Laursen, Munk-Olsen, Nordentoft and Bo22,Reference Mortensen, Pedersen, Melbye, Mors and Ewald50,Reference Nickerson, Bryant, Aderka, Hinton and Hofmann51 Refuting our initial hypothesis, the effect of childhood adversity on type 2 bipolar disorder, compared with type 1 disorder, did not reach statistical significance. Although the analysis included only four studies, it is possible that childhood adversity is associated with both the more severe and attenuated bipolar profiles.
Rates of childhood adversity were significantly greater in bipolar disorder when compared with major depression. However, this effect became non-significant when controlling for hypothetically missing studies. The absence of a stronger effect may have been due to the overrepresentation of studies considering parental loss, which did not appear to be elevated in bipolar disorder more generally. When repeating the analysis without effects pertaining to parental loss, individuals with bipolar disorder presented with higher levels of adversity compared with the major depression group. Nevertheless, it is difficult to draw firm conclusions concerning the specificity of childhood adversity in the two disorders.
The results showed no significant difference in the rates of childhood adversity between individuals diagnosed with bipolar disorder and those with schizophrenia. A wealth of research has focused on the role of childhood adversity in the development of psychosis; Reference Varese, Smeets, Drukker, Lieverse, Lataster and Viechtbauer15 our findings suggest similar levels of adversity in bipolar disorder. Interestingly, correlational studies have showed associations between childhood adversity and psychotic experiences in bipolar disorder. Reference Upthegrove, Chard, Jones, Gordon-Smith, Forty and Jones8 Future research should explore the exact pathways by which specific forms of adversity lead to particular symptom clusters.
The analysis revealed high levels of statistical heterogeneity, which allows for less confidence in the estimated effect sizes, but is not surprising given the methodological and analytic variances in the identified studies. For example, measures of childhood adversity included national registers, questionnaires, survey items and semi-structured interviews. Furthermore, studies differed in terms of diagnostic assessments (e.g. the Structured Clinical Interview for DSM Disorders, the Composite International Diagnostic Interview) and inclusion criteria (e.g. adolescents, adults), with two studies restricting their analysis to type 1 bipolar disorder. Reference Gilman, Ni, Dunn, Breslau, McLaughlin and Smoller26,Reference Hlastala and McClellan42 Although the analyses allowed for the examination of some potential sources of heterogeneity (e.g. the impact of study design), the limited number of identified studies prevented the authors from testing the impact of other methodological differences on effect sizes. In the presence of further publications, future reviews might wish to explore whether such methodological and clinical variations moderate the association between childhood adversity and bipolar disorder.
There are some limitations of this meta-analysis and of the research literature more generally. Recall bias and illness representations may confound retrospective reporting of childhood adversity. Reference Fisher and Hosang17 In the absence of long-term prospective research it is impossible to reach a definitive conclusion on the causal link between childhood adversity and bipolar disorder. It is feasible that, in some cases, early or prodromal symptoms in childhood might place greater strain on parenting, which could contribute to dysfunctional relationships. Therefore, a genetic predisposition to bipolar disorder might increase levels of childhood adversity. Similarly, we note the absence of studies carefully examining graded (i.e. dose–response) relationships, which in conjunction with the investigation of putative biological and psychosocial mechanisms might enable the identification of plausible pathogenic pathways linking adversity to bipolar psychopathology. Last, the adversity subtypes were not statistically independent, making it difficult to draw firm conclusions on the specificity of adversity subtypes on bipolar disorder.
Given the association between childhood adversity and bipolar disorder, practitioners should carefully enquire about their clients' past adverse experiences, including emotional abuse. Read et al have provided guidance on how clinicians might conduct these conversations and respond sensitively to and deal with disclosures. Reference Read, Hammersley and Rudegeair52 Identification of childhood adversity should then lead to its integration into personalised formulations of clients' difficulties and the provision of appropriate support and interventions.
In conclusion, childhood adversity appears to be associated with the development of bipolar disorder. Rates of childhood adversity in bipolar disorder appear to be similar to those observed in psychosis and major depression. Researchers should explore the ways in which childhood adversity interacts with cognitive, behavioural and biological factors. They should also investigate the potential impact of alternative forms of adversity such as bullying and witnessing domestic violence. Further prospective research exploring dose–response and accounting for genetic effects would help to elucidate the nature of the relationship between childhood adversity and bipolar symptoms. The findings have implications for the study and treatment of bipolar disorder.
The authors thank Louise Laverty and Rebecca Harrop for their assistance with reliability checking; and Alex Fowke, Daniela Caldirola, Delfina Janiri, Gianfranco Spalletta, Jonathan Savitz, Lauren Alloy, Lize van der Merwe, Luciano de Souza, Monica Aas, Netta Horesh, Po-Hsiu Kuo, Stuart Watson and Toshi Furukawa for the information they supplied.