Because of the complex sampling procedures used in both studies, we conducted analyses using SUDAAN software (Version 9; Research Triangle Institute, 2004
). SUDAAN software is avail able as an adjunct to the SAS/STAT software package (Version 9.1; SAS Institute, 2003
) and allows for the application of sampling weights, commonly used to adjust large epidemiologic data sets that employ complex sampling, to the data prior to computing common statistics. Specifically, a sample design is specified to allow the program to estimate standard errors. We specified a Taylor Linearization Method Without Replacement (Wolter, 1985
) for the computation of standard errors in our analyses. Preliminary analyses using SUDAAN with this particular sample design specification resulted in replication of previously published prevalence estimates and corresponding standard errors for NESARC data.
Using DSM–IV diagnostic scoring, the overall prevalence of past-12-month BP disorder in the NESARC sample was 2.02% (SE = 0.09) for BPI and 0.82% (SE = 0.06) for BPII, in contrast to a past-12-month prevalence of BPI of 0.6% (SE = 0.10) and 0.8% (SE = 0.10) for BPII in the NCS–R using strict diagnostic scoring.
When the NESARC sample is stratified by age, the past-12-month prevalence using DSM–IV diagnoses was found to be higher among younger participants (see ) for BPI, χ2(6, N = 43,093) = 22.38, p <.001; BPII, χ2(6, N = 43,093) = 13.45, p<.001; and BPI and BPII combined, χ2(6, N = 43,093) = 26.23, p <.001, with similar age trends evident for lifetime diagnoses, χ2(6, N = 43,093) = 22.36, 15.30, 26.14 for BPI, BPII and BPI/III, respectively (all ps <.001). In particular, the prevalence of past-12-month BPI for 18–20-year-olds (4.41%, SE = 0.52) and BPII (2.87%, SE = 0.46) was significantly greater than among participants over 25 years old. Compared with the 18–20-year-old cohort, odds ratios (ORs) for the 25–29 cohort and all older cohorts are significant, with cohorts over 25 years old being at decreased odds of diagnosing compared with 18–20 year-olds (ORs ranging from 0.11 vs. the 60+ group to 0.80 vs. the 25–29 group; all ps <.01). Results in the NCS–R sample using strict scoring showed similar age trends for Strict-BPI, χ2(6, N = 9,282) = 3.13, p =.013; Strict-BPII, χ2(6, N = 9,282) = 5.01, p <.001; Other BP, χ2(6, N = 9,282) = 4.58, p = .001; and BP spectrum, χ2(6, N = 9,282) = 8.20, p <.001. Although these age gradients are similar, it should be noted that the NESARC sample tended to have higher rates of bipolar disorders across all age groups.
Because the original scoring of the NESARC and the strict NCS–R samples yield different overall prevalences and different degrees of an age gradient, despite the fact that both are population-based, national samples, we explored the extent to which these large study differences could be attributable to different scoring algorithms. We therefore recoded NESARC diagnoses adapting the NCS–R Strict criteria and recalculated prevalence as a function of age. The resulting prevalence rates of BP disorder in the NESARC sample are more comparable to those in the NCS–R sample (see ). Specifically, the recoded Strict NESARC diagnoses yielded a reduction in both BPI (from 2.02% to 1.15%) and BPII (from 0.82% to 0.46%), although NESARC-Strict BPI diagnoses remained systematically more prevalent than NCS–R diagnoses (see ). However, despite these changes in the overall rates of diagnoses, when NESARC was scored using the “strict” algorithm, the age gradient remained for Strict-BPI, χ2(6, N = 43,093) = 18.35, p <.001, and Strict-BPII, χ2(6, N = 43,093) = 11.90, p< .001 (see ).
Figure 2 Prevalence of past-12-month bipolar spectrum disorders using the strict criteria from National Comorbidity Survey—Replication (NCS–R; Kessler & Merikangas, 2004) in the NCS–R sample (N = 9,278) and the National Epidemiological (more ...)
Rates of Strict-BPI in the NCS–R sample (see ) are slightly lower than those for NESARC, with the decline in prevalence occurring after the third decade of life (30–39), slightly later than the decline in NESARC at 25–29. Despite this, the age gradient is still observed in NCS–R for Strict-BPI (see ). Although the rates of Strict-BPII in the NCS–R sample appear to be relatively constant until after the sixth decade of life (>59), with the exception of 25–29-year-olds, we believe this may be due to a smaller sample size relative to the NESARC sample (note the large confidence intervals at younger age groups). It is also possible that there is greater inconsistency in Strict-BPII diagnoses generally, because of the impact of depressive episodes as required by the diagnostic criteria. When we examined rates of past-12-month hypomania regardless of major depressive episodes, the pattern across age groups was much more consistent (1.02%, SE = 0.55; 2.03%, SE = 0.81; 0.57%, SE = 0.29; 0.49%, SE = 0.19; 0.11%, SE = 0.07; 0.17%, SE = 0.12; and 0.00% among 18 –20, 21–24, 25–29, 30 –39, 40 – 49, 50 –59, and >59 age groups, respectively). The prevalence of hypomania demonstrates a decline at age 25–29, similar to that in NESARC, χ2(6, N = 9,282) = 2.90, p = .019. Thus, at least in NCS–R, criteria for past-12-month Strict-BPII appear to be more heavily influenced by past-12-month major depression than by past-12-month hypomania in older cohorts. This suggests that the age gradient in Strict-BP diagnosis, especially Strict-BPII, is more specific to the mania/hypomania pole than to the depressive pole.
Age Differences in the Pattern of Symptom Endorsement
To understand better if the age gradient for NESARC Strict-BPI is attributable, in part, to age-related endorsement of specific symptoms, we examined the prevalence of criterion symptoms among those diagnosed with Strict-BPI. There were no significant differences across age groups in endorsement of elation, irritability, or past-12-month depression among participants with Strict-BPI, χ2(6, N = 494) = 0.12, 1.24, and 0.59 for elation, irritability, and depression, respectively (all ps >.05). However, participants diagnosed with Strict-BPI who were 18 –20 years old were less likely to endorse racing thoughts than were age groups over 25, χ2(6, N = 494) = 4.07, p = .002. Compared with the 18 –20-year-old cohort, ORs for the 25–29 cohort and older cohorts up to age 59 were significant (the ORs for >59 year-old cohort could not be computed because of small cell sizes), with cohorts over 25 years old being at increased odds of reporting racing thoughts when compared with 18 –20 year olds (ORs ranging from 3.54 vs. the 30 –39 group to 13.52 vs. the 25–29 group; all ps < .05). There were no other significant differences in criteria endorsement. These analyses do suggest, however, that the absence of racing thoughts could be a correlate of a developmentally limited form of BP.
Age Differences in Impairment and Treatment Experiences
NESARC measured impairment due to symptoms across five areas: personal discomfort, interpersonal, work or school, productivity, and legal. Participants who were diagnosed with Strict-BPI and were 18–20 years old endorsed more legal impairment than did the 25–29 (OR = 4.69; 95% CI = 1.59, 13.85) and 50–59 (OR = 5.71; 95% CI = 2.03, 16.09) year-old age groups, χ2(6, N = 494) = 2.58, p = .026. No additional differences were found for impairment among participants with Strict-BPI. Differences across age groups in lifetime prevalence of help-seeking were found for participants diagnosed with Strict-BPI. Specifically, younger participants (age 18–39) were less likely to endorse seeing a health care provider for manic symptoms, χ2(6, N = 494) = 2.34, p = .042, than were older participants; and 18–29-year-olds were less likely to endorse using medications for manic symptoms, χ2(6, N = 494) = 4.52, p <.001, than were older participants.
On the basis of examination of age differences in symptoms and impairment, there appear to be relatively few differences between participants who diagnose between the ages of 18 and 20 and those who diagnose at older ages. However, the younger Strict-BPI participants reported fewer racing thoughts and more legal problems. Thus, with these two exceptions, at least internally, the diagnoses in NESARC appear to be generally consistent across age.
If the apparent excess of BPI diagnoses in emerging adulthood represents Type I error in diagnosis, we might expect to see either greater or lesser comorbidity with conditions that frequently co-occur with BPI and BPII. If the age gradient is a result of disorder-specific misdiagnoses in the younger age groups (i.e., simple false positives), then we would expect to find that younger individuals have less comorbidity with other disorders that commonly co-occur with BP disorder. Conversely, relatively high levels of comorbidity could result from common reporting biases or relevant age-related criterial bias in BP and near neighbor diagnoses. Following this logic, rates of comorbidity that are statistically equivalent across age groups may suggest that the rates of Type I error in the NESARC study are random and not related to age. Thus, to assess age differences in the external validity of these diagnoses, we examined comorbidity with past-12-month anxiety disorders (5.73%, SE = 0.19 overall; 35.91%, SE = 1.75 among those with a BP spectrum diagnosis), lifetime ASPD (3.63%, SE = 0.15 overall; 20.97%, SE = 1.49 among those with a BP spectrum diagnosis), childhood conduct disorder (4.69%, SE = 0.16 overall; 23.51%, SE = 1.54 among those with a BP spectrum diagnosis), and past-12-month drug (2.00%, SE = 0.10 overall; 11.69%, SE = 1.36 among those with a BP spectrum diagnosis) and alcohol use disorders (8.46%, SE = 0.24 overall; 23.41%, SE = 1.58 among those with a BP spectrum diagnosis) across age strata. Overall, there is significant comorbidity with each of these conditions across all possible BP disorder diagnoses, with significant ORs for childhood conduct disorder, adult ASPD, substance use disorder diagnoses, and anxiety disorders (see ) across age groups.
Childhood conduct disorder (see , upper panel) and adult ASPD (see , lower panel) were consistently more likely across all age groups and all levels of BP spectrum disorder, with few exceptions (ORs ranged from 3.92 to 11.69 across all age groups for BP spectrum diagnosis; ps < .05).
Drug and alcohol use disorders were not as consistent as conduct disorder and ASPD but, nonetheless, appear to be more likely among those with a BP spectrum disorder in general. Drug use disorders were most likely to occur in participants with a Strict-BPI diagnosis (ORs ranged from 5.76 to 26.03 across age groups; all ps < .05; see , upper panel). Alcohol use disorders were also most consistently associated with Strict-BPI diagnoses (ORs ranged from 1.46 to 7.60; all ps < .05, except among participants 50 and older; see , lower panel).
Anxiety disorders were consistently more likely among all levels of BP spectrum disorders and age groups (ORs ranged from 3.16 for 25–29 year-olds with Strict-BPII to 30.98 for 60 and older with Strict-BPII; all ps < .05; see ). Overall, results from these analyses suggest that a diagnosis of BP spectrum disorder at any age is associated with significant impairment and illness burden.
As noted earlier, Merikangas et al. (2007)
recently highlighted both the prevalence and clinical relevance of subthreshold BP symptomatology in the general population but failed to address the question of the age gradient of this type of symptomatology. To determine whether subthreshold symptomatology follows the same basic age gradient we observed for Strict-BPI, we estimated the prevalence of Other BP across age strata in both the NCS–R and NESARC, using the recalibrated criteria developed by NCS–R and used by Merikangas et al. (2007)
; these estimates are provided in . Similar rates of past-12-month Other BP were obtained across the two samples with similar age gradients in both the NCS–R, χ2
= 9,282) = 4.58, p
<.01, and the NESARC, χ2
= 43,093) = 13.44, p
<.001 (see ). Rates of Other BP disorder averaged across the 18–20- and 21–24-year-old cohorts are nearly twice as high (NESARC, 2.99%, SE
= 0.30; NCS–R, 3.4%, SE
= 0.63), as in the 25–29-year-old cohort (NESARC, 1.57%, SE
= 0.25; NCS–R, 1.5%, SE
= 0.38) for both samples. The rates of Other BP disorder in the 18–20 age group in NCS–R appear to be lower than expected, given the rates in the 21–24 age group. This is exacerbated by the small number of participants that comprise these small age intervals (3–5 years), resulting in bigger standard errors, compared with older age groups that contain larger age intervals (10 years) and, thus, relatively large numbers of participants. Further, the discrepancy between rates observed in the NCS–R and the NESARC data for this particular age group may be related to the sampling frame used by both studies to capture students residing in group housing units, such as residence halls and fraternity/sorority housing. Although both studies attempted to sample from such group housing units, their methods were different (for NCS–R procedures, see Kessler et al., 2004
; for NESARC procedures, see online documentation at http://www.nesarc.niaaa.nih.gov/
). When we examined the percentage of 18–20-year-olds who reported being students, it was significantly lower in NCS–R (13%) than in NESARC (44%), with rates in NESARC comparable to national statistics (39% of 18–24-year-olds in 2005; United States Department of Education, 2008
). This discrepancy may explain some of the divergence in prevalence rates among this age group.
Figure 5 Prevalence of past-12-month Other Bipolar disorder using strict criteria from National Comorbidity Survey—Replication (NCS–R; Kessler & Merikangas, 2004) in the NCS–R sample (N = 9,278) and the NESARC baseline sample ( (more ...)
When combining all BP spectrum diagnoses (Strict-BPI, Strict-BPII, and Other BP), rates of any BP disorder are similar across samples, with the exception of 18–20-year-olds in NCS–R (see possible explanations above), and display similar age trends, NESARC χ2(6, N = 43,093) = 26.23, p <.001, and NCS–R χ2(6, N = 9,282) = 8.20, p <.001 (see ). It is important to note that the Other BP category in the NESARC sample has relatively high levels of comorbidity, although they are somewhat lower than those for Strict-BPI and Strict-BPII (see ).
Examination of Offset in Prospective Data
On the basis of results from the baseline NESARC and NCS–R data, we anticipated that offset of BP disorder would be greatest among younger age groups, particularly for Other BP disorder. To determine whether the observed age gradient could be attributable to differential offset rates as a function of chronological age, we conducted a series of logistic regression analyses predicting the presence or absence of a 12-month BP diagnosis at follow-up among NESARC participants who were diagnosed at baseline and completed the follow-up survey (n = 1,042). Contrary to our expectation, offset at follow-up was not related to age (spectrum disorder β = 0.00, p = .442; Strict-BPI β = 0.00, p = .603; Strict-BPII β = −0.02, p = .225; Other BP β = −0.01, p = .517; controlling for sex in all models). In , we illustrate these findings for Other BP, where we found the strongest age gradient and anticipated finding differential offset. documents the various BP-related outcomes at follow-up among those with an Other BP diagnosis at baseline as a function of age. Although there is a drop in rate of offset from the 18–20- (70%) to the 21–24-year-old (54%) age groups, overall, the rate of offset from Other BP disorder is fairly constant across age groups (M = 64%, with a range of 52% to 70%). This is also true for Strict-BPI and Strict-BPII (M = 55% for both disorders, with a range for Strict-BPI of 41% to 67% and for Strict-BPII of 40% to 69%; data not presented). These analyses are extremely informative in that they suggest that most of the age gradient is attributable to high hazard rates in late adolescence and early adulthood and not to differential remission or recovery rates.
Figure 6 Patterns of remission, stability, and progression among participants with past-12-month Other Bipolar disorder using strict National Comorbidity Survey—Replication (Kessler & Merikangas, 2004) criteria in the National Epidemiological Survey (more ...)
Can the Gradient Be Explained by BP-Related Attrition From the Sample?
We conducted basic attrition analyses in the NESARC sample determine whether follow-up was related to BP diagnoses, age, sex. An overall diagnosis of BP spectrum disorder at baseline was associated with increased odds of completing the follow-up survey (OR = 1.37; 95% CI = 1.12, 1.66). Although follow-up was not significantly associated with a specific diagnosis of Strict-BPI Other BP at baseline, participants with a Strict-BPII diagnosis baseline were more likely to complete the follow-up survey (OR 1.61; 95% CI = 1.01, 2.58). In general, older age groups, Wald F(6, N = 43,093) = 44.75, p <.001, were more likely to complete the follow-up survey, as were women (OR = 1.61; 95% CI 1.11, 1.25). On the basis of these analyses, we do not believe that attrition in NESARC could explain the rates of offset observed at follow-up.