- Email: [email protected]
Commingling analysis of age-of-onset in bipolar I disorder and the morbid risk for major psychoses in first degree relatives of bipolar I probands
Journal of Affective Disorders 168 (2014) 197–204
Contents lists available at ScienceDirect
Journal of Affective Disorders journal homepage: www.elsevier.com/locate/jad
Research report
Commingling analysis of age-of-onset in bipolar I disorder and the morbid risk for major psychoses in first degree relatives of bipolar I probands Maria Grigoroiu-Serbanescu a,n, Marcella Rietschel b, Joanna Hauser c, Piotr M. Czerski c, Stefan Herms d, Xianqing Sun e, Priya Wickramaratne f, Robert C. Elston e a
Biometric Psychiatric Genetics Research Unit, Alexandru Obregia Clinical Psychiatric Hospital, 10, Sos. Berceni, R-041914 Bucharest, Romania Central Institute for Mental Health, Division Genetic Epidemiology in Psychiatry, Mannheim, Germany Laboratory of Psychiatric Genetics, Department of Psychiatry, Poznan University of Medical Sciences, Poznan, Poland d University Hospital Basel, Research Group Genomics, Medical Genetics, Basel, Switzerland e Case Western Reserve University School of Medicine, Department of Epidemiology and Biostatistics, Cleveland, OH, USA f Department of Psychiatry, College of Physicians and Surgeons, and Department of Biostatistics, Joseph L. Mailman School of Public Health, Columbia University; Division of Clinical and Genetic Epidemiology, New York State Psychiatric Institute, New York, New York, USA b c
art ic l e i nf o
a b s t r a c t
Article history: Received 16 February 2014 Received in revised form 29 June 2014 Accepted 30 June 2014 Available online 9 July 2014
Background: Age-of-onset (AO) is increasingly used in molecular genetics of bipolar I disorder (BP-I) as a phenotypic specifier with the goal of reducing genetic heterogeneity. However, questions regarding the cut-off age for defining early onset (EO), as well as the number of onset groups characterizing BP-I have emerged over the last decade with no definite conclusion. The aims of this paper are: 1) to see whether a mixture of three distributions better describes the AO of BP-I than a mixture of two distributions in different independent samples; 2) to compare the morbid risk (MR) for BP-I and for major affective disorders and schizophrenia in first degree relatives of BP-I probands by proband onset group derived from commingling analysis, since the MR to relatives is a trait with strong genetic background. Methods: We applied commingling (admixture) analysis to the AO of three BP-I samples from Romania (n ¼621), Germany (n ¼ 882), and Poland (n ¼ 354). Subsequently, the morbid risk (MR) for BP-I and for major psychoses (BP-I, BP-II, Mdd-UP, schizoaffective disorders, schizophrenia) was estimated in first degree relatives by proband AO-group derived from admixture analysis in the Romanian sample. Results: In the three independent samples and in the combined sample two- and three-AO-group distributions fitted the empirical data equally well. The upper EO limit varied between 21 and 25 years from sample to sample. The MR for both BP-I and for all major psychoses was similar in first degree relatives of EO probands (AO r21) and in relatives of intermediate-onset probands (AO¼ 22–34). Significant MR differences appeared only when comparing the EO group to the late-onset (LO) group (AO4 34). Similar to Mdd-UP and schizophrenia, a significant MR decrease in proband first degree relatives was visible after proband AO of 34 years. Under the three-AO-group classification the MR for both BP-I and all major psychoses in first degree relatives did not differ by relative sex in any proband AO-group. Under the two-AO-group classification female relatives of LO probands (AO 424) had a significantly higher MR for all major psychoses than male relatives, while there was no sex difference for the relatives of EO probands. Limitations: MR was not computed in the German and Polish samples because family data were not available and 34% of the relatives of the Romanian probands were not available for direct interview. Conclusion: Similar to other clinical traits, the MR for major psychoses to relatives failed to support a three-AO-group classification in BP-I suggesting that this is not more useful for the molecular analysis than a two-AO-group classification. & 2014 Elsevier B.V. All rights reserved.
Keywords: Admixture analysis Morbid risk Sex differences Molecular analysis Onset Affective disorders
1. Introduction n
Corresponding author. Tel./fax: þ 40 21 3323929. E-mail address: [email protected] (M. Grigoroiu-Serbanescu).
http://dx.doi.org/10.1016/j.jad.2014.06.054 0165-0327/& 2014 Elsevier B.V. All rights reserved.
Age-of-onset (AO) is considered a phenotypic specifier that might reduce the genetic heterogeneity of the disorders, being
198
M. Grigoroiu-Serbanescu et al. / Journal of Affective Disorders 168 (2014) 197–204
increasingly used in the molecular genetics of bipolar I disorders (BP-I). In this context, a debate about the cut-off point for defining the early onset (EO) versus the late onset (LO) in BP-I disorder, as well as about the number of onset groups characterizing this disorder, has emerged over the last decade, especially after the publication of two papers by Bellivier et al. (2001, 2003) that applied admixture analysis to the AO of two small BP samples finding a mixture of three normal distributions as the best fit to the observed data. Some of the subsequent studies (Lin et al., 2006; Manchia et al., 2008, Hamshere et al., 2009; Tozzi et al., 2011, Ortiz et al., 2011) have replicated the three AO-group model as a best fit for the AO distribution in BP-I disorder, although with variable upper limits for the early onset (18–22 years). However, when attempting to find clinical differences among the AO groups derived by admixture analysis, with few exceptions, the majority of studies restricted the comparison to two groups – the EO and the LO groups – the intermediate onset group being usually ignored. The proportion of cases included in the three onset groups determined through admixture analysis varies widely from study to study and even within the EO group defined as AOo 21 years the variation ranges from 21% (Bellivier et al., 2001) to 79% (Lin et al., 2006) in different samples. But not all studies applying admixture analysis found the best fit of BP-I onset under a three-group model. Kennedy et al. (2005) found the best fit for a two-AO-group model with age 40 as cut-off for EO with a peak incidence for mania in the age band 21–25 years. Javaid et al. (2011) found the two onset group model the most adequate for the AO distribution of BP-I in a Canadian sample. The cut-off between EO and LO was set at age 22, the intersection point between the density distribution curves. Summarizing the formal aspects of the results provided by admixture studies so far, we may observe that the cut-off point for the early onset BP-I varies from 18 (Ortiz et al., 2011) to 40 years (Kennedy et al., 2005) and the proportion of cases included from 21 to 79%, depending on the sample. So, no universal standards could be validated for BP-I disorder. We note that the differences in the number of identified onset groups and age limits of these groups in the cited studies are not due to differences in AO definition. All studies used the retrospective measurement based on patient interview and clinical records considering the age at which the patients first met either DSM-IV criteria for BP-I disorder (the majority of the studies) or DSM-III-R criteria (Lin et al., 2006) as the illness AO. The DSM-IV criteria do not differ significantly from DSM-III-R criteria for BP-I disorder.
1.1. Clinical differences in onset groups derived by admixture analysis 1.1.1. Clinical differences in three-AO-group classifications in BP-I Bellivier et al. (2001) compared certain clinical features of BP-I among the three onset groups found in their sample. They found no significant difference between the EO and the intermediate onset groups in terms of psychosis during affective episodes, family history of affective disorders, or number of suicide attempts. Differences emerged only in the comparison of EO versus LO group. Lin et al. (2006) compared the three onset groups to one another in a sample of 211 BP-I patients and found no clinical difference between the intermediate group (AO¼ 22–28) and the LO group (AO4 28) with respect to psychosis, rapid cycling, comorbidity with panic and obsessive compulsive disorders, drug and alcohol abuse, suicide attempts, or episode frequency. These clinical features were significant only in the comparison of the EO group (79.7% cases) with the LO group (13.1% cases).
Hamshere et al. (2009) also contrasted all three AO groups revealed by admixture analysis. They found significant differences for clinical variables (positive family history for affective disorders, rapid-cycling, suicide attempts) only for comparisons between the EO (AO r22 years) and the LO group (AO Z40 years) but not for comparisons between the intermediate onset (AO¼25–37 years) and the LO group. In the study by Ortiz et al. (2011), the intermediate AO group was not considered for clinical comparison. The EO group (AOo 19) was associated with a stronger family history of affective disorders, psychotic symptoms in the manic episode, suicidal behavior, co-morbid anxiety disorders, and more females compared to the LO group (AO 432). The AO in females was significantly younger than in males in the total sample (p o0.01). Tozzi et al. (2011) found that the three-AO-group model was marginally better than the two-AO-group model and graphically the intermediate group was completely overlapped by the other two groups. So, they subdivided the sample (n ¼954) into two groups for the analysis of clinical differences choosing the age 25 as cut-off for the EO because the probability of belonging both to the EO and the LO group was equal at this age. They failed to find differences regarding family history of affective disorders, psychosis, and illness severity between the EO and LO group. Only the suicide attempts were more frequent in the EO group. In this sample the AO was younger in females than in males (p ¼0.02). Coryell et al. (2013) applied mean AO thresholds resulting from previous admixture studies to subdivide their sample into three AO groups (r20, 21–29, Z30 years). There were no differences in terms of psychotic features, frequency of manic/hypomanic episodes, functional impairment, or lifetime comorbidity among the AO groups. Only the drug abuse and panic attacks were more frequent in the EO group when compared with the LO group. In summary, all studies that investigated clinical characteristics of BP-I according to a three-AO-group model derived from admixture analysis failed to evidence significant differences either between the intermediate onset group and the LO group or between the EO and the intermediate onset group.
1.1.2. Clinical differences in two-AO-group classifications in BP-I Kennedy et al. (2005) reported clinical differences between early onset (AO o40 years) and late onset (AO440 years) with regard to family history of affective disorders (po 0.01) and psychosis (p o0.05). Females had a younger onset than males (p o0.01). Javaid et al. (2011) found no significant difference between the EO and the LO group with respect to the presence of psychosis and axis I comorbidity when comparing the EO and the LO groups defined by the cut-off age of 22 years. No admixture study has addressed the issue of the morbid risk (MR) for major affective disorders or for all major psychoses to first degree relatives of probands to differentiate among onset groups generated by admixture analysis. In this context, the objectives of our study are as follows: 1) to see whether a mixture of three distributions better describes the AO of BP-I than a mixture of two distributions in different independent samples, recruited under similar conditions (consecutive hospital admissions); 2) to compare the MR for BP-I and for major affective disorders and schizophrenia in first degree relatives of BP-I probands by proband onset group derived from commingling analysis, since the MR to relatives is a trait with strong genetic background. We consider the MR both for BP-I and for all major psychoses together (BP-I, BP-II, recurrent unipolar major depression, schizoaffective disorders, schizophrenia) since genome-wide association meta-analyses have shown an overlapping molecular basis for all these disorders (Craddock et al., 2009;
M. Grigoroiu-Serbanescu et al. / Journal of Affective Disorders 168 (2014) 197–204
Cross-Disorder Group of PGC, 2013) and epidemiologic studies have revealed increased risk of schizoaffective disorders and unipolar major depression among relatives of both BP-I and schizophrenia probands (Gershon et al., 1982; Barnett and Smoller, 2009; Lichtenstein et al., 2009).
2. Methods
199
German and Polish patients were administered the SCID-I interview and the OPCRIT checklist. All diagnoses were based on DSM-IV criteria (APA, 1994). The final diagnosis was based on the best estimate procedure considering all available information sources, i.e. patient interviews, medical records, and also, for the Romanian patients, the information provided by close relatives. 2.3. Definition and diagnosis of proband age-of-onset
Written informed consent was obtained from all patients of the three national samples and their interviewed relatives participating in the study after explanation of the study goals and methods. The Grant Committee of the Romanian Ministry for Education and Research, as granting agency, approved the ethical standards for the Romanian sample. For the German and Polish samples the Institutional Review Boards of the Central Institute of Mental Health, Mannheim and University Clinic of Bonn, as well as the Ethics Committee of the Medical University of Poznan approved the study. All patients were recruited from consecutive hospital admissions at the three sites without regard to familial psychopathology.
AO was defined as the age at which the probands first met DSM-IV criteria for a manic, mixed or major depressive episode. For all three samples at least two information sources were considered (patient interview, medical records) and for the Romanian sample also a third source (close relative interview) when estimating the proband's AO. The best-estimate diagnostic procedure based on multiple information sources ensured the most accurate estimation of AO possible, since no other more objective method is available in psychiatric research. 2.4. Diagnosis of proband relatives in the Romanian sample
2.1. BP-I samples The demographic characteristics of the three samples are shown in Table 1. The Romanian sample comprised 621 BP-I probands who had at least two hospitalizations in the Obregia Hospital. The mean number of hospitalization per patient was 7.19 (SD¼4.33, range 2– 30). The German sample comprised 882 BP-I patients and the Polish sample consisted of 354 BP-I patients; also these patients were hospitalized in different clinics from Germany and Poland. The AO means of the German and Romanian samples did not differ significantly (t ¼1.12, p ¼0.26) and the medians were identical, while the mean AO of the Polish sample was significantly different from both the mean of the Romanian sample (t ¼4.13, p ¼0.000) and the mean of the German sample (t¼4.05; p ¼0.001). The AO distributions of all three samples significantly deviated from normal Gaussian distribution (po 0.001) but they did not significantly differ from one another (Kolmogorov–Smirnov test for the comparison Romanian–German: p¼ 0.10; Romanian–Polish: p ¼0.10; German–Polish: p¼ 0.09). In the Romanian sample the females had a significantly younger AO than the males (p¼0.001), in the German sample females tended to have a younger AO (p¼0.07) and in the Polish sample there was no significant AO difference between females and males (p¼0.51). 2.2. Diagnostic procedure of the probands The Romanian patients were administered the Diagnostic Interview for Genetic Studies (DIGS), the Family Interview for Genetic Studies (FIGS), the SCID-I interview and the OPCRIT checklist. The
The 621 probands had 2938 first-degree relatives. 66% (1939/ 2938) of them were directly interviewed with DIGS and FIGS. The psychopathological information about the rest of the relatives was collected through family history method with the FIGS administered to all probands and relatives directly interviewed. The medical records of many relatives who were hospitalized in the Obregia Hospital could also be traced. The final DSM-IV diagnosis of the relatives was based on all available information and it was consensual involving a blind rater and the direct interviewers (kappa coefficient for the agreement among raters¼0.92, 95%CI: 0.88–0.97). 2.5. Statistical analysis We performed commingling analysis (MacLean et al., 1976) as implemented in the program SEGREG of S.A.G.E. 6.3. (S.A.G.E., 2012). The purpose of commingling analysis is to distinguish an inherent skewed distribution of empirical data from a mixture of several distributions. The commingling (mixture) of several distributions is rejected when a power transformation of a smaller number of distributions provides an acceptable alternative (MacLean et al., 1976). Because the distribution of the raw data was skewed in all three samples (Table 1) and skewness may give evidence of spurious commingling if Gaussian component distributions are assumed (MacLean et al., 1976), different types of transformations included in the Box–Cox family of transformations (Box and Cox, 1964) were applied. One, two and three component models were compared under the same type of data transformation.
Table 1 Description of the demographic characteristics of the Romanian, German, and Polish samples.
Age at interview (mean, SD) Age of onset
Kolmogorov –Smirnov test (K–S)
Female/male ratio AO men AO females AO difference by sex
Romanian sample (n¼621)
German sample (n ¼882)
Polish sample (n¼ 354)
43.43 (13.37) mean ¼ 25.97 (9.68) median ¼24 range: 9–59 K–S¼ 3.13 p o 0.001 skewness¼ 0.86 354 (57%)/267 (43%) 27.42 (10.10) 24.88 (9.21) t ¼3.27 p ¼ 0.001
44.03 (13.41) mean¼ 27.18 (9.84) median ¼ 24 range: 9–66 K–S¼ 3.80 p o0.001 skewness¼ 1.10 551 (62.47%)/331(37.53) 27.40 (11.05) 25.91 (10.96) t¼ 1.89 p ¼0.07
45.0 (14.11) mean ¼ 29.45 (10.48) median ¼27 range: 10–63 K–S¼ 2.35 p o 0.001 skewness¼0.72 203 (57.34%)/151 (42.66%) 29.03 (10.60) 29.77 (10.41) t ¼0.66 p ¼ 0.51
200
M. Grigoroiu-Serbanescu et al. / Journal of Affective Disorders 168 (2014) 197–204
Mixtures of power-normal distributions were fitted by maximum likelihood simultaneously estimating regression coefficients and the power parameter λ1 of the Box and Cox transformation. The SEGREG program adds a constant λ2 to each AO-value prior to power transformation and directly maximizes the likelihood numerically. In our study the choice of the best model fitting the data was guided by the value of Akaike's A Information Criterion (AIC) (Akaike, 1974). When close values of the Akaike AIC were obtained for two models with a different number of components under the same type of data transformation, a χ2 goodness of fit test was additionally considered for model selection substracting -2ln (likelihood) of the model estimated with more parameters from -2ln (likelihood) of the model estimated with fewer parameters. We also visually compared the fitted distributions with cumulative empirical plots of the AO data. Since in the Romanian sample there was a significant difference in AO between female and male probands all commingling analyses in this sample were performed with sex included as covariate in the model. We also performed commingling analysis both with and without the proband's sex and ethnicity as covariates in the combined Romanian–German sample. In the German and in the combined Romanian–German samples the covariate sex was removed after showing no significant influence on the Akaike's AIC-score in preliminary analyses. Likewise the covariate “ethnicity” was not significant in any analysis whatever the data transformation. The morbid risks for BP-I and for all major psychoses in first degree relatives by probands’ AO group were estimated using the survival analysis (SPSSv.19) that computes age-corrected cumulative lifetime rates of disorders.
3. Results 3.1. Fitting empirical AO distributions to theoretical distributions Table 2 summarizes the best models that fitted the AO distributions in the three independent samples and in the combined Romanian–German sample (N ¼1503; mean AO¼ 26.71, SD ¼10.75, median ¼24). The Polish sample was not combined
with the other two samples owing to the significant AO mean and median difference against the other samples. In all three samples the AIC-scores for the two-AO-group (component) model and the three-AO-group (component) model were very close to one another and at the 5% level there was no significant χ2 value for the difference between 2ln(L) of the model with less parameters and 2ln (L) of the model with more parameters. This implies that both a three-AO-group distribution and a two-AO-group distribution fit the data equally well. In the pooled Romanian–German sample the theoretical threeAO-group distribution fitted marginally better than the two-AOgroup distribution according to the χ2 test of goodness of fit (p ¼0.05). Fig. 1shows the plot between the empirical data against the theoretical models, visually depicts this finding, showing that a mixture of distributions fits the empirical data much better than a single distribution, whereas they do not deviate too much from either a two or three component model. The plot shows that the empirical data do not deviate too much from either model although being slightly closer to the three component model. The upper age limits of EO in the three independent samples and in the combined Romanian–German sample given by the intersection points of the onset group curves resulting from commingling analysis varied under the three-AO-group model; the Romanian sample had a limit of 20–21 years (33% cases), the German sample had the limit at 25 years (46% cases), the Polish sample at 24–25 years (44% cases) and the combined sample had the limit also at 24–25 years (54% cases) (Table 2). In all the independent samples the means and SDs of the first group (EO) remained quite stable, regardless of the number of model components (two or three). In the Romanian, the German, and the combined Romanian– German samples there were significantly more women than males included in the EO group both under the three-AO-group and the two-AO-group classifications. 3.2. The morbid risk in first degree relatives by AO group of probands No previous study concerned with clinical differences in AO groups generated by admixture analysis examined the MR for
Table 2 Best models in the separate Romanian, German, and Polish samples and in the combined Romanian–German sample. Number of components (groups)
Akaike's AIC
Component (group) means (M) and SD
Case proportion
χ2 for the comparison between models
Romanian sample (n¼621) Two λ1 ¼ 0.22; λ2 ¼ 1 Three λ1 ¼ 0.40; λ2 ¼ 1
AIC ¼4167.73 2Ln(L) ¼4151,13 AIC ¼4166 2 Ln(L) ¼4146
M1 ¼ 17.55 (3.23) M2 ¼ 29.86 (8.18) M1 ¼ 17.25 (2.84) M2 ¼ 25.61 (6.29) M3 ¼ 40.89 (5.31)
0.43 0.57 0.33 0.46 0.21
χ2 ¼ 5.13, df¼ 2, p ¼0.08
German sample (n¼ 882) Two λ1 ¼ 0.08; λ2 ¼ 1 Three λ1 ¼ 0.04; λ2 ¼ 1
AIC ¼8968.85 2Ln(L) ¼8956.85 AIC ¼8968.56 2Ln(L) ¼8954.56
M1 ¼ 20.71 (5.98) M2 ¼ 38.40 (6.52) M1 ¼ 19.27 (5.52) M2 ¼ 28.49 (7.11) M3 ¼ 45.43 (4.69)
0.67 0.33 0.46 0.41 0.13
χ2 ¼2.29, df ¼2, p 40.30
Polish sample (n¼ 354) Two λ1 ¼ 0.57; λ2 ¼ 1 Three λ1 ¼ 0.97; λ2 ¼1
AIC ¼2604.86 2Ln(L) ¼2590.0 AIC ¼2603.92 2Ln(L) ¼2586.92
M1 ¼ 20.47 ( 3.91) M2 ¼ 33.57 ( 9.12) M1 ¼ 20.65 (3.69) M2 ¼ 33.00 ( 6.05) M3 ¼ 49.02 (5.27)
0.65 0.35 0.44 0.45 0.11
χ2 ¼ 3.08, df¼ 2, p 40.20
M1 ¼ M2 ¼ M1 ¼ M2 ¼ M3 ¼
0.64 0.36 0.54 0.39 0.07
χ2 ¼ 6.06, df¼ 2, p ¼ 0.05
Romanian–German sample (n¼ 1503) Two AIC ¼ 12772.4 λ1 ¼ 0; λ2 ¼ 0 2 LN(L) ¼12756.4 Three AIC ¼12769.8 λ1 ¼ 0; λ2 ¼ 0 2 LN(L) ¼12749.8
20.10 ( 5.83) 36.50, ( 6.11) 19.27 (5.37) 31.83 (6.24) 47.25 (3.76)
M. Grigoroiu-Serbanescu et al. / Journal of Affective Disorders 168 (2014) 197–204
affective disorders or all major psychoses in first degree relatives of BP-I probands, although this is perhaps the most genetic trait characterizing the disorder, since the MR for BP-I in first degree relatives of BP-I patients is ten times higher than in the general population (Barnett and Smoller, 2009) and an additional 10–20% risk for unipolar major depression (Craddock and Jones, 1999; Smoller and Finn, 2003) and schizophrenia (2.4%) (Lichtenstein et al., 2009) also exists. Data on psychopathology of the probands’ first degree relatives were available only for the Romanian sample. The MR in first degree relatives was estimated separately for BP-I and combined for all major psychoses (BP-I, BP-II, recurrent unipolar major depression, schizoaffective disorders, and schizophrenia). For defining the limits of proband AO groups we inspected the intersection points of the fitted curves depicting the AO groups generated by commingling analysis under the best three-AO-group and two-AO-group models in the Romanian sample (Fig. 2-A and -B). Thus, the three AO groups had the following limits: early onset ¼AOr 21 years; intermediate onset ¼ AO between 22 and 34 years; late onset ¼ AO 434 years. The limits of the AO groups under the two-AO-group model were: early onset ¼AO r24 years, late onset ¼AO4 24 years. In the total sample of first degree relatives the MR for BP-I was 7% (122/2938) and the MR for all affective psychoses, schizoaffective psychoses and schizophrenia was 17% (337/2938). The overall χ2 comparisons among the groups both in the three-AO-group classification and the two-AO-group classification were significant indicating group differences (Table 3).
201
In the three-AO-group classification we observed a lack of difference in MR for both BP-I and for all major psychoses between first degree relatives of EO probands and first degree relatives of intermediate-onset probands, while the MR differences between the relatives of EO probands and the relatives of LO probands were highly significant (p ¼0.0006). This suggests that probands’ intermediate onset (AO ¼22–34) does not confer a specific MR to relatives, distinct from the risk conferred by the EO in probands (AO r21). When comparing the MR for BP-I and for all major psychoses between the relatives of intermediate-onset probands and the relatives of LO probands, we also found significant differences; the MR was significantly lower in the relatives of LO probands. Under the two-AO-group model the difference in MR between the relatives of EO probands (AOr24) and the relatives of LO probands (AO424) was significant for both BP-I (p ¼0.0006) and for all major psychoses (p ¼0.010). We also observe that the MR for BP-I and for all major psychoses in relatives does not change in the proband AO interval 22–24 years if we compare the MR figures in the onset group with AOr21 years with the onset group with AO r24 years. Since the comparison between the EO and the intermediateonset group led to no significant difference in MR to proband relatives, we computed the MR to relatives by several AO bands in probands with the intention to detect the AO band in which the MR significantly changes (Table 4). Any comparison that dichotomized the sample into one group “less than” a certain cut-off and a group “greater than” the same cut-off was significant. But the comparison of successive AO bands showed that the MR to first degree relatives significantly decreases only after age 34 for all major psychoses (χ2 ¼ 3.85, p¼ 0.050 for the comparison between the AO-band 4 29 years and AO-band 434 years). The MR for BPI smoothly decreases across successive proband AO bands.
3.3. The morbid risk by gender of first degree relatives and AO-group of probands
Fig. 1. Fitting of the empirical AO data to the theoretical two-AO-component and three-AO-component distributions in the combined Romanian–German sample (n¼ 1503).
Under the three-AO-group classification the MR for both BP-I and all major psychoses in first degree relatives did not differ by relative gender in any AO-group. Under the two-AO-group classification the MR for BP-I was similar for male and female relatives of probands in the EO group, as well as in the LO group. As regards the MR for all psychoses, this was similar for male and female relatives of EO probands but significantly higher for female relatives (17%) than for male relatives (11%) of LO probands due to a higher prevalence of unipolar major depression in women (χ2 ¼ 6.46, df ¼1, p ¼0.004).
Fig. 2. Theoretical two-AO-component distribution (panel A) and three-AO-component distribution (panel B) fitted to the empirical AO data in the Romanian sample.
202
M. Grigoroiu-Serbanescu et al. / Journal of Affective Disorders 168 (2014) 197–204
Table 3 Morbid risk for BP-I and for major affective psychoses and schizophrenia in first degree relatives of BP-I probands by AO groups in the Romanian sample. Proband AO group
Morbid risk in first degree relatives (MR)*
No. probands
χ2
Three AO groups AOr 21 AO¼22–34 AO434
270 229 122
AOr r 21 AO¼22–34 AO434
270 229 122 Two AO groups
AOr 24 AO424
325 296
AOr 24 AO424
325 296
n
Overall χ2 ¼13.54, df¼ 2, p ¼ 0.001 a 2 χ ¼ 3.22, df¼ 1, p ¼ 0.077 b 2 χ ¼11.8, df ¼1, p ¼0.0006 c 2 χ ¼4.34, df ¼ 1, p ¼ 0.037 Overall χ2 ¼13.89, df¼ 2, p ¼ 0.001 d 2 χ ¼ 0.090, df ¼1, p¼ 0.76 e 2 χ ¼ 13.67, df ¼ 1, p ¼0.0002 f 2 χ ¼ 11.71, df¼ 1, p ¼ 0.0006
MR for BP-I MR ¼7%) a b (62/1089) MR ¼5% a c (45/1133) MR ¼3% b c (15/716) MR for all major psychoses MR ¼18% d e (142/1089) MR ¼17% d f (142/1133) MR ¼11% e f (53/716) MR for BP-I MR ¼7% g (75/1347) MR ¼5% g (47/1591) MR for all major psychoses MR ¼18% h (177/1347) MR ¼ 14% h (160/1591)
g
χ2 ¼ 11.87 df¼ 1, p ¼ 0.0006
h
χ2 ¼ 6.53, df¼ 1, p¼ 0.010
The figures in parantheses represent raw frequencies, while the MRs represent age corrected cumulative lifetime rates.
Table 4 Morbid risk for BP-1 and for all major psychoses in first degree relatives of BP-I probands by AO intervals in probands. Proband AO group
Morbid risk (MR) in relatives
AOr 21 AOr 24 AOr 29 AOr 34
7% 7% 7% 6%
years years years years
AOr 21 years AOr 24 years AOr 29 years AOr r 34 years
(62/1089) (75/1347) (94/1822) (107/2222)
18% 18% 17% 17%
(142/1089) (177/1347) (223/1822) (284/2222)
Proband AO group
MR for BP-I AO421 years AO 424years AO429 years AO 434 years MR for all major psychoses AO421 years AO424years AO429 years AO434 years
As a general conclusion based on MR to first degree relatives we may state that the intermediate AO of BP-I probands (AO band 22– 34 years) does not seem to confer a different MR to relatives distinct from the MR conferred by the AO band r21 years. The MR for affective disorders and schizophrenia remains constant up to proband AO 29 and visibly decreases after AO 34.
4. Discussion There is no doubt that the onset of BP-I disorder can be subdivided into groups (see Figs. 1 and 2) due to variable genetic, epigenetic and environmental constellations causing the disorder. Their existence is supported by AO differences between familial and sporadic cases (Moorhead and Young, 2003), by the AO variation among BP-I patients with different types of familial loading (BP-I; schizoaffective, unipolar major depression, etc.) (Gershon et al., 1982; Grigoroiu-Serbanescu et al., 2005) and by different MR for BP-I in first degree relatives of BP-I patients subdivided into EO and LO patients ( Taylor and Abrams, 1981; Rice et al., 1987; Grigoroiu-Serbanescu et al., 2001; Somanath et al., 2002). But the number of valid AO groups that can be detected in a BP-I sample through commingling/admixture analysis is not necessarily a three AO group mixture. Alternatively, although having a skewed distribution in all published samples, the AO may be also treated as a continuous variable in the genetic or phenotypic analysis of BP-I or other psychiatric disorder, but this requires very large samples in order to detect significant associations and to overcome the power problem. A first genome-wide investigation of BP-I that used the AO as continuous trait ( Belmonte Mahon et al., 2011) came to the
Morbid risk (MR) in relatives
5% 5% 4% 3%
(60/1849) (47/1591) (28/1116) (15/716)
15% 15% 14% 11%
(195/1849) (160/1591) (114/1116) (53/716)
χ2
χ2 ¼ 9.72, p ¼ 0.0018; df ¼ 1 χ2 ¼ 11.87, p ¼0.0006; df ¼1 χ2 ¼ 9.60, p¼ 0.002; df ¼1 χ2 ¼ 9.40, p¼ 0.002; df ¼1 χ2 ¼ 3.95, p¼ 0.047 χ2 ¼ 6.53, df¼ 1, p ¼ 0.010 χ2 ¼ 4.87, df¼ 1, p ¼ 0.027 χ2 ¼ 14.90, df ¼1, p ¼0.0001
conclusion that AO itself has a very complicated genetic architecture and revealed only trends of association between some SNPs located on different chromosomes and AO. A study on unipolar major depression (Power et al., 2012) showed that the strength of the association of several SNPs tending to genome-wide significance did not vary over the AO span 8–52 years. All admixture analyses of the AO in BP-I disorder so far have assumed a mixture of Gaussian distributions; these can be considered special cases of a more general commingling analysis that specifically allows for skewed distributions. The number of AO groups that may be detected depends on the sample characteristics. The Romanian and the German samples with AO means around 25 years and median at 24 years were similar to other published large samples of hospitalized patients (Tondo et al., 2010; Baldessarini et al., 2012) and to the world sample investigated by Merikangas et al. (2011). Similar to our preliminary report on a patient subgroup of the samples analyzed in the present study (Grigoroiu-Serbanescu et al., 2010, Priebe et al., 2012) and to Tozzi et al. (2011), we showed that both a three-AO-group distribution and a two-AOgroup distribution may fit the data equally well, as the AIC-score differences between the two models were close, only marginally favoring the three-AO-group model. We also showed that the upper limit of the EO group varies from sample to sample by two to four years, although the patients were recruited under similar conditions at the three sites, were diagnosed with the same instruments and the samples were large enough (354–882 cases). Under the three-AO-group model the upper limit of the EO group was around 21 years in the Romanian sample, 25 years in the German and Polish samples, and 25 years
M. Grigoroiu-Serbanescu et al. / Journal of Affective Disorders 168 (2014) 197–204
in the combined Romanian–German sample. Under the three-AO group model, the Romanian EO group had the same mean and upper age limit as the samples published by Bellivier et al. (2003) and Hamshere et al. (2009). When assessing the validity of the AO grouping, beyond the magnitude of the Akaike score, the graphical representation of the groups, as well as the case proportions should also be considered. A classification where the intermediate group is nearly completely overlapped by one of the other two groups should not be considered a valid classification. This was exemplified by the studies of Tozzi et al. (2011) and Lin et al. (2006), where the intermediate group was overlapped either by the EO or the LO group and unrealistic case proportions emerged in their middle onset groups. The authors of commingling analysis (MacLean et al., 1976) warned about the possibility of spurious admixture caused by biased selection and highly skewed distributions. In BP-I disorder the AO distribution is always skewed and sometimes thresholds based on quartiles or histograms may be more useful than thresholds derived by an admixture analysis that simplistically assumes all component distributions are Gaussian. The morbid risk for BP-I in first degree relatives of BP-I probands of 7% in the total Romanian sample is in line with previously published studies (Rice et al. 1987, (5.74%); Maier et al., 1993, (7%), Lichtenstein et al., 2009, (6.4%)) and with our risk figure (5.3%) previously found in a subgroup of the current sample (Grigoroiu-Serbanescu et al., 2001). MR figures for BP-I and for any major affective psychosis and schizophrenia in first degree relatives of BP-I probands computed on the basis of three AO categories provided by admixture analysis or based on predetermined cut-offs are not available in the literature to undertake a comparison. For our two-AO-group classification we may compare our MR figures with figures reported in two studies that used predetermined cut-offs. Taylor and Abrams (1981) found 9.18% risk for BP-I in first degree relatives of EO probands (AO r29) and 3.91% in relatives of LO probands (AO 429). Somanath et al. (2002) found a MR of 8.90% for BP-I in the first degree relatives of probands with AO o25 years and a MR of 3.14% in the relatives of probands with AO 425 years. This study had the same cut-off for the onset groups as our two-AO-group model. Using a predetermined AO cut-off (age 25) Helenius et al. (2013) dichotomized a Danish registry-based sample of 1204 BP-I probands and found a significantly higher psychopathology load in families of BP-I probands with early AO (o25 years) than in families of BP-I probands with late AO ( Z25 years). Rice et al. (1987) also found the highest MR for BP-I in first degree relatives of probands with AOo25 years. This finding is consistent with our result showing a significant MR difference both for BP-I and for all affective psychoses and schizophrenia between relatives of EO probands and relatives of LO probands in the two-AO-group classification generated by commingling analysis. The MR for all affective psychoses and schizophrenia started to decrease after age 29 but a significant decrease of the MR to first degree relatives was observed only after age 34. This finding is consistent with the observation of Gershon et al. (1982) according to whom the MR for BP-I þBP-II þschizoaffective disorders in first degree relatives of BP-I probands does not change up to age 30. Our results are also in accordance with studies reporting lack of difference in positive family history for affective disorders in comparisons involving three onset groups resulting from admixture analysis (Bellivier et al., 2001; Bellivier et al., 2003; Hamshere et al., 2009; Tozzi et al., 2011). In conclusion we showed that the number of valid AO groups that can be detected in a BP-I sample through commingling/ admixture analysis is not necessarily three and finding a universally valid age limit for defining early onset in BP-I disorder based
203
only on statistical analysis seems to be a difficult task. No previous study could document robust clinical differences in terms of psychotic traits, psychiatric family history, or rapid cycling among three onset groups derived by admixture analysis in BP-I. In additional, our study showed that the MR for all major psychoses to first degree relatives also fails to differentiate among three onset groups but differentiates between two onset groups with the most visible risk declining after age 34. The same finding was reported for unipolar major depression (Weissman et al., 1984) and schizophrenia (Byrne et al., 2002). It seems that the statement of Panariello et al. (2010) with respect to schizophrenia holds true also for BP-I disorder; they stated that “the intermediate onset group is probably an artifact overlapping the late-onset group” (p. 278). Although an intermediate onset group may exist between an EO and a LO group, its clinical usefulness seems to be doubtful. 4.1. Limitations We could not compute MR in the German and Polish samples because the family data were not available. A drawback of all studies concerning disease AO including our study consists in the retrospective estimation of the AO in probands, which might have introduced a recall bias, especially in those cases who were not hospitalized at the first episode or were older when entering the study. 34% of the first degree relatives of the Romanian probands were assessed through family history method that may result in inaccurate diagnostic information. 4.2. Strengths of the study No previous study investigated how well the theoretical AO distributions derived from admixture/commingling analysis fit the empirical data and none tested the same AO models in independent samples of Caucasian origin recruited under similar conditions (consecutive hospital admission) and diagnosed with the same diagnostic instruments. Our study is the first one that compared MR in first degree relatives of BP-I patients by onset groups generated by commingling/admixture analysis, not just percentages of positive family history. 66% of the first degree relatives of probands were directly interviewed. The MR in first degree relatives is the most genetic trait of a disorder and the effort of defining onset groups of BP-I not only through commingling/admixture analysis but also through clinical correlates is mostly intended to reduce the genetic heterogeneity of subgroups used in molecular analysis.
Role of funding source Funding agencies: The Romanian Academy, Grant no. 112/2005-2006 to M. Grigoroiu-Serbanescu; The Romanian Ministry for Education and Research – UEFISCDI, Bucharest, Romania, Grant nos. 122/2006; 42-151/2008, and 89/2012 to Maria Grigoroiu-Serbanescu.
Conflict of interest All authors declare that they have no conflicts of interests.
Acknowledgments The first author is indebted to drs. Dan Prelipceanu, Anca Tanase, Marina Codreanu, Dorina Sima, Mihaela Grimberg, Dana Cojocaru, Carmen Udrea, Anca Talasman, Mihail Gherghel, Radu Mihailescu þ from the Obregia Clinical Psychiatric Hospital, who participated to the diagnostic process of the patients involved in the study, and to the patients and their relatives.
References Akaike, H., 1974. A new look at the statistical model identification. IEEE Trans. Automat. Contr. AC-19, 716–723.
204
M. Grigoroiu-Serbanescu et al. / Journal of Affective Disorders 168 (2014) 197–204
Baldessarini, R.J., Tondo, L., Vázquez, G.H., Undurraga, J., Bolzani, L., Yildiz, A., Khalsa, H.M.K., Lai, M., Lepri, B., Lolich, M., Maffei, P.M., Salvatore, P., Faedda, G. L., Vieta, E., Tohen, M., 2012. Age at onset versus family history and clinical outcomes in 1665 international bipolar-I disorder patients. World Psychiatry 11, 40–46. Barnett, J.H., Smoller, J.W., 2009. The genetics of bipolar disorder. Neuroscience 164, 331–343. Bellivier, F., Golmard, J.L., Henry, C., Leboyer, M., Schurhoff, F., 2001. Admixture analysis of age at onset in bipolar I affective disorder. Arch. Gen. Psychiatry 58, 510–512. Bellivier, F., Golmard, J.L., Rietschel, M., Schulze, T.G., Malafosse, A., Preisig, M., McKeon, P., Mynett-Johnson, L., Henry, C., Leboyer, M., 2003. Age at onset in bipolar I affective disorder: Further evidence for three subgroups. Am. J. Psychiatry 160, 999–1001. Belmonte Mahon, P, Pirooznia, M, Goes, F.S., Seifuddin, F., Steele, J., Lee, P.H., Huang, J., Hamshere, M.L., Bipolar Genome Study (BiGS) Consortium, The Wellcome Trust Case Control Consortium Bipolar Disorder GroupDepaulo Jr, J.R., Kelsoe, J. R., Rietschel, M., Nöthen, M., Cichon, S., Gurling, H., Purcell, S., Smoller, J.W., Craddock, N., Schulze, T.G., McMahon, F.J., Potash, J.B., Zandi, P.P., 2011. Genome-wide association analysis of age at onset and psychotic symptoms in bipolar disorder. Am. J. Med. Genet. B Neuropsychiatr. Genet. 156B, 370–378. Box, G.E.P., Cox, D.R., 1964. An analysis of transformations (with discussion). J. R. Statist. Soc., Ser. B 26, 211–252. Byrne, M., Agerbo, E., Mortensen, P.B., 2002. Family history of psychiatric disorders and age at first contact in schizophrenia: an epidemiological study. Br. J. Psychiatry Suppl. 43, s19–25. Coryell, W., Fiedorowicz, J., Leon, A.C., Endicott, J., Keller, M.B., 2013. Age of onset and the prospectively observed course of illness in bipolar disorder. J. Affect. Disord. 146, 34–38. Craddock, N, Jones, I., 1999. Genetics of bipolar disorder. J. Med. Genet. 36, 585–594. Craddock, N., O'Donovan, M.C., Owen, M.J., 2009. Psychosis genetics: modeling the relationship between schizophrenia, bipolar disorder, and mixed (or “schizoaffective”) psychoses. Schizophr. Bull. 35, 482–490. Cross-Disorder Group of the Psychiatric Genomics Consortium, 2013. Identification of risk loci with shared effects on five major psychiatric disorders: a genomewide analysis. Lancet 381, 1371–1379 (April 20). Gershon, E.S., Dehisi, L.E., Hamovit, J., Nurnberger Jr., J.I., Maxwell, M.E., Schrelber, J., Davphinais, D., Dingman 2nd, C.W., Guroff, J.J., 1982. A controlled family study of chronic psychoses. schizophrenia and schizoaffective disorder. Arch. Gen. Psychiatry 45, 328–336 (Arch. Gen. Psychiatry 39, 1157-1167). Grigoroiu-Serbanescu, M., Martinez, M., Nöthen, M.M., Grinberg, M., Sima, D., Propping, P., 2001. Different familial transmission patterns in bipolar I disorder with onset before and after age 25. Am. J. Med. Genet.( B) - Neuropsychiatr. Genet. 105B, 765–773. Grigoroiu-Serbanescu, M., Nöthen, MM, Ohlraun, S, Propping, P, Maier, W, Wickramaratne, P, Georgescu, MJ, Prelipceanu, D, Grimberg, M, Sima, D, Rietschel, M., 2005. Family history influences age of onset in bipolar I disorder in females but not in males. Am. J. Med. Genet. (B) – Neuropsychiatr. Genet. 133B, 6–11. Grigoroiu-Serbanescu, M., Rietschel, M., Paul, T., Schulze, T.G., Noethen, M.M., Cichon, S., Elston, RC., 2010. Two or three age-of-onset groups in bipolar I disorder? Findings of commingling analysis in Romanian and German bipolar I patients. Eur. Psychiatry 25 (Suppl 1), 1428. Hamshere, M.L., Gordon-Smith, K., Forty, L., Jones, L., Caesar, S., Fraser, C., Hyde, S., Tredget, J., Kirov, G., Jones, I., Craddock, N., Smith, D.J., 2009. Age-at-onset in bipolar-I disorder: mixture analysis of 1369 cases identifies three distinct clinical sub-groups. J. Affect. Disord. 116, 23–29. Helenius, D., Jørgensen, P.M., Steinhausen, H.C., 2013. A three generations nationwide population study of family load estimates in bipolar disorder with different age at onset. J Affect Disord. 150, 146–151. Javaid, N., Kennedy, J.L., De Luca, V., 2011. Ethnicity and age at onset in bipolar spectrum disorders. C.N.S. Spectr., 16 (ePub). Kennedy, N., Everitt, B., Boydell, J., Van Os, J., Jones, P.B., Murray, R.M., 2005. Incidence and distribution of first-episode mania by age: results from a 35-year study. Psychol. Med. 35, 855–863.
Lichtenstein, P., Yip, B.H., Björk, C., Pawitan, Y., Cannon, T.D., Sullivan, P.F., Hultman, C.M., 2009. Common genetic determinants of schizophrenia and bipolar disorder in Swedish families: a population-based study. Lancet 373 (9659), 234–239. Lin, P.I., McInnis, M.G., Potash, J.B., Willour, V., MacKinnon, D.F., DePaulo, J.R., Zandi, P.P., 2006. Clinical correlates and familial aggregation of age at onset in bipolar disorder. Am. J. Psychiatry 163, 240–246. MacLean, C.J., Morton, N.E., Elston, R.C., Yee, S., 1976. Skeness in commingled distributions. Biometrics 32, 695–699. Manchia, M., Lampus, S., Chillotti, C., Sardu, C., Ardau, R., Severino, G., Del Zompo, M., 2008. Age at onset in Sardinian bipolar I patients: evidence for three subgroups. Bipolar. Disord. 10, 443–446. Maier, W., Lichtermann, D., Minges, J., Hallmayer, J., Heun, R., Benkert, O., Levinson, D.F., 1993. Continuity and discontinuity of affective disorders and schizophrenia. Results of a controlled family study. Arch. Gen. Psychiatry 50, 871–873. Merikangas, KR, Jin, R, He, J.P., Kessler, R.C., Lee, S., Sampson, N.A., Viana, M.C., Andrade, L.H., Hu, C., Karam, E.G., Ladea, M., Medina-Mora, M.E., Ono, Y., Posada-Villa, J., Sagar, R, Wells, JE, Zarkov, Z., 2011. Prevalence and correlates of bipolar spectrum disorder in the world mental health survey initiative. Arch. Gen. Psychiatry 68, 241–251. Moorhead, SR, Young, AH., 2003. Evidence for a late onset bipolar-I disorder subgroup from 50 years. J. Affect. Disord. 73, 271–277. Ortiz, A., Bradler, K., Slaney, C., Garnham, J., Ruzickova, M., O'Donovan, C., Hajek, T., Alda, M., 2011. An admixture analysis of the age at index episodes in bipolar disorder. Psychiatr. Res. 188, 34–39. Panariello, F., O'Driscoll, L., De Souza, R.P., Tiwari, A., Manchia, M., Kennedy, J., De Luca, V., 2010. Age at onset in Canadian Schizophrenia patients: admixture analysis. Schizophr. Res. 122, 278–279. Power, RA, Keers, R, Ng, MY, Butler, AW, Uher, R, Cohen-Woods, S, Ising, M, Craddock, N, Owen, MJ, Korszun, A, Jones, L, Jones, I, Gill, M, Rice, JP, Hauser, J, Henigsberg, N, Maier, W, Zobel, A, Mors, O, Placentino, AS, Rietschel, M, Souery, D, Kozel, D, Preisig, M, Lucae, S, Binder, EB, Aitchison, KJ, Tozzi, F, Muglia, P, Breen, G, Craig, IW, Farmer, AE, Müller-Myhsok, B, McGuffin, P, Lewis, CM., 2012. Dissecting the genetic heterogeneity of depression through age at onset. Am. J. Med. Genet (B) – Neuropsychiatr. Genet. 159B, 859–868. Priebe, L., Degenhardt, F., Herms, S., Haenisch, B., Mattheisen, M., Nieratschker, V., Weingarten, M., Witt, S., Breuer, R., Paul, T., Alblas, M., Moebus, S., Lathrop, M., Leboyer, M., Schreiber, S., Grigoroiu-Serbanescu, M., Maier, W., Propping, P., Rietschel, M., Nöthen, M.M., Cichon, S., Mühleisen, T.W., 2012. Genome-wide survey implicates the influence of copy number variants (CNVs) on the development of early-onset bipolar disorder. Mol. Psychiatry 17, 421–432. Rice, J., Reich, T., Andreasen, N.C., Endicott, J., Van Eerdewegh, M., Fishman, R., Hirschfeld, R.M., Klerman, G.L., 1987. The familial transmission of bipolar illness. Arch. Gen. Psychiatry, 441–447. S.A.G.E., 2012. Statistical Analysis for Genetic Epidemiology, Release 6.3.〈http:// darwin.cwru.edu〉. Smoller, J.W., Finn, C.T., 2003. Family, twin, and adoption studies of bipolar disorder. Am. J. Med. Genet. (C) – Semin. Med. Genet. 123C, 48–58. Somanath, C.P., Jain, S., Reddy, Y.C., 2002. A family study of early-onset bipolar I disorder. J. Affect. Disord. 70, 91–94. Taylor, M.A., Abrams, R., 1981. Early- and late-onset bipolar illness. Arch. Gen. Psychiatry 38, 58–61. Tondo, L., Lepri, B., Cruz, N., Baldessarini, R.L., 2010. Age at onset in 3014 Sardinian bipolar and major depressive disorder patients. Acta Psychiatr. Scand. 121, 446–452. Tozzi, F., Manchia, M., Galwey, N.W., Severino, G., Del Zompo, M., Day, R., Matthews, K., Strauss, J., Kennedy, J.L., McGuffin, P., Vincent, J.B., Farmer, A., Muglia, P., 2011. Admixture analysis of age at onset in bipolar disorder. Psychiatr. Res. 185, 27–32. Weissman, M.M., Wickramaratne, P., Merikangas, K.R., Leckman, J.F., Prusoff, B.A., Caruso, K.A., Kidd, K.K., Gammon, G.D., 1984. Onset of major depression in early adulthood. Increased familial loading and specificity. Arch. Gen. Psychiatry 41, 1136–1143.
Contents lists available at ScienceDirect
Journal of Affective Disorders journal homepage: www.elsevier.com/locate/jad
Research report
Commingling analysis of age-of-onset in bipolar I disorder and the morbid risk for major psychoses in first degree relatives of bipolar I probands Maria Grigoroiu-Serbanescu a,n, Marcella Rietschel b, Joanna Hauser c, Piotr M. Czerski c, Stefan Herms d, Xianqing Sun e, Priya Wickramaratne f, Robert C. Elston e a
Biometric Psychiatric Genetics Research Unit, Alexandru Obregia Clinical Psychiatric Hospital, 10, Sos. Berceni, R-041914 Bucharest, Romania Central Institute for Mental Health, Division Genetic Epidemiology in Psychiatry, Mannheim, Germany Laboratory of Psychiatric Genetics, Department of Psychiatry, Poznan University of Medical Sciences, Poznan, Poland d University Hospital Basel, Research Group Genomics, Medical Genetics, Basel, Switzerland e Case Western Reserve University School of Medicine, Department of Epidemiology and Biostatistics, Cleveland, OH, USA f Department of Psychiatry, College of Physicians and Surgeons, and Department of Biostatistics, Joseph L. Mailman School of Public Health, Columbia University; Division of Clinical and Genetic Epidemiology, New York State Psychiatric Institute, New York, New York, USA b c
art ic l e i nf o
a b s t r a c t
Article history: Received 16 February 2014 Received in revised form 29 June 2014 Accepted 30 June 2014 Available online 9 July 2014
Background: Age-of-onset (AO) is increasingly used in molecular genetics of bipolar I disorder (BP-I) as a phenotypic specifier with the goal of reducing genetic heterogeneity. However, questions regarding the cut-off age for defining early onset (EO), as well as the number of onset groups characterizing BP-I have emerged over the last decade with no definite conclusion. The aims of this paper are: 1) to see whether a mixture of three distributions better describes the AO of BP-I than a mixture of two distributions in different independent samples; 2) to compare the morbid risk (MR) for BP-I and for major affective disorders and schizophrenia in first degree relatives of BP-I probands by proband onset group derived from commingling analysis, since the MR to relatives is a trait with strong genetic background. Methods: We applied commingling (admixture) analysis to the AO of three BP-I samples from Romania (n ¼621), Germany (n ¼ 882), and Poland (n ¼ 354). Subsequently, the morbid risk (MR) for BP-I and for major psychoses (BP-I, BP-II, Mdd-UP, schizoaffective disorders, schizophrenia) was estimated in first degree relatives by proband AO-group derived from admixture analysis in the Romanian sample. Results: In the three independent samples and in the combined sample two- and three-AO-group distributions fitted the empirical data equally well. The upper EO limit varied between 21 and 25 years from sample to sample. The MR for both BP-I and for all major psychoses was similar in first degree relatives of EO probands (AO r21) and in relatives of intermediate-onset probands (AO¼ 22–34). Significant MR differences appeared only when comparing the EO group to the late-onset (LO) group (AO4 34). Similar to Mdd-UP and schizophrenia, a significant MR decrease in proband first degree relatives was visible after proband AO of 34 years. Under the three-AO-group classification the MR for both BP-I and all major psychoses in first degree relatives did not differ by relative sex in any proband AO-group. Under the two-AO-group classification female relatives of LO probands (AO 424) had a significantly higher MR for all major psychoses than male relatives, while there was no sex difference for the relatives of EO probands. Limitations: MR was not computed in the German and Polish samples because family data were not available and 34% of the relatives of the Romanian probands were not available for direct interview. Conclusion: Similar to other clinical traits, the MR for major psychoses to relatives failed to support a three-AO-group classification in BP-I suggesting that this is not more useful for the molecular analysis than a two-AO-group classification. & 2014 Elsevier B.V. All rights reserved.
Keywords: Admixture analysis Morbid risk Sex differences Molecular analysis Onset Affective disorders
1. Introduction n
Corresponding author. Tel./fax: þ 40 21 3323929. E-mail address: [email protected] (M. Grigoroiu-Serbanescu).
http://dx.doi.org/10.1016/j.jad.2014.06.054 0165-0327/& 2014 Elsevier B.V. All rights reserved.
Age-of-onset (AO) is considered a phenotypic specifier that might reduce the genetic heterogeneity of the disorders, being
198
M. Grigoroiu-Serbanescu et al. / Journal of Affective Disorders 168 (2014) 197–204
increasingly used in the molecular genetics of bipolar I disorders (BP-I). In this context, a debate about the cut-off point for defining the early onset (EO) versus the late onset (LO) in BP-I disorder, as well as about the number of onset groups characterizing this disorder, has emerged over the last decade, especially after the publication of two papers by Bellivier et al. (2001, 2003) that applied admixture analysis to the AO of two small BP samples finding a mixture of three normal distributions as the best fit to the observed data. Some of the subsequent studies (Lin et al., 2006; Manchia et al., 2008, Hamshere et al., 2009; Tozzi et al., 2011, Ortiz et al., 2011) have replicated the three AO-group model as a best fit for the AO distribution in BP-I disorder, although with variable upper limits for the early onset (18–22 years). However, when attempting to find clinical differences among the AO groups derived by admixture analysis, with few exceptions, the majority of studies restricted the comparison to two groups – the EO and the LO groups – the intermediate onset group being usually ignored. The proportion of cases included in the three onset groups determined through admixture analysis varies widely from study to study and even within the EO group defined as AOo 21 years the variation ranges from 21% (Bellivier et al., 2001) to 79% (Lin et al., 2006) in different samples. But not all studies applying admixture analysis found the best fit of BP-I onset under a three-group model. Kennedy et al. (2005) found the best fit for a two-AO-group model with age 40 as cut-off for EO with a peak incidence for mania in the age band 21–25 years. Javaid et al. (2011) found the two onset group model the most adequate for the AO distribution of BP-I in a Canadian sample. The cut-off between EO and LO was set at age 22, the intersection point between the density distribution curves. Summarizing the formal aspects of the results provided by admixture studies so far, we may observe that the cut-off point for the early onset BP-I varies from 18 (Ortiz et al., 2011) to 40 years (Kennedy et al., 2005) and the proportion of cases included from 21 to 79%, depending on the sample. So, no universal standards could be validated for BP-I disorder. We note that the differences in the number of identified onset groups and age limits of these groups in the cited studies are not due to differences in AO definition. All studies used the retrospective measurement based on patient interview and clinical records considering the age at which the patients first met either DSM-IV criteria for BP-I disorder (the majority of the studies) or DSM-III-R criteria (Lin et al., 2006) as the illness AO. The DSM-IV criteria do not differ significantly from DSM-III-R criteria for BP-I disorder.
1.1. Clinical differences in onset groups derived by admixture analysis 1.1.1. Clinical differences in three-AO-group classifications in BP-I Bellivier et al. (2001) compared certain clinical features of BP-I among the three onset groups found in their sample. They found no significant difference between the EO and the intermediate onset groups in terms of psychosis during affective episodes, family history of affective disorders, or number of suicide attempts. Differences emerged only in the comparison of EO versus LO group. Lin et al. (2006) compared the three onset groups to one another in a sample of 211 BP-I patients and found no clinical difference between the intermediate group (AO¼ 22–28) and the LO group (AO4 28) with respect to psychosis, rapid cycling, comorbidity with panic and obsessive compulsive disorders, drug and alcohol abuse, suicide attempts, or episode frequency. These clinical features were significant only in the comparison of the EO group (79.7% cases) with the LO group (13.1% cases).
Hamshere et al. (2009) also contrasted all three AO groups revealed by admixture analysis. They found significant differences for clinical variables (positive family history for affective disorders, rapid-cycling, suicide attempts) only for comparisons between the EO (AO r22 years) and the LO group (AO Z40 years) but not for comparisons between the intermediate onset (AO¼25–37 years) and the LO group. In the study by Ortiz et al. (2011), the intermediate AO group was not considered for clinical comparison. The EO group (AOo 19) was associated with a stronger family history of affective disorders, psychotic symptoms in the manic episode, suicidal behavior, co-morbid anxiety disorders, and more females compared to the LO group (AO 432). The AO in females was significantly younger than in males in the total sample (p o0.01). Tozzi et al. (2011) found that the three-AO-group model was marginally better than the two-AO-group model and graphically the intermediate group was completely overlapped by the other two groups. So, they subdivided the sample (n ¼954) into two groups for the analysis of clinical differences choosing the age 25 as cut-off for the EO because the probability of belonging both to the EO and the LO group was equal at this age. They failed to find differences regarding family history of affective disorders, psychosis, and illness severity between the EO and LO group. Only the suicide attempts were more frequent in the EO group. In this sample the AO was younger in females than in males (p ¼0.02). Coryell et al. (2013) applied mean AO thresholds resulting from previous admixture studies to subdivide their sample into three AO groups (r20, 21–29, Z30 years). There were no differences in terms of psychotic features, frequency of manic/hypomanic episodes, functional impairment, or lifetime comorbidity among the AO groups. Only the drug abuse and panic attacks were more frequent in the EO group when compared with the LO group. In summary, all studies that investigated clinical characteristics of BP-I according to a three-AO-group model derived from admixture analysis failed to evidence significant differences either between the intermediate onset group and the LO group or between the EO and the intermediate onset group.
1.1.2. Clinical differences in two-AO-group classifications in BP-I Kennedy et al. (2005) reported clinical differences between early onset (AO o40 years) and late onset (AO440 years) with regard to family history of affective disorders (po 0.01) and psychosis (p o0.05). Females had a younger onset than males (p o0.01). Javaid et al. (2011) found no significant difference between the EO and the LO group with respect to the presence of psychosis and axis I comorbidity when comparing the EO and the LO groups defined by the cut-off age of 22 years. No admixture study has addressed the issue of the morbid risk (MR) for major affective disorders or for all major psychoses to first degree relatives of probands to differentiate among onset groups generated by admixture analysis. In this context, the objectives of our study are as follows: 1) to see whether a mixture of three distributions better describes the AO of BP-I than a mixture of two distributions in different independent samples, recruited under similar conditions (consecutive hospital admissions); 2) to compare the MR for BP-I and for major affective disorders and schizophrenia in first degree relatives of BP-I probands by proband onset group derived from commingling analysis, since the MR to relatives is a trait with strong genetic background. We consider the MR both for BP-I and for all major psychoses together (BP-I, BP-II, recurrent unipolar major depression, schizoaffective disorders, schizophrenia) since genome-wide association meta-analyses have shown an overlapping molecular basis for all these disorders (Craddock et al., 2009;
M. Grigoroiu-Serbanescu et al. / Journal of Affective Disorders 168 (2014) 197–204
Cross-Disorder Group of PGC, 2013) and epidemiologic studies have revealed increased risk of schizoaffective disorders and unipolar major depression among relatives of both BP-I and schizophrenia probands (Gershon et al., 1982; Barnett and Smoller, 2009; Lichtenstein et al., 2009).
2. Methods
199
German and Polish patients were administered the SCID-I interview and the OPCRIT checklist. All diagnoses were based on DSM-IV criteria (APA, 1994). The final diagnosis was based on the best estimate procedure considering all available information sources, i.e. patient interviews, medical records, and also, for the Romanian patients, the information provided by close relatives. 2.3. Definition and diagnosis of proband age-of-onset
Written informed consent was obtained from all patients of the three national samples and their interviewed relatives participating in the study after explanation of the study goals and methods. The Grant Committee of the Romanian Ministry for Education and Research, as granting agency, approved the ethical standards for the Romanian sample. For the German and Polish samples the Institutional Review Boards of the Central Institute of Mental Health, Mannheim and University Clinic of Bonn, as well as the Ethics Committee of the Medical University of Poznan approved the study. All patients were recruited from consecutive hospital admissions at the three sites without regard to familial psychopathology.
AO was defined as the age at which the probands first met DSM-IV criteria for a manic, mixed or major depressive episode. For all three samples at least two information sources were considered (patient interview, medical records) and for the Romanian sample also a third source (close relative interview) when estimating the proband's AO. The best-estimate diagnostic procedure based on multiple information sources ensured the most accurate estimation of AO possible, since no other more objective method is available in psychiatric research. 2.4. Diagnosis of proband relatives in the Romanian sample
2.1. BP-I samples The demographic characteristics of the three samples are shown in Table 1. The Romanian sample comprised 621 BP-I probands who had at least two hospitalizations in the Obregia Hospital. The mean number of hospitalization per patient was 7.19 (SD¼4.33, range 2– 30). The German sample comprised 882 BP-I patients and the Polish sample consisted of 354 BP-I patients; also these patients were hospitalized in different clinics from Germany and Poland. The AO means of the German and Romanian samples did not differ significantly (t ¼1.12, p ¼0.26) and the medians were identical, while the mean AO of the Polish sample was significantly different from both the mean of the Romanian sample (t ¼4.13, p ¼0.000) and the mean of the German sample (t¼4.05; p ¼0.001). The AO distributions of all three samples significantly deviated from normal Gaussian distribution (po 0.001) but they did not significantly differ from one another (Kolmogorov–Smirnov test for the comparison Romanian–German: p¼ 0.10; Romanian–Polish: p ¼0.10; German–Polish: p¼ 0.09). In the Romanian sample the females had a significantly younger AO than the males (p¼0.001), in the German sample females tended to have a younger AO (p¼0.07) and in the Polish sample there was no significant AO difference between females and males (p¼0.51). 2.2. Diagnostic procedure of the probands The Romanian patients were administered the Diagnostic Interview for Genetic Studies (DIGS), the Family Interview for Genetic Studies (FIGS), the SCID-I interview and the OPCRIT checklist. The
The 621 probands had 2938 first-degree relatives. 66% (1939/ 2938) of them were directly interviewed with DIGS and FIGS. The psychopathological information about the rest of the relatives was collected through family history method with the FIGS administered to all probands and relatives directly interviewed. The medical records of many relatives who were hospitalized in the Obregia Hospital could also be traced. The final DSM-IV diagnosis of the relatives was based on all available information and it was consensual involving a blind rater and the direct interviewers (kappa coefficient for the agreement among raters¼0.92, 95%CI: 0.88–0.97). 2.5. Statistical analysis We performed commingling analysis (MacLean et al., 1976) as implemented in the program SEGREG of S.A.G.E. 6.3. (S.A.G.E., 2012). The purpose of commingling analysis is to distinguish an inherent skewed distribution of empirical data from a mixture of several distributions. The commingling (mixture) of several distributions is rejected when a power transformation of a smaller number of distributions provides an acceptable alternative (MacLean et al., 1976). Because the distribution of the raw data was skewed in all three samples (Table 1) and skewness may give evidence of spurious commingling if Gaussian component distributions are assumed (MacLean et al., 1976), different types of transformations included in the Box–Cox family of transformations (Box and Cox, 1964) were applied. One, two and three component models were compared under the same type of data transformation.
Table 1 Description of the demographic characteristics of the Romanian, German, and Polish samples.
Age at interview (mean, SD) Age of onset
Kolmogorov –Smirnov test (K–S)
Female/male ratio AO men AO females AO difference by sex
Romanian sample (n¼621)
German sample (n ¼882)
Polish sample (n¼ 354)
43.43 (13.37) mean ¼ 25.97 (9.68) median ¼24 range: 9–59 K–S¼ 3.13 p o 0.001 skewness¼ 0.86 354 (57%)/267 (43%) 27.42 (10.10) 24.88 (9.21) t ¼3.27 p ¼ 0.001
44.03 (13.41) mean¼ 27.18 (9.84) median ¼ 24 range: 9–66 K–S¼ 3.80 p o0.001 skewness¼ 1.10 551 (62.47%)/331(37.53) 27.40 (11.05) 25.91 (10.96) t¼ 1.89 p ¼0.07
45.0 (14.11) mean ¼ 29.45 (10.48) median ¼27 range: 10–63 K–S¼ 2.35 p o 0.001 skewness¼0.72 203 (57.34%)/151 (42.66%) 29.03 (10.60) 29.77 (10.41) t ¼0.66 p ¼ 0.51
200
M. Grigoroiu-Serbanescu et al. / Journal of Affective Disorders 168 (2014) 197–204
Mixtures of power-normal distributions were fitted by maximum likelihood simultaneously estimating regression coefficients and the power parameter λ1 of the Box and Cox transformation. The SEGREG program adds a constant λ2 to each AO-value prior to power transformation and directly maximizes the likelihood numerically. In our study the choice of the best model fitting the data was guided by the value of Akaike's A Information Criterion (AIC) (Akaike, 1974). When close values of the Akaike AIC were obtained for two models with a different number of components under the same type of data transformation, a χ2 goodness of fit test was additionally considered for model selection substracting -2ln (likelihood) of the model estimated with more parameters from -2ln (likelihood) of the model estimated with fewer parameters. We also visually compared the fitted distributions with cumulative empirical plots of the AO data. Since in the Romanian sample there was a significant difference in AO between female and male probands all commingling analyses in this sample were performed with sex included as covariate in the model. We also performed commingling analysis both with and without the proband's sex and ethnicity as covariates in the combined Romanian–German sample. In the German and in the combined Romanian–German samples the covariate sex was removed after showing no significant influence on the Akaike's AIC-score in preliminary analyses. Likewise the covariate “ethnicity” was not significant in any analysis whatever the data transformation. The morbid risks for BP-I and for all major psychoses in first degree relatives by probands’ AO group were estimated using the survival analysis (SPSSv.19) that computes age-corrected cumulative lifetime rates of disorders.
3. Results 3.1. Fitting empirical AO distributions to theoretical distributions Table 2 summarizes the best models that fitted the AO distributions in the three independent samples and in the combined Romanian–German sample (N ¼1503; mean AO¼ 26.71, SD ¼10.75, median ¼24). The Polish sample was not combined
with the other two samples owing to the significant AO mean and median difference against the other samples. In all three samples the AIC-scores for the two-AO-group (component) model and the three-AO-group (component) model were very close to one another and at the 5% level there was no significant χ2 value for the difference between 2ln(L) of the model with less parameters and 2ln (L) of the model with more parameters. This implies that both a three-AO-group distribution and a two-AO-group distribution fit the data equally well. In the pooled Romanian–German sample the theoretical threeAO-group distribution fitted marginally better than the two-AOgroup distribution according to the χ2 test of goodness of fit (p ¼0.05). Fig. 1shows the plot between the empirical data against the theoretical models, visually depicts this finding, showing that a mixture of distributions fits the empirical data much better than a single distribution, whereas they do not deviate too much from either a two or three component model. The plot shows that the empirical data do not deviate too much from either model although being slightly closer to the three component model. The upper age limits of EO in the three independent samples and in the combined Romanian–German sample given by the intersection points of the onset group curves resulting from commingling analysis varied under the three-AO-group model; the Romanian sample had a limit of 20–21 years (33% cases), the German sample had the limit at 25 years (46% cases), the Polish sample at 24–25 years (44% cases) and the combined sample had the limit also at 24–25 years (54% cases) (Table 2). In all the independent samples the means and SDs of the first group (EO) remained quite stable, regardless of the number of model components (two or three). In the Romanian, the German, and the combined Romanian– German samples there were significantly more women than males included in the EO group both under the three-AO-group and the two-AO-group classifications. 3.2. The morbid risk in first degree relatives by AO group of probands No previous study concerned with clinical differences in AO groups generated by admixture analysis examined the MR for
Table 2 Best models in the separate Romanian, German, and Polish samples and in the combined Romanian–German sample. Number of components (groups)
Akaike's AIC
Component (group) means (M) and SD
Case proportion
χ2 for the comparison between models
Romanian sample (n¼621) Two λ1 ¼ 0.22; λ2 ¼ 1 Three λ1 ¼ 0.40; λ2 ¼ 1
AIC ¼4167.73 2Ln(L) ¼4151,13 AIC ¼4166 2 Ln(L) ¼4146
M1 ¼ 17.55 (3.23) M2 ¼ 29.86 (8.18) M1 ¼ 17.25 (2.84) M2 ¼ 25.61 (6.29) M3 ¼ 40.89 (5.31)
0.43 0.57 0.33 0.46 0.21
χ2 ¼ 5.13, df¼ 2, p ¼0.08
German sample (n¼ 882) Two λ1 ¼ 0.08; λ2 ¼ 1 Three λ1 ¼ 0.04; λ2 ¼ 1
AIC ¼8968.85 2Ln(L) ¼8956.85 AIC ¼8968.56 2Ln(L) ¼8954.56
M1 ¼ 20.71 (5.98) M2 ¼ 38.40 (6.52) M1 ¼ 19.27 (5.52) M2 ¼ 28.49 (7.11) M3 ¼ 45.43 (4.69)
0.67 0.33 0.46 0.41 0.13
χ2 ¼2.29, df ¼2, p 40.30
Polish sample (n¼ 354) Two λ1 ¼ 0.57; λ2 ¼ 1 Three λ1 ¼ 0.97; λ2 ¼1
AIC ¼2604.86 2Ln(L) ¼2590.0 AIC ¼2603.92 2Ln(L) ¼2586.92
M1 ¼ 20.47 ( 3.91) M2 ¼ 33.57 ( 9.12) M1 ¼ 20.65 (3.69) M2 ¼ 33.00 ( 6.05) M3 ¼ 49.02 (5.27)
0.65 0.35 0.44 0.45 0.11
χ2 ¼ 3.08, df¼ 2, p 40.20
M1 ¼ M2 ¼ M1 ¼ M2 ¼ M3 ¼
0.64 0.36 0.54 0.39 0.07
χ2 ¼ 6.06, df¼ 2, p ¼ 0.05
Romanian–German sample (n¼ 1503) Two AIC ¼ 12772.4 λ1 ¼ 0; λ2 ¼ 0 2 LN(L) ¼12756.4 Three AIC ¼12769.8 λ1 ¼ 0; λ2 ¼ 0 2 LN(L) ¼12749.8
20.10 ( 5.83) 36.50, ( 6.11) 19.27 (5.37) 31.83 (6.24) 47.25 (3.76)
M. Grigoroiu-Serbanescu et al. / Journal of Affective Disorders 168 (2014) 197–204
affective disorders or all major psychoses in first degree relatives of BP-I probands, although this is perhaps the most genetic trait characterizing the disorder, since the MR for BP-I in first degree relatives of BP-I patients is ten times higher than in the general population (Barnett and Smoller, 2009) and an additional 10–20% risk for unipolar major depression (Craddock and Jones, 1999; Smoller and Finn, 2003) and schizophrenia (2.4%) (Lichtenstein et al., 2009) also exists. Data on psychopathology of the probands’ first degree relatives were available only for the Romanian sample. The MR in first degree relatives was estimated separately for BP-I and combined for all major psychoses (BP-I, BP-II, recurrent unipolar major depression, schizoaffective disorders, and schizophrenia). For defining the limits of proband AO groups we inspected the intersection points of the fitted curves depicting the AO groups generated by commingling analysis under the best three-AO-group and two-AO-group models in the Romanian sample (Fig. 2-A and -B). Thus, the three AO groups had the following limits: early onset ¼AOr 21 years; intermediate onset ¼ AO between 22 and 34 years; late onset ¼ AO 434 years. The limits of the AO groups under the two-AO-group model were: early onset ¼AO r24 years, late onset ¼AO4 24 years. In the total sample of first degree relatives the MR for BP-I was 7% (122/2938) and the MR for all affective psychoses, schizoaffective psychoses and schizophrenia was 17% (337/2938). The overall χ2 comparisons among the groups both in the three-AO-group classification and the two-AO-group classification were significant indicating group differences (Table 3).
201
In the three-AO-group classification we observed a lack of difference in MR for both BP-I and for all major psychoses between first degree relatives of EO probands and first degree relatives of intermediate-onset probands, while the MR differences between the relatives of EO probands and the relatives of LO probands were highly significant (p ¼0.0006). This suggests that probands’ intermediate onset (AO ¼22–34) does not confer a specific MR to relatives, distinct from the risk conferred by the EO in probands (AO r21). When comparing the MR for BP-I and for all major psychoses between the relatives of intermediate-onset probands and the relatives of LO probands, we also found significant differences; the MR was significantly lower in the relatives of LO probands. Under the two-AO-group model the difference in MR between the relatives of EO probands (AOr24) and the relatives of LO probands (AO424) was significant for both BP-I (p ¼0.0006) and for all major psychoses (p ¼0.010). We also observe that the MR for BP-I and for all major psychoses in relatives does not change in the proband AO interval 22–24 years if we compare the MR figures in the onset group with AOr21 years with the onset group with AO r24 years. Since the comparison between the EO and the intermediateonset group led to no significant difference in MR to proband relatives, we computed the MR to relatives by several AO bands in probands with the intention to detect the AO band in which the MR significantly changes (Table 4). Any comparison that dichotomized the sample into one group “less than” a certain cut-off and a group “greater than” the same cut-off was significant. But the comparison of successive AO bands showed that the MR to first degree relatives significantly decreases only after age 34 for all major psychoses (χ2 ¼ 3.85, p¼ 0.050 for the comparison between the AO-band 4 29 years and AO-band 434 years). The MR for BPI smoothly decreases across successive proband AO bands.
3.3. The morbid risk by gender of first degree relatives and AO-group of probands
Fig. 1. Fitting of the empirical AO data to the theoretical two-AO-component and three-AO-component distributions in the combined Romanian–German sample (n¼ 1503).
Under the three-AO-group classification the MR for both BP-I and all major psychoses in first degree relatives did not differ by relative gender in any AO-group. Under the two-AO-group classification the MR for BP-I was similar for male and female relatives of probands in the EO group, as well as in the LO group. As regards the MR for all psychoses, this was similar for male and female relatives of EO probands but significantly higher for female relatives (17%) than for male relatives (11%) of LO probands due to a higher prevalence of unipolar major depression in women (χ2 ¼ 6.46, df ¼1, p ¼0.004).
Fig. 2. Theoretical two-AO-component distribution (panel A) and three-AO-component distribution (panel B) fitted to the empirical AO data in the Romanian sample.
202
M. Grigoroiu-Serbanescu et al. / Journal of Affective Disorders 168 (2014) 197–204
Table 3 Morbid risk for BP-I and for major affective psychoses and schizophrenia in first degree relatives of BP-I probands by AO groups in the Romanian sample. Proband AO group
Morbid risk in first degree relatives (MR)*
No. probands
χ2
Three AO groups AOr 21 AO¼22–34 AO434
270 229 122
AOr r 21 AO¼22–34 AO434
270 229 122 Two AO groups
AOr 24 AO424
325 296
AOr 24 AO424
325 296
n
Overall χ2 ¼13.54, df¼ 2, p ¼ 0.001 a 2 χ ¼ 3.22, df¼ 1, p ¼ 0.077 b 2 χ ¼11.8, df ¼1, p ¼0.0006 c 2 χ ¼4.34, df ¼ 1, p ¼ 0.037 Overall χ2 ¼13.89, df¼ 2, p ¼ 0.001 d 2 χ ¼ 0.090, df ¼1, p¼ 0.76 e 2 χ ¼ 13.67, df ¼ 1, p ¼0.0002 f 2 χ ¼ 11.71, df¼ 1, p ¼ 0.0006
MR for BP-I MR ¼7%) a b (62/1089) MR ¼5% a c (45/1133) MR ¼3% b c (15/716) MR for all major psychoses MR ¼18% d e (142/1089) MR ¼17% d f (142/1133) MR ¼11% e f (53/716) MR for BP-I MR ¼7% g (75/1347) MR ¼5% g (47/1591) MR for all major psychoses MR ¼18% h (177/1347) MR ¼ 14% h (160/1591)
g
χ2 ¼ 11.87 df¼ 1, p ¼ 0.0006
h
χ2 ¼ 6.53, df¼ 1, p¼ 0.010
The figures in parantheses represent raw frequencies, while the MRs represent age corrected cumulative lifetime rates.
Table 4 Morbid risk for BP-1 and for all major psychoses in first degree relatives of BP-I probands by AO intervals in probands. Proband AO group
Morbid risk (MR) in relatives
AOr 21 AOr 24 AOr 29 AOr 34
7% 7% 7% 6%
years years years years
AOr 21 years AOr 24 years AOr 29 years AOr r 34 years
(62/1089) (75/1347) (94/1822) (107/2222)
18% 18% 17% 17%
(142/1089) (177/1347) (223/1822) (284/2222)
Proband AO group
MR for BP-I AO421 years AO 424years AO429 years AO 434 years MR for all major psychoses AO421 years AO424years AO429 years AO434 years
As a general conclusion based on MR to first degree relatives we may state that the intermediate AO of BP-I probands (AO band 22– 34 years) does not seem to confer a different MR to relatives distinct from the MR conferred by the AO band r21 years. The MR for affective disorders and schizophrenia remains constant up to proband AO 29 and visibly decreases after AO 34.
4. Discussion There is no doubt that the onset of BP-I disorder can be subdivided into groups (see Figs. 1 and 2) due to variable genetic, epigenetic and environmental constellations causing the disorder. Their existence is supported by AO differences between familial and sporadic cases (Moorhead and Young, 2003), by the AO variation among BP-I patients with different types of familial loading (BP-I; schizoaffective, unipolar major depression, etc.) (Gershon et al., 1982; Grigoroiu-Serbanescu et al., 2005) and by different MR for BP-I in first degree relatives of BP-I patients subdivided into EO and LO patients ( Taylor and Abrams, 1981; Rice et al., 1987; Grigoroiu-Serbanescu et al., 2001; Somanath et al., 2002). But the number of valid AO groups that can be detected in a BP-I sample through commingling/admixture analysis is not necessarily a three AO group mixture. Alternatively, although having a skewed distribution in all published samples, the AO may be also treated as a continuous variable in the genetic or phenotypic analysis of BP-I or other psychiatric disorder, but this requires very large samples in order to detect significant associations and to overcome the power problem. A first genome-wide investigation of BP-I that used the AO as continuous trait ( Belmonte Mahon et al., 2011) came to the
Morbid risk (MR) in relatives
5% 5% 4% 3%
(60/1849) (47/1591) (28/1116) (15/716)
15% 15% 14% 11%
(195/1849) (160/1591) (114/1116) (53/716)
χ2
χ2 ¼ 9.72, p ¼ 0.0018; df ¼ 1 χ2 ¼ 11.87, p ¼0.0006; df ¼1 χ2 ¼ 9.60, p¼ 0.002; df ¼1 χ2 ¼ 9.40, p¼ 0.002; df ¼1 χ2 ¼ 3.95, p¼ 0.047 χ2 ¼ 6.53, df¼ 1, p ¼ 0.010 χ2 ¼ 4.87, df¼ 1, p ¼ 0.027 χ2 ¼ 14.90, df ¼1, p ¼0.0001
conclusion that AO itself has a very complicated genetic architecture and revealed only trends of association between some SNPs located on different chromosomes and AO. A study on unipolar major depression (Power et al., 2012) showed that the strength of the association of several SNPs tending to genome-wide significance did not vary over the AO span 8–52 years. All admixture analyses of the AO in BP-I disorder so far have assumed a mixture of Gaussian distributions; these can be considered special cases of a more general commingling analysis that specifically allows for skewed distributions. The number of AO groups that may be detected depends on the sample characteristics. The Romanian and the German samples with AO means around 25 years and median at 24 years were similar to other published large samples of hospitalized patients (Tondo et al., 2010; Baldessarini et al., 2012) and to the world sample investigated by Merikangas et al. (2011). Similar to our preliminary report on a patient subgroup of the samples analyzed in the present study (Grigoroiu-Serbanescu et al., 2010, Priebe et al., 2012) and to Tozzi et al. (2011), we showed that both a three-AO-group distribution and a two-AOgroup distribution may fit the data equally well, as the AIC-score differences between the two models were close, only marginally favoring the three-AO-group model. We also showed that the upper limit of the EO group varies from sample to sample by two to four years, although the patients were recruited under similar conditions at the three sites, were diagnosed with the same instruments and the samples were large enough (354–882 cases). Under the three-AO-group model the upper limit of the EO group was around 21 years in the Romanian sample, 25 years in the German and Polish samples, and 25 years
M. Grigoroiu-Serbanescu et al. / Journal of Affective Disorders 168 (2014) 197–204
in the combined Romanian–German sample. Under the three-AO group model, the Romanian EO group had the same mean and upper age limit as the samples published by Bellivier et al. (2003) and Hamshere et al. (2009). When assessing the validity of the AO grouping, beyond the magnitude of the Akaike score, the graphical representation of the groups, as well as the case proportions should also be considered. A classification where the intermediate group is nearly completely overlapped by one of the other two groups should not be considered a valid classification. This was exemplified by the studies of Tozzi et al. (2011) and Lin et al. (2006), where the intermediate group was overlapped either by the EO or the LO group and unrealistic case proportions emerged in their middle onset groups. The authors of commingling analysis (MacLean et al., 1976) warned about the possibility of spurious admixture caused by biased selection and highly skewed distributions. In BP-I disorder the AO distribution is always skewed and sometimes thresholds based on quartiles or histograms may be more useful than thresholds derived by an admixture analysis that simplistically assumes all component distributions are Gaussian. The morbid risk for BP-I in first degree relatives of BP-I probands of 7% in the total Romanian sample is in line with previously published studies (Rice et al. 1987, (5.74%); Maier et al., 1993, (7%), Lichtenstein et al., 2009, (6.4%)) and with our risk figure (5.3%) previously found in a subgroup of the current sample (Grigoroiu-Serbanescu et al., 2001). MR figures for BP-I and for any major affective psychosis and schizophrenia in first degree relatives of BP-I probands computed on the basis of three AO categories provided by admixture analysis or based on predetermined cut-offs are not available in the literature to undertake a comparison. For our two-AO-group classification we may compare our MR figures with figures reported in two studies that used predetermined cut-offs. Taylor and Abrams (1981) found 9.18% risk for BP-I in first degree relatives of EO probands (AO r29) and 3.91% in relatives of LO probands (AO 429). Somanath et al. (2002) found a MR of 8.90% for BP-I in the first degree relatives of probands with AO o25 years and a MR of 3.14% in the relatives of probands with AO 425 years. This study had the same cut-off for the onset groups as our two-AO-group model. Using a predetermined AO cut-off (age 25) Helenius et al. (2013) dichotomized a Danish registry-based sample of 1204 BP-I probands and found a significantly higher psychopathology load in families of BP-I probands with early AO (o25 years) than in families of BP-I probands with late AO ( Z25 years). Rice et al. (1987) also found the highest MR for BP-I in first degree relatives of probands with AOo25 years. This finding is consistent with our result showing a significant MR difference both for BP-I and for all affective psychoses and schizophrenia between relatives of EO probands and relatives of LO probands in the two-AO-group classification generated by commingling analysis. The MR for all affective psychoses and schizophrenia started to decrease after age 29 but a significant decrease of the MR to first degree relatives was observed only after age 34. This finding is consistent with the observation of Gershon et al. (1982) according to whom the MR for BP-I þBP-II þschizoaffective disorders in first degree relatives of BP-I probands does not change up to age 30. Our results are also in accordance with studies reporting lack of difference in positive family history for affective disorders in comparisons involving three onset groups resulting from admixture analysis (Bellivier et al., 2001; Bellivier et al., 2003; Hamshere et al., 2009; Tozzi et al., 2011). In conclusion we showed that the number of valid AO groups that can be detected in a BP-I sample through commingling/ admixture analysis is not necessarily three and finding a universally valid age limit for defining early onset in BP-I disorder based
203
only on statistical analysis seems to be a difficult task. No previous study could document robust clinical differences in terms of psychotic traits, psychiatric family history, or rapid cycling among three onset groups derived by admixture analysis in BP-I. In additional, our study showed that the MR for all major psychoses to first degree relatives also fails to differentiate among three onset groups but differentiates between two onset groups with the most visible risk declining after age 34. The same finding was reported for unipolar major depression (Weissman et al., 1984) and schizophrenia (Byrne et al., 2002). It seems that the statement of Panariello et al. (2010) with respect to schizophrenia holds true also for BP-I disorder; they stated that “the intermediate onset group is probably an artifact overlapping the late-onset group” (p. 278). Although an intermediate onset group may exist between an EO and a LO group, its clinical usefulness seems to be doubtful. 4.1. Limitations We could not compute MR in the German and Polish samples because the family data were not available. A drawback of all studies concerning disease AO including our study consists in the retrospective estimation of the AO in probands, which might have introduced a recall bias, especially in those cases who were not hospitalized at the first episode or were older when entering the study. 34% of the first degree relatives of the Romanian probands were assessed through family history method that may result in inaccurate diagnostic information. 4.2. Strengths of the study No previous study investigated how well the theoretical AO distributions derived from admixture/commingling analysis fit the empirical data and none tested the same AO models in independent samples of Caucasian origin recruited under similar conditions (consecutive hospital admission) and diagnosed with the same diagnostic instruments. Our study is the first one that compared MR in first degree relatives of BP-I patients by onset groups generated by commingling/admixture analysis, not just percentages of positive family history. 66% of the first degree relatives of probands were directly interviewed. The MR in first degree relatives is the most genetic trait of a disorder and the effort of defining onset groups of BP-I not only through commingling/admixture analysis but also through clinical correlates is mostly intended to reduce the genetic heterogeneity of subgroups used in molecular analysis.
Role of funding source Funding agencies: The Romanian Academy, Grant no. 112/2005-2006 to M. Grigoroiu-Serbanescu; The Romanian Ministry for Education and Research – UEFISCDI, Bucharest, Romania, Grant nos. 122/2006; 42-151/2008, and 89/2012 to Maria Grigoroiu-Serbanescu.
Conflict of interest All authors declare that they have no conflicts of interests.
Acknowledgments The first author is indebted to drs. Dan Prelipceanu, Anca Tanase, Marina Codreanu, Dorina Sima, Mihaela Grimberg, Dana Cojocaru, Carmen Udrea, Anca Talasman, Mihail Gherghel, Radu Mihailescu þ from the Obregia Clinical Psychiatric Hospital, who participated to the diagnostic process of the patients involved in the study, and to the patients and their relatives.
References Akaike, H., 1974. A new look at the statistical model identification. IEEE Trans. Automat. Contr. AC-19, 716–723.
204
M. Grigoroiu-Serbanescu et al. / Journal of Affective Disorders 168 (2014) 197–204
Baldessarini, R.J., Tondo, L., Vázquez, G.H., Undurraga, J., Bolzani, L., Yildiz, A., Khalsa, H.M.K., Lai, M., Lepri, B., Lolich, M., Maffei, P.M., Salvatore, P., Faedda, G. L., Vieta, E., Tohen, M., 2012. Age at onset versus family history and clinical outcomes in 1665 international bipolar-I disorder patients. World Psychiatry 11, 40–46. Barnett, J.H., Smoller, J.W., 2009. The genetics of bipolar disorder. Neuroscience 164, 331–343. Bellivier, F., Golmard, J.L., Henry, C., Leboyer, M., Schurhoff, F., 2001. Admixture analysis of age at onset in bipolar I affective disorder. Arch. Gen. Psychiatry 58, 510–512. Bellivier, F., Golmard, J.L., Rietschel, M., Schulze, T.G., Malafosse, A., Preisig, M., McKeon, P., Mynett-Johnson, L., Henry, C., Leboyer, M., 2003. Age at onset in bipolar I affective disorder: Further evidence for three subgroups. Am. J. Psychiatry 160, 999–1001. Belmonte Mahon, P, Pirooznia, M, Goes, F.S., Seifuddin, F., Steele, J., Lee, P.H., Huang, J., Hamshere, M.L., Bipolar Genome Study (BiGS) Consortium, The Wellcome Trust Case Control Consortium Bipolar Disorder GroupDepaulo Jr, J.R., Kelsoe, J. R., Rietschel, M., Nöthen, M., Cichon, S., Gurling, H., Purcell, S., Smoller, J.W., Craddock, N., Schulze, T.G., McMahon, F.J., Potash, J.B., Zandi, P.P., 2011. Genome-wide association analysis of age at onset and psychotic symptoms in bipolar disorder. Am. J. Med. Genet. B Neuropsychiatr. Genet. 156B, 370–378. Box, G.E.P., Cox, D.R., 1964. An analysis of transformations (with discussion). J. R. Statist. Soc., Ser. B 26, 211–252. Byrne, M., Agerbo, E., Mortensen, P.B., 2002. Family history of psychiatric disorders and age at first contact in schizophrenia: an epidemiological study. Br. J. Psychiatry Suppl. 43, s19–25. Coryell, W., Fiedorowicz, J., Leon, A.C., Endicott, J., Keller, M.B., 2013. Age of onset and the prospectively observed course of illness in bipolar disorder. J. Affect. Disord. 146, 34–38. Craddock, N, Jones, I., 1999. Genetics of bipolar disorder. J. Med. Genet. 36, 585–594. Craddock, N., O'Donovan, M.C., Owen, M.J., 2009. Psychosis genetics: modeling the relationship between schizophrenia, bipolar disorder, and mixed (or “schizoaffective”) psychoses. Schizophr. Bull. 35, 482–490. Cross-Disorder Group of the Psychiatric Genomics Consortium, 2013. Identification of risk loci with shared effects on five major psychiatric disorders: a genomewide analysis. Lancet 381, 1371–1379 (April 20). Gershon, E.S., Dehisi, L.E., Hamovit, J., Nurnberger Jr., J.I., Maxwell, M.E., Schrelber, J., Davphinais, D., Dingman 2nd, C.W., Guroff, J.J., 1982. A controlled family study of chronic psychoses. schizophrenia and schizoaffective disorder. Arch. Gen. Psychiatry 45, 328–336 (Arch. Gen. Psychiatry 39, 1157-1167). Grigoroiu-Serbanescu, M., Martinez, M., Nöthen, M.M., Grinberg, M., Sima, D., Propping, P., 2001. Different familial transmission patterns in bipolar I disorder with onset before and after age 25. Am. J. Med. Genet.( B) - Neuropsychiatr. Genet. 105B, 765–773. Grigoroiu-Serbanescu, M., Nöthen, MM, Ohlraun, S, Propping, P, Maier, W, Wickramaratne, P, Georgescu, MJ, Prelipceanu, D, Grimberg, M, Sima, D, Rietschel, M., 2005. Family history influences age of onset in bipolar I disorder in females but not in males. Am. J. Med. Genet. (B) – Neuropsychiatr. Genet. 133B, 6–11. Grigoroiu-Serbanescu, M., Rietschel, M., Paul, T., Schulze, T.G., Noethen, M.M., Cichon, S., Elston, RC., 2010. Two or three age-of-onset groups in bipolar I disorder? Findings of commingling analysis in Romanian and German bipolar I patients. Eur. Psychiatry 25 (Suppl 1), 1428. Hamshere, M.L., Gordon-Smith, K., Forty, L., Jones, L., Caesar, S., Fraser, C., Hyde, S., Tredget, J., Kirov, G., Jones, I., Craddock, N., Smith, D.J., 2009. Age-at-onset in bipolar-I disorder: mixture analysis of 1369 cases identifies three distinct clinical sub-groups. J. Affect. Disord. 116, 23–29. Helenius, D., Jørgensen, P.M., Steinhausen, H.C., 2013. A three generations nationwide population study of family load estimates in bipolar disorder with different age at onset. J Affect Disord. 150, 146–151. Javaid, N., Kennedy, J.L., De Luca, V., 2011. Ethnicity and age at onset in bipolar spectrum disorders. C.N.S. Spectr., 16 (ePub). Kennedy, N., Everitt, B., Boydell, J., Van Os, J., Jones, P.B., Murray, R.M., 2005. Incidence and distribution of first-episode mania by age: results from a 35-year study. Psychol. Med. 35, 855–863.
Lichtenstein, P., Yip, B.H., Björk, C., Pawitan, Y., Cannon, T.D., Sullivan, P.F., Hultman, C.M., 2009. Common genetic determinants of schizophrenia and bipolar disorder in Swedish families: a population-based study. Lancet 373 (9659), 234–239. Lin, P.I., McInnis, M.G., Potash, J.B., Willour, V., MacKinnon, D.F., DePaulo, J.R., Zandi, P.P., 2006. Clinical correlates and familial aggregation of age at onset in bipolar disorder. Am. J. Psychiatry 163, 240–246. MacLean, C.J., Morton, N.E., Elston, R.C., Yee, S., 1976. Skeness in commingled distributions. Biometrics 32, 695–699. Manchia, M., Lampus, S., Chillotti, C., Sardu, C., Ardau, R., Severino, G., Del Zompo, M., 2008. Age at onset in Sardinian bipolar I patients: evidence for three subgroups. Bipolar. Disord. 10, 443–446. Maier, W., Lichtermann, D., Minges, J., Hallmayer, J., Heun, R., Benkert, O., Levinson, D.F., 1993. Continuity and discontinuity of affective disorders and schizophrenia. Results of a controlled family study. Arch. Gen. Psychiatry 50, 871–873. Merikangas, KR, Jin, R, He, J.P., Kessler, R.C., Lee, S., Sampson, N.A., Viana, M.C., Andrade, L.H., Hu, C., Karam, E.G., Ladea, M., Medina-Mora, M.E., Ono, Y., Posada-Villa, J., Sagar, R, Wells, JE, Zarkov, Z., 2011. Prevalence and correlates of bipolar spectrum disorder in the world mental health survey initiative. Arch. Gen. Psychiatry 68, 241–251. Moorhead, SR, Young, AH., 2003. Evidence for a late onset bipolar-I disorder subgroup from 50 years. J. Affect. Disord. 73, 271–277. Ortiz, A., Bradler, K., Slaney, C., Garnham, J., Ruzickova, M., O'Donovan, C., Hajek, T., Alda, M., 2011. An admixture analysis of the age at index episodes in bipolar disorder. Psychiatr. Res. 188, 34–39. Panariello, F., O'Driscoll, L., De Souza, R.P., Tiwari, A., Manchia, M., Kennedy, J., De Luca, V., 2010. Age at onset in Canadian Schizophrenia patients: admixture analysis. Schizophr. Res. 122, 278–279. Power, RA, Keers, R, Ng, MY, Butler, AW, Uher, R, Cohen-Woods, S, Ising, M, Craddock, N, Owen, MJ, Korszun, A, Jones, L, Jones, I, Gill, M, Rice, JP, Hauser, J, Henigsberg, N, Maier, W, Zobel, A, Mors, O, Placentino, AS, Rietschel, M, Souery, D, Kozel, D, Preisig, M, Lucae, S, Binder, EB, Aitchison, KJ, Tozzi, F, Muglia, P, Breen, G, Craig, IW, Farmer, AE, Müller-Myhsok, B, McGuffin, P, Lewis, CM., 2012. Dissecting the genetic heterogeneity of depression through age at onset. Am. J. Med. Genet (B) – Neuropsychiatr. Genet. 159B, 859–868. Priebe, L., Degenhardt, F., Herms, S., Haenisch, B., Mattheisen, M., Nieratschker, V., Weingarten, M., Witt, S., Breuer, R., Paul, T., Alblas, M., Moebus, S., Lathrop, M., Leboyer, M., Schreiber, S., Grigoroiu-Serbanescu, M., Maier, W., Propping, P., Rietschel, M., Nöthen, M.M., Cichon, S., Mühleisen, T.W., 2012. Genome-wide survey implicates the influence of copy number variants (CNVs) on the development of early-onset bipolar disorder. Mol. Psychiatry 17, 421–432. Rice, J., Reich, T., Andreasen, N.C., Endicott, J., Van Eerdewegh, M., Fishman, R., Hirschfeld, R.M., Klerman, G.L., 1987. The familial transmission of bipolar illness. Arch. Gen. Psychiatry, 441–447. S.A.G.E., 2012. Statistical Analysis for Genetic Epidemiology, Release 6.3.〈http:// darwin.cwru.edu〉. Smoller, J.W., Finn, C.T., 2003. Family, twin, and adoption studies of bipolar disorder. Am. J. Med. Genet. (C) – Semin. Med. Genet. 123C, 48–58. Somanath, C.P., Jain, S., Reddy, Y.C., 2002. A family study of early-onset bipolar I disorder. J. Affect. Disord. 70, 91–94. Taylor, M.A., Abrams, R., 1981. Early- and late-onset bipolar illness. Arch. Gen. Psychiatry 38, 58–61. Tondo, L., Lepri, B., Cruz, N., Baldessarini, R.L., 2010. Age at onset in 3014 Sardinian bipolar and major depressive disorder patients. Acta Psychiatr. Scand. 121, 446–452. Tozzi, F., Manchia, M., Galwey, N.W., Severino, G., Del Zompo, M., Day, R., Matthews, K., Strauss, J., Kennedy, J.L., McGuffin, P., Vincent, J.B., Farmer, A., Muglia, P., 2011. Admixture analysis of age at onset in bipolar disorder. Psychiatr. Res. 185, 27–32. Weissman, M.M., Wickramaratne, P., Merikangas, K.R., Leckman, J.F., Prusoff, B.A., Caruso, K.A., Kidd, K.K., Gammon, G.D., 1984. Onset of major depression in early adulthood. Increased familial loading and specificity. Arch. Gen. Psychiatry 41, 1136–1143.