Preliminary studies suggested that the age at onset of bipolar affective disorder can be used to define homogeneous disorder subtypes, which can be used to identify vulnerability genes. Early-onset bipolar disorder is associated with a greater familial risk of bipolar disorder, higher rates of comorbidity, poor prognosis, and poor response to lithium R1605BGBHCBCA–R1605BGBCIBCE. Age at onset is also correlated in affected siblings, suggesting that some familial vulnerability factors are specific to the age at onset R1605BGBCHAGA.

On the basis of admixture analysis in bipolar disorder, we recently proposed a model with three age-at-onset subgroups (5). We tested the validity of this model in two new groups of patients recruited from an international collaboration. In a group of consecutively admitted patients, we tested whether the age-at-onset function differed significantly from that identified in the previous study R1605BGBIIHIF. In a group of bipolar sibships, we investigated whether affected siblings were more likely to belong to the same theoretical subgroup as identified by the admixture analysis.

All consecutive patients with bipolar I disorder (N=368) admitted to three centers (in France, Switzerland, and Germany) between 1994 and 2001 were interviewed with the Diagnostic Interview for Genetic Studies R1605BGBIEJHI. We also recruited 130 bipolar patients belonging to 59 sibships from centers in Ireland, Germany, and France between 1992 and 1997 (previously described in reference R1605BGBCHAGA).

Age at onset was defined as the age at which the patient first met the DSM-IV criteria for a thymic episode. Age at onset was assessed by the interviewer (F.B., T.G.S., M.P., L.M.-J. and C.H. were among the interviewers) and then blindly rated by an independent psychiatrist according to medical case notes and the Diagnostic Interview for Genetic Studies. Written informed consent was obtained from all subjects. All interviewers were blind to the hypothesis tested.

We previously used admixture analysis to determine the model that best fit the observed distribution of the ages at onset R1605BGBIIHIF. A model with three subgroups was identified. We tested the validity of this model in two ways. First, we used admixture analysis to determine the model that best fit the observed distribution of the ages at onset in our prospectively recruited patients with bipolar I disorder. The Kolmogorov-Smirnov test was used to determine whether the age-at-onset distribution observed in the new study group was consistent with the probability function modeled previously.

Second, for the affected sib pairs, we used a Monte Carlo test with 10,000 simulations to determine whether the age-at-onset classes identified by the model were independent of sibship (null hypothesis) or whether affected siblings were more likely to belong to the same age-at-onset subgroup (alternative hypothesis). The probability that all siblings within the same family belonged to the same theoretical age-at-onset subgroup was calculated for the subject group. This probability was also calculated in 10,000 simulations in which bipolar patients were randomly assigned to families. The Monte Carlo p value is given by the number of times (in the 10,000 simulations) that the probability for simulated study groups exceeded the probability calculated for the real group.

For the group of 368 bipolar I patients, the likelihood ratio indicated that the model with three distributions fit significantly better the observed distribution of age at onset than did the model with two distributions (χ²=25.1, df=3, p<0.0001). No further improvement was obtained with a four-component model. Thus, the model that best fit the observed age-at-onset distribution was a mixture of three Gaussian distributions: mean=17.6 years (SD=1.8), 21.4% of the patients; mean=24.6 years (SD=6.1), 57.3% of the patients; and mean=39.2 years (SD=9.6), 21.2% of the patients.

The age-at-onset distribution observed in the present group (N=368) was not statistically different from that modeled previously (N=211) R1605BGBIIHIF. The Kolmogorov-Smirnov value was 0.06, which corresponds to a p value of 0.08, with N=368. As no significant difference between the two groups was observed, they were pooled to obtain the best-fitting model for analysis of the affected sib pairs. For the combined group (N=579), the model that best fit the observed distribution of ages at onset was a mixture of three distributions: mean=17.4 years (SD=2.3), 27.9% of the patients; mean=25.1 years (SD=6.2), 50.1% of the patients; and mean=40.4 years (SD=11.3), 21.9% of the patients. The probability of belonging to a given age-at-onset subgroup, i, is where x is the age at onset for a given subject, G_i and G_j are the age-at-onset subgroups, and the subgroups are defined by μ (mean), σ (standard deviation), and P(G) (proportion of the population). The probability density functions are shown in F1.

We calculated the probability that all members of a given family belonged to the same class. In the Monte Carlo test, the probability for the simulation (random assignment to families) exceeded that for the real study group only once in 10,000 simulations (p=0.0001). Thus, affected siblings were more likely to belong to the same theoretical subgroup. The probability that all siblings within a family belonged to the same theoretical age-at-onset subgroup was calculated for each family. The families were then grouped, with each family assigned to the age-at-onset subgroup to which it had the highest probability of belonging (1, 2, or 3). Twenty-eight (47.5%) of the 59 families belonged to the early-onset group, 29 (49.2%) to the intermediate-onset group, and two (3.4%) to the late-onset group.

Our main finding was that the age-at-onset model produced for the original group of 211 patients with bipolar I disorder fit the age-at-onset function observed in an independent group of consecutively recruited bipolar I patients (N=368). This strongly suggests the existence of three genuine age-at-onset subgroups of patients with bipolar affective disorder. For the first subgroup, the mean age at onset was 17.4 years (SD=2.3), and 27.9% of the subjects were included. The second subgroup had a mean age at onset of 25.1 years (SD=6.2) and a 50.1% share of the patients. The mean age at onset for the third subgroup was 40.4 years (SD=11.3), and 21.9% of the patients were included.

We also found that bipolar siblings are much more likely to belong to the same theoretical age-at-onset subgroup than to different subgroups. This finding confirms and extends the results of previous studies of sib pairs and twins showing an intrafamilial correlation for age at onset in bipolar affective disorder R1605BGBCHAGA, R1605BGBEDHGC. The existence of three age-at-onset subgroups and the clustering of siblings in the same subgroup might be attributable to different genetic vulnerability factors and/or the same genetic mechanism differently expressed in various environments at different periods in life. This genetic heterogeneity in bipolar subgroups defined according to age at onset is consistent with the results of a recent segregation analysis of bipolar affective disorder, which showed that a major gene with a polygenic component is involved in early-onset bipolar disorder, whereas late-onset bipolar disorder is compatible with a multifactorial model R1605BGBDCCBA. Preliminary results have been obtained in genetic association studies on subgroups of bipolar patients defined according to age at onset R1605BGBJBEIH, R1605BGBEFFAB.

Comparison of the estimated ages at onset and frequencies of the three subgroups in the present study group and in the previous group R1605BGBIIHIF deserves comment. In the current study, group 1 had a mean age of 17.6 years (SD=1.8) and a frequency of 21.4%; in the earlier study the mean age and frequency were 16.9 years (SD=2.7) and 41.4%, respectively. Group 2 in the current study had a mean age of 24.6 years (SD=6.1) and a frequency of 57.3%; in the previous study the mean age and frequency were 26.9 years (SD=5.0) and 41.9%. Group 3 in the current study had a mean age of 39.2 years (SD=9.6) and a frequency of 21.2%; in the previous study the mean age was 46.2 years (SD=8.0) and the frequency was 16.6%. The mean ages and standard deviations of each theoretical subgroup were similar in the two study groups. However, the proportions of the patients in each subgroup differed considerably, and there was a lower proportion of patients with an early onset in the present study. This may be due to differences in recruitment procedures. The first study group (N=211) was recruited by two French research teams, using similar protocols. However, the patients in the current study (N=368) were gathered by pooling independently recruited patients. Between-center heterogeneity in recruitment procedures (hospital versus community, lithium clinic, adolescent units, etc.) may account for these differences in the proportions of patients with early-, intermediate-, and late-onset bipolar disorder. If some genetic vulnerability factors are specific to age at onset, that may account for the conflicting results obtained in some genetic studies.

Since the three age-at-onset subgroups are observed at different ages, several confounding factors may have influenced the estimated proportions of patients in each subgroup: effect of birth cohort R1605BGBJGEEC, differential censoring due to suicide, and differential validity of evaluation of age at onset (recall bias is more common in patients who are interviewed at a later age and have no living parents who can provide valid information on age at onset).

Further studies are required to confirm the existence of three age-at-onset subgroups in bipolar affective disorder and to determine whether the use of such subgroups will facilitate the identification of 1) more genetic subgroups, 2) specific genetic vulnerability factors underlying each subgroup, and 3) the vulnerability factors involved in onset of the disorder.

Received Dec. 19, 2001; revisions received June 5 and Sept. 24, 2002; accepted Oct. 17, 2002. From the Department of Psychiatry, Hôpital Henri Mondor et Albert Chenevier, Assistance Publique-Hôpitaux de Paris; the Unité Neurobiologie et Psychiatrie (Institut National de la Santé et de la Recherche Médicale Unité 513), Faculté de Médecine de Créteil, Créteil, France; the Département de Biostatistiques et Informatique et Institut National de la Santé et de la Recherche Médicale Unité 436, University Hospital Pitié-Salpétrière, Université Paris VI, Paris; the Department of Psychiatry, University of Bonn, Bonn, Germany; the Department of Psychiatry, University of Chicago; the Division de Neuropsychiatrie, Hôpital Belle-Idée, Hôpitaux Universitaire de Genève, Geneva; the University Department of Adult Psychiatry, Hôpital de Cery, Lausanne, Switzerland; the Department of Psychiatry, St. Patrick’s Hospital, Dublin; and the Department of Psychiatry, Unité d’Investigation Clinique Approfondie, Bordeaux, France. Address reprint requests to Dr. Bellivier, Service de Psychiatrie, Hôpital Henri Mondor et Albert Chenevier, 51, avenue du Mal de Lattre de Tassigny, 94010 Créteil, Cedex, France; bellivier@im3.inserm.fr (e-mail). Supported by Contrat de Recherche Clinique 94232 from Assistance Publique-Hôpitaux de Paris, by Programme Hospitalier de Recherche Clinique AOM98152 from the Ministère de la Recherche of France, by grants from the Institut National de la Santé et de la Recherche Médicale, and by grants 32-40677.94 and 32-049746.96 from the Swiss National Science Foundation.