Portal de Periódicos da CAPES

Bipolar vs. borderline – diagnosis is prognosis once again

2016; Wiley; Volume: 133; Issue: 3 Linguagem: Inglês

10.1111/acps.12560

1600-0447

S. Nassir Ghaemi,

Mental Health and Psychiatry

In the course of the DSM-5 debates, a key question was how to avoid ‘false-positive’ diagnoses, in other words, saying that someone had a psychiatric condition when they did not, or when they had a different one. The leadership of the DSM-IV and DSM-5 task forces held to the view that broadening the definition of bipolar illness, such as reducing hypomanic episode duration from 4 to 2 days, would produce a high rate of ‘false-positive’ diagnoses 1. DSM-IV and 5 leaders told their respective committees to avoid broadening diagnostic definitions as much as possible, based on the above rationale. Yet there is zero scientific evidence to support the 4-day threshold for the definition of hypomania, and there are a number of studies which support a 2- to 3-day cutoff as being diagnostically valid 2. As one of the DSM-IV task force leaders once told me, the 4-day threshold was invented, without any scientific evidence for it, as a number that was less than 7 days (the definition of mania) but more than a day or two. A common claim is that type II bipolar illness is often misdiagnosed in persons who have borderline personality 3. The latter condition involves ‘affective instability’, so if the definition of hypomania or the bipolar spectrum was broadened, the bipolar diagnosis would be made in persons who have the affective instability of borderline personality. This opinion ignores the fact that a clinician or researcher who only focuses on symptoms is not a good clinician or researcher. Symptoms by themselves are not sufficient to distinguish between diagnoses, whether in psychiatry or any aspect of medicine. This is why in psychiatry three other classic diagnostic validators are invoked: course of illness, genetics, and biological markers, which we have suggested are essential in this differential diagnosis in particular 3. Indeed, a major problem in my view with the DSM definition of borderline personality disorder is that it ignores the major course factor of childhood sexual abuse, and it relegates the key course factor of repetitive self-harm to one of nine criteria. In other words, borderline personality is so defined that mere affective instability and unstable interpersonal relationships along with anger and impulsivity would suffice for the DSM diagnosis, a definition that would encompass many patients with bipolar illness. In the study in this issue of Acta by Bayes et al., we are presented with a very creative and elegant analysis of broad vs. narrow definitions of bipolar illness, compared with borderline personality, with special attention to the diagnostic validators of course and genetics. The study initially presents univariate comparisons between the two diagnoses and identifies the following key clinical differences. Childhood sexual abuse was present in DSM-defined bipolar illness in 12.2% (10/82) vs. 42.3% (22/52) of DSM-defined borderline personality, producing a relative risk (RR) of 3.47 (95% confidence intervals 1.79, 6.72). When using broad clinical criteria for bipolar illness, similar rates are found (14.5% in bipolar illness vs. 51.9% for borderline personality; RR = 3.57, 95% CIs 2.17, 5.90). On more broadly defined ‘developmental trauma’, there was about a two-fold increased risk with borderline personality (65.4%, 34/52) vs. bipolar illness (34.9%, 29/83; RR = 1.87, 95% CIs 1.31, 2.67) (RRs and CIs were calculated by me). Acts of self-harm were still about two-fold higher with DSM-defined borderline personality (RR = 1.70, 95% CI 1/32, 2.19). More importantly, absolute rates were very high (83.0%, 44/53) in borderline personality vs. only about one-half in bipolar illness (48.8%, 40/82). Broader clinical definitions produced similar differences. Suicide attempts were also similarly different, about twice as frequent in borderline personality as in bipolar illness (59.6% vs. 30.1% respectively), with a somewhat higher rate in the broader clinical definition of borderline personality (73.1%) but not bipolar illness (30.4%). Family history of bipolar illness was twice as frequent in bipolar subjects vs. borderline personality, whether using DSM or broader criteria (41.3% with DSM bipolar illness vs. 20.0% in DSM-based borderline personality, RR = 2.19, 95% CIs 1.18, 4.05). Family history of ‘depression’ did not differentiate groups, which makes sense given that ‘depression’ or ‘major depressive disorder’ may not be a scientifically valid diagnostic construct 4. Using DSM definitions, rates were 69.2% for borderline personality vs. 75.9% for bipolar illness, with similar rates in broad clinical definitions. Childhood depersonalization and parent rejection were about 2–3 times more common in borderline personality than in bipolar illness, however defined, and, as expected, bipolar illness occurred equally in both genders, while borderline personality occurred in women in 85% of cases. In sum, the key distinction between these two conditions is childhood sexual abuse, followed by repeated acts of self-harm and depersonalization. If those features are present, then other classic borderline constructs, like help rejection, would be seen. But without those classic features, then simply rejecting help or being female is not specific to borderline personality. The authors conducted regression models, correcting for confounding bias and effect modification, to spell out how these features interacted with each other and which ones best predicted diagnosis. They found that the largest predictor was childhood sexual abuse, with an even larger difference than seen in simple univariate comparison. In multivariate regression, the odds of childhood sexual abuse was found to be almost 10 times more common in DSM-defined borderline personality than in DSM-defined bipolar illness (odds ratio, OR = 9.38). It should be emphasized that this effect size is very large, similar to the impact of cigarette smoking on lung cancer, and yet the DSM system excludes childhood sexual abuse as a criterion of borderline personality, and many borderline experts minimize its relevance to diagnosis. The childhood sexual abuse effect size was not as huge, though still large, using broader clinical definitions of bipolar illness (OR = 3.62). After this major difference, all other effects besides gender were around two or three times more frequent with one diagnosis vs. another. Again, the key differences in favor of borderline personality tended to be more depersonalization, self-harm acts, and much less family history of bipolar illness (OR = 0.17 for DSM-defined probands). Other features were more non-specific or inconsistent. Lastly, the authors conducted ROC analyses for sensitivity and specificity and predictive values of the diagnostic features assessed. This is where the clinically broad definitions of bipolar illness seemed most useful. Using DSM definitions, the presence of manic or hypomanic symptoms produced a sensitivity of 66.7% and a specificity of 100%, translating to a positive predictive value (PPV) of 100% and a negative predictive value (NPV) of 83%. These results were not changed at all by adding in to the model ‘relationship difficulties’, which contradicts a central focus of many experts on borderline personality. However, the addition of childhood sexual abuse increased sensitivity to 83% and improved the NPV to 90%, while still producing a high PPV of 95%. The further addition of rejection avoidance produced no change in these numbers, again contradicting a central psychoanalytic interpretation of the borderline personality construct. Using the broader clinical definitions, manic/hypomanic symptoms produced similar sensitivity and predictive value effects (63.3% sensitivity, 91.2% PPV, 86.5% NPV), which did not improve by adding depersonalization, but improved notably with the addition of self-harm (90% sensitivity, 90% PPV, 96% NPV). Again rejection avoidance had no effect. These analyses confirm a prior review 5 that broad bipolar diagnoses lead to lower false-positive diagnoses (better negative predictive value) when combined with known and easily identifiable course of illness features, in this case absence of parasuicidal self-harm and/or childhood sexual abuse. In short, these analyses provide an algorithm to differentiate bipolar illness from borderline personality, whether you strictly adhere to DSM definitions, or whether you use broader bipolar definitions. Using strict DSM definitions: First, assess the presence of manic or hypomanic episodes or even brief manic symptoms. If present, you will correctly predict bipolar illness about 90% of the time (with only 10% false positives), and you will misdiagnose bipolar illness in persons who have borderline personality about 10% of the time. Next, assess the presence of childhood sexual abuse. If absent, you will increase your true-positive bipolar diagnosis to 95% and still have about a 10% misdiagnosis of bipolar illness in those with borderline personality. Using broader bipolar definitions: Begin again with the presence of manic or hypomanic episodes, producing a 91% correct diagnosis rate and a 14% rate of misdiagnosis of bipolar illness as borderline personality. Next, instead of childhood sexual abuse, assess parasuicidal self-harm. If present, you will still have 90% correct bipolar diagnosis (only 10% false positive misdiagnosis) and only 4% misdiagnosis of bipolar illness as borderline personality. To conclude Kraepelin famously taught that diagnosis is prognosis. This dictum is supported by the findings of this study, which teaches us that course features, such as childhood sexual abuse and parasuicidal self-harm, effectively and successfully distinguish between bipolar illness and borderline personality. Even with broadly defined bipolar illness, we find the same effect. DSM ideology notwithstanding, we can correctly diagnose bipolar illness with less than four days of manic symptoms, and we can avoid overdiagnosis in relation to borderline personality by assessing relevant borderline course features, especially childhood sexual abuse.