Demographic and diagnostic characteristics
STEP-BD enrolled a total of 4,360 individuals, of whom 4,107 fully completed baseline evaluations; 3,750 (91%) were diagnosed with bipolar I disorder or bipolar II disorder. The remainder, who met criteria for bipolar disorder NOS, cyclothymia, or schizoaffective disorder, bipolar type, were excluded from the present analysis because by definition their mood episodes were briefer or less clearly delineated, and thus onset of mood disorder was more difficult to define. Of the 3,750 individuals identified, age of onset was reported for 3,658, forming the cohort analyzed here, which included 1,068 (29.2%) with childhood/prepubertal onset, 1,403 (38.4%) with adolescent onset, and 1,187 (32.5%) with adult onset. A histogram of onset-age distribution is shown in ; 2,786 (76.2%) experienced onset before age 21.
Age at illness onset in bipolar disorder (n = 3,658).
Characteristics of the full STEP-BD cohort by age of onset are presented in . Consistent with cross-sectional findings from a subset of 1,000 patients (3
), subjects with childhood onset were younger and significantly more likely to be female, to have bipolar II disorder, to have rapid cycling in the prior year, and to have substance use disorder and comorbid anxiety disorder in the past year. Those with earlier onset were also less likely to be euthymic at study entry [χ2
(2 df) = 72.0, p < 0.001], more likely to be depressed [χ2
(2 df) = 10.8, p = 0.004], and more likely to be elevated (hypomanic, manic, or mixed) [χ2
(2 df) = 58.1, p < 0.001] . As expected, age at onset was inversely correlated with illness duration (r
= −0.34, p < 0.0001). Use of lithium, valproate, or antipsychotics at recovery did not differ significantly between onset-age groups (p
0.10 for all comparisons).
Comparison of baseline features of onset-age groups (n = 3,658)
We first examined the hypothesis that those individuals with early onset of bipolar disorder would experience shorter times to recurrence. illustrates the Kaplan- Meier survival curves adjusted for illness duration for the 2,237 subjects, with at least one prospectively observed recovery period of eight or more weeks (i.e., remission). These analyses included a mean of 8.6 visits (SD 6.7, range 2–45) over a mean follow-up of 309 days (median 211 days). Those with childhood onset experienced the earliest recurrence (log-rank χ2 = 14.98, p = 0.0001, childhood versus adult-onset), and those with adolescent onset also experienced recurrence sooner than those with adult onset (χ2 = 6.87, p = 0.01, adolescent versus adult onset). No significant difference was identified between child- and adolescent-onset groups (χ2 = 1.79, p = 0.18). Median days to recurrence were 308, 418, and 542 for the child-, adolescent- , and adult-onset age groups, respectively. In Cox regression adjusting for illness duration and bipolar I versus II status, onset at or before age 18 was associated with earlier recurrence than onset after age 18 [hazard ratio (HR) 1.17 (1.03–1.38)], but onset before 13 did not confer significant additional hazard [HR 1.08 (0.92–1.26)]. In all three groups, the majority of recurrences were major depressive episodes [74.4%, 74.1%, and 72.1% for child, adolescent, and adult groups, respectively; χ2(2 df) = 0.52, p = 0.78]; manic recurrences comprised 6.9%, 7.4%, and 10.9% of all recurrences in the three groups [χ2(2 df) = 4.16, p = 0.12]; mixed episodes comprised 6.2%, 4.1%, and 4.1% [χ2(2 df) = 2.07, p = 0.36]; and the remainder were hypomanic. Incorporating individual clinical and sociodemographic features which differed significantly between onset-age groups () in no case led to a change in HR for recurrence by 10% or greater, suggesting an absence of confounding.
Time to recurrence of depression, mania, hypomania or mixed episodes during up to two years of follow-up, by retrospectively-obtained onset-age group, adjusted for illness duration.
We next examined the hypothesis that early onset would also be associated with greater chronicity, in terms of proportion of days well in the first year of follow-up. Individuals had a mean of 7.2 visits (SD 5.9) in addition to the baseline and quarterly visits. For childhood-, adolescent-, and adult-onset patients, the percent of days euthymic was 42.0 (SD 0.3), 47.1 (SD 0.3), and 54.0 (SD 0.3), respectively; F (2 df) = 35.9, p < 0.0001. Post hoc pairwise comparisons with Bonferroni correction indicated significant differences between all three groups (p < 0.001). These differences remained significant (p < 0.05) in a regression model incorporating terms for illness at study entry, illness duration, and bipolar I versus II status (not shown).
Finally, we examined functional status and quality of life at study entry and quarterly follow-ups over two years (). A univariate ANOVA indicated significant differences for LIFE-RIFT total scores at study entry (F = 22.49, p < 0.0001). Post-hoc pairwise comparisons indicated that the childhood-onset and adolescent-onset groups were each more impaired than the adult-onset group (p < 0.0001). The childhood-onset group functioned significantly more poorly than the adolescent-onset group at baseline (p < 0.0001), but the adolescent- and adult-onset groups did not differ significantly (p = 0.13). Likewise, an ANOVA indicated significant differences for QLESQ scores at study entry (F = 24.13, p < 0.0001), with significantly poorer quality of life for the earlier-onset group in all pairwise comparisons (p < 0.0001 for childhood-onset versus the two other groups, p = 0.005 for the adolescent- versus adult-onset group). Differences remained significant (p < 0.05) in regression models after adjustment for illness at study entry, illness duration, and bipolar I versus II status.
Longitudinal measures of quality of life (A) and functional status (B) among individuals with retrospectively-obtained childhood, adolescent, and adult onset of bipolar disorder.
In mixed-effects regression models, we examined change in LIFE-RIFT and QLESQ scores during 24-month follow-up, adjusting for baseline values as well as bipolar subtype and mood state at study entry. No significant differences in slopes were observed for the three groups (p
0.05 for childhood- and adolescent-onset groups).