How to Understand Divergent Views on Bipolar Disorder in Youth

History of the Controversy Over the Diagnosis of Mania/Bipolar Disorder in Youth

The nature of mania, and therefore bipolar I (BP I) disorder, in children (versus adolescents) has been a cause of debate for many years. It was most clearly outlined in a “point-counterpoint” between Biederman and Gittleman-Klein in 1998 (Biederman et al. 1998), but others have addressed the question before (Carlson 1990) and since (Galanter & Leibenluft 2008, Harrington & Myatt 2003, Kent & Craddock 2003). The debate has even extended to adults (Wingo & Ghaemi 2007). The core of the disagreement was and remains whether mania in children should be diagnosed narrowly, with discrete episodes that represent a clear departure from the person’s prior function and usual state with symptoms that are the same as one observes for adults (the conservative or narrow viewpoint), or if it should be diagnosed more broadly (the liberal or broad viewpoint). The broader definition appears to embody early childhood onset of chronic emotion dysregulation, usually superimposed on symptoms of attention-deficit hyperactivity disorder (ADHD) and oppositional defiant disorder with multiple other comorbidities. It includes severe irritability manifested by intense and protracted rage outbursts (Mick et al. 2005), a “bipolar phenotype” as measured by the Child Behavior Checklist (Biederman et al. 1995), as well as by modified manic symptoms as described in the Washington University version of the Schedule of Affective Disorders and Schizophrenia for Children (WASH-U-KSADS; Geller et al. 2001). Episodes are extremely protracted. The cases meet the letter of the mania criteria identified in the American Psychiatric Association’s Diagnostic and Statistical Manual of Mental Disorders (DSM)-III to DSM-IV-R (Am. Psychiatr. Assoc. 1980, 1987, 1994, 2000) as measured by the investigators. The question is whether children with this condition have a complicated, very early onset form of adult bipolar disorder or experience mood problems concomitant with their many other comorbidities (Perlis et al. 2004).

After the National Institute of Mental Health’s Bipolar Roundtable in 2000 (Natl. Inst. Ment. Health 2001), some investigators started to describe the different mania/bipolar phenomenologies as narrow and broad phenotypes. Although some researchers focused on the importance of elation and grandiosity as central to the narrow phenotype (Geller et al. 2002), others identified discrete episodes as the primary distinguishing feature (Carlson 1990, Leibenluft et al. 2003). This second approach is the one used in this critical review. That is, the narrow or conservative view of mania/BP I is identified by the clarity of episodes; the broad or liberal view by manic symptoms without well-defined episodicity.

Disagreement about the boundaries of mania in children is not a new problem, as both mania and childhood have a variety of definitions (Carlson & Glovinsky 2009). The German physician and psychologist, Theodor Ziehen, in 1917 defined mania as

Ziehen’s rather classical description, however, applied to children older than age 11. What differentiated mania in children from the behaviors of typical young children was the fact that the symptoms reflected a rapid change from the child’s characteristic behavior.

Clinical interest in manic-depression in youth began in greater earnest after Kraepelin’s (1921) monograph on manic-depressive insanity in the early twentieth century. Surveys of mentally ill youths, culminating in a special issue of the journal The Nervous Child in 1952 (e.g., Barton-Hall 1952, Harms 1952), concluded that depression was more common than mania and occurred mostly in adolescents but that there may be “temperamental variants in children” (for a review, see Carlson & Glovinsky 2009). Anthony & Scott (1960), however, applied a conservative view to a literature search for mania in children, and requiring strictly defined, classical manic-depressive symptoms in preadolescents, they found the condition to be rare.

Ironically, the introduction of explicit criteria for the definition of psychiatric disorders in general, and mania and depression in particular (Feighner et al. 1972), laid the foundation for a broader view of pediatric bipolar disorder. Inspired by the Feighner criteria, Weinberg & Brumback (1976), two pediatric neurologists from Washington University, systematically examined the children in their learning disabilities clinic and identified a number as having mania rather than being hyperkinetic, the DSM-II (Am. Psychiatr. Assoc. 1968) name for ADHD.

The Emergence of Two Views of Mania/Bipolar Disorder

Changes in Conceptualizations of Attention-Deficit Hyperactivity Disorder and Oppositional Behavior

It is clear that considerable overlap exists between symptoms of ADHD and mania. The question is why, in recent years, has there been such an increased focus on these cases (Blader & Carlson 2007, Moreno et al. 2007), and how has this impacted our understanding of mania? Reasons emerge from decisions made since 1980 in the DSM criteria both for what previously was called “hyperkinetic reaction of childhood” (Am. Psychiatr. Assoc. 1968) and for mania/bipolar disorder.

Hyperactive children, diagnosed prior to the DSM-III (Am. Psychiatr. Assoc. 1980) as having hyperkinetic reaction of childhood, have a condition that we still recognize. Hyperactivity is the most striking symptom.

However, struck from the DSM-III (Am. Psychiatr. Assoc. 1980) and subsequent iterations are what appear to be mood symptoms and behavioral fluctuations seen in many hyperactive children. “Variability also is frequent, with the child being described as quite unpredictable and with wide fluctuations in performance. Outstandingly also these children seem unable to tolerate any delay in gratification of their needs and demands. They are irritable and explosive, with low frustration tolerance” (Laufer et al. 1957, p. 38). These symptoms were placed in “associated symptoms” in DSM-III:

The emotional symptoms seen in hyperactive children were also described in “oppositional disorder,” which was defined as a “pattern of disobedient, negativistic and provocative opposition to authority” but which included “temper tantrums” as one of the criteria (Am. Psychiatr. Assoc. 1980). The high rate (up to 65%; e.g., Petty et al. 2009) of comorbidity between ADHD and oppositional/oppositional defiant disorder may stem from the cleavage of symptoms that previously were part of the same condition.

The description of ADHD-associated features was not changed in DSM-III-R (Am. Psychiatr. Assoc. 1987, p. 51), but oppositional disorder began to include more of what Stringaris & Goodman (2009) considered “mood” symptoms: “Often loses temper, is often touchy or easily annoyed by others, is often angry and resentful.” Unfortunately, the structured and semistructured interviews that emerged after DSM-III was published focused only on the primary criteria of psychiatric disorders like ADHD. It became impossible to test whether deleting the mood symptoms from the construct of ADHD was a valid decision. Some investigators felt that it was a wise decision because those children were thought to have a mood disorder, and the creation of a more liberal definition of mania/bipolar disorder began to surface (e.g., Biederman et al. 1995, 1996; Weinberg et al. 1997); however, others disagreed (Barkley & Fischer 2010).

Changes in the Conceptualization of Mania/Bipolar Disorder

Mania and the condition in which it occurs (initially “manic depressive illness” and then bipolar disorder) also had both a name and a concept change from DSM-II to DSM-III. DSM-II (Am. Psychiatr. Assoc. 1968) provided little diagnostic guidance, but the spirit of the disorder’s definition prior to 1980 was kept in the clinical description in the tenth revision of the International Classification of Diseases (ICD-10; World Health Organ. 1992), which contrasts somewhat with the symptom-based DSM approach. That is, DSM-III and subsequent iterations operationalize specific symptoms that occur during the course of mania, emphasizing a “distinct period [lasting at least a week] during which expansive, irritable or elevated mood symptoms occur with associated symptoms.” When an individual endorses enough of these symptoms, s/he is given the diagnosis of a depressive and/or manic episode. However, three important qualifiers were not made explicit for mania, as they were for depression. That is, criteria for major depression specify that symptoms should occur most of the day every day; depressive symptoms themselves should co-occur; and, finally, symptoms should represent a change from previous functioning. This has led to different interpretations of how persistent manic symptoms should be and whether or not mania should represent a change, as well as how that change should be established. We submit that more conservative bipolar diagnosticians believed that mania should follow the same rules as depression; those who interpreted the criteria literally did not.

In addition, the course of illness was not built into the definition of bipolar disorder. This approach is somewhat different from the ICD-10 clinical version (World Health Organ. 1992), which relies on pattern recognition to define bipolar disorder. The ICD-10 text definition of bipolar disorder requires that the patient experience repeated (i.e., at least two) episodes in which mood, energy, and activity levels are significantly disturbed; that recovery is usually complete between episodes; and that manic episodes usually begin abruptly and last for between two weeks and four to five months (median duration about four months). Depressions tend to last longer (median length about six months), although rarely for more than a year, except in the elderly. The first episode may occur at any age from childhood to old age, with a variable pattern of remission and relapse. The ICD-10 definition is clearly a more narrow and conservative view of mania/BP I disorder than that of the DSM.

Subsequent alterations in DSM mania criteria also had a downstream impact. Irritability, encompassed in the concept of “dysphoric mood” in DSM-III (Am. Psychiatr. Assoc. 1980), was eliminated from the depression criteria in DSM-III-R (Am. Psychiatr. Assoc. 1987) and later DSMs and was only ascertained in mania (except for children and adolescents, where irritable mood still counts toward a diagnosis of depression). Simultaneously, the duration requirement of at least one week for the “distinct period” was eliminated, meaning that a distinct period could last from minutes to hours, thus confusing mood lability with manic episodes. Finally, the criterion of at least one week of a distinct period does not necessarily clarify how to define the offset of an episode. That is, if symptoms start but never end, they have lasted at least one week. Without an offset, there is no real episode. A focus on symptoms without a comprehensive longitudinal history that defines episodes makes it difficult to distinguish a hyperactive, distractible, impulsive, emotionally labile, overly reactive, and irritable child with ADHD from one with mania.

Indeed, the development of a condition called severe mood dysregulation (SMD) emerged as the result of a research effort to distinguish chronic irritability from episodic irritability (Leibenluft et al. 2003, 2011) that may suggest mania. Symptoms of SMD include both chronic irritability and frequent and intense explosive outbursts. However, also included in the definition are the “B” symptoms of mania, many of which are also symptoms of ADHD (e.g., restlessness, distractibility, intrusiveness, pressured speech). It should be no surprise, then, that 75% to 80% of children with SMD in clinical samples carry diagnoses of comorbid ADHD (from the “B” mania symptoms) and oppositional defiant disorder (often loses temper, touchy, and easily annoyed) (Brotman et al. 2006). Disruptive mood dysregulation disorder, the DSM-5 committee’s reformulation of SMD (Am. Psychiatr. Assoc. 2013), has put the whole issue on the clinical map for the coming years. Notably, however, although the committee eliminated the ADHD-like symptoms from the disruptive mood dysregulation disorder criteria, rates of ADHD are still high (Axelson et al. 2012, Margulies et al. 2012). There are many important exclusionary criteria for disruptive mood dysregulation disorder, too, and it remains to be seen whether it will provide a useful alternative for erroneously diagnosed bipolar disorder or cause clinicians to miss the diagnosis of ADHD.

We submit, then, that it is not the symptoms that are so different between mania and ADHD but rather the illness course. ADHD is a developmental disorder beginning in early childhood that, although fluctuating, is chronic. Mania occurs in episodes defined as clear differences from one’s previous behavior and function (Carlson 1990). Changes from DSM-II (Am. Psychiatr. Assoc. 1968) to DSM-III (Am. Psychiatr. Assoc. 1980) that de-emphasized emotionality unintentionally caused the relabeling of some children who previously would have been considered as having hyperkinesis/ADHD, placing them in a bipolar spectrum where episodicity was inadvertently neglected.

Structured and Semistructured Interviews Collect Different Information

Another issue relevant to understanding the confusion that has developed over identifying children with mania is that the criteria that were established emerged from a very long history of understanding the condition in adults. There was no guidance on how to assess the condition in children and adolescents and thus no wrestling with developmental or assessment considerations (Carlson & Meyer 2006). The DSM adds developmental qualifiers for most conditions that begin in childhood (e.g., the symptom of inattention in ADHD is described as maladaptive and inconsistent with developmental level). However, there are no developmental qualifiers for mania or mood regulation, probably because we don’t know what the developmental qualifiers are rather than because they are unnecessary. Nevertheless, if investigators don’t know that they need to collect such information, we are going to be (and are) no wiser in the future.

Child and adolescent psychiatric assessment has wrestled with informant issues for years (De los Reyes & Kazdin 2005; Youngstrom et al. 2003, 2006). The general consensus has been to believe adult informants (parents and teachers) for behaviors that are observable (e.g., hyperactivity, inattention, impulsivity, and aggression) and to believe the child (or person experiencing the internal symptom) for symptoms such as depression or anxiety. Even if that heuristic is entirely valid, mania presents a particularly complicated set of issues because some of the symptoms are observable, and some are more internal (Carlson & Meyer 2006).

Systematic assessments of symptoms have provided a way to increase the reliability of diagnoses (Endicott & Spitzer 1978). By asking questions in the same or similar ways, the hope has been to eliminate the variability that occurs when interviewers decide arbitrarily what to ask or how to ask it. However, structured and semistructured interviews make no attempt to collect a history systematically (Andreasen 2007), and interviews are inconsistent in how they operationalize the DSM criteria and manic symptoms (Carlson 2011, Galanter et al. 2012).

At least six child and adolescent interviews have been designed to assess DSM-IV (Am. Psychiatr. Assoc. 1994) diagnoses in general and mania/bipolar disorder in particular (Galanter et al. 2012). Where there are parent and child informants, the question of whose information should be counted is addressed by one of several different algorithms (Youngstrom et al. 2006). However, the choice of algorithm has significant implications for diagnosis. Differences exist in whether both informants must agree on the presence of a symptom versus only one informant and whether the total symptom count can be contributed to by two informants. One way to get around making largely arbitrary choices between algorithms is to end the interview by bringing both informants together to reconcile differences. However, this is a luxury that often is not possible in large studies. Moreover, the order of interviewing informants is also important. The K-SADS (Kaufman et al. 1997) strongly recommends interviewing the teen before interviewing the parent. However, respecting the youth’s confidentiality prevents reconciling disparate information from the parent. Interviewing the youth after the parent allows the possibility of asking the child/teen what s/he felt like on the occasion that the parent reported the youth had experienced manic symptoms. Similarly, if a youth describes a manic episode where s/he was up all night, wore provocative clothes, and started lots of projects and a parent never noticed, the interviewer can tell the teen that the parent didn’t mention the episode and ask why it went unnoticed.

Some interviews don’t use child information at all. This addresses the problem of reconciling discrepancies by fiat. However, it does not address the concern that some experiences are likely to be described only by the child (e.g., hallucinations, racing thoughts). Teacher information is almost never solicited in studies of children with bipolar disorder, and where the issue has been addressed, teacher information has been discounted because it is felt that teachers are not qualified to identify mania (Youngstrom et al. 2003). On the other hand, teachers are qualified to recognize ADHD, so in situations where parents describe both manic symptoms and ADHD symptoms, or simply manic symptoms with hyperactivity, distractibility, and rapid speech, and teachers observe no behavior problems, questions clearly exist about whether the child could be actually experiencing manic episodes (Carlson & Blader 2011).

Most germane to the liberal-versus-conservative evolution of mania/bipolar disorder is how episodes are operationalized and how descriptions of episodes are elicited. Galanter and colleagues (2012) have detailed the degree to which interviews, which many assume are virtually identical to each other, are quite different. Differences include how episodes are ascertained in interviews, whether behaviors must be observed by others or represent a change from previous behavior, and whether the various manic symptoms co-occur. The WASH-U-KSADS (Geller et al. 2002), which elicits information on a current episode, tries to get the parent and child to date the onset and offset of each symptom, including ADHD symptoms (symptoms that count toward both ADHD and mania). The point at which the dates overlap constitutes an episode. Strictly speaking, though, during an episode, symptoms should start and stop at around the same time and not be defined cross-sectionally by the point at which they overlap.

The K-SADS has numerous versions (Ambrosini 2000), some of which are better than others in delineating episodes. In general, the mood question asks, “Has there ever been a time when you felt XXX? How long did that last?” On the one hand, subsequent symptom screenings do not specify whether symptoms occurred when the elated, expansive, or irritable mood occurred; on the other hand, the mania symptoms in the subsequent supplement do specify whether they occurred at the same time as the mood symptoms. In some, but not all, versions of the K-SADS, episodes and time frames are supposed to be recorded to provide the interviewer with a chronology of the behaviors to be discussed. The chronology is probably necessary to distinguish ADHD with oppositional defiant disorder from the onset of a manic episode. In the 1996 version of the K-SADS-Present and Lifetime (PL) version, the current episode (if present) and the most severe past episode are inquired about, although the age at the time of the worst episode is never actually solicited.

Researchers determine who to interview, and they elicit comorbidity in different ways. For example, investigators from Massachusetts General Hospital (e.g., Biederman et al. 2000) use the K-SADS-Epidemiological (E) 1994 version (Orvaschel 1994) and do not include children under 12 in the interview due to concerns that the information provided is not reliable. Investigators double-count symptoms across all disorders. Those who double-count symptoms believe that it is simply not possible to know which condition is accounting for the psychopathology and thus count the item toward both disorders. Not surprisingly, rates of comorbidity are much higher in these investigations. It is our contention that the result yields a broader conceptualization of mania and a sample that is different from investigations that take a more conservative approach and require manic symptoms to be over and above ADHD symptoms. Finally, clinicians are not used as interviewers, although the investigators provide many checks and balances to overcome this (Wozniak et al. 2003).

Galanter and colleagues (2012) imply, rather than state explicitly, that there is every likelihood that the heterogeneity within samples and across samples, because of different interviews and criteria interpretations, is quite large. The issue of heterogeneity is ironic because structured and semistructured interviews were designed to mitigate just this problem.

In summary, it is not unusual for method variance to create significant heterogeneity in diagnostic results (Youngstrom et al. 2006). For instance, in a large meta-analysis that explored methodological variables in ADHD, variables such as using a single informant versus summing criteria from either informant where there was more than one, and using DSM-IV (Am. Psychiatr. Assoc. 1994) rather than earlier DSM criteria, increased rates of ADHD (Polanczyk et al. 2007). However, reviews and meta-analyses have virtually ignored these issues in studies of bipolar disorder in children and adolescents (Faraone et al. 2012, Kowatch et al. 2005). The following section aims to explicate the bipolar controversy and reconcile disparities in findings of bipolar disorder in youth by highlighting some of the conservative and liberal biases. Admittedly, however, reconciliation is not always possible because unidentified heterogeneity is occurring within as well as across samples.

How Methodological Issues Impact Epidemiologic, Comorbidity, High-Risk, and Longitudinal Studies

Epidemiologic Studies

Rates of bipolar disorder in community samples are influenced by all of the aforementioned methodological caveats with some added ones. The time frame studied (lifetime, past year, past three months), whether parent and/or child (usually teen or older child) is interviewed and whether mania and/or hypomania is addressed, yields different rates. An important question has been whether rates in youths are the same as in adults. On the one hand, because bipolar disorder is an early-onset condition, rates should be similar. On the other hand, because first onsets continue to accumulate through adulthood, lifetime rates should increase with age.

To simplify the discussion, we focus on rates of mania/BP I. A comprehensive meta-analysis of rates of bipolar disorder in youths was recently published by Van Meter et al. (2011), who questioned whether rates of early-onset BP I disorder are disproportionately higher in the United States than in Europe. The authors conclude that rates are similar. They also conclude that rates of mania are about 1.2%, similar to rates in at least some studies of BP I in adults (Kessler et al. 1994; Merikangas et al. 2007, 2011).

A closer look at the articles in the meta-analysis, however, reveals that very few studies provide satisfying information, and those that do have found lower rates. The study by Kim-Cohen et al. (2003), done on the New Zealand Dunedin cohort, was included in error; it did not measure mania in youth under age 18. In the Netherlands study (Verhulst et al. 1997), the ∼2% rate was achieved by adding the parent rate (1.1%) and the teen rate (0.9%). As the authors noted, there was very little overlap in symptom reports, and the authors themselves were quite skeptical of their results. Stringaris and colleagues (2010) required convergence between parent and child reports to diagnose bipolar disorder because they found a very low kappa (0.02%) between parent and child. Doing so yielded a rate of mania of 0.1%. In a much smaller community study using parent information to corroborate child data, Carlson & Kashani (1988) also found low rates (0.7%). Although the Methods for the Epidemiology of Child and Adolescent Mental Disorders study (Goodman et al. 1998) was cited as having a rate of mania of 1.3%, symptoms from either parent or child were used. Reliability for mania was never examined because the condition was too rare. Rates were likely much lower than 1.3%—closer to 0.4% to 0.6% when convergence reports were required (P. Fisher, personal communication). Nevertheless, rates are variable even when using a single informant. Using only teen reports, the lifetime rate for mania was found to be 0.1% in the Oregon Adolescent Depression Project (Lewinsohn et al. 1995) and 1.7% in the National Comorbidity Survey-Adolescents (Merikangas et al. 2012). In summary, studies designed from a liberal perspective have generally produced rates of BP I that differ from the more conservative studies. Liberal studies, where there is an absence of confirming symptoms from another informant, found rates in youths to be similar to those of adults in most instances. Studies using more narrowly defined mania and convergent informants reported lower rates.

Even when teens report symptoms that meet mania criteria, there are questions about the validity of their reports. Cicero and colleagues (2009) have raised concerns about the significance of self-reported manic symptoms in young adults. They used publicly available data from the National Epidemiological Survey on Alcohol and Related Conditions, conducted during 2001 and 2002, and a follow-up sample collected three years later (Natl. Inst. Alcohol Abuse Alcohol. 2001–2005). The sample included 43,093 noninstitutionalized, civilian respondents age 18 and over; 80% of the original sample completed the follow-up survey. The investigators found that there was a significant drop over the three-year time frame in the respondents reporting lifetime mania. Kessler et al. (2009) used the same algorithm as the National Comorbidity Survey-Adolescents and found an overall rate of 1.15% for BP I. There was a very significant age gradient by age group, however, with highest rates being in those younger than age 25. In addition, between 50% and 75% of the sample no longer endorsed manic symptoms three years later (although there does not appear to be an age gradient in that observation). The authors concluded that there is a “developmentally limited” form of bipolar disorder (i.e., young people report a manic episode during young adulthood but never have another episode, hence the decline in rates). That conclusion certainly contradicts data from clinical samples suggesting that early-onset bipolar patients are more likely to be chronic or to have multiple episodes (Birmaher et al. 2009a, Carlson et al. 2012, Geller et al. 2008, Perlis et al. 2004). The alternative conclusion is that young people from epidemiologic studies who endorse manic symptoms (along with other comorbidities) may have mental health problems, but they do not correspond to the problems of manic patients seen in clinical practice. If that is true, then studies reporting high rates of bipolar disorder in youths need to be understood in that context.

Comorbidity with Attention-Deficit Hyperactivity Disorder

Community studies of ADHD examining rates of bipolar disorder.

The relationship between ADHD and bipolar disorder is at the heart of the debate about bipolar disorder phenomenology in youths (for a review, see Pataki & Carlson 2013). How often are the two conditions comorbid? Are they confused because of methodological problems (Carlson 1998, Kent & Craddock 2003)? Is ADHD a risk factor for subsequent bipolar disorder in adults (Carlson & Weintraub 1993), or are the emotional symptoms simply part of ADHD (Barkley & Fischer 2010, Sobansky et al. 2010)? Although well-sampled community studies should be the gold standard for identifying conditions without the bias of treatment referral, and there have been many community studies of ADHD in children (for a review, see Döpfner et al. 2008), we could find only one published report where comorbidity of ADHD with bipolar disorder was examined. In a Finnish study of ADHD in teens (Smalley et al. 2007), in which 6,622 parents returned an ADHD screening form and a subsample was evaluated with the K-SADS-PL, 6.7% of teens were identified with ADHD, 1% with bipolar disorder, and none with both. However, these were conservative diagnoses insofar as parent and teen data were examined together and discrepancies were resolved.

Community studies of bipolar disorder examining rates of ADHD.

Several community studies have examined rates of ADHD in teens with mania or manic symptoms. In a study of 150 Missouri teens using the DSM-III criteria and the Diagnostic Interview for Children and Adolescents (Herjanic & Reich 1982), 20 teens (13.3%) said they had impairing manic symptoms (symptoms of mania but not in episodes nor corroborated by parent; only 1 teen had parent-confirmed BP I) (Carlson & Kashani 1988). Of these 20, 45% had ADHD, a rate significantly higher than in symptomatic teens without manic symptoms. In the Oregon Adolescent Depression Study (Lewinsohn et al. 1995), which used the DSM-III-R criteria (Am. Psychiatr. Assoc. 1987) and the K-SADS-E, 18 teens (0.9%) had lifetime BP I, BP II, or cyclothymia. Only 11.1% of the 18 teens had ADHD—a rate that was perhaps low because only the teens were interviewed (Barkley et al. 2002); however, this rate was much higher than in the rest of the sample (2.7%). Rates of all disruptive behavior disorders were 22.2%.

Most recently, the National Comorbidity Survey-Adolescents used the Composite International Diagnostic Interview (Kessler et al. 2009) and DSM-IV criteria (Am. Psychiatr. Assoc. 1994) with diagnostic information from teens only (Merikangas et al. 2012). Rates of ADHD in youths with bipolar disorder varied from 18% to 24%, depending on whether the sample included both unipolar mania and mania plus depression, or only the latter. With the exception of the small Carlson & Kashani (1988) study, the above studies lacked corroborating information from another informant; only the affected teen was interviewed about ADHD symptoms. A parent informant increases the rates of ADHD diagnosis over information from the subject alone (Barkley et al. 2002) and decreases the rates of mania diagnosis (Carlson & Kashani 1988, Stringaris et al. 2010). Methodology matters.

Clinical samples of ADHD examining rates of bipolar disorder.

Rates of co-occurring bipolar disorder in children with ADHD vary widely, again depending on the assessment approach and specific procedures used by the research group (Arnold et al. 2011, Carlson 2011, Galanter et al. 2012, Galanter & Leibenluft 2008, Kent & Craddock 2003) and the age of the sample. Recent work includes a study from the United Kingdom on an ADHD sample (youths ages 6 to 18) collected for genetic research and using a three-month prevalence with the Child and Adolescent Psychiatric Assessment (Angold et al. 1995). The study found no child with mania and only one who met DSM-IV criteria for hypomania (0.5% of the sample) (Hassan et al. 2011). Only 1.8% of youths in the Multimodal Treatment Study of ADHD, ages 15 to 18 at the eight-year follow up, were reported to have even hypomanic symptoms (Molina et al. 2009) using the Diagnostic Interview Schedule for Children (Shaffer et al. 2000). A study from Italy using the K-SADS-PL (Kaufman et al. 1997) in 129 children with ADHD found a rate of 16.7% (Masi et al. 2006), though most cases had bipolar disorder-not otherwise specified (BP-NOS), and only 1.7% had BP I. In contrast, studies taking the position that judgments cannot be made about whether shared ADHD, or other, symptoms and manic symptoms belong to one or the other condition (e.g., Biederman et al. 1996, 2004; Tillman & Geller. 2006; Wozniak et al. 1995) report rates between 11% and 23% of BP I/mania using the K-SADS-E (Orvaschel 1994) or WASH-U-KSADS (Geller et al. 2001). Perhaps not surprisingly, then, if symptoms count toward two conditions, comorbidity rates will be higher.

Perhaps the most compelling finding against the conclusion that people with ADHD develop bipolar disorder comes from long-term studies of children with ADHD, in which rates of bipolar disorder in adulthood are no higher than rates in the general population (Carlson et al. 2000, Fischer et al. 2002, Klein et al. 2012, Weiss et al. 1985). However, those studies diagnose bipolar disorder conservatively.

In summary, if the base rate of BP I disorder in teens is <1%, then conservatively defined BP I in youths with ADHD is not much higher. Rates are high when ADHD and manic symptoms are double-counted and episodes are de-emphasized.

Clinical studies of bipolar disorder examining rates of ADHD.

In clinical samples of children diagnosed with bipolar disorder, rates of ADHD also vary considerably and for the usual methodological reasons. For instance, Kowatch and colleagues (2005) combined data from seven very different studies of bipolar disorder (not necessarily BP I) and reported rates of comorbid ADHD ranging from about 10% to over 80%. Methodological differences (age of sample, quality of study quality, and number of informants) increased rates of comorbid ADHD. Not surprisingly, rates of ADHD are higher in bipolar samples where symptoms of ADHD and mania are double-counted and no judgment is made regarding whether the ADHD symptoms worsened during the episode. Thus, studies of children (versus teens) using the WASH-U-KSADS (Geller et al. 2001) typically find high rates of ADHD, ranging from 70% to 95% [e.g., 95% in the baseline episode (Geller et al. 2008), 92.8% in the Treatment of Early Age Mania study (Geller et al. 2012), approximately 70% in the divalproex-extended release study (Wagner et al. 2009), and 75% in the oxcarbazepine study (Wagner et al. 2006)]. Similarly, studies using DSM-III-R criteria (Am. Psychiatr. Assoc. 1987) and the 1994 version of the K-SADS-E (e.g., Biederman et al. 1996, 2004; Faraone et al. 1997; West et al. 1995; Wozniak et al. 1995) find rates as high as 87%. Studies using the K-SADS- PL (Kaufman et al. 1997) find slightly lower rates of ADHD in bipolar samples, i.e., 60% to 70% (Axelson et al. 2006, Brotman et al. 2006). To the degree that investigators believe in the importance of clear episodes (see http://www.aacap.org/cs/BipolarDisorder.ResourceCenter) and train their interviewers accordingly, rates of ADHD appear to be somewhat lower.

Informant corroboration on the presence of symptoms in more than one setting is important for the diagnosis of ADHD. Parents and teachers are the usual informants because they represent different and clinically relevant settings. Optimally, investigators obtain information from both rather than rely on one source to speculate on what is happening elsewhere. Requiring documentation of symptoms from both parents and teachers substantially lowers rates found for ADHD. For mania, the change in the person’s behavior and function that identifies the actual episode should be confirmed by an informant other than the subject, who may experience symptoms but lack insight on their impact. As noted previously, however, not all interviews ask whether a change has been noticed by others or clarify whether the behavior change is clearly due to mania. Not surprisingly, when information is obtained from a second source, rates found for mania are lower. For instance, Carlson & Blader (2011) examined children whose parents endorsed severe manic symptoms but who were not identified as manic or even hyperactive by teacher ratings. In those instances, the children were actually found by best-estimate diagnosis to have depressive or anxiety disorders and not mania (Carlson & Blader 2011).

The effect of other informants may also be seen in the multisite Longitudinal Assessment of Manic Symptoms study (Arnold et al. 2011). Parents of 707 outpatients rated their child’s level of manic symptoms on the General Behavior Inventory (Youngstrom et al. 2008). Parents and children were subsequently interviewed with the K-SADS-PL (Kaufman et al. 1997). LAMS investigators found that 69% of children with BP I had comorbid ADHD and, on the basis of parent reports, groups of children were identified as having BP I, ADHD, or both. The BP + ADHD children had the most severe parent symptom ratings. Teacher information on the same children, however, indicated that the children with BP + ADHD by parent report did not differ on manic symptoms from those with ADHD without BP (i.e., symptoms were less severe than parents reported), and most importantly, children whose parents endorsed BP only with severe symptoms did not differ by teacher report from those without any diagnosis. In other words, many children with parent-reported severe manic symptoms do not have those symptoms in school and therefore didn’t have them most of the day every day. Conservatively speaking then, they would not get a BP I diagnosis.

A developmental relationship between ADHD and bipolar disorder likely accounts for the varying rates with which they seem to co-occur at different ages. As originally noted in a study by Faraone and colleagues (1997), the reported rate of comorbid ADHD in youths with bipolar disorder was 93% in children, 88% in teens with a childhood onset, and 59% in teens with adolescent-onset mania. The Course and Outcome of Bipolar Youth study similarly reported rates of ADHD in their bipolar subjects (including all of the bipolar spectrum, not just BP I) that varied by age as follows: 71.7% in children under age 12, 69.3% in teens with childhood-onset bipolar disorder, and 31.1% in adolescents with adolescent-onset bipolar disorder (Birmaher et al. 2009c). In adults with bipolar disorder, rates of ADHD/childhood externalizing disorders are lower still and have been generally reported to be between 10% and 20% (Bernardi et al. 2010, Carlson et al. 2002, Perlis et al. 2004, Sachs et al. 2000).

In summary, rates of bipolar disorder in children with ADHD depend on how mania has been elicited and defined, the comparison group, and the age of the sample studied. If multiple informants are required and mania is described as a clear episode superimposed on ADHD, as one sees in teens and adults, the rates are much lower. On the other hand, as noted in the community studies, data are consistent with finding elevated rates of ADHD in youths with mania/bipolar disorder. Some hypothesize that this represents a specific subtype (Arnold et al. 2011, Biederman et al. 2000, Carlson 1998).

Liberal and Conservative Views of Top-Down High-Risk Studies

Rates of bipolar disorder are significantly higher in the offspring of bipolar parents than in the offspring of normal controls. In contrast, studies differ on whether ADHD is more common in the offspring of bipolar parents than controls. For both disorders, rates vary widely, again depending on the age of the sample; whether mania or the bipolar spectrum is measured; whether the concept of mania/bipolar disorder is construed broadly or narrowly, and for this issue, the origin of the proband sample; and the nature of the comparison group (e.g., representative or “supernormal”; Duffy et al. 2011).

A meta-analysis of studies of bipolar offspring completed before 1995 (LaPalme et al. 1997) found rates of BP I disorder to be 5.4% compared to 0% in the offspring of parents without bipolar disorder. Since then, a number of studies have been done, all but one of which used a prospective follow-up design, and the findings are much more variable (Carlson 2013). Conservatively diagnosed assessment approaches reported rates of 2% to 6% for BP I disorder (Duffy et al. 2007, Mesman et al. 2013, Meyer et al. 2004, Nurnberger et al. 2011, Shaw et al. 2005). These rates contrast with another group of high-risk studies using either the WASH-U-KSADS or the KSADS-E in which rates of bipolar I disorder were 10% to 33% in offspring (Chang et al. 2000, Henin et al. 2005, Hirshfeld-Becker et al. 2006, Singh et al. 2007).

Similar differences are found in rates of ADHD, which were between 0% and 8% in conservatively diagnosed samples (i.e., similar to population rates; see review by Duffy 2012). With two exceptions (Birmaher et al. 2009b, 2010), studies finding high rates of bipolar disorder in the offspring of bipolar parents also found rates of ADHD that were high (15% to 30%) and significantly greater than rates of ADHD in comparison groups (Chang et al. 2000, Henin et al. 2005, Hirshfeld-Becker et al. 2006, Singh et al. 2007).

Duffy and co-investigators (2011, 2012) reviewed the high-risk literature to explain the range of rates of bipolar disorder and ADHD in various studies. They found that the source of the parent probands made a significant difference. Where parents were identified through neurobiologic research, with parent and child diagnosis obtained through best-estimate procedures, rates of parent comorbidity were lower, rates of psychopathology in the nonproband parent were lower, and rates of BP I were lower in the offspring, who themselves had an adolescent or older age of onset and low rates of ADHD. Self-referred bipolar probands recruited from advertisements or recruited from tertiary care clinics, who were diagnosed with structured interviews and whose children were assessed only with structured interviews, were found to have more offspring with psychopathology, including higher rates of ADHD and a younger age of onset of mental health problems.

The meta-analysis by Faraone and colleagues (2012) of 27 family genetic studies examining the comorbidity between ADHD and BP I disorder illustrates how findings are shaped by interpretation, study design, and selection of studies. The meta-analysis concluded that the relative risk of ADHD in bipolar probands is 2.2 (95% CI 2.1–3.2), with rates of ADHD in offspring of bipolar versus comparison probands of 27% versus 9.6%, in siblings 30.1% versus 11.0%, and in parents 16.5% versus 4.5%. Conversely, the relative risk of bipolar disorder in ADHD probands is 1.8 (95% CI 1.3–2.6), with rates of bipolar disorder in offspring of ADHD probands versus comparison probands being 6.8% versus 3.5%, in siblings 5.9% versus 2.8%, and in parents 5.1% versus 3.1%.

Pertinent to this discussion, the highest relative risks of ADHD in offspring and other relatives of probands with bipolar disorder were from studies using the WASH-U-KSADS and the K-SADS-E. Closer inspection revealed that in two of the studies (i.e., Carlson & Weintraub 1993, Duffy et al. 2007), offspring had learning problems or ADHD symptoms but not an ADHD diagnosis. Additionally, the study by Zahn-Waxler et al. (1988) stated that two of seven offspring of bipolar parents had ADHD. Not stated was that this conclusion was drawn from a parent rating scale. Follow-up interviews of the parent and child yielded no children with ADHD. Neither of the Dutch bipolar high-risk studies (e.g., Hillegers et al. 2005, Wals et al. 2001) nor the Amish study (Shaw et al. 2005), all of which reported low rates of ADHD, were included in the meta-analysis. All told, it would appear that rates of ADHD in bipolar offspring were about the same as population rates (around 10% when all studies are considered). What made the odds ratio high were the low rates in comparison groups that were usually selected to have no psychopathology (“supernormals”).

Finally, in Faraone’s (2012) meta-analysis, just as studies of ADHD among relatives of bipolar disorder found the highest relative risks using the WASH-U-KSADS or K-SADS-E, so did studies of bipolar disorder among relatives of ADHD probands. The one exception came from a study where 2 of 112 (1.78%) fathers with ADHD were said to have mania, basically the same as the population rate (Bhatia et al. 1991). No mention was made of rates of ADHD in the comparison group, which precludes an accurate risk ratio.

In summary, rates of BP I in offspring of BP I probands are almost always higher than in a nonbipolar comparison group. How high they are depends on whether mania is defined conservatively or liberally and on how complicated and strictly diagnosed the bipolar disorder in the parent probands are. These conclusions apply equally to the relationship of ADHD and bipolar disorder, whether the focus is on rates of BP I disorder in samples with ADHD or on rates of ADHD in samples of BP I disorder.

Longitudinal Course of Bipolar Disorder

The most pressing question about childhood-onset bipolar disorder (and especially mania/BP I disorder) is whether it is continuous with bipolar disorder in adulthood. Manic symptoms, whether or not they occur in episodes, are clearly impairing in children. The question is one of homotypic versus heterotypic continuity: Is mania the same disorder starting younger, does it evolve into something else, was it something else all along, or, in some cases, does it remit completely? These issues are further complicated by the fact that BP I disorder in adults is heterogeneous in its phenomenology and outcome, and although there are many short-term outcome studies, very few are longer than four years. In child- and adolescent-onset studies, a minimum of a four-year follow-up is necessary for the samples to enter adolescence and in some cases young adulthood.

A number of follow-up studies of adults with BP I disorder have examined overall outcome. Although when the study took place, the sample composition and definitions of outcome differ and some studies have sample overlap, it is possible to combine information into (a) recovered, remitted, good outcome; (b) incomplete remission or fair outcome; and (c) chronic or poor outcome. With this metric, a little less than half of the samples do fairly well (they experience no recurrences, or if there is a recurrence, good functioning resumes when the episodes remit), approximately one-quarter of the samples do poorly (they are chronically ill in one or another mood state, they attempt or commit suicide, or they become chronically psychotic), and outcomes for the remaining one-quarter of the samples are in between (Angst & Sellaro 2000; Angst et al. 2003; Carlson et al. 1974, 2012; Coryell et al. 1998; Goldberg & Harrow 2011; Morrison et al. 1973; Stephens & McHugh 1991; Tsuang & Dempsey 1979).

Only three of the studies cited above have followed participants for four or more years and provide the kind of sample and follow-up detail that allows some comparison with child and adolescent follow-up studies: Angst’s Zurich sample, the Collaborative Depression Study, and the Suffolk County Mental Heath Project (Angst et al. 2003, Solomon et al. 2010, and Carlson et al. 2012, respectively). In these studies, where follow-up duration ranges from 4 to more than 30 years, 40% to 50% of adult participants were male, 89% to 100% had been initially hospitalized, and 55% to 100% were psychotic at their index episode. Their age at the time of study entry was late 20s to early 30s; age of onset was approximately 25 years. As noted above, overall outcome is decent for almost half the sample, and the rate of chronicity and poor outcome is approximately 20% to 25%. Comorbidity was not a focus of these studies, although in two studies that used data from other papers on the same samples, ADHD/externalizing disorders occurred in a little over 20% of the adult samples (Carlson et al. 2002, Winokur et al. 1993). Median manic episode duration was 7 to 16 weeks, and median time ill over follow-up was about 20%.

The impact of childhood psychopathology—either prior externalizing disorders (ADHD and/or conduct disorders) or other childhood psychopathology (anxiety disorders, depressive symptoms, adjustment disorders)—on outcome in bipolar disorder has been examined in the Suffolk County Mental Health Project (Carlson et al. 2002, 2012). The study reported that childhood psychopathology prior to the onset of BP I contributed substantially to the global assessment of functioning at both the 24-month and 48-month follow-ups and was more predictive of functioning and length of time ill than was age of onset. In terms of influence on episode duration and recurrence, among the significant predictors of time to remission were young age of onset and childhood psychopathology. Almost two-thirds of remitters relapsed by four-year follow-up, mostly to a manic episode. The predictors of time to relapse were depressive symptoms during childhood (Bromet et al. 2005).

Adolescent-onset bipolar studies have smaller samples and shorter-term follow-up (four to five years) (Jairam et al. 2004, Srinath et al. 1998, Strober et al. 1995). As with adult studies, half the patients in these studies were male and were hospitalized for their index episode, which was usually manic or mixed. They were ages 14 to 16 and had an adolescent onset. Rates of psychosis varied from 28% to 63%. Rates of externalizing comorbidity were low (4% to 15%). The index manic episode duration was 8 to 16 weeks. Recovery rates were almost 100% if follow-up was long enough, but relapse over follow-up ranged from 44% to 67%—higher if subthreshold symptoms are counted (Strober et al. 1995). In the two samples from India, about half the teens had a good recovery, about 20% were chronically ill, and the remainder of the teens were somewhat impaired. These adolescent samples, then, appeared to be roughly similar to their adult counterparts in terms of their relatively classic description and outcome.

There is one follow-up study of children (Wozniak et al. 2011) and two follow-up studies of children and adolescents (Birmaher et al. 2009a, Geller et al. 2008) that are four years or longer. In studies by Wozniak and co-investigators (2011) and Geller and colleagues (2008), the samples were diagnosed with current mania/BP I at the time of study entry. In the Course of Bipolar Youth study (Axelson et al. 2006, Birmaher et al. 2009a), it is unclear how many subjects were symptomatic with mania at study entry because worst episode rather than current status was reported. Age comparisons included the entire bipolar spectrum (BP I, BP II, and BP-NOS), but in the four-year follow-up data, BP I subjects were described separately from the BP II and BP-NOS subjects.

Compared to the adolescent-onset and adult studies reported above, children in the Massachusetts General Hospital (Wozniak et al. 2011) and Washington University (Geller et al. 2008) bipolar outcome studies were all outpatients, predominantly male, and had very high rates of ADHD and oppositional defiant disorder (>80%). These studies did not delineate episodes and counted the same symptoms toward ADHD and mania. Most of the children did not recover or had very long first episodes of mania (mean 142.7 ± 139.4 weeks). Most of the Massachusetts General Hospital sample (93.6%), who were still children and young adolescents at the four-year follow-up, either continued to meet full diagnostic criteria for BP I disorder (73.1%), or continued to have persistent subthreshold BP I disorder (6.4%). In the Washington University study, where more than half the sample was over age 18 at follow-up, respondents spent 60% of the follow-up period with mood disorder symptoms. As is usual with data on percent time ill, it isn’t clear how this was distributed across the sample (e.g., whether most youths were ill for 60% of the time or if some were continuously well and others ill constantly). By the age of 18, 44% had either been continuously ill or had recurrence of the same symptoms. Although tables in the manuscript (Geller et al. 2008, p. 1130) describe the sample in their first, second, and third episodes, no data described whether the children remitted completely from their mania, remained somewhat ill, or were depressed between episodes. No functional follow-up data were described (Geller et al. 2008).

In liberally diagnosed samples, follow-up studies of children with ADHD reveal comparatively high rates of development of bipolar disorder. For instance, the Washington University mania sample had an ADHD comparison group: In a six-year follow-up of these children, 22.2% developed BP I or BP II (morbidity risk of 28.5%). Interestingly, the use of stimulants was found to be a protective factor (Tillman & Geller 2006). In the Biederman et al. (2006) 10-year follow-up study of children with ADHD, 29% developed bipolar disorder compared to 3% of controls. By contrast, in other long-term follow-up studies of ADHD (Barkley et al. 2002, Carlson et al. 2000, Klein et al. 2012, Weiss et al. 1985), conservatively diagnosed bipolar disorder occurred at population rates. One interpretation of the findings in the Washington University and Massachusetts General samples is that untreated ADHD is, in fact, worsening, and some children are developing higher rates of aggressive, irritable, dysregulated behavior rather than mania.

In the Course and Outcome of Bipolar Youth study (Birmaher et al. 2009c), which used a more conservative definition of mania, the childhood-onset BP-spectrum group had more males and higher rates of ADHD than did the teen-onset group, although rates of oppositional defiant disorder were much lower in the childhood-onset group, suggesting that chronic irritability was less of a problem. The duration of mania was at least one year. The youths spent 56% of the time over the course of the follow-up with either full-syndromal or subsyndromal mood symptoms. Those with the earliest onsets had more episodes, polarity changes, and comorbidity than the teen-onset samples. Even when episodes are more central to the diagnosis of mania, childhood onset, comorbidity, and chronicity appear to be inextricably intertwined.

The best way to explain the high comorbidity of ADHD and mania in some studies, then, is to suggest that it results from counting ADHD symptoms toward both mania and ADHD. Thus, the liberal studies show poor outcome for very-early-onset bipolar disorder because many of their bipolar subjects really have severe ADHD, which is chronic. Although the co-occurrence of mania and ADHD symptoms may represent a complex bipolar subtype, the co-occurrence may also simply represent severe ADHD with mood lability. Regardless, it is both stable and chronically impairing. Interestingly, ADHD researchers are finally paying attention to the separate impact of emotion dysregulation on outcome, and the impact is not favorable (Barkley & Fischer 2010, Martel 2009, Sobanski et al. 2010).

Although the conservative studies report lower rates of comorbidity and outcomes that are not uniformly poor, even some individuals conservatively diagnosed with bipolar disorder have poor outcomes. Individuals with poor outcomes are likely to have comorbid disorders (or, in adults, a childhood history of nonbipolar psychopathology).

Conclusion

The premise of this review is that conservative and liberal philosophies have influenced how investigators design and interpret their studies on child- and adolescent-onset bipolar disorder. The conservative viewpoint is that BP I disorder is defined with distinct episodes of mania that are clearly different from the person’s usual functioning and behavior. The difference is best identified either by a convergence of identified patient information with parent reports or by some other source documenting that there has been a change in functioning and behavior. Within the episode, although developmental factors and contexts must be considered, the manic symptoms are basically as they are defined in adults. Although comorbidity with ADHD may occur, unless there is significant change from baseline, ADHD symptoms should not be used toward the diagnosis of mania. Irritability also must be part of the episode of mania and not a symptom that is borrowed from other conditions and merged into the mania criteria. Defined this way, mania is uncommon prior to puberty and increases in frequency thereafter.

A more liberal viewpoint would argue that the above definition is too narrow. Any informant can identify manic symptoms, or multiple informants can each provide unique information, supporting the “OR” rule (i.e. symptoms can come from one informant OR the other without their necessarily agreeing) and leading to higher prevalence rates. In addition, given how difficult it is to determine which symptoms, in children with many problems, are due to which condition (and there is no question that this is difficult), symptoms shared by comorbid conditions should be counted toward all of them. For example, irritability can be counted toward depression, generalized anxiety, mania, oppositional defiant disorder, etc. Episodes are given less weight than the DSM-IV symptoms because episodes have, in fact, been poorly defined by the criteria and the structured interviews designed to elicit these criteria.

Although it is important to compare the relative validity of the liberal and conservative positions, the larger lesson is that the samples and results acquired with these divergent philosophies are dissimilar. Differences are apparent, not surprisingly, in prevalence, rates of co-occurring ADHD, rates of bipolar disorder and ADHD in family/genetic studies, and outcome. The way information is obtained in structured or semistructured interviews precludes being able to re-examine one’s sample and determine which child was liberally versus conservatively diagnosed, as there is no way to reconstruct the episode. Both perspectives can claim evidence for reliability and validity that supports their position. That is to say, investigators with the same philosophy agree about what they are defining, and they use the same philosophy to identify symptoms in family members and in follow-up studies. Thus, the good news may be that each philosophy is internally consistent. The bad news is that the conservative and liberal philosophies are so different that it is extremely difficult to compare studies conducted from these different perspectives.