Differences in Diagnostic Rules Used to Determine Borderline Personality Disorder Impact Prevalence and Associations with Clinically Relevant Variables: Findings from the National Epidemiologic Survey on Alcohol and Related Conditions-III

Abstract

Borderline personality disorder (BPD) is a serious and understudied mental health condition associated with profound personal and public health consequences. Methodological differences in characterizing BPD may limit understanding the scope of the disorder’s prevalence and effect. For example, using different diagnostic rules for BPD can affect apparent prevalence, comorbidity, and clinical presentation. This study examined how differences in diagnostic rules used to assign BPD diagnosis impacted its prevalence and associations with clinically relevant variables (e.g., demographics, comorbidity, treatment-seeking). Participants were a nationally representative sample of 36,309 non-institutionalized US adults. All variables were assessed via clinical interview (AUDADIS-5). Six diagnostic rules determined BPD status. We used frequencies to examine prevalence rates of and associations between BPD and other clinical variables, and logistic regressions to examine the associations between each BPD variable and the other outcomes. The prevalence of BPD ranged widely—from 0.5 to 11.4%—per the diagnostic rule used. Associations between BPD diagnosis and various outcomes and clinical variables generally remained stable across all diagnostic rules, though effects became more extreme as diagnostic rules became more restrictive. Additionally, meaningful differences emerged as a function of the number of items used (30 versus 18 items) even with no other changes to diagnostic rules. The field examining BPD and associated problem behaviors should critically consider how to most effectively characterize BPD to understand these problems more accurately and optimize generalizability of findings.

INTRODUCTION

Borderline personality disorder (BPD) is a serious mental health condition characterized by affective instability, impulsivity, impairment in self-image, and intense, unstable interpersonal relationships (American Psychiatric Association [APA], 2013). BPD is associated with high rates of morbidity and mortality; between 23–76% of individuals with BPD attempt suicide (Grilo & Udo, 2021; Kuehn et al., 2020) and roughly 10% die by suicide (Black et al., 2004). Moreover, the social, familial, and economic sequalae of BPD are substantial. Individuals with BPD experience increased psychiatric comorbidity (El-Gabalawy et al., 2010; Jacobi et al., 2021; Pagura et al., 2010), disability (El-Gabalawy et al., 2010; Kjær et al., 2020; Zanarini et al., 2004), social/occupational impairment (Morgan et al., 2013; Wagner et al., 2022; Zimmerman, 2015), and diminished quality of life (Botter et al., 2021). With respect to treatment, estimates suggest that BPD patients account for between 9–22% of outpatient (Korzekwa et al., 2008; Zimmerman et al., 2008) and 20–25% (Iliakis et al., 2019; van Asselt et al., 2007) of inpatient psychiatric admissions. Monetary costs associated with BPD are estimated at more than double those associated with depression (Iliakis et al., 2019). Taken together, these facts indicate BPD constitutes a substantial source of public health burden.

Despite the seriousness of this disorder, BPD remains understudied relative to other mental health conditions (Gunderson et al., 2018), and BPD research remains grossly underfunded (Zimmerman & Gazarian, 2014). One of the most fundamental issues limiting growth in BPD research is methodological. Specifically, differences in methodology (e.g., sampling, study design) may add to the variability in estimated prevalence, limiting understanding of the true population prevalence and scope of the problem. Researchers and public health officials alike traditionally characterize BPD as a low prevalence disorder, citing estimates varying from >1–5.9% in the US (e.g., Grant et al., 2008; Korzekwa et al., 2008; Lenzenweger et al., 2007; Trull et al., 2010; Zanarini et al., 2004; Zimmerman, 2015). Historically, researchers have concluded that stigma may play a role in the assignment and reporting of BPD diagnosis, leading to underreporting, biased sampling, and variability estimation. However, these reasons are not the only issues at play. Recently, two separate research groups examined BPD prevalence using the same large nationally representative sample with markedly different results: 2.7 versus 5.9% (Grant et al., 2008; Trull et al., 2010). Understanding how and why their findings differed and what factors bias estimation of prevalence rates are crucial to informing future work on the prevalence and clinical characterization of BPD.

Grant and colleagues (2008) and Trull and colleagues (2010) examined BPD prevalence using Wave 2 of the National Epidemiologic Survey on Alcohol and Related Conditions (NESARC), a dataset of 34,653 non-institutionalized US adults interviewed between 2004–2005 (Grant et al., 2008; Trull et al., 2010). BPD was assessed with 18 items in the Alcohol Use Disorder and Associated Disabilities Interview Schedule-IV (AUDADIS-IV). Grant and colleagues (2008) first examined the prevalence and correlates (disability, comorbidity) of lifetime BPD diagnosis using a set of diagnostic rules including endorsing at least five symptoms (consistent with the DSM) with at least one symptom associated with reported social or occupational distress/impairment. They observed an unprecedentedly high prevalence of 5.9% of US adults endorsing symptoms consistent with lifetime BPD compared to previous examinations of BPD prevalence (e.g., Lenzenweger et al., 2007). Motivated in part by concern over the high prevalence rate of BPD reported by Grant and colleagues (2008), Trull and colleagues (2010) examined BPD prevalence with the same sample using different diagnostic rules, which produced vastly different results. Trull and colleagues (2010) required endorsement of at least five BPD symptoms, and all had to be associated with impairment. With these revised rules, they found a lifetime BPD prevalence of only 2.7%—a substantial difference with meaningful implications from a public health standpoint given the frequency with which these papers have been cited as evidence for the US prevalence of BPD. As of December 2022, 1,752 papers cited Grant and colleagues (2008) and 708 cited Trull and colleagues (2010).

Notably, subsequent studies utilizing the most recent wave of the NESARC—NESARC-III, collected between 2012–2013 (N=36,309)—have exhibited similar variability in applying diagnostic rules for BPD. While no study has explicitly examined BPD prevalence with the NESARC-III, several have explored related questions. For example, Elliott and Ragsdale (2022) examined comorbidity between BPD and bipolar disorder and utilized two different strategies for identifying BPD diagnosis. First, using the NESARC-III variable for BPD diagnosis (derived using diagnostic rules defined by Grant and colleagues [2008]), they reported a lifetime prevalence of 10.5% (n=3,800). They deemed this estimate as “high” versus previous research and adopted more stringent diagnostic rules, requiring more symptoms associated with functional impairment (in line with Trull et al., 2010). With revised rules, they reported a prevalence of 5.9% (n=2,146; Elliott & Ragsdale, 2022). Similarly, Grilo and Udo (2021) utilized several different sets of diagnostic rules, including two provided by NESARC-III, to examine whether lifetime BPD diagnosis and specific criteria were associated with lifetime and past-year suicide attempt (SA). Analyses were completed with all three diagnostic rules, which required that a different number of BPD criteria were endorsed and associated with functional impairment. In examining differential diagnosis as a function of sexual minority status, Rodriquez-Seijas and colleagues (2021) utilized a similar strategy by conducting analyses with both the NESARC-III-provided BPD diagnostic variables and a more stringent set requiring more symptoms be associated with impairment. Additionally, these researchers used only the 18 (of 30 total) items assessing for BPD in the AUDADIS-5 that overlapped with the AUDADIS-IV from the Wave 2 NESARC (Rodriguez-Seijas et al., 2021). Thus, variability in the rules used to define BPD continues to affect estimates and clinical correlates of the disorder.

These studies highlight the importance of carefully and thoughtfully defining diagnostic rules when making population-level estimates, especially for conditions with substantial impact on public health like BPD. As these studies demonstrate, minor differences in diagnostic rules result in substantial and meaningful differences in prevalence and have implications for understanding the disorder and its effect on public health. Older studies have demonstrated that when even slightly different criteria (including minor variations in questions) are used to characterize disorders with the same assessment administered in different areas, vastly different prevalence rates are found (e.g., Coryell et al., 1981; Robins et al., 1984). Despite such studies, researchers continue to use different diagnostic criteria to determine prevalence rates for different psychiatric problems, including BPD. No systematic investigation of the implication of different diagnostic rules for BPD has been conducted. In light of differing methods characterizing BPD by researchers who used the NESARC-III dataset (e.g., Elliot & Ragsdale, 2022; Grilo & Udo, 2021; Rodriguez-Seijas et al., 2021), the purpose of this study was to examine how systematic changes in the diagnostic rules for BPD impact its understanding, specifically, 1) to provide updated population prevalence estimates for BPD, replicating previous diagnostic rules for defining BPD in a more recent dataset, the NESARC-III, and 2) to demonstrate the effect of differences in the diagnostic rules for BPD on prevalence and clinical characteristics (e.g., comorbidity, suicide behavior, treatment use).

METHOD

Participants

A nationally representative sample of 36,309 U.S. noninstitutionalized civilian adults participated in the NESARC-III. Participants were randomly selected from the United States, with Black, Hispanic/Latinx, and Asian individuals sampled at a higher rate to ensure reliable estimates of these groups (Grant et al., 2014). Unweighted, approximately 56% of the sample was female (n=20,447), and the sample had an average age of 45.6 years old with an age range spanning from 18 to 90. A majority of the sample were identified as White, non-Hispanic (52.9%; n=19,194), with 21.4% identified as Black, non-Hispanic (n=7,766), 19.4% identified as Hispanic, any race (n=7,037), 5.0% identified as Asian/Native Hawaiian/“other” Pacific Islander, non-Hispanic (n=1,801), and 1.4% identified as American Indian/Alaska Native, non-Hispanic (n=511). Detailed procedures concerning the assignment of racial/ethnic groups can be found in the NESARC-III Source and Accuracy Statement (Grant et al., 2014). Analyses were weighted with population-attributable fractions (PAFs).

Measures

All demographic information, assignment of BPD and other psychiatric diagnoses, and history of suicide behavior, treatment-seeking behavior, and receiving disability benefits in the past 12 months were assessed with the National Institute on Alcohol Abuse and Alcoholism AUDADIS-5. Psychiatric disorders assessed with the AUDADIS-5 generally have good reliability and validity (Grant et al., 2015; Hasin et al., 2015a, 2015b).

BPD

The AUDADIS-5 expanded on the 18-item BPD assessment included in AUDADIS-IV (used in the Wave 2 NESARC). In the AUDADIS-5, BPD was assessed with 30 items. Details concerning differences between these measures are provided in the online supplemental materials. Participants were instructed to respond with respect to how they felt or acted most of the time—excluding instances when they were not themselves or when they acted differently than usual because they were depressed, hyper, anxious, drinking heavily or using drugs or medicines or were in withdrawal from these substances, or times when they were physically ill—since early adulthood regardless of the situation or person/people they were with. If any of the 30 items were endorsed, a follow-up question regarding whether the item troubled them or caused problems at work, school, or with family or others (i.e., social or occupational dysfunction/distress) was asked.

The test-retest reliability of BPD in the AUDADIS-5 was fair (Grant et al., 2015). Without the requirement for social or occupational dysfunction/distress, the kappa statistic was 0.54 (SE=0.04); with the requirement for social or occupational dysfunction/distress, the kappa statistic was slightly lower (κ=0.46, SE not reported).

Other Psychiatric Disorders

The AUDADIS-5 also assessed the presence of past-year/lifetime psychiatric disorders. Mood disorders included major depressive episode, major depressive disorder, and dysthymia. Bipolar disorders included manic episode, hypomanic episode, and bipolar I disorder. Anxiety disorders included specific phobia, social phobia, panic disorder, agoraphobia, and generalized anxiety disorder, with posttraumatic stress disorder (PTSD) assessed separately.

Substance use disorders (SUDs) assessed problems with alcohol, sedatives, cannabis, opioids, cocaine, stimulants, hallucinogens, inhalants/solvents, club drugs, heroin, other drugs, and tobacco. Endorsement of diagnostic criteria consistent with any past-year/lifetime SUDs (except alcohol or tobacco use disorder, examined separately) identified a participant as having a past-year/lifetime drug use disorder.

History of Suicide Behavior

History of engaging in suicide behavior was assessed with six variables: history of SA; the age at first SA; the age at most recent SA; SA occurring in the past year, past 5 years; and the number of lifetime SAs. Items in the AUDADIS-5 assessed whether participants had ever attempted suicide, the age at first SA, the age at most recent SA, and the number of lifetime SA, and other variables were calculated.

History of Treatment and Disability

Participants who endorsed symptoms for any psychiatric disorder were asked whether they engaged in a range of treatment-seeking behaviors specific to those disorders during and prior to the last 12 months. For analyses, all such variables were combined into two composites: “sought help in the last 12 months (i.e., past-year)” and “sought help prior to 12 months ago.”

The AUDADIS-5 also assessed a range of disability benefits (e.g., Social Security, Supplemental Security Income, etc.). For analyses, the receipt of any disability benefits in the past 12 months was combined into one item (“Received disability benefits, past-year”).

Procedure

The NESARC-III was preregistered (ClinicalTrials.gov Identifier: NCT01273220). Data were collected between April 2012 and June 2013. Participants first completed a computer-assisted personal interviewing (CAPI) screener, which was conducted to collect household information and to select one or two individuals from the household to participate in the AUDADIS-5, including saliva collection (Grant et al., 2014). The screener began with a confirmation of the sampled address. Then, the interviewer recorded the first name, as well as the sex, age, race, and ethnicity, as well as active-duty military status, of all household members for a household enumeration procedure; to note, individuals on active duty in the US Armed Forces, Military Reserves, or National Guard were not eligible to participate in the NESARC-III. The selection criteria for eligible participants were programmed into the CAPI screener and thus selected individuals into the NESARC-III sample; the interviewer did not have discretion about whom to include in the sample. A Spanish version of the CAPI screener was administered by a bilingual interviewer in households where the members spoke only Spanish.

Following screening procedures, participants completed informed consent, the first incentive module, the AUDADIS-5 interview, the second incentive module, the recontact module (which was composed of collecting/verifying the best time and contact information for standard control purposes; informed consent for the Reliability or Validity follow-up studies; and additional sample person information for follow-up study procedures), and then collection of the saliva sample. The AUDADIS-5 was administered through face-to-face interviews by trained interviewers, who participated in extensive training prior to conducting the assessment. Training was composed of home study (including a 9-hr study package to be completed prior to attending the in-person NESARC-III training session, hands-on practice with the laptop used for study procedures, and a 4-hr in-person training for general interviewer techniques among interviewers new to Westat), demonstrations (i.e., viewing a videotaped demonstration of the full NESARC-III procedures), interactive lectures (including detailed instructions for administering the data collection instruments), practice exercises (e.g., written exercises to test trainees’ comprehension of concepts), and dyad role-playing (for hands-on practice of procedures). All training materials were scripted. Additionally, trainees participated in practices with a paid respondent to conduct an unscripted interview in a “safe” training environment; training staff observed these practice interviews and provided feedback. Full information regarding specific methods for the NESARC-III, including sample selection procedures, can be found in the NESARC-III Source and Accuracy Statement (Grant et al., 2014). The Institutional Review Boards of the National Institutes of Health and Westat, Inc. approved all procedures for NESARC-III data collection. Procedures of the present study were not preregistered. Additionally, they were deemed as exempt under 45CFR46.104 by the Institutional Review Board of Yale University.¹

Data Analytic Plan

SAS (version 9.4) and SPSS (version 28.0) were used to conduct statistical analyses.

BPD Variables

We used six sets of diagnostic rules to define BPD and compare across definitions to examine the effects of differences in these rules on prevalence and clinical correlates of BPD. See Figure 1 for a visual representation of the diagnostic rules and how they relate and differ, Supplemental Table 1 for item content, and Supplemental Table 2 for a description of each set of diagnostic rules and corresponding BPD variable.

Figure 1. Diagnostic Rules for Describing the Six BPD Variables.

Notes. BPD-A: ≥5 BPD criteria, ≥1 criterion associated with functional impairment (FI). BPD-A-18: Similar to BPD-A but composed of the 18 items that overlap between the AUDADIS-IV and −5. BPD-B: ≥5 BPD criteria, all criteria associated with FI. BPD-B-18: Similar to BPD-B but composed of the 18 items that overlap between the AUDADIS-IV and −5. BPD-C: All items for the ≥5 BPD criteria endorsed, ≥1 criterion associated with FI. BPD-D: All items for the ≥5 BPD criteria, all criteria associated with FI.

First, we identified which of the 30 BPD items of the AUDADIS-5 corresponded to each of the nine BPD DSM-5 criteria (see Supplemental Table 1; e.g., Grilo & Udo, 2021). We then determined whether a participant met criteria for BPD using six sets of diagnostic rules. These varied along two primary dimensions: (a) the number of criteria required to be associated with impairment, and (b) the number of items required to be endorsed. Regarding criterion endorsement, consistent with the DSM-5, at least five BPD criteria had to be endorsed in all rules. BPD-A, BPD-B, BPD-A-18, and BPD-B-18 required at least one item must have been endorsed to count as overall criterion endorsement, except for impulsive behaviors where two items were required (i.e., DSM Criterion 4). With respect to impairment, for BPD-A/BPD-A-18, at least one of the endorsed criteria must have been associated with impairment. For BPD-B/BPD-B-18, all had to be associated with impairment. The difference between BPD-A/BPD-B and BPD-A-18/BPD-B-18 (respectively) was the number of items used. To replicate Grant et al. (2008) and Trull et al. (2010) and evaluate differences in prevalence as a function of item use, we recreated BPD-A and BPD-B with the 18 items that overlapped the AUDADIS-IV and −5, creating BPD-A-18 and BPD-B-18.

The last two sets of diagnostic rules required that all items be endorsed for the criterion to count toward BPD diagnosis. For BPD-C, at least one endorsed criterion had to be associated with impairment, whereas all had to be associated with impairment for BPD-D.

Primary Analyses

We first used descriptive analyses (frequencies) to examine the prevalence of BPD in the NESARC-III sample using the six diagnostic rules. We used logistic regression to examine the relationships between BPD and sociodemographic characteristic, psychiatric, suicide, treatment, and disability variables. Analyses were repeated for each diagnostic rule. Additionally, analyses were repeated controlling for sociodemographics and psychiatric diagnoses (to replicate Grant et al., 2008). Reference groups for each analysis were selected based on previous literature (Grant et al., 2008; Tomko et al., 2014).

RESULTS

Prevalence of BPD

We first calculated the prevalence of BPD given the six diagnostic rules. With BPD-A, lifetime BPD prevalence was 11.4% (n=4,301). Of note, despite a slight difference due to weighting between BPD-A and the NESARC-III-derived BPD variable (bpddx1; NESARC-III Diagnostic Variable Codebook, n.d.), the number of participants meeting diagnostic criteria was the same between BPD-A and bpddx1. With BPD-B, prevalence was 5.4% (n=2,069). Both prevalence rates were markedly higher than observed using the Wave 2 NESARC (i.e., 5.9% [Grant et al., 2008], 2.7% [Trull et al., 2010]). To understand the discrepancy, we examined BPD prevalence using only the 18 AUDADIS-5 items that overlapped AUDADIS-IV (see Supplemental Table 1). With the 18 items, prevalence was 6.9% (n=2,695; BPD-A-18) and 3.8% (n=1,446; BPD-B-18).

The final BPD variables were calculated based on maximally restrictive diagnostic rules (see Figure 1). Consistent with these, prevalence was 1.1% (n=408) with BPD-C and 0.5% (n=220) with BPD-D. Taken together, these results indicate BPD prevalence decreases as the definition of BPD becomes more restrictive.

Sociodemographic Characteristics

Table 1 provides prevalence rates and associations between each BPD and sociodemographic variable. With respect to sex, BPD prevalence increased in females and decreased in males as diagnostic rules became more restrictive, from BPD-A (female=52.7%; male=47.3%) to BPD-D (female=62.6%; male=37.5%; see Figure 2). Concerning odds ratios (ORs), all BPD variables were significantly associated with sex (females 1.14–1.56 times more likely to have BPD), except for BPD-A, where sex differences were not significant. Across definitions, those in the oldest age group (65+) had lower odds of meeting criteria for BPD relative to the youngest age group (18–19; e.g., OR_BPD-A=0.45). Regarding race/ethnicity, as the diagnostic rule for BPD became more restrictive, the prevalence of BPD increased among American Indian/Alaska Native individuals (e.g., BPD-A=2.8%; BPD-D=4.4%) and decreased among Asian individuals (e.g., BPD-A=3.0%; BPD-D=1.3%). Observed differences were consistently significant in both groups relative to White individuals (e.g., American Indian/Alaska Native individuals OR_BPD-A=1.92; Asian individuals OR_BPD-A=0.47). No differences were observed in White, Black, and Hispanic individuals across definitions. As rules became more restrictive, prevalence also decreased among people who lived in an urban area (e.g., BPD-A=76.9%; BPD-D=69.4%) and increased among those who lived in a rural area (e.g., BPD-A=23.1%; BPD-D=30.6%), though observed differences were not significant.

Table 1.

Prevalence Rates and Odds Ratios (OR) of Borderline Personality Disorder and Sociodemographic Characteristics (Uncontrolled)

Sociodemographic characteristics	Full sample	BPD variables
		BPD-A		BPD-B		BPD-A-18		BPD-B-18		BPD-C		BPD-D
	% (SE)	% (SE)	OR (99% CI)	% (SE)	OR (99% CI)	% (SE)	OR (99% CI)	% (SE)	OR (99% CI)	% (SE)	OR (99% CI)	% (SE)	OR (99% CI)
Sex
Malea	48.1(0.30)	47.3(0.98)	1.0	42.7(1.33)	1.0	45.1(1.29)	1.0	41.6(1.81)	1.0	39.7(3.10)	1.0	37.5(3.43)	1.0
Female	51.9(0.30)	52.7(0.98)	1.03 (0.93–1.16)	57.3(1.33)	1.26*** (1.08–1.47)	54.9(1.29)	1.14* (0.98–1.32)	58.4(1.81)	1.31*** (1.07–1.62)	60.3(3.10)	1.41* (0.98–2.03)	62.6(3.43)	1.56* (0.97–2.47)
Age, years
18–19b	3.4(0.12)	3.8(0.41)	1.0	3.2(0.48)	1.0	4.3(0.51)	1.0	3.0(0.55)	1.0	3.3(1.14)	1.0	3.4(1.59)	1.0
20–29	18.3(0.36)	23.1(0.85)	1.15 (0.81–1.62)	23.5(1.46)	1.41 (0.87–2.28)	24.6(1.10)	1.08 (0.74–1.57)	25.3(1.68)	1.62* (0.92–2.85)	30.7(3.15)	1.75 (0.64–4.81)	28.4(3.63)	1.56 (0.36–6.83)
30–44	25.7(0.33)	28.9(0.97)	1.00 (0.72–1.39)	30.4(1.44)	1.29 (0.82–2.02)	29.5(1.26)	0.91 (0.62–1.32)	31.1(1.87)	1.40 (0.82–2.40)	31.6(2.62)	1.28 (0.47–3.45)	32.6(2.89)	1.27 (0.33–4.86)
45–64	35.0(0.32)	34.6(0.90)	0.86 (0.62–1.21)	36.1(1.46)	1.11 (0.72–1.73)	34.3(1.27)	0.77* (0.54–1.09)	34.7(1.65)	1.14 (0.68–1.93)	31.6(2.46)	0.93 (0.35–2.53)	33.6(3.54)	0.96 (0.24–3.86)
65+	17.6(0.37)	9.5(0.64)	0.45*** (0.31–0.65)	6.8(0.67)	0.40*** (0.24–0.66)	7.4(0.72)	0.32*** (0.20–0.49)	6.0(0.77)	0.38*** (0.21–0.70)	2.8(0.91)	0.16*** (0.05–0.59)	2.0(0.86)	0.11** (0.01–0.88)
Race-ethnicity
White, non-Hispanica	66.2(0.77)	69.2(1.12)	1.0	69.8(1.41)	1.0	67.7(1.47)	1.0	70.7(1.75)	1.0	68.4(2.65)	1.0	68.1(3.54)	1.0
Black, non-Hispanic	11.8(0.66)	12.3(0.88)	0.99 (0.84–1.18)	11.4(1.03)	0.91 (0.73–1.13)	13.5(1.09)	1.13 (0.92–1.39)	11.7(1.10)	0.93 (0.72–1.19)	12.0(1.61)	0.98 (0.66–1.46)	11.5(2.12)	0.95 (0.52–1.73)
American Indian/Alaska Native, non-Hispanic	1.6(0.12)	2.8(0.36)	1.92*** (1.41–2.60)	2.6(0.43)	1.66** (1.06–2.60)	2.8(0.37)	1.86*** (1.28–2.70)	2.9(0.50)	1.78** (1.10–2.87)	4.8(1.17)	3.02*** (1.35–6.75)	4.4(1.10)	2.75** (1.14–6.61)
Asian/Native Hawaiian/Other Pacific Islander, non-Hispanic	5.7(0.47)	3.0(0.54)	0.47*** (0.33–0.67)	2.8(0.56)	0.44*** (0.28–0.70)	2.5(0.49)	0.42*** (0.27–0.63)	2.6(0.61)	0.42*** (0.24–0.74)	1.9(0.66)	0.32*** (0.14–0.74)	1.3(0.09)	0.22* (0.04–1.34)
Hispanic, any race	14.7(0.67)	12.8(0.83)	0.81** (0.68–0.96)	13.5(0.99)	0.86 (0.70–1.06)	13.5(1.00)	0.89 (0.72–1.10)	12.1(1.17)	0.76* (0.57–1.01)	12.9(2.02)	0.85 (0.52–1.39)	14.8(2.84)	0.98 (0.50–1.90)
Family income, $
0–19,999	22.8(0.50)	33.7(0.92)	2.84*** (2.36–3.42)	35.6(1.42)	3.28*** (2.64–4.06)	37.7(1.20)	3.74*** (3.00–4.67)	37.8(1.72)	3.89*** (2.86–5.28)	44.6(2.76)	6.02*** (3.08–11.76)	46.4(3.39)	6.20*** (2.74–14.06)
20,000–34,999	18.9(0.35)	22.3(0.79)	2.17*** (1.78–2.64)	23.2(1.17)	2.52*** (2.00–3.18)	22.7(0.94)	2.62*** (2.08–3.30)	23.6(1.56)	2.87*** (2.09–3.95)	22.3(2.26)	3.59*** (1.79–7.21)	22.5(2.80)	3.61*** (1.45–8.99)
35,000–69,999	27.2(0.33)	25.9(0.79)	1.71*** (1.42–2.04)	25.4(1.25)	1.89*** (1.49–2.39)	24.7(1.08)	1.95*** (1.57–2.40)	24.8(1.61)	2.08*** (1.51–2.86)	22.8(2.56)	2.55*** (1.31–4.97)	20.7(2.97)	2.31* (0.88–6.03)
≥70,000a	31.1(0.66)	18.2(0.99)	1.0	15.7(1.03)	1.0	15.0(1.01)	1.0	13.9(1.27)	1.0	10.3(2.30)	1.0	10.3(2.67)	1.0
Marital status
Married/cohabitatinga	57.8(0.51)	46.1(1.06)	1.0	43.8(1.25)	1.0	44.0(1.38)	1.0	41.6(1.68)	1.0	36.9(3.28)	1.0	35.7(3.23)	1.0
Separated/divorced/widowed	19.7(0.32)	25.7(0.75)	1.75*** (1.57–1.96)	27.9(1.02)	1.95*** (1.68–2.27)	26.2(0.97)	1.83*** (1.57–2.12)	28.1(1.27)	2.05*** (1.67–2.50)	27.4(2.20)	2.20*** (1.46–3.33)	26.7(2.75)	2.22*** (1.32–3.71)
Never married	22.5(0.44)	28.2(1.04)	1.67*** (1.44–1.93)	28.3(1.23)	1.71*** (1.43–2.04)	29.7(1.35)	1.81*** (1.49–2.19)	30.3(1.57)	1.92*** (1.52–2.43)	35.7(2.98)	2.51*** (1.63–3.88)	37.6(3.74)	2.72*** (1.59–4.66)
Education
<High school	13.0(0.42)	16.0(0.84)	1.52*** (1.29–1.80)	16.5(1.09)	1.57*** (1.26–1.96)	18.4(1.12)	1.85*** (1.52–2.26)	17.6(1.27)	1.73*** (1.34–2.24)	23.2(2.49)	2.66*** (1.76–4.03)	22.8(2.85)	2.67*** (1.53–4.67)
High school	25.8(0.51)	32.1(0.92)	1.55*** (1.38–1.73)	32.9(1.34)	1.59*** (1.34–1.87)	32.7(1.10)	1.64*** (1.43–1.89)	33.4(1.51)	1.65*** (1.35–2.03)	35.3(2.74)	2.04*** (1.42–2.93)	36.8(3.36)	2.17*** (1.31–3.60)
≥Some collegea	61.2(0.76)	51.9(0.96)	1.0	50.6(1.33)	1.0	48.9(1.18)	1.0	48.9(1.70)	1.0	41.5(2.77)	1.0	40.4(3.54)	1.0
Urbanicity
Urban	78.7(1.54)	76.9(2.02)	0.88 (0.71–1.10)	74.3(2.46)	0.77** (0.60–0.98)	75.1(2.23)	0.80* (0.63–1.02)	74.0(2.76)	0.76* (0.56–1.03)	72.2(3.03)	0.70** (0.49–1.00)	69.4(4.09)	0.61* (0.37–1.01)
Rurala	21.3(1.54)	23.1(2.02)	1.0	25.7(2.46)	1.0	24.9(2.23)	1.0	26.0(2.76)	1.0	27.8(3.03)	1.0	30.6(4.09)	1.0
Region
Northeast	18.2(0.51)	17.8(0.99)	0.98 (0.79–1.21)	18.9(1.12)	1.03 (0.80–1.32)	17.8(1.26)	1.03 (0.79–1.35)	19.4(1.39)	1.07 (0.81–1.41)	18.3(2.92)	1.14 (0.63–2.06)	17.9(2.07)	0.97 (0.49–1.90)
Midwest	21.5(0.44)	21.3(1.22)	0.99 (0.79–1.25)	22.0(1.47)	1.02 (0.78–1.33)	21.5(1.43)	1.06 (0.81–1.39)	21.5(1.90)	1.00 (0.72–1.40)	27.3(2.88)	1.45(0.92–2.29)	24.6(3.37)	1.13 (0.61–2.09)
South	37.1(0.89)	37.8(1.83)	1.03 (0.84–1.25)	35.8(1.97)	0.96 (0.76–1.21)	38.6(2.00)	1.11 (0.89–1.38)	35.9(2.37)	0.97 (0.74–1.27)	33.9(3.01)	1.04 (0.70–1.55)	33.9(3.40)	0.90 (0.53–1.53)
Westa	23.2(0.91)	23.2(1.55)	1.0	23.4(1.68)	1.0	22.0(1.58)	1.0	23.2(1.79)	1.0	20.4(1.93)	1.0	23.5(2.61)	1.0
Sexual orientation
Only opposite sex (heterosexual)a	90.9(0.25)	84.1(0.78)	1.0	83.7(1.13)	1.0	82.7(0.99)	1.0	81.7(1.30)	1.0	76.0(2.49)	1.0	73.8(3.04)	1.0
Mostly opposite sex	3.6(0.16)	7.9(0.64)	2.84*** (2.25–3.57)	7.7(0.79)	2.53*** (1.92–3.34)	8.5(0.75)	2.93*** (2.32–3.70)	8.6(1.06)	2.83*** (2.02–3.98)	14.7(2.41)	5.10*** (3.08–8.43)	14.3(2.90)	5.01*** (2.57–9.79)
Equally males/females	1.4(0.10)	3.3(0.34)	3.21*** (2.30–4.48)	4.1(0.58)	3.61*** (2.34–5.57)	3.9(0.46)	3.67*** (2.52–5.35)	4.6(0.72)	4.07*** (2.57–6.45)	4.8(1.13)	4.30*** (2.25–8.25)	5.9(1.73)	5.37*** (2.35–12.27)
Mostly same sex	0.6(0.05)	0.8(0.15)	1.49 (0.88–2.54)	0.9(0.23)	1.65 (0.84–3.24)	0.9(0.21)	1.71* (0.92–3.18)	0.9(0.24)	1.61 (0.76–3.39)	1.2(0.27)	2.24* (0.78–6.41)	1.5(0.39)	2.93 (0.57–15.00)
Only same sex (homosexual)	2.7(0.11)	3.3(0.31)	1.39** (1.05–1.83)	3.0(0.49)	1.25 (0.80–1.95)	3.3(0.42)	1.41* (0.99–2.01)	3.5(0.63)	1.51* (0.93–2.46)	2.7(0.90)	1.21 (0.45–3.24)	3.5(1.09)	1.63 (0.58–4.60)
Sexual partners
Only opposite sexa	90.5(0.24)	85.7(0.62)	1.0	84.4(0.99)	1.0	84.8(0.74)	1.0	82.0(1.22)	1.0	73.4(2.32)	1.0	71.1(2.81)	1.0
Only same sex	2.2(0.10)	2.3(0.25)	1.12 (0.81–1.53)	2.2(0.39)	1.10 (0.68–1.78)	2.3(0.34)	1.12 (0.75–1.68)	2.7(0.55)	1.38 (0.79–2.39)	2.2(0.80)	1.24 (0.41–3.70)	1.2(0.71)	0.69 (0.14–3.43)
Both sexes	3.6(0.14)	9.7(0.49)	3.69*** (3.06–4.46)	11.6(0.87)	3.99*** (3.02–5.28)	11.0(0.68)	3.91*** (3.13–4.89)	13.6(1.03)	4.72*** (3.57–6.24)	22.5(2.15)	8.21*** (5.62–11.99)	26.5(2.67)	9.76*** (6.13–15.54)
Never had sex	2.8(0.13)	1.8(0.28)	0.64** (0.42–0.98)	1.3(0.33)	0.49** (0.26–0.95)	1.3(0.29)	0.48** (0.26–0.86)	1.0(0.36)	0.37** (0.14–1.00)	1.2(0.82)	0.51 (0.08–3.25)	-	N/A

Notes.

^*= p < .05

^**= p < .01

^***= p < .001.

^aThis group was used as the reference group for analyses.

^bAs opposed to the oldest age group (i.e., 65+ years old) as the reference group for the logistic regression analysis examining the associations between each BPD variable and age as in Grant et al. (2008), we used the youngest age group (i.e., 18 to 19 years old) as the reference group.

Figure 2. Prevalence Rates and Odds Ratios (OR) of Females and Males with BPD across the Six BPD Variables.

Notes. A) Prevalence rates of females and males with BPD for each of the six diagnostic rules. Solid bars=Males; Open=Females. B) ORs of females meeting criteria for BPD (compared to males) across the six BPD variables.

Across diagnostic rules, those with a high school education or less had greater odds of meeting criteria for BPD compared to those who had some college or higher (e.g., high school OR_BPD-A=1.55; less than high school OR_BPD-A=1.52). Both ORs and differences in prevalence increased in magnitude as the diagnostic rule became more restrictive. People who earned <$70,000 had significantly greater odds of meeting criteria for BPD across all six diagnostic rules compared to those who earned $70,000 or more. As with education, ORs and differences in prevalence increased as rules became more restrictive (e.g., $0–19,000: BPD-A=33.7%, BPD-D=46.4%; ≥$70,000: BPD-A=18.2%; BPD-D=10.3%).

As the diagnostic rule of BPD became more restrictive, significant differences were observed in the association between BPD diagnosis and marital status, sexual orientation, and sexual partners. First, across definitions, the prevalence of BPD was lower among people who were married/cohabitating (e.g., BPD-A=46.1%; BPD-D=35.7%) and higher among those who were never married (e.g., BPD-A=28.2%; BPD-D=37.6%). ORs confirmed those who were married had consistently lower odds of BPD than any other distinction. People reportedly only attracted to the opposite sex were less likely to meet criteria than those mostly attracted and equally attracted to the opposite sex. Likewise, across definitions those with sexual partners of the opposite sex only were less likely to meet criteria for BPD than those with sexual partners who were both male and female. Associations increased in magnitude as the diagnostic rule became more restrictive.

Psychiatric Disorders

Table 2 provides prevalence rates and ORs associated with each BPD variable in relation to past-year and lifetime psychiatric disorders. Findings were consistent across disorders: as diagnostic rules became more restrictive, rates of comorbidity increased (e.g., any lifetime PTSD BPD-A=27.8%; BPD-D=69.7%; see Figure 3). Of note, rates of comorbidity were significantly larger for BPD-A-18 and BPD-B-18 versus BPD-A and BPD-B (respectively; e.g., any lifetime PTSD BPD-A=27.8% vs. BPD-A-18=35.1%). All ORs were significant, and associations increased in magnitude as diagnostic rules became more restrictive. Those who endorsed symptoms of any psychiatric disorder had greater odds of BPD.

Table 2.

Prevalence Rates and Odds Ratios (OR) of Borderline Personality Disorder and Psychiatric Disorders (Uncontrolled)

Psychiatric disorders	Full sample	BPD variables
		BPD-A		BPD-B		BPD-A-18		BPD-B-18		BPD-C		BPD-D
	% (SE)	% (SE)	OR (99% CI)	% (SE)	OR (99% CI)	% (SE)	OR (99% CI)	% (SE)	OR (99% CI)	% (SE)	OR (99% CI)	% (SE)	OR (99% CI)
Any mood, past-year	12.5(0.27)	42.0(0.95)	7.64*** (6.78–8.60)	52.4(1.35)	9.67*** (8.26–11.32)	48.9(1.18)	8.87*** (7.74–10.17)	55.6(1.60)	10.33*** (8.54–12.50)	68.9(2.41)	16.45*** (11.94–22.67)	73.1(2.50)	19.69*** (13.14–29.50)
Any mood, lifetime	23.5(0.40)	60.7(0.96)	6.71*** (5.98–7.54)	72.6(1.46)	10.16*** (8.31–12.44)	67.7(1.16)	8.27*** (7.08–9.66)	74.4(1.56)	10.61*** (8.45–13.32)	82.3(2.26)	15.62*** (10.00–24.39)	83.3(2.24)	16.50*** (9.53–28.58)
Any bipolar, past-year	2.1(0.09)	10.9(0.54)	13.01*** (10.09–16.77)	15.5(0.99)	13.87*** (10.41–18.48)	14.9(0.86)	15.53*** (11.83–20.38)	18.2(1.23)	15.18*** (11.36–20.28)	27.9(2.69)	21.05*** (14.28–31.04)	32.0(3.90)	24.07*** (13.96–41.51)
Any bipolar, lifetime	2.8(0.12)	13.8(0.57)	10.97*** (8.92–13.49)	19.0(1.00)	12.01*** (9.43–15.30)	18.2(0.87)	12.89*** (10.22–16.25)	22.1(1.25)	13.34*** (10.34–17.21)	32.1(2.95)	18.25*** (12.26–27.18)	37.3(4.32)	21.82*** (12.37–38.47)
Any anxiety, past-year	13.1(0.24)	43.5(0.95)	7.65*** (6.82–8.58)	55.3(1.18)	10.37*** (9.08–11.84)	49.5(1.08)	8.48*** (7.48–9.62)	58.6(1.48)	11.14*** (9.37–13.26)	66.8(2.45)	14.10*** (10.24–19.42)	78.8(2.33)	25.46*** (16.35–39.63)
Any anxiety, lifetime	17.0(0.31)	50.6(1.08)	7.08*** (6.33–7.92)	62.5(1.24)	9.94*** (8.59–11.50)	55.9(1.07)	7.74*** (6.88–8.72)	65.8(1.51)	10.86*** (9.03–13.07)	71.6(2.56)	12.88*** (9.00–18.41)	80.0(2.25)	20.13*** (12.94–31.31)
Any PTSD, past-year	4.7(0.17)	22.9(0.79)	12.56*** (10.54–14.97)	32.9(1.22)	15.60*** (13.04–18.66)	29.2(1.03)	14.12*** (11.93–16.72)	37.1(1.55)	16.80*** (13.70–20.60)	49.7(3.27)	22.68*** (15.43–33.34)	61.1(3.31)	34.47*** (22.40–53.03)
Any PTSD, lifetime	6.1(0.21)	27.8(0.82)	11.18*** (9.63–12.98)	38.7(1.18)	14.14*** (12.03–16.62)	35.1(1.08)	13.08*** (11.16–15.32)	43.2(1.66)	15.53*** (12.64–19.07)	55.2(3.26)	20.78*** (14.11–30.62)	69.7(3.11)	37.46*** (23.90–58.72)
Alcohol use disorder, past-year	13.9(0.31)	30.8(0.95)	3.38*** (2.98–3.82)	34.2(1.34)	3.56*** (3.03–4.19)	32.9(1.22)	3.45*** (2.95–4.03)	35.9(1.74)	3.75*** (3.05–4.62)	41.9(3.08)	4.59*** (3.23–6.54)	43.5(3.14)	4.85*** (3.29–7.16)
Alcohol use disorder, lifetime	29.1(0.48)	58.5(0.93)	4.15*** (3.75–4.59)	64.1(1.32)	4.80*** (4.14–5.56)	61.0(1.27)	4.29*** (3.72–4.94)	66.5(1.64)	5.21*** (4.27–6.36)	71.6(2.43)	6.30*** (4.45–8.91)	75.0(2.49)	7.40*** (4.48–12.23)
Drug use disorder, past-year	3.9(0.13)	14.6(0.68)	6.71*** (5.48–8.21)	17.5(0.96)	6.65*** (5.34–8.28)	17.2(0.97)	6.97*** (5.65–8.61)	19.6(1.28)	7.22*** (5.67–9.21)	27.1(2.61)	9.86*** (6.81–14.26)	29.2(3.32)	10.62*** (6.60–17.11)
Drug use disorder, lifetime	9.9(0.27)	32.3(1.03)	6.34*** (5.43–7.41)	39.4(1.35)	7.26*** (6.11–8.62)	36.8(1.24)	6.80*** (5.80–7.97)	42.7(1.63)	7.89*** (6.43–9.69)	51.5(2.74)	10.17*** (7.42–13.92)	54.7(4.01)	11.28*** (7.05–18.06)
Tobacco use disorder, past-year	20.0(0.41)	43.2(1.07)	3.71*** (3.28–4.20)	48.2(1.30)	4.12*** (3.60–4.71)	48.8(1.12)	4.37*** (3.87–4.94)	51.2(1.53)	4.53*** (3.85–5.34)	61.6(2.49)	6.60*** (4.84–8.98)	62.1(3.03)	6.65*** (4.33–10.20)
Tobacco use disorder, lifetime	27.9(0.52)	54.8(1.06)	3.76*** (3.36–4.21)	58.8(1.23)	4.03*** (3.56–4.57)	58.6(1.12)	4.13*** (3.66–4.66)	61.0(1.54)	4.33*** (3.67–5.10)	69.5(2.61)	6.03*** (4.23–8.60)	69.9(2.81)	6.09*** (3.87–9.58)

Notes.

^*= p < .05

^**= p < .01

^***= p < .001.

Absence of the indicated psychiatric disorder was used as the reference group for analyses.

Figure 3. Prevalence Rates and Odds Ratios (OR) of BPD and Any Lifetime Posttraumatic Stress Disorder.

Notes. A) Prevalence rate of lifetime PTSD among people with BPD across the six diagnostic rules. B) ORs of meeting diagnostic criteria for BPD across all six diagnostic rules among people with any lifetime PTSD (compared to those without lifetime PTSD).

History of Suicide Behavior

Table 3 provides prevalence rates and ORs for associations between each BPD variable (set of diagnostic rules) and reported suicide behavior. As the diagnostic rules for BPD became more restrictive, the likelihood of SA history and recency (past-year, past 5 years) increased (e.g., ever attempted suicide BPD-A=22.7%; BPD-D=55.6%), as did the magnitude of the associations between each BPD and suicide variables. Mean reported lifetime SAs increased significantly as the diagnostic rules for BPD became more restrictive; multiple attempters were more likely to meet criteria for BPD irrespective of definition, with the likelihood increasing with each reported attempt. With respect to age, as diagnostic rules became more restrictive, prevalence of BPD decreased among those who first attempted suicide between 18–29 years old (e.g., BPD-A=33.8%; BPD-D=28.7%). Unsurprisingly, the mean number of SAs was markedly higher than the general population across all BPD definitions (whole sample=0.09; BPD-A=0.48; BPD-D=1.62). Associations between BPD-A-18 and BPD-B-18 and each suicide outcome were higher than those between BPD-A and BPD-B and each suicide outcome (respectively; e.g., ever attempted suicide BPD-A=22.7% vs. BPD-A-18=29.7%).

Table 3.

Prevalence Rates and Odds Ratios (OR) of Borderline Personality Disorder and History of Suicide Behavior (Uncontrolled)

History of suicide behavior	Full sample	BPD variables
		BPD-A		BPD-B		BPD-A-18		BPD-B-18		BPD-C		BPD-D
	% (SE)	% (SE)	OR (99% CI)	% (SE)	OR (99% CI)	% (SE)	OR (99% CI)	% (SE)	OR (99% CI)	% (SE)	OR (99% CI)	% (SE)	OR (99% CI)
Ever attempted suicide
Noa	94.8(0.18)	77.3(0.78)	1.0	68.7(1.18)	1.0	70.3(1.00)	1.0	64.2(1.46)	1.0	48.0(3.23)	1.0	44.4(4.12)	1.0
Yes	5.2(0.18)	22.7(0.78)		31.3(1.18)		29.7(1.00)		35.8(1.46)		52.0(3.23)		55.6(4.12)
			9.81*** (8.50–11.31)		11.94*** (10.08–14.16)		12.26*** (10.56–14.22)		13.48*** (11.11–16.35)		22.16*** (15.41–31.88)		24.29*** (14.54–40.59)
Age at first attempt
5–17a	44.5(1.33)	46.5(1.96)	1.0	46.7(2.42)	1.0	45.9(2.19)	1.0	47.0(2.74)	1.0	55.6(3.49)	1.0	54.6(3.30)	1.0
18–29	35.5(1.49)	33.8(2.11)	0.83 (0.58–1.19)	32.7(2.21)	0.83 (0.59–1.15)	33.6(2.19)	0.87 (0.63–1.21)	30.5(2.37)	0.76 (0.52–1.11)	28.6(3.33)	0.61* (0.37–1.02)	28.7(3.72)	0.64 (0.31–1.35)
30–44	14.1(1.00)	14.7(1.46)	1.00 (0.68–1.47)	14.3(1.71)	0.96 (0.61–1.49)	14.7(1.40)	1.02 (0.72–1.46)	14.8(1.78)	0.99 (0.62–1.57)	11.1(2.69)	0.59 (0.27–1.31)	11.9(2.94)	0.67 (0.26–1.70)
45+b	5.9(0.62)	5.0(0.87)	0.66 (0.33–1.32)	6.3(1.27)	1.04 (0.51–2.10)	5.8(1.05)	0.93 (0.47–1.83)	7.7(1.56)	1.36 (0.66–2.78)	4.8(1.77)	0.61 (0.21–1.80)	4.8(2.32)	0.65 (0.16–2.71)
Age at most recent attempt
5–17a	30.3(1.34)	26.6(1.84)	1.0	24.2(2.13)	1.0	25.7(1.98)	1.0	24.2(2.50)	1.0	28.7(3.85)	1.0	27.5(3.91)	1.0
18–29	38.7(1.51)	38.0(2.20)	1.23 (0.83–1.84)	36.6(2.54)	1.27 (0.89–1.81)	37.2(2.32)	1.22 (0.85–1.74)	33.1(2.80)	1.09 (0.71–1.69)	33.0(3.68)	0.89 (0.50–1.60)	35.3(4.99)	1.01 (0.44–2.32)
30–44	20.9(1.24)	24.9(1.76)	1.87*** (1.23–2.84)	26.2(2.07)	1.97*** (1.29–3.00)	25.7(1.88)	1.87*** (1.28–2.74)	28.0(2.25)	2.04*** (1.30–3.19)	26.5(2.84)	1.39 (0.71–2.72)	29.3(2.91)	1.59 (0.66–3.85)
45+b	10.0(0.69)	10.4(1.04)	1.40 (0.81–2.41)	13.1(1.50)	2.11*** (1.22–3.66)	11.4(1.29)	1.63* (0.95–2.81)	14.7(1.81)	2.37*** (1.34–4.21)	11.7(3.10)	1.28 (0.46–3.53)	8.0(2.38)	0.88 (0.28–2.79)
Attempted suicide, past-year (n=152)
Noa	92.7(0.71)	90.1(1.04)	1.0	89.1(1.45)	1.0	89.6(1.23)	1.0	88.8(1.60)	1.0	85.4(2.40)	1.0	85.2(2.58)	1.0
Yes	7.3(0.71)	9.9(1.04)	2.18*** (1.21–3.91)	10.9(1.45)	2.07*** (1.22–3.51)	10.4(1.23)	2.08*** (1.22–3.55)	11.2(1.60)	1.99** (1.16–3.43)	14.6(2.40)	2.50*** (1.29–4.83)	14.8(2.58)	2.37* (0.98–5.72)
Attempted suicide, past five-years (n=509)
Noa	72.5(1.25)	64.9(1.72)	1.0	61.1(2.25)	1.0	62.0(1.95)	1.0	57.4(2.33)	1.0	56.9(4.59)	1.0	50.4(4.27)	1.0
Yes	27.5(1.25)	35.1(1.72)	2.18*** (1.55–3.09)	38.9(2.25)	2.26*** (1.61–3.17)	38.0(1.95)	2.36*** (1.68–3.30)	42.6(2.33)	2.59*** (1.89–3.56)	43.1(4.59)	2.20*** (1.27–3.81)	49.6(4.27)	2.79*** (1.51–5.16)
	M (SE)	M (SE)	OR (99% CI)	M (SE)	OR (99% CI)	M (SE)	OR (99% CI)	M (SE)	OR (99% CI)	M (SE)	OR (99% CI)	M (SE)	OR (99% CI)
Number of lifetime suicide attempts	0.09(0.00)	0.48(0.03)	2.90*** (2.44–3.44)	0.73(0.05)	2.46*** (2.07–2.91)	0.66(0.04)	2.74*** (2.29–3.27)	0.88(0.06)	2.31*** (1.96–2.73)	1.45(0.15)	1.72*** (1.31–2.26)	1.62(0.21)	1.48*** (1.22–1.79)

Notes.

^*= p < .05

^**= p < .01

^***= p < .001.

There was no reference group for the number of lifetime suicide attempts.

^aThis group was used as the reference group for analyses.

^bDue to missing data, we combined the “45–64” and “65+” age-groups.

Mean reported lifetime SAs increased significantly as the diagnostic rules for BPD became more restrictive; multiple attempters were more likely to meet criteria for BPD irrespective of definition, with the likelihood increasing with each reported attempt. Associations between age at attempt and BPD diagnosis varied by BPD definition. As with comorbidity, associations between suicide outcomes and BPD-A-18 and BPD-B-18 were more robust than with BPD-A and BPD-B.

History of Treatment-Seeking Behavior and Receiving Disability Benefits

Table 4 provides prevalence rates and ORs for associations between the BPD variables, treatment-seeking behavior, and receiving disability benefits. Those who endorsed receiving treatment and those who reported receiving disability had consistently greater odds of BPD. As the diagnostic rule of BPD became more restrictive, the likelihood of both seeking treatment and receiving disability benefits increased (e.g., received professional services for a psychiatric problem BPD-A=66.8%; BPD-D=89.5%). Additionally, the associations between BPD-A-18 and BPD-B-18, each treatment-seeking outcome, and receiving disability benefits were generally more robust than observed associations with BPD-A and BPD-B.

Table 4.

Prevalence Rates and Odds Ratios (OR) of Borderline Personality Disorder and History of Treatment-seeking Behavior and Receiving Disability Benefits (Uncontrolled)

History of treatment-seeking behavior and receiving disability benefits	Full sample	BPD variables
		BPD-A		BPD-B		BPD-A-18		BPD-B-18		BPD-C		BPD-D
	% (SE)	% (SE)	OR (99% CI)	% (SE)	OR (99% CI)	% (SE)	OR (99% CI)	% (SE)	OR (99% CI)	% (SE)	OR (99% CI)	% (SE)	OR (99% CI)
Received professional services for a psychiatric problem	28.2(0.45)	66.8(0.94)	6.66*** (5.92–7.48)	76.6(1.15)	9.60*** (8.11–11.36)	72.1(1.12)	7.79*** (6.66–9.10)	79.2(1.28)	10.70*** (8.64–13.27)	87.7(1.58)	18.68*** (11.67–29.92)	89.5(2.02)	21.93*** (9.60–50.12)
Sought self-help services for a psychiatric problem	14.3(0.30)	38.4(0.94)	4.95*** (4.42–5.55)	46.7(1.28)	6.16*** (5.40–7.03)	42.8(1.28)	5.41*** (4.68–6.24)	48.6(1.55)	6.36*** (5.39–7.50)	58.2(3.39)	8.70*** (5.89–12.85)	62.0(3.42)	9.98*** (6.31–15.78)
Stayed overnight in a hospital	8.3(0.21)	30.4(0.94)	7.55*** (6.48–8.80)	40.2(1.26)	9.67*** (8.25–11.34)	36.5(1.13)	8.69*** (7.43–10.16)	44.0(1.39)	10.60*** (8.90–12.62)	57.4(2.84)	15.96*** (11.40–22.33)	62.8(3.13)	19.40*** (12.50–30.09)
Went to the emergency department for a psychiatric problem	7.4(0.21)	27.7(0.85)	7.65*** (6.63–8.83)	36.2(1.27)	9.30*** (7.93–10.90)	33.7(1.11)	8.88*** (7.63–10.34)	40.0(1.44)	10.23*** (8.60–12.17)	53.2(2.93)	15.35*** (11.01–21.41)	57.0(3.93)	17.29*** (10.68–27.99)
Prescribed medication for a psychiatric problem	21.8(0.39)	55.0(0.99)	5.75*** (5.16–6.42)	65.8(1.29)	8.04*** (6.84–9.47)	61.0(1.27)	6.72*** (5.75–7.86)	68.2(1.67)	8.59*** (6.88–10.72)	78.3(2.53)	13.37*** (8.65–20.68)	84.0(2.56)	19.12*** (10.07–36.33)
Received treatment services, past-year	16.6(0.34)	45.4(0.95)	5.66*** (5.02–6.38)	54.6(1.42)	7.16*** (6.07–8.44)	51.4(1.13)	6.53*** (5.69–7.50)	58.1(1.81)	7.90*** (6.36–9.81)	69.8(2.81)	12.14*** (8.30–17.76)	78.3(3.28)	18.64*** (10.11–34.34)
Received treatment services, lifetime	32.3(0.47)	71.2(0.93)	6.61*** (5.86–7.47)	81.5(1.09)	10.58*** (8.73–12.82)	76.8(1.12)	8.12*** (6.80–9.69)	83.8(1.08)	11.91*** (9.51–14.91)	90.4(1.60)	20.26*** (11.23–36.53)	91.2(2.17)	22.02*** (8.22–59.04)
Received disability benefits, past-year	35.9(0.66)	46.0(1.19)	1.61*** (1.46–1.78)	50.3(1.54)	1.87*** (1.62–2.17)	49.8(1.35)	1.85*** (1.64–2.09)	52.4(1.76)	2.02*** (1.70–2.41)	62.0(3.23)	2.95*** (2.02–4.30)	65.5(3.92)	3.41*** (1.99–5.86)

Notes.

^*= p < .05

^**= p < .01

^***= p < .001.

Those who did not endorse treatment or disability were used as the reference group for analyses.

Controlled Analyses

Supplemental Tables 3–5 provide ORs with sociodemographic characteristics and lifetime psychiatric disorders controlled for in parallel to analyses by Grant and colleagues (2008). With few exceptions (e.g., lifetime tobacco use for BPD-D), associations with demographics, psychiatric comorbidities, treatment, and disability remained significant. However, fewer ORs were significant for the variables examining history of suicide behavior.

DISCUSSION

This study examined how differences in diagnostic rules used to assess BPD impacted prevalence and clinical presentation in a nationally representative sample. Across the six sets of diagnostic rules used, BPD prevalence ranged widely: 0.5%–11.4%. Whereas the associations between BPD and clinical variables—other psychiatric disorders, treatment-seeking, and disability—generally remained stable across rules, associations became more robust as rules became more restrictive. Given the striking effect of differences in diagnostic rules on prevalence and clinical presentation, it is imperative for the field to critically consider what variables affect characterization of BPD and how to most effectively characterize the disorder to increase generalizability and clinical relevance of findings moving forward.

All adjustments to the diagnostic rules (e.g., at least one vs. all endorsed BPD criteria associated with impairment; one vs. all items required to be endorsed for a criterion) used to define BPD meaningfully altered prevalence and associations of BPD with related problems. For example, using BPD-A, the sex distribution mirrored that of the general population (57.5% women/42.5% men). With the similar but more restrictive BPD-C, the rate of BPD was significantly higher among women than men (60.3%/39.7%). Historically, BPD had been thought of as a “gendered” disorder; in fact, the DSM-5 indicates that BPD is predominantly diagnosed in females (APA, 2013, p. 666). However, large-scale studies and epidemiological research consistently observe equal rates of BPD across sex (Grant et al., 2008; Tomko et al., 2014). Of note, assessments like the McLean Screening Instrument for BPD (Zanarini et al., 2003) used frequently in clinical settings more closely resemble the more restrictive definitions (e.g., fewer items to assess each criterion for BPD; requiring all items for a given criterion be endorsed). Findings from clinical settings using such measures may tend to report a higher female/male ratio, thus diagnosing more females with BPD. Additionally, as more BPD criteria were required to be associated with impairment, the rates of comorbidity with other psychiatric problems generally increased. These findings indicate that requiring impairment is meaningful. However, though it may seem that more impairment leads to greater severity of mental health problems, many assessments do not directly assess for whether each item assessed is associated with impairment (e.g., PTSD Checklist for DSM-5, Blevins et al., 2015; Beck Depression Inventory-II, Beck et al., 1996). In addition to disagreements of requiring one or all endorsed items to be associated with impairment, assessing for impairment may raise the issue of participant insight. Overall, our results stress the importance of critically considering diagnostic rules when examining BPD to identify the true prevalence of the disorder and scope of impact and need from a public health perspective.

Relatedly, using all 30 items intended to assess BPD in the AUDADIS-5 versus only the 18 items that overlapped with the AUDADIS-IV altered (reduced) BPD prevalence. Otherwise stated, the prevalence of BPD differed between two BPD variables with identical diagnostic rules except for the number of items used (e.g., BPD-A and BPD-A-18). Some may argue that using a shorter assessment of BPD is preferable (e.g., takes less time; is less burdensome). However, results suggest assessments with fewer items may also tend to bias those diagnosed with the disorder. To that point, though BPD prevalence was lower using 18 versus 30 items, rates of comorbidity and treatment-seeking were higher, suggesting the nature of the subsample assigned diagnosis differed in clinically meaningful respects. Meaningfully, compared to the 30-item assessment, where each BPD criterion was assessed with at least two items in the AUDADIS-5, four of the nine BPD criteria were assessed with only one item in the 18-item BPD assessment in the AUDADIS-IV (Supplemental Table 1). A participant not endorsing any of these four items with the 18-item BPD assessment could endorse comparable items with the 30-item assessment, arguably equally representative of the underlying constructs, and be diagnosed with BPD. To illustrate, regarding unstable and intense interpersonal relationships, although “Have your relationships with people you really care about had lots of extreme ups and downs?” (in both the AUDADIS-IV and AUDADIS-5) may not be endorsed, either “Have you usually gotten attached to people very quickly?” or “Have you often started out thinking that someone was a great person only to be disappointed when they didn’t live up to your expectations?” (only in the AUDADIS-5) could be endorsed. A BPD assessment where only one item assesses any number of criteria may not capture BPD as accurately as an assessment where at least two items assess each criterion; such an assessment is less comprehensive and likely to produce more false negatives.

In sum, the field researching BPD would benefit from greater consistency in defining the disorder to derive a more accurate understanding of how many people are affected and how. Moreover, results from this study suggest that the prevalence of BPD in the United States may be increasing. Using the same diagnostic rule as Grant et al. (2008), who observed a 5.9% prevalence with data from 2004 to 2005, we observed an 11.4% prevalence (6.9% with the 18 items [BPD-A-18]) in 2012–2013. Similarly, using more restrictive diagnostic rules, Trull et al. (2010) observed a 2.7% prevalence in 2004–2005, whereas we observed a 5.4% prevalence (3.8% with the 18 items [BPD-B-18]). Increases in BPD prevalence over the 8 years between studies suggest that the U.S. population is changing in meaningful ways. For example, a study examining trends in SAs between the Wave 2 NESARC and the NESARC-III found those with a recent SA in 2012–2013 were more likely to be younger and have a depressive disorder, antisocial personality disorder, and a history of violent behavior compared to those reporting recent SA in 2004–2005 (Olfson et al., 2017). Economic inequality has evolved, such as with the 2007–2009 financial crisis, a period of serious economic difficulties and unemployment for countless in the United States (Schoen, 2017). Economic hardship negatively impacts parents, and increases risk for mental health problems among children (Solantaus et al., 2004). Additionally, although social media use was not at its zenith during the period between the Wave 2 NESARC and NESARC-III data collection, there was an upward trend of use (Perrin, 2015). Among a nationally representative sample of U.S. young adults assessed in 2014, up to 44% of participants reported problematic social media use, which was positively associated with depressive symptoms (Shensa et al., 2017). Additionally, negative interactions and social comparison on social network sites (SNSs), as well as “SNS addiction” and problematic SNS use, were associated with higher levels of depression and anxiety (Seabrook et al., 2016). On another note, with the implementation of the Mental Health Parity and Addiction Equity Act in 2008, outpatient services for mental health and substance use problems are accessed and utilized more than prior to its enactment. Additionally, rates of prolonged length of stay among children who present to the emergency department for mental health needs has increased over time (Nash et al., 2021). Furthermore, perhaps more public and open discussion of mental health problems and continued destigmatization of these issues encourage people to disclose experiencing psychiatric problems. As the U.S. population continues to change, particularly following significant stressful events (e.g., the COVID-19 pandemic, increasing political fragmentation), it is crucial that the field assesses BPD accurately and standardly to understand its true prevalence and associations with other clinically relevant problems.

Limitations and Future Directions

This study has limitations. First, a formal diagnostic instrument was not utilized to assess the presence or absence of BPD. However, the 30 items of the AUDADIS-5 were based on the DSM-5 criteria, and the basic specification that at least five criteria must be endorsed parallels DSM procedures. Second, although the AUDADIS-5 was administered by a trained interviewer, completion of the interview required a level of self-awareness and self-reflection by the participant of their general behavior. A potential future study could compare NESARC-III-derived BPD and clinical observations by a trained clinician to confirm BPD diagnosis. Third, despite the 30-item assessment of BPD indicating fair test-retest reliability outcomes (see Grant et al., 2015), a future study could examine the factor structure of the BPD assessment using the 30 items, as well as 18 items, assessing BPD in the AUDADIS-5.

Conclusions

Results indicate the importance of critically considering methodology when examining BPD, as prevalence estimates and associations of BPD to other variables change with even slight adjustments to diagnostic rules. The field should optimally determine and use a single, consistent set of diagnostic rules for BPD to maximize the generalizability of findings, including whether BPD would be more effectively captured dimensionally instead of as a taxon (e.g., Rothschild et al., 2003; Trull et al., 1990). Despite the seriousness of this disorder, including the scope of its manifest effects concerning both personal and societal costs, BPD remains largely understudied and arguably underdiagnosed. Research focused on BPD accounts for <1% of the National Institutes of Health annual budget, which is one-tenth of the funds allocated for bipolar disorder, a psychiatric disorder believed to have a similar prevalence with BPD (Gunderson et al., 2018; Zimmerman, 2015). Consequently, the understanding of BPD and the advancement of treatments for the disorder have been limited. In a 5-year follow-up review on the pharmacology of BPD, Stoffers-Winterling et al. (2021) noted that little additional evidence had accumulated concerning drug treatment effects in BPD since their 2015 study. Of further concern, a recent study examining available resources across 22 countries found significant shortages in providers available, certified, and willing to provide treatment for BPD relative to the number of patients seeking treatment (Iliakis et al., 2019). BPD is a substantial source of public health concern—a disorder prevalent enough to warrant the allocation of resources to understand its nature and enhance treatment development. Careful assessment across settings, particularly when determining prevalence, is essential to highlight the nature and extent of the need for resource allocation to this underserved and understudied population.