The Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition, Alternative Model Conceptualization of Borderline Personality Disorder: A Review of the Evidence

Abstract

In the 10 years following the publishing of the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM–5), studies have accumulated testing the validity and utility of the alternative model for personality disorders (AMPD) in the context of borderline personality disorder (BPD). In this article, we review the studies that have tested how well the AMPD conceptualization of BPD captures the traditional (DSM–5, Section II) conceptualization of BPD. Although we note that studies that measure the full conceptualization of the AMPD-BPD are limited compared with studies focusing on a single aspect of the AMPD, studies reviewed suggest that the AMPD conceptualization of BPD largely overlaps with Section II, is associated with a similar range of external constructs, and can be measured with similar levels of interrater reliability. This evidence is promising in terms of the goal of the AMPD developers to not lose relevant and clinically meaningful information associated with traditional conceptualizations of BPD. However, further applied research is needed to understand how the AMPD may improve upon our existing categorical conceptualization of BPD.

There have long been criticisms of the categorical diagnostic system for personality disorders introduced in the Diagnostic and Statistical Manual of Mental Disorders, Third Edition (DSM–III), which highlight the limited reliability and validity of a borderline personality disorder (BPD) diagnosis in research and practice. For one, there is extensive comorbidity of BPD both with other PDs and non-PD psychiatric diagnoses (Grant et al., 2008). Second, high heterogeneity is observed among patients with a BPD diagnosis and any two individuals with BPD could share very few features. Finally, the diagnostic threshold also excludes individuals who may only present with one or two criteria who demonstrate clinically meaningful impairment.

However, due to concerns about the clinical utility of a fully dimensional model, diagnostic formulations for specific disorders (including BPD) were retained in the alternative model for personality disorders (AMPD) in Section III of the DSM–5 that was introduced to address the shortcomings of the categorical system. The AMPD combines core impairment in personality functioning (i.e., at least moderate impairment in two or more of the domains of identity, self-direction, empathy, and intimacy; Criterion A) with specific configurations of pathological personality traits (Criterion B). To receive a diagnosis of BPD, one must present with four of the following traits: emotional lability, anxiousness, separation insecurity, depressivity, impulsivity, risk-taking, and hostility (American Psychiatric Association, 2013).

Research has accumulated testing various aspects of this model in relation to BPD. Given concerns regarding the potential loss of information based on years of research on BPD, a large majority of studies on DSM-5-III BPD focus on convergence between the categorical conceptualization of BPD (listed in DSM-5-II) and the AMPD conceptualization. Research also focused on the construct validity of the DSM-5-III BPD (how it overlaps with the nomological net of DSM-5-II BPD). Lastly, a limited number of studies have compared interrater reliability.

The aim of the current article was to systematically review empirical studies that evaluated how well the DSM-5-III BPD captures the categorical (DSM-5-II) conceptualization. Details about the literature review and study details are described in the online supplemental materials, whereas our article briefly narrates the overall findings of these studies organized by the aims to (a) evaluate coverage or convergence, (b) construct validity, and (c) interrater reliability.

Convergence Between Section II and AMPD Conceptualizations of BPD

Evaluation of the convergence, or overlap, between DSM-5-II and DSM-5-III BPD was studied in n = 9 studies using both Criteria A and B (Table S1 in the online supplemental materials), n = 14 with Criterion A alone (Table S2 in the online supplemental materials), and n = 21 with Criterion B alone (Table S3 in the online supplemental materials). For studies including both criteria, bivariate correlations between composite scores of both operationalizations (i.e., summing all elements of Criteria A and B together with criterion counts of DSM-5-II BPD) were large (.76–.80) based on clinician ratings (Morey & Skodol, 2013) and self-report (McCabe & Widiger, 2020) and diagnostic agreement of the two disorders was 78.6% (Morey, 2019). When measures of Criteria A and B were entered simultaneously in a regression, they accounted for variance in DSM-5-II BPD (typically using criterion counts) ranging from .40 to .73. Although this conclusion should be tempered by the overall small number of studies reviewed, it appears as if the greater variance in Section II BPD was accounted for when using interviewer rated features (vs. self-report); however, the overall diversity of measures and methods used to operationalize DSM-5-II and DSM-5-III BPD made it difficult to make any further conclusions driven by methodology.

Regressions testing whether Criteria A and B incremented each other in predicting DSM-5-II BPD revealed that Criterion B tended to account for a greater amount of incremental variance in DSM-5-II BPD (13%–27% more variance), which was always significant; however, with the exception of two studies (Anderson & Sellbom, 2018; Few et al., 2013), Criterion A still predicted significant unique variance in DSM-5-II BPD (0%–1% more variance). Results were the same when using a single composite score each to represent Criteria A and B, as when multiple scores were entered to represent Criteria A (four domain scores) and B (seven facet scores).

In studies testing overlap between measures of Criterion A and DSM-5-II BPD, multiple measures of Criterion A were used including those designed specifically based on the DSM-5-III (see Table S2 in the online supplemental materials) and other self-report measures that mirrored the structure and content represented in Criterion A. These studies largely used either clinical (forensic included) or mixed clinical and community patients. Measures of Criterion A (total score, self- and other-impairment, and the four domain scores) were associated with dimensional scores of DSM-5-II BPD to a medium-to-large degree, with the exception of one study conducted with Iranian patients (Amini et al., 2015). The Levels of Personality Functioning Scale–Self Report (LPFS-SR) demonstrated the strongest associations with measures of DSM-5-II BPD (Hemmati et al., 2020; Hopwood et al., 2018; Morey, 2017; Sleep et al., 2019), with slightly lower magnitude correlations for interpersonal versus self-impairment. However, these studies included the lowest proportion of participants from clinical settings, and estimates may be inflated due to mono-method bias (all studies used self-report measures of DSM-5-II BPD). A brief screening measure developed specifically for the DSM–5 (LPFS-BF and Version 2.0) exhibited smaller associations with DSM-5-II BPD; however, these were based on categorical diagnoses of BPD using the Structured Clinical Interview for the Diagnostic and Statistical Manual of Mental Disorders Axis II disorders (SCID-II) and were still moderate in size (Hutsebaut et al., 2016; Weekers et al., 2019). When using dimensional SCID-II scores, the LPF interview demonstrated the highest overlap with DSM-5-II BPD (large-sized correlations that were slightly smaller for interpersonal scales; Few et al., 2013) followed by the Semi-Structured Interview for Personality Functioning DSM–5, which demonstrated medium-sized correlations, similar in magnitude across self- and interpersonal-domains (Hutsebaut et al., 2017; Weekers et al., 2020).

The largest body of studies evaluated the convergent validity of Criterion B traits with DSM-5-II BPD (18 of 21 studies used the Personality Inventory for the DSM–5 [PID-5] as their measure of Criterion B). A recent meta-analysis of 24 correlation matrices provides an empirical synthesis of the bivariate associations between individual facets and DSM-5-II BPD (Watters et al., 2019). Their results showed that all BPD facets except risk-taking demonstrated meaningful associations (three facets demonstrated large magnitude associations with DSM-5-II BPD: depressivity, emotional lability, and hostility). Results also showed poor discriminant validity for traits, with 11 out of 18 facets not included in the DSM-5-III formulation of BPD reaching meaningful levels of associations. In fact, the facet of cognitive and perceptual dysregulation demonstrated a large association, stronger than observed with risk-taking. Although studies included in the meta-analysis largely used self-report across DSM-5-II and DSM-5-III, a recent study using the SCID interviews developed for the DSM–5 found a similar pattern of associations (Somma et al., 2020). Similar conclusions were drawn based on studies using hierarchical regression to understand (a) which traits accounted for unique variance in dimensional and (b) which traits not included in the Criterion B facets for BPD accounted for incremental variance. Most studies included all seven BPD facets predicting criterion counts of DSM-5-II BPD based mostly on self-report and primarily in clinical samples. Findings were mixed, with the two studies using the SCID-II interview finding that only facets of impulsivity and emotional lability explained unique variance in DSM-5-II BPD (Bach et al., 2017; Orbons et al., 2019). Evidence for each facet predicting unique variance was stronger in the remainder of studies with the exception of risk-taking and anxiousness (Anderson et al., 2014; Bastiaens et al., 2016; Yam & Simms, 2014). When including facets not a part of DSM-5-III BPD in a second block of predictors, suspiciousness (Anderson et al., 2014), cognitive and perceptual dysregulation, and deceitfulness (Bastiaens et al., 2016) predicted additional variance in DSM-5-II BPD.

A few studies examined Criterion B as a whole, with either latent factors or composites calculated by summing scores for the seven BPD trait facets/DSM-5-II criteria separately. Correlations between composites/factors were high (.81 and .86; Miller et al., 2015; Sellbom et al., 2014) and diagnostic prevalence of BPD based on the Criterion B composite was similar to rates seen in analogous samples (Yam & Simms, 2014). However, there was also strong overlap between BPD and antisocial personality disorder across operationalizations in one study (Miller et al., 2015).

Construct Validity of the AMPD Conceptualization Compared With Section II BPD

One concern raised during the reconceptualization of PD diagnoses was the loss of clinically relevant information for specific PDs. Therefore, establishing the construct validity of DSM-5-III BPD is needed. Several studies in our review examined how DSM-5-III BPD was associated with constructs within the nomological net of DSM-5-II BPD, one using both Criteria A and B (Munro & Sellbom, 2020) and seven using only Criterion B. The study using both Criteria A and B (combined score of the LPFSBF total score and seven PID-5 facet scores) found similar magnitude correlations between int4erpersonal violence and DSM-5-II and DSM-5-III BPD (r = .13–.35 with DSM-5-II and r = .08–.28 with DSM-5-III); however, DSM-5-II BPD was more strongly correlated with frequency of psychological aggression, minor acts of violation, and all acts of violence.

Studies focused on Criterion B were conducted in a mix of samples with patients (n = 2), community members (n = 1), undergraduates (n = 3), and mixed patient/undergrads (n = 1). All but one study (Miller et al., 2015) used the PID-5 and all but two studies (Fossati et al., 2017; Miller et al., 2015) used self-report measures of Section II BPD (Personality Diagnostic Questionnaire-4 and SCID-II-PQ). Outcome measures included five-factor model personality traits, internalizing symptoms (e.g., depression and anxiety), externalizing symptoms (e.g., drug use, crime/antisocial behavior, attention deficit hyperactivity disorder symptoms), social cognition, broad functional domains (e.g., total personality pathology, general functioning), self-harm, and abuse history. Although the majority of outcomes were measured with self-report, Fossati et al. (2016) used an interview measure of personality pathology, and Smith and Samuel (2017) included informant reports (mostly parent report) of their outcome measures.

Bivariate correlations between DSM-5-II criterion counts and either composite measures of the Criterion B trait facets for BPD(i.e., summed or latent factor) for each of the facets were tested. Associations between the two operationalizations of BPD and external measures were similar (e.g., intraclass correlation coefficient [ICC] = .98 between two sets of correlations; Miller et al., 2015) such that correlation coefficients did not significantly differ from each other (when tested). There were some exceptions to this finding—correlations between Section III trait composites were more strongly correlated with social functioning, social and physical quality of life, antisocial behavior, dysfunction negative emotions, and anger proneness in two undergraduate samples (Anderson et al., 2016; Boland et al., 2018). However, in a patient sample, correlations were stronger with DSM-5-II BPD for antisocial behavior and self-harm (Anderson et al., 2016). Notably, risk-taking showed minimal correlations (statistically insignificant or small in magnitude) with several outcome measures that were correlated moderate-to-high with DSM-5-II BPD in multiple studies. Two studies tested whether DSM-5-II and DSM-5-III incremented each other in predicting external correlates (internalizing and externalizing; self-harm/suicidal ideation; risky sexual behavior; hospitalizations) using hierarchical regression. Although limited conclusions can be drawn from the small number of studies conducted, evidence was stronger that DSM-5-II BPD did not significantly improve the amount of variance accounted for in these outcomes, with the exception of self-harm/suicidal ideation, number of sexual partners with which the individual used a condom, and number of hospitalizations (Anderson et al., 2016; Boland et al., 2018).

Comparisons of Interrater Reliability Across the DSM-5-II and DSM-5-III Conceptualizations

Two studies compared interrater reliability of BPD using parallel measures. Based on case vignettes, clinicians demonstrated similar rates of reliability across conceptualizations (ICC for DSM-5-II = .392; ICC for DSM-5-III = .333; Morey, 2019). Using the new SCID-5, similar levels of reliability were found (ICC for DSM-5-II = .91; ICC for Criterion A = .87; ICC for Criterion B = .67–.89 for BPD trait facets; Somma et al., 2020).

Conclusion and Directions for Future Research

Since the introduction of the AMPD, numerous authors have sought to test its use. Although these studies are informative, our review suggests that only a limited number of studies have evaluated the AMPD conceptualization of BPD using all criteria (i.e., both Criteria A and B). Despite this, questions regarding how well this new conceptualization of BPD can capture DSM-5-II BPD have been examined in a range of samples using various techniques and results suggest that a large amount of variance in DSM-5-II BPD is accounted for by AMPD constructs. Similarly, DSM-5-III (mostly Criterion B, though) accounts for similar variance in external correlates of DSM-5-II BPD, and interrater reliability appears to be similar. That being said, the facet configuration comprising DSM-5-III BPD requires revision, there is overlap in the way Criteria A and B are operationalized, and more research is needed that examines external criteria in longitudinal studies using measures other than self-report. Finally, we argue that the question we should be asking is not whether DSM-5-III BPD can capture DSM-5-II BPD, but whether it can improve upon our existing conceptualization of BPD in terms of facilitating clinical decision making and improving outcomes for individuals with personality pathology.

Our review aligns with what other authors suggest—that DSM-5-III BPD largely overlaps with DSM-5-II BPD. This is not surprising. Maladaptive self-functioning and interpersonal has long been considered a central mechanism of BPD (Bender & Skodol, 2007) and Criterion B is based off the Big Five traits, under which most psychological traits can be organized (Bainbridge et al., 2022). In this way, the DSM-5-III construct of BPD could be seen as an effective reorganization of the symptom profile of DSM-5-II BPD after removing the more acute dysfunctional behaviors (i.e., self-harm/suicidal behavior). However, the accumulation of research suggests that risk-taking be replaced with cognitive and perceptual dysregulation due to the consistent failure of the risk-taking facet to predict variance in DSM-5-II BPD and external correlates of BPD (the opposite being true for cognitive and perceptual dysregulation).

A second point of contention is whether Criteria A and B are themselves overlapping and whether a diagnosis using one of them is sufficient. Our review was somewhat mixed in this regard, with two studies concluding that Criterion A did not explain additional unique variance in DSM-5-II BPD after accounting for Criterion B and most studies finding that Criterion B accounted for a greater amount of variance in DSM-5-II BPD. However, as other have suggested, this may be because the maladaptive nature of Criterion B traits leads to a saturation of impairment in these measures, thus conflating with Criterion A, which was originally designed to capture the impairing nature of personality pathology (Morey et al., 2015). One study has tested and refuted this hypothesis, finding that Criterion A had similar overlap with general and pathological personality traits (Sleep et al., 2020); however, more research should explore how to separate personality severity and style (see Hopwood et al., 2018 for a discussion). In practice, rather than using Criteria A and B on the same level, it may be useful to use Criterion A as a screener. Alternatively, it may be fruitful to evaluate whether level of impairment (Criterion A) moderates clinical decision-making when the BPD trait profile is present. For example, whether Criterion A can reliably be used to choose between levels of care (e.g., group vs. comprehensive dialectical behavior therapy programs).

Despite the promise of a hybrid system to move the field away from the problems inherent in a categorical diagnostic system, our review suggests that the major issues of DSM-5-II BPD are largely still present and more information is needed regarding utility. Specifically, DSM-5-III BPD suffers the poor discriminant validity seen in DSM-5-II BPD. This could be an unavoidable consequence of placing boundaries around psychiatric illness, but it also could be because this research is designed to map DSM-5-III BPD onto its predecessor. For this reason, we recommend that future research focus less on the cross-sectional concordance between diagnostic constructs and more on the utility and implementation of this new diagnostic model. Without sufficient evidence that the AMPD improves upon the categorical system, it may be difficult to motivate the daunting task of overhauling a diagnostic system that has become entrenched in the mental health field. For this reason, we urge future research to focus on the practical use of the AMPD and include perspectives of patients and service providers who are typically excluded from this research (i.e., social workers, counselors, and psychiatrists) who see the bulk of patients in clinical practice. This is especially the case for BPD, which is the only PD for which empirically based treatments have been developed and is the most common PD seen in clinical practice. Although initial studies of clinician perceptions (Milinkovic & Tiliopoulos, 2020) suggest acceptability and optimism for the use of the AMPD, how its use plays out in clinical practice is needed. Future research should be conducted that directly compares the two diagnostic systems in terms of implementation and clinically meaningful outcomes (suicide, arrests, and hospitalization) should be tracked longitudinally to determine whether the AMPD can lead to improvements in selecting treatments, adherence, symptom improvement, and other long-term patient outcomes.