Clinical correlates of age at onset distribution in bipolar disorder: a comparison between diagnostic subgroups

Bipolar disorder (BD) is a heritable psychiatric illness typically characterized by cyclic mood episodes of opposite polarity al- ternating with intervals of well-being. As in other psychiatric complex genetic diseases, the relatively high clinical heterogeneity of BD might have hindered the identification of molecular and clinical determinants of risk as well as of predictors of treatment outcome. The magnitude of heterogeneity might be modifiable by studying subgroups of BD patients sharing specific clinical char- acteristics such as, for instance, patterns of treatment response, mood incongruent psychosis or early illness onset. Indeed, the extensive analysis of age at onset (AAO) BD subgroups through admixture analysis has shown clinical and genetic characteristics specific, particularly, to early onset (EO) BD. As the vast majority of studies investigated BD type 1 (BD1) samples, it remains to be established, however, whether this clinical delineation of EO is present also in BD type 2 (BD2) patients. Furthermore, the distributional properties of AAO have never been investigated in sample exclusively comprised of BD2 patients. This study aims to test whether bipolar disorder type 1 (BD1) and bipolar disorder type 2 (BD2) differed in terms of age at onset (AAO) distributions and whether early onset (EO) BD patients had clinical characteristics specific for each diagnostic-subgroup. We studied 515 BD patients (279 BD1 and 224 BD2 and 12 BDNOS) diagnosed according to DSM-IV criteria. AAO was defined as the first reliably diagnosed hypo/manic or depressive episode according to diagnostic criteria. We used normal distribution mixture analysis to test whether we could identify subgroups of patients according to the AAO. We investigated a range of number of AAO groups (1 to 9). The choice of the mixture model that best fit the distribution of AAO was made according to the Schwarz’s Bayesian information criteria (BIC). Specifically, the analysis performed with the “Mclust” package implemented in R indicates the best model as the one with the highest BIC among the fitted models. Cut off points were derived using the theoretical AAO function and calculating each data point’s probability of belonging to each class. Clinical correlates of early onset were analyzed using univariate analysis (t-test or contingency tables as appropriate). Multivariate logistic regression analysis was used to account for intercorrelations. A two normal components model best fitted the observed distribution of AAO in BD1 (BIC = −1599.3), BD2 (BIC = −2158.4), and in the whole sample (BIC = −3854.9). Early onset (EO) BD1 had a lower age at interview and a longer duration of illness than late onset (LO) BD1. Early onset BD2 had also a lower age at interview and longer illness duration than LO BD2, as well as higher rate of comorbidity with alcohol dependence. A higher number of EO BD2 presented with a DMI course, whilst a higher rate of MDI course was found in EO BD1. We showed the presence of similar, bimodal, AAO distributions in BD1 and BD2, confirming however significant differences in terms of clinical characteristics of the different onset subgroups