BRIDGING THE GAP BETWEEN REMISSION AND RECOVERY IN BPD: QUALITATIVE VERSUS QUANTITATIVE PERSPECTIVES

Abstract

The 10-year outcome for patients with borderline personality disorder (BPD) is diagnostic remission in 85% to 93%; however, less than half achieve good social and vocational functioning, and few attain full psychosocial recovery. To assess the gap between diagnostic remission and psychosocial recovery, quantitative measures of outcome were compared with narrative reports of psychosocial functioning in 150 BPD subjects followed prospectively from 2 to 31 years (mean 9.94 years). Subjects with the best and the worst outcomes were compared on symptom changes over time, and on efforts to improve psychosocial functioning. At intake, poor outcome subjects were more impaired than those with good outcomes, with more borderline psychopathology, hospitalizations, and poverty. At follow-up, 53.8% of good outcome subjects complained of continuing problems with depression, 33.3% with anger and impulse control, and 25.6% with unstable relationships. Despite objective measures of improvement, narrative reports documented residual BPD symptoms, comorbidity, and unemployment interfering with psychosocial recovery.

The long-term outcome of borderline personality disorder (BPD) is for symptomatic and diagnostic remission. Two landmark studies, the Collaborative Longitudinal Personality Disorders Study (CLPS) and the McLean Study of Adult Development (MSAD) reported diagnostic remission in 85% to 93% of BPD subjects over a 10-year follow-up period (Gunderson et al., 2011; Zanarini, Frankenburg, Reich, & Fitzmaurice, 2010a). A recent prospective study of suicidal behavior in BPD found diagnostic remission in 69% of subjects, followed for 10 or more years. (Soloff & Chiappetta, 2018a). Despite diagnostic and symptomatic improvement, all three studies reported poor psychosocial outcomes for a large number of subjects. The MSAD study found that only half of BPD subjects met criteria for psychosocial recovery at 10-year follow-up, (where recovery was defined by remission of BPD, a good interpersonal relationship, full-time employment, and a Global Assessment of Functioning (GAF) score ≥ 61) (Zanarini, Frankenburg, Reich, & Fitzmaurice, 2010b). Failure to achieve full-time employment, not relationship failure, was the primary cause of poor psychosocial outcome. In the CLPS, the mean GAF score among BPD subjects at 10-year follow-up was 57, with 64% unemployed. Rates of employment among BPD subjects were significantly less than among clinical comparison groups (Gunderson et al., 2011). Among BPD subjects completing 10 years of prospective study for suicidal behavior, Soloff and Chiappetta (2018a) found a mean GAF score of 61.4, with 44% scoring less than 61. At 10-year follow-up, the unemployment rate was 43.2%, with 41.5% living below the 2015 federal poverty level for household incomes (www.hhs.gov), and 32.5% dependent on government assistance.

The path to psychosocial recovery is not smooth for most subjects. Diagnostic and symptomatic relapses were frequent in the MSAD study. Among subjects who had attained a stable diagnostic remission for 2 years, 30% experienced a recurrence by 10-year follow-up. Loss of good psychosocial function was even more striking. Among subjects who had good psychosocial function at intake, 87% had lost their baseline good functioning by 10-year follow-up (Zanarini, Frankenburg, Reich, & Fitzmaurice, 2010a, 2010b, 2012).

The great disparity between high rates of diagnostic remission and poor psychosocial outcomes is unexplained. Early reports from the CLPS found that improvement in BPD psychopathology predicted functional improvement in the first 2 years of follow-up. BPD psychopathology improved more than functional impairment, especially for measures of employment, recreation, and global functioning (e.g., GAF) (Skodol et al., 2005). The relationship between diagnostic remission from BPD and good psychosocial functioning is less apparent in longer-term studies. Among BPD subjects followed prospectively from 2 to 31 years (mean 9.9 years.), Soloff and Chiappetta (2018b) found that diagnostic remission from BPD was neither necessary nor sufficient to achieve good interpersonal relationships or full-time employment, two pillars of psychosocial recovery. Using baseline and time-varying follow-up data as predictors, they reported that full-time employment as well as social and vocational adjustment at follow-up predicted good psychosocial outcome, but not remission of BPD. Psychiatric comorbidity with major depressive disorder (MDD), substance use disorder (SUD), and anxiety disorders decreased the likelihood of good psychosocial outcomes.

This study differs from our previous work in that it seeks to identify factors responsible for the disparity between high rates of diagnostic remission and low rates of psychosocial recovery in BPD subjects followed prospectively in long-term studies. We hypothesized that the standardized quantitative reports of diagnostic remissions did not fully capture the clinical reality of BPD, where patients often require long-term treatment despite apparent diagnostic remission. To address this discrepancy, we contrasted standardized quantitative measures of psychosocial outcome with subjective clinical narratives, comparing subjects with the best and the worst psychosocial outcomes. Narrative reporting is an unconventional method in the current literature, but one that is most relevant to the realities of clinical practice.

METHOD

This study was approved by the Institutional Review Board of the University of Pittsburgh, and funded by the National Institute of Mental Health (NIMH; RO1 MH 048463). Subjects were recruited from inpatient, outpatient, and community sources to participate in a longitudinal study of suicidal behavior in BPD. Informed consent was obtained from all participants. The longitudinal sample included 150 subjects who were followed prospectively for 2 to 31 years (mean [SD]: 9.9 [6.3] years) and interviewed biannually using standardized, multidimensional assessment measures. Given the broad range of follow-ups (2 to 31 years), the actual number of assessments varied greatly between subjects. Subjects were screened by master’s-level-prepared interviewers using the International Personality Disorders Examination (IPDE; Loranger, 1999). A lifetime diagnosis of probable or definite BPD was required to continue. Inclusion in the study required a current diagnosis of BPD on the Diagnostic Interview for Borderline Patients (DIB scaled score ≥ 7) (Gunderson, Kolb, & Austin, 1981) or, after 2001, the Diagnostic Interview for Borderlines–Revised (DIB-R total score ≥ 8) (Zanarini, Gunderson, Frankenburg & Chauncey, 1989). Both DIB and DIB-R have a 2-year time frame and were scored concurrently at each follow-up. Results of both interviews are reported separately. (Though the DIB-R was derived from the DIB, it included substantial revisions in content and structure, including a different number of scored summary statements (DIB-R 22 vs. DIB 29) and section scores (DIB-R 4 vs. DIB 5). The DIB section on social adaptation was eliminated entirely in the DIB-R. Additionally, the wording and content of many summary statements were changed in the DIB-R). DIB/DIB-R interviews were conducted by the principal investigator (PI), independent of all other assessments at both intake and follow-up. Axis I disorders were determined using the Structured Clinical Interview for DSM IV-TR (SCID) (First, Spitzer, Gibbon & Williams, 2005). Intake diagnoses were confirmed by a consensus of raters, using a best estimate process and all available data. All subjects were newly reassessed for this analysis. Diagnostic remission was determined by comparison with the subject’s initial DIB/DIB-R intake score (e.g., DIB to DIB, DIB-R to DIB-R), with remission defined as a DIB scaled score < 7; DIB-R total score < 8).

MEASURES

The core assessment battery for this longitudinal study was adapted from the Mental Health Clinical Research Center (MHCRC) for the Study of Suicidal Behavior (J. J. Mann, MD, PI), and has been presented elsewhere (Soloff, Lynch, Kelly, Malone, & Mann, 2000). Standardized assessments included: (1) the MHCRC demographic history; (2) Axis I and II diagnoses (using SCID and IPDE interviews, respectively); (3) clinical state scales (Beck Depression Inventory [BDI; Beck, Ward, Mendelson, Mock, & Erbaugh, 1961]. Hamilton Rating Scale for Depression–24-item format [HamD-24; Guy, 1976], Beck Hopelessness Scale [Beck & Steer, 1988], and Global Assessment Scale [GAS; Endicott, Spitzer, Fleiss, & Cohen, 1976]); (4) suicidal behavior (MHCRC Suicide History and Lethality Rating Scale [Oquendo, Halberstam, & Mann, 2003], Suicide Intent Scale, Scale for Suicidal Ideation [Beck, Schuyler, & Herman, 1974; Beck, Beck, & Kovacs, 1975], and Reasons for Living Scale [RFL; Linehan, Goodstein, Nielson, & Chiles, 1983]); (5) personality traits (Barratt Impulsiveness Scale, version 11 [BIS-11; Barratt, 1965; Barratt & Stanford, 1995], Buss-Durkee Hostility Inventory [BDHI; Buss & Durkee, 1957], Life History of Aggression, adult score [LHA; Brown & Goodwin, 1986], MMPI– Psychopathic Deviate subscale [MMPI–Pd; Hathaway & Meehl, 1951], and NEO Five-Factor Inventory-3 [Costa & McCrae, 2010]); (6) MHCRC family history and history of childhood abuse (Soloff, Lynch, & Kelly, 2002); (7) self-rated social adjustment (Social Adjustment Scale–Self-Report [SAS-SR; Weissman & Bothwell, 1976]); and (8) MHCRC hospitalization and treatment history. All Axis I diagnoses were reassessed at each follow-up using an abbreviated interview for DSM criteria. Using current baseline data from this sample, nine of the most prevalent SCID diagnoses are reported in this article: MDD, dysthymia, alcohol abuse/dependence (ALC), substance use disorder (SUD), panic disorder, social phobia, specific phobia, post-traumatic stress disorder (PTSD), and generalized anxiety disorder (GAD). For some analyses, all five anxiety disorders were pooled in one variable, termed “Any Anxiety Disorder.” Eating disorders (ED) were not a significant source of comorbidity in this sample.

EXCLUSION CRITERIA

Subjects were excluded for a lifetime (past or current) Axis I diagnosis of any psychotic disorder including: schizophrenia, delusional (paranoid) disorder, psychotic depression, or schizoaffective disorder. Subjects with any bipolar diagnosis were excluded, as were any subjects with clinical evidence of central nervous system pathology of any etiology (including seizure disorders, acquired brain injury, or developmental deficits), medical disorders or treatments with psychiatric consequences (e.g., hypothyroidism, steroid medication), or borderline intellectual functioning.

GOOD VERSUS POOR PSYCHOSOCIAL OUTCOMES

A frequency distribution of final follow-up GAS scores clearly defined upper and lower peaks, roughly corresponding to the top and bottom 20% of the distribution. Poor psychosocial outcomes were defined by a final follow-up GAS score less than or equal to 50, good outcomes by a final follow-up GAS score greater than or equal to 70, resulting in 31 poor and 39 good outcome subjects. Final follow-up GAS scores referred to current functioning. Subjects with good outcomes were compared to those with poor outcomes on baseline intake scores and final follow-up variables using t-tests and chi square analyses. The number of interval assessments or assessment intervals was not relevant for this analysis. Groups were compared on prevalence of BPD psychopathology and diagnostic remission using DIB/DIB-R scores, on Axis I comorbidity, personality traits, social and vocational attainment, economic status, interim suicide attempts, and treatment histories. Bonferroni corrections were calculated by dividing the a-value (.05) by the number of comparisons for each method of analysis (i.e., p < .005 for DIB/DIB-R changes within groups in Table 1, and p < .003 for differences between groups in Table 2).

TABLE 1.

Changes Within Groups From Baseline (BL) to Follow-Up (FU)

DIB/DIB-R scores	Good Outcome Subjects			Poor Outcome Subjects
	BL	FU	t, df, p	BL	FU	t, df, p
DIB Total Section Scorea	28.6 (4.5)	17.5 (7.6)	t = 7.3, df = 37, p ≤ .001	29.2 (4.3)	25.1 (8.2)	t = 2.71, df = 28, p = .01
Social Adaptation	5.3 (1.4)	4.6 (1.5)	t = 2.2, df = 37, p = .04	4.8 (1.7)	3.8 (1.7)	t = 3.00, df = 28, p = .006
Impulse Action	7.1 (2.2)	3.1 (2.5)	t = 7.4, df = 37, p ≤ . 001	8.0 (1.9)	6.2 (3.2)	t = 2.56, df = 28, p = .016
Affects	6.1 (1.4)	4.1 (3.0)	t = 4.4, df = 37, p ≤ .001	6.4 (1.4)	5.8 (1.6)	t = 1.62, df = 28, p = .12
Psychosis	3.2 (2.2)	1.6 (2.0)	t = 3.1, df = 37, p = .004	3.4 (2.3)	3.3 (2.1)	t = .068, df = 28, p = .95
Interpersonal Relations	7.0 (2.1)	4.1 (3.0)	t = 5.1, df = 37, p ≤ .001	6.6 (1.9)	6.0 (3.3)	t = 1.04, df = 28, p = .31
DIB-R Total Section Scoreb	28.8 (4.7)	14.6 (5.7)	t = 8.6, df = 27, p ≤ .001	31.5 (3.6)	20.0 (5.3)	t = 9.05, df = 21, p ≤ .001
Affect	8.5 (2.0)	6.0 (3.3)	t = 3.0, df = 27, p = .006	9.7 (.6)	8.9 (1.3)	t = 3.15, df = 21, p = .005
Cognition	3.0 (1.9)	2.1 (1.9)	t = 1.4, df = 27, p = .16	3.6 (1.7)	3.1 (2.0)	t = 1.00, df = 21, p = .33
Impulse Action	7.1 (2.1)	3.2 (2.5)	t = 5.8, df = 27, p ≤ .001	7.7 (1.6)	6.1 (2.8)	t = 2.21, df = 21, p = .04
Interpersonal Relations	10.1 (2.7)	6.4 (3.8)	t = 4.4, df = 27, p ≤ .001	10.5 (2.0)	9.6 (3.5)	t = 1.40, df = 21, p = .18

Note.

^aDIB (Diagnostic Interview for Borderline Patients), n = BL 38, FU 29;

^bDIB-R (Diagnostic Interview for Borderlines– Revised), sum of section scores, n = BL 26, FU 22. Bonferroni correction p < .005.

TABLE 2.

Good Versus Poor Psychosocial Outcomes at Follow-Up (FU)

	Good Outcome	Poor Outcome	statistic, df, p
A. BPD Diagnostic Criteria: DIB/DIB-R	n = 39	n = 31
DIB Section Score Total	17.6 (7.5)	24.8 (8.5)	t = 3.77, df = 68, p ≤ .001
Social Adaptation	4.6 (1.4)	3.8 (1.8)	t = 1.90, df = 68, p = .06
Impulse Action Patterns	3.1 (2.4)	6.1 (3.1)	t = 4.52, df = 68, p ≤ .001
Affects	4.1 (2.6)	5.8 (1.6)	t = 3.36, df = 64.3, p = .002
Psychosis	1.6 (1.9)	3.4 (2.4)	t = 3.43, df = 68, p = .001
Interpersonal Relations	4.2 (3.0)	5.7 (3.4)	t = 2.08, df = 68, p = .04
DIB-R Total Section Scorea	15.7 (9.4)	26.2 (8.4)	t = 4.87, df = 68, p ≤ .001
Affect	5.3 (3.4)	8.6 (1.7)	t = 5.20, df = 58.8, p ≤ .001
Cognition	1.7 (1.8)	3.3 (2.2)	t = 3.39, df = 68, p = .001
Impulse Action Patterns	2.8 (2.4)	5.8 (2.8)	t = 4.89, df = 68, p ≤ .001
Interpersonal Relations	5.9 (3.8)	8.5 (4.3)	t = 2.67, df = 68, p = .01
Remitted (% yes)	71.8	35.5	χ2 = 9.23, df = 1, p = .002
B. Axis I Comorbidity (% yes)
MDD	12.8	54.8	χ2 = 14.15, df = 1, p < .001
Alcohol Abuse/Dependency	10.3	25.8	χ2 = 2.94, df = 1, p = .09
Substance Use	2.5	16.1	χ2 = 2.51, df = 1, p = .11
Panic Disorder	15.4	51.6	χ2 = 10.52, df = 1, p = .001
Social Phobia	0	24.0	χ2 = 5.97, df = 1, p = .006
Specific Phobia	10.3	19.4	χ2 = 0.54, df = 1, p = .46
PTSD	2.6	25.8	χ2 = 6.38, df = 1, p = .008
General Anxiety Disorder	17.9	41.9	χ2 = 4.87, df = 1, p = .03
Any Anxiety Disorder	28.2	80.6	χ2 = 19.01, df = 1, p ≤ .001
C. Socioeconomic Variables
% Gov’t/non-self support	35.9	74.2	χ2 = 8.50, df = 1, p = .004
Household Income (% < $20K)	13.9	46.4	χ2 = 8.25, df = 1, p = .004
Employed (% yes)	71.8	35.5	χ2 = 9.23, df = 1, p ≤ .002
SAS-SR total score	1.93 (0.40)	2.60 (0.66)	t = 5.00, df = 46.7, p ≤ .001
Significant Relations (% yes)	79.5	51.6	χ2 = 6.08, df = 1, p = .014
Total Recovery (% yes)b	46.2	0	χ2 = 16.90, df = 1, p ≤ .001

Note.

^aSum of section scores,

^bRemitted + FT Employed + Significant Relationship + GAS ≥ 61. No FU dysthymia diagnoses in either group. DIB: Diagnostic Interview for Borderline Patients, DIB-R = Diagnostic Interview for Borderlines–Revised; MDD: major depressive disorder; PTSD: posttraumatic stress disorder; SAS-SR: Social Adjustment Scale–Self-Report. Bonferroni correction p < .003.

SUBJECTIVE REPORTS

At the end of each DIB/DIB-R interview, subjects were asked to contrast their current life circumstances with those at the beginning of the study (e.g., “What got better, What got worse, What remained unchanged, What is a problem today?”). They were then asked to discuss what efforts helped or did not help effect this outcome (“What did you try that helped, “What did not help?”). Questions were open-ended without further elaboration. Process notes taken during these interviews constitute the clinical narrative data for this report. This data was not qualitatively coded into pre-defined categories at the time of assessment. For purposes of analysis and presentation, the clinical narrative data were grouped by thematic content after the study was completed. Patient descriptions of persisting symptoms were grouped in clinical terms related to borderline psychopathology (e.g., impulsivity, unstable interpersonal relations). Patient reports of interventions were grouped into categories: self-help measures, professional interventions, and medication experience, which were not mutually exclusive.

RESULTS

SAMPLE CHARACTERISTICS

The overall sample consisted of 150 subjects (26 male, 124 female), with a mean (SD) age of 38.1 (9.6) years, and 9.9 (6.3) years in the longitudinal study. Recruitment was evenly balanced between inpatient, outpatient, and community sources. Subjects were predominately of lower socioeconomic status (65.8% in Hollingshead-Redlich [H-R] Classes IV and V). At follow-up, the mean (SD) GAS score of the overall sample was 59.5 (10.7). There were 31 poor outcome and 39 good outcome subjects. The poor outcome sample consisted of 5 male and 26 female subjects, with a mean (SD) age of 39.0 (10.0) years and 10.9 (6.6) years in the study. The good outcome group had 7 male and 32 female subjects, with a mean (SD) age of 36.9 (9.9) years and 9.8 (7.4) years in the study. Mean (SD) GAS scores at follow-up were 45.0 (5.0) for poor outcome and 72.9 (4.7) for good outcome subjects. There were no significant differences between groups in age, gender, or time spent in the study.

GOOD VERSUS POOR OUTCOME SUBJECTS, QUANTITATIVE DATA (TABLES 1 AND 2)

At the time of recruitment, good outcome subjects had a mean (SD) GAS score of 57.4 (10.2), higher than poor outcome subjects, who had a baseline GAS score of 52.0 (12.7), t = 1.98, 68 df, p = .05. Good outcome subjects tended to be single (ChiSq = 3.01, 1 df, p = .08) and childless compared to poor outcome subjects (ChiSq = 4.92, 1 df, p = .03), though these differences did not persist to follow-up.

At baseline, the two groups did not differ significantly on any of the five component scores of the DIB, or on the DIB total section score, representing the sum of the five section scores. On the DIB-R, the good outcome group endorsed significantly less baseline pathology only on the affect section score, t = 3.05, 33.5 df, p = .004, with a trend on the calculated total section score, t = 2.27, 48 df, p = .03 (representing the sum of the four component section scores).

Within-group interval changes are presented in Table 1 to facilitate contrast with subjective narrative reports (below). By follow-up, the poor outcome group had improved on two DIB sections scores—social adaptation, and impulse action patterns—and on the DIB total section score. Although clinically noteworthy, these changes fell short of statistical significance after Bonferroni correction. On the DIB-R, poor outcome subjects improved significantly on the affect section score, and the calculated total section score, with a trend on the impulse action patterns section score. In contrast, the good outcome group improved significantly on four of five DIB section scores (and the DIB total section score), and on two of the four DIB-R section scores (and the calculated DIB-R total section score.) The DIB-R cognition section score was minimally endorsed at both baseline and follow-up assessments in both groups.

Comparisons between groups at follow-up are given in Table 2. At follow-up, the poor outcome group endorsed significantly more borderline pathology on both summary measures, the DIB total section score, and calculated DIB-R total section score, with significant differences between groups on three of five DIB section scores (impulse action, affects, psychosis) and three of four DIB-R section scores (impulse action, affect, cognition) after Bonferroni correction. In the total sample, 66% of subjects achieved diagnostic remission. Among good outcome subjects, 71.8% achieved diagnostic remission; however, 28.2% achieved good psychosocial outcome without diagnostic remission. Among poor outcome subjects, 35.5% achieved diagnostic remission despite poor psychosocial outcomes (ChiSq = 9.23, 1 df, p = .002).

AXIS I COMORBIDITY (TABLE 2)

At baseline, there were no statistical differences between groups in prevalence of any Axis I disorders. MDD was highly prevalent in both groups, in 43.6% of good outcome subjects and 45.2% of poor outcome subjects. At follow-up, fewer good outcome subjects (12.8%) and more poor outcome subjects (54.8%) met criteria for MDD (Table 2). Similarly, at baseline there was no difference between groups in prevalence of panic disorder, but at follow-up, 15.4% of good outcome and 51.6% of poor outcome subjects were diagnosed with panic disorder. This pattern was repeated for social phobia, GAD, and PTSD. That is, the groups did not differ at baseline, but by follow-up, no subject with good outcome reported social phobia, only seven good outcome subjects reported GAD, and only one reported PTSD. There were no significant differences in prevalence at baseline or follow-up for specific phobia. Among poor outcome subjects, 80% met criteria for Any Anxiety Disorder at follow-up, compared to only 28.2% of good outcome subjects. Group differences at follow-up in the prevalence of MDD, panic disorder, and Any Anxiety Disorder were highly significant. In contrast, the prevalence of alcohol dependence or abuse and substance use disorder did not differ between groups at baseline or follow-up. No subject met criteria for dysthymic disorder at follow-up.

A diagnosis of bipolar disorder (BP) or schizoaffective disorder, bipolar type, was an exclusionary criterion at intake; however, by follow-up, nine subjects from the full sample (6.0%) met formal SCID criteria for bipolar disorder type I (n = 6) or type II (n = 3), and four subjects (2.7%) met SCID criteria for schizoaffective disorder, bipolar type. These new onset SCID diagnoses were confirmed by review of interval psychiatric records and follow-up clinical interviews. Contrary to expectation, none of the nine subjects with new onset BP diagnoses were in the poor outcome sample. The mean GAS score of the nine new onset BP subjects at follow-up was 64.3, all scoring 60 or above, with three subjects in the good outcome sample. However, all four subjects with new onset schizoaffective disorder had psychotic symptoms at the time of follow-up and poor psychosocial outcomes.

At baseline, a past history of psychiatric hospitalization was more prevalent among poor outcome subjects (83.9%) compared to good outcome subjects (48.7%) (ChiSq = 9.30, 1 df, p = .002). In the follow-up interval, 30.8% of good outcome subjects had psychiatric hospitalizations compared to 61.3% of poor outcome subjects (ChiSq = 6.52, 1 df, p = .01). Good outcome subjects were also less likely to be taking psychiatric medications at baseline (30.8%) compared to poor outcome subjects (61.3%) (ChiSq = 6.52, 1 df, p = .01). The frequency of medication use increased dramatically over time in both groups during the study. At follow-up, all poor outcome subjects were taking medications, while medication use increased to 71.8% among good outcome subjects (ChiSq = 8.35, 1 df, p = .004).

Suicidal behavior was a major focus of this longitudinal study. In the overall sample of 150 subjects, 68% had histories of medically significant suicide attempts at baseline and 34% attempted in the follow-up interval. At baseline, there was no relationship between attempter status and future outcome group; however, by follow-up, 51.6% of poor outcome subjects made suicide attempts compared to only 30.8% of good outcome subjects, a suggestive but non-significant trend (ChiSq = 3.13, 1 df, p = .08).

PERSONALITY MEASURES

Temperament was assessed using the NEO-FFI-3 (Costa & McCrae, 2010). Poor outcome subjects had significantly higher degrees of Neuroticism compared to good outcome subjects (t = 3.74, 55 df, p = < .001), but did not significantly differ on Extraversion, Openness, Agreeableness or Conscientiousness. The groups did not differ significantly on the Life History of Aggression at baseline or follow-up. On a measure of trait impulsiveness (BIS-11), the groups did not differ at baseline, but by follow-up, poor outcome subjects endorsed a trend for more impulsivity (t = 2.13, 68 df, p = .04.)

SOCIAL AND VOCATIONAL ADJUSTMENT (TABLE 2)

There were no differences between groups at baseline on a self-rated measure of social and vocational adjustment (Social Adjustment Scale–Self-Report; SAS-SR), but by follow-up, poor outcome subjects reported significantly worse overall functioning compared to good outcome subjects (Table 2). At baseline, only 29% of poor outcome subjects were employed, compared to 53.8% of good outcome subjects, a suggestive trend (ChiSq = 4.32, 1 df, p = .04). At follow-up, poor outcome subjects were significantly less likely to be employed (35.5%), compared to 71.8% of good outcome subjects (Table 2). Good outcome subjects were also more likely to have at least one meaningful interpersonal relationship at follow-up (79.5%), compared to poor outcome subjects (51.6%).

SOCIOECONOMIC VARIABLES

At baseline, more good outcome subjects (65.8%) had household incomes above the 2015 federal household poverty guideline of $20,000/year, compared to poor outcome subjects (40%) (ChiSq 4.49, 1 df, p = .03). Only 13.9% of good outcome subjects had household incomes at or below the poverty level by follow-up, compared to 46.4% of poor outcome subjects (Table 2).

Dependence on government assistance programs is a useful indicator of socioeconomic function in BPD (Zanarini, Jacoby, Frankenburg, Reich, & Fitzmaurice, 2009). At baseline, over half (55.3%) of all study subjects were receiving some form of government support (SSI, SSDI, welfare benefits), with no significant difference between good and poor outcome subjects. By follow-up, 52.0% of all subjects still relied on government assistance; however, a smaller proportion of good outcome subjects (35.9%) remained on government assistance compared to poor outcome subjects (71%) (see Table 2).

CLINICAL NARRATIVE REPORTS: GOOD VERSUS POOR OUTCOME SUBJECTS

1. “WHAT GOT BETTER? WHAT GOT WORSE? WHAT REMAINED UNCHANGED? WHAT IS A PROBLEM TODAY?”

Good outcome subjects most frequently reported improvement in anger management and impulse control (38.5%), and decreased abuse of alcohol and drugs (38.5%). Examples of impulsive behaviors included promiscuous sex, shoplifting, gambling, overspending, and overeating. Improvement in mood swings and intolerance of being alone were each cited by 17.9% of subjects. Improvement does not imply remission. Among symptoms that remained problematic at follow-up, 53.8% of good outcome subjects specifically mentioned depressed or anxious moods, 33.3% problems with anger management and impulse control, and 25.6% problems with unstable interpersonal relationships (including episodes of physical assault towards a partner). Each of these symptom areas was scored as significantly improved on the DIB/DIB-R interviews. Alcohol and drug abuse were cited as problematic for 25.6% of good outcome subjects. Although 30.8% of good outcome subjects reported a suicide attempt in the follow-up interval, none of these behaviors occurred within the 2-year time frame of the DIB/DIB-R interviews at follow-up.

Poor outcome subjects reported significant difficulties with most BPD symptoms. Though some reported, “certain things are better,” improvement was minimal for most. Depression and anxiety were, by far, the most prevalent, persisting symptoms, reported by 87.1% of poor outcome subjects. Suicidal or self-injurious behaviors were recent issues for 58.1%. Impulsivity, aggression, and anger were problematic for 54.8%. The same prevalence was noted for abuse of alcohol and drugs. Psychoticism, including hallucinations, paranoid ideation, and ideas of reference, were reported by 45.2%. Unstable interpersonal relationships were reported by 35.5% of poor outcome subjects.

2. SELF-HELP MEASURES

In the overall sample, 59.3% of subjects cited self-help measures as contributing positively to their outcomes. Helpful measures were described as: “conscious effort,” “self-reflection,” “self-education” (e.g., reading about psychology and BPD), “physical self-care,” and “life lessons.” Changes in behavior were internally motivated. Subjects saw their improvement as a personal accomplishment and were proud that they “did it themselves.” Among the 39 good outcome subjects, self-help measures were credited with improvement by 21 subjects (53.8%) but only by 7 of 31 poor outcome subjects (22.6%).

Avoidance of all intimate relationships was the most frequently cited helpful measure in the overall sample, reported by 31 subjects (20.7%). Based on life experience, these subjects saw intimate relationships as a trigger for emotional instability, disruptive and suicidal behavior, and had made a conscious choice to avoid them. Duration of avoidance ranged widely, from 2 to 25 years. In most cases, friends, family, or mental health support resources were enlisted to minimize isolation. Avoidance of intimate relationship was a coping style found equally among good outcome (n = 6) and poor outcome subjects (n = 6).

In contrast to avoidance, some subjects in the overall sample (12.7%) spontaneously credited their adult relationships as key to improved outcomes. For some, a stable partner allowed the subject to maintain a comfortably dependent role. Stability and consistency were cited as the most important qualities in these relationships, though the partners in some cases were quite controlling. In other cases, the subject assumed caretaker responsibility for a dependent person, including those medically or psychiatrically ill. Same-sex relationships were reported by 17 subjects (11.3%) in the overall sample. Having young children was described as giving purpose to one’s life, setting a limit on any suicidal ideation or behavior.

Other useful self-help measures included structured activity, including school or jobs (“keeping busy”). In the overall sample, 34 subjects (22.7%) had notable lifetime academic or vocational achievements, defined as attaining greater than a bachelor’s degree in college, and/or employment in a managerial or professional capacity. Half of these “high achievers” (55.9%), though not all, were good outcome subjects. High lifetime academic or vocational achievement did not guarantee good psychosocial outcome. Many high achievers reported persistence of borderline symptoms or ongoing problems with drug and alcohol use and fell short of our cut-off for good outcome. Five high achievers were unemployed at follow-up, two with disabling physical illnesses. Four high achievers (11.8%) met criteria for poor outcome.

A lifestyle choice to give up drugs, alcohol, and sexual promiscuity was frequently cited as an important self-help measure. Use of drugs and alcohol was most often recognized as “self-medication,” which was incompatible with other self-help efforts. Importantly, the prevalence of alcohol and substance use disorders did not significantly differ between groups at baseline or follow-up. It was a major impediment to good psychosocial functioning in both groups.

Self-help measures were described as unhelpful by 19.3% of subjects in the overall sample, including four good outcome subjects and one poor outcome subject. The most prominent reasons included attempts to self-medicate with drugs and alcohol and poor relationship choices (e.g., “wrong people,” “critical people”).

3. PROFESSIONAL INTERVENTIONS

Access to professional help was limited to public clinics for the majority of our subjects, who were predominately of lower socioeconomic status (SES). Professional interventions were cited as helpful by only 37.3% of subjects in the overall sample. In most cases, this referred to a good relationship with a trusted therapist, “someone who listened.” DBT was by far the most frequently named form of helpful psychotherapy (i.e., by 14 subjects), and was a common referral for BPD subjects in public clinics. Cognitive behaviorial therapy (CBT) and Alcoholics Anonymous (AA) were each cited by a few subjects. Among good outcome subjects, 15 (38.5%) spontaneously credited professional help for their improvement, compared to only 9 (29.0%) poor outcome subjects.

In the overall sample, a smaller number of subjects (19.3%) offered examples of professional interventions that were unhelpful. These included adverse experiences with a wide variety of group and insight-oriented therapies, including dialectical behavior therapy (DBT), CBT, eye movement desensitization and reprocessing (EMDR), 12-step programs, intensive outpatient programs, and hospitalization. No clear pattern was apparent. Subjects objected to insight-oriented therapies “stirring up negative memories.” Objections to group therapies included feeling that no one could possibly understand their specific issues. Professional interventions were considered unhelpful by three good and five poor outcome subjects.

4. MEDICATION

Medication use was spontaneously reported as helpful by 25.3% of the total sample, and unhelpful by 16.7%. Since multiple medication trials were commonly tried, only the subject’s final judgment was reported. A therapeutic relationship with the prescribing physician was rarely mentioned. Psychiatrists who performed rote “checklist” evaluations were viewed as uncaring. Subjects were unequivocal and very emphatic about medication effects, whether positive or negative, though treatment targets were generally not specified. (i.e., treatment for depression or anxiety could refer either to the borderline patient’s negative affectivity or to Axis I comorbidity). Among good outcome subjects, 28.2% reported helpful effects, as did 30% of poor outcome subjects. Similarly, 15.4% of good outcome and 16.7% of poor outcome subjects found medication use unhelpful.

DISCUSSION

The primary intent of this study was to contrast data from standardized quantitative research interviews with clinical narrative reports of psychosocial functioning in order to identify factors contributing to the great disparity in rates of diagnostic remission and psychosocial recovery in BPD. We found that diagnostic remission from BPD is neither necessary nor sufficient for attainment of good interpersonal relationships or full-time employment, two pillars of good psychosocial outcome. Clinical narratives highlight the adverse effects on psychosocial outcome of residual borderline symptoms, comorbid psychiatric disorders, unemployment, and poverty, including limited access to treatment.

The contrast between standardized interview data and clinical narrative reports is most striking in the assessment of borderline psychopathology, defined by both DIB/DIB-R and addressed in clinical narratives. Over half of good outcome subjects spontaneously complained of current depressive and anxious moods at follow-up, despite significant improvement in DIB/DIB-R Affect section scores. Poor outcome subjects also showed significant improvement in Affect scores (DIB-R) at follow-up, though current depressive and anxious symptoms were highly prevalent (87.1%) in narrative reports. A high prevalence of both MDD (54.8%) and Any Anxiety Disorder (80.6%) were also noted in poor outcome subjects at follow-up. These findings support the MSAD study, which reported that affective symptoms were among the most refractory to change over time (Zanarini et al., 2007). Both good and poor outcome groups also complained of current problems with impulse control, which were among the most likely symptoms to remit among the inpatient sample in the MSAD study. Suicidal behavior and self-injury, SUD, and sexual promiscuity all decreased dramatically in the MSAD study over time (Zanarini et al., 2007, 2008). While this trend was also true for our overall sample and for good outcome subjects, the frequency of suicide attempts changed little among poor outcome subjects from baseline to follow-up despite significant improvement on the DIB/DIB-R Impulse Actions scores, which assess suicidal and self-injurious behaviors. Similarly, among poor outcome subjects, comorbid diagnoses of alcohol abuse/dependence (in 25.8%) and SUD (in 16.1%) remained prominent. Unstable interpersonal relationships were also widely reported as current problems in both good and poor outcome subjects. Clinical narrative reports suggest that for many subjects statistically significant improvement in DIB/DIB-R scores does not reflect clinically meaningful changes in these symptom domains.

Our study affirms the importance of psychiatric comorbidity as a major factor in poor psychosocial outcomes. Psychiatric comorbidity is highly prevalent in subjects with BPD, especially disorders of mood, such as MDD, anxiety disorders, and impulsivity, including SUD (Zanarini et al., 1998.) At study intake, the MSAD study reported a very high prevalence of mood disorders (97%), SUD (62%), anxiety disorders and PTSD (89%), and eating disorders (54%) among their BPD inpatients (Zanarini, Frankenburg, Hennen, Reich, & Silk, 2004). In early reports, the CLPS found significant longitudinal associations between MDD and PTSD among their BPD sample at 6-, 12-, and 24-month follow-ups (Shea et al., 2004). Prevalence for Axis I disorders decreased by the 6-year follow-up in the MSAD study, but only for those BPD subjects who experienced diagnostic remission. The absence of SUD was the strongest predictor of time-to-remission among Axis I disorders (Zanarini, Frankenburg, Hennen, Reich, & Silk, 2004). By 10-year follow-up, prevalence rates for all Axis I comorbidities had decreased markedly in the entire MSAD sample. Using baseline data to predict time-to-remission at 10-year follow-up, the MSAD study found little evidence that psychiatric comorbidity was predictive of early time-to-remission. Absence of PTSD and anxious cluster disorders were the only comorbid psychiatric disorders among 16 variables that predicted earlier time-to-remission (Zanarini, Frankenburg, Hennen, Reich, & Silk, 2006). (An analysis of baseline data to predict time-to-recovery at 16 years yielded similar findings, with the addition of “no ADHD” as a favorable predictor; Zanarini et al., 2014).

Differences between our results and those of the MSAD study may be attributed, in part, to sample characteristics. Our results reflect a prevalence of Axis I disorders in a sample well balanced between inpatient, outpatient, and non-patient community sources. For example, in contrast with the inpatient MSAD study, eating disorders (ED) were not a significant source of comorbidity in our sample. At intake, current SCID diagnoses of ED were made in only 2 subjects with anorexia nervosa, 7 with bulimia nervosa, and 7 with binge eating disorder, or 10.7% of the total sample. Among the 70 subjects chosen for the good versus poor outcome analyses, there were only 5 subjects with a current ED diagnosis (7.1%), all in the poor outcome group. In contrast, the MSAD study reported a baseline prevalence of 54% for ED among their 290 BPD inpatients (Zanarini, Frankenburg, Hennen, Reich, & Silk, 2004).

Use of baseline data to predict outcome variables in long-term studies ignores the effects of time-varying symptoms and inter-current relapses, including recurrent episodes of MDD, SUD, and anxiety disorders. Using both baseline and time-varying data, these disorders were predictive of poor psychosocial outcomes in BPD subjects with 2- to 31-year follow-ups (Soloff & Chiappatta, 2018b). Effects of interval suicide attempts are also discounted in predictive models using only baseline variables. While borderline symptoms diminish in intensity over time for most subjects in longitudinal studies, the same is not true for some time-varying Axis I comorbidities. The prevalence of Axis I disorders such as MDD and anxiety disorders actually increased over time among the poor outcome subjects in our study. In clinical narrative reports, both good and poor outcome groups described ongoing struggles with alcohol and drug abuse, and difficulties in achieving and maintaining sobriety. For these disorders, standardized structured interviews and clinical narratives were in agreement. Alcohol and substance use disorders were equally prevalent in both groups at baseline and follow-up, and played a major role in social and vocational outcomes.

In a longitudinal study, the onset of new psychiatric or medical illnesses in the interim influences psychosocial outcomes, independent of BPD. A diagnosis of new onset bipolar disorder was unexpected, as this disorder was an exclusionary criterion at intake. None of the bipolar disorder subjects were acutely symptomatic at the time of the follow-up SCID interview. Surprisingly, the new onset diagnosis had little adverse effect on psychosocial outcomes. The MSAD study also reported new onsets of bipolar disorder in 15% of BPD subjects in their 16-year follow-up. Any new-onset diagnosis of bipolar disorder raises the question of whether these subjects were misdiagnosed at baseline. Did they have a sub-syndromal form of bipolar disorder that was misdiagnosed as BPD? Although there is a small true comorbidity of these two disorders, community studies indicate that when diagnoses of BPD and bipolar disorder are confounded, the bipolar diagnosis is more often given in error (Paris, Gunderson, & Weinberg, 2007; Ruggero, Zimmerman, Chelminski, & Young, 2010; Zimmerman et al., 2010). New onset psychotic disorders and chronic disabling medical illnesses were diagnosed in a small number of poor outcome subjects. Adverse effects of new onset psychiatric and medical illnesses are missed in predictive studies of outcome that use only baseline data.

Standardized, structured research assessments do not generally address subjects’ own efforts, their successes, and failures in achieving good psychosocial outcomes. Self-help measures were most prominently cited in achieving good psychosocial outcomes. Among good outcome subjects, self-help measures were credited more often than professional interventions. Self-help measures were inwardly motivated, requiring sustained, disciplined effort. In contrast, poor outcome subjects tended to be more dependent on external sources of motivation, focused on practical measures of support, including help with finances and housing, consistent with a reality of poverty, dependence on government support, and socioeconomic disadvantage. These adverse socioeconomic factors play a significant role in the educational and vocational disadvantage of poor outcome subjects. The ability to hold full-time employment (or full-time school attendance) and attain social and vocational adjustment (SAS-SR) has been shown to be predictive of good psychosocial outcome in subjects with BPD (Soloff & Chiappetta, 2018b). Although acting indirectly, poverty is a relevant factor associated with poor psychosocial outcomes.

More subjects reported that self-help measures were useful than professional help. The efficacy of professional intervention in BPD cannot be assessed in naturalistic longitudinal studies. Treatment-seeking introduces a well-known bias for illness severity, as only the more impaired subjects seek treatment. For example, in a 10-year follow-up study of suicidal behavior in BPD, treatment utilization (prior to any attempt) was a predictor of subsequent suicide attempts (Soloff & Chiappetta, 2018a). Treatment utilization was not a predictor of good psychosocial outcomes among BPD subjects in this population (Soloff & Chiappetta, 2018b). In both the MSAD study and the CLPS, treatment was not a predictor of diagnostic remission or good psychosocial outcomes, despite high rates of treatment utilization in both studies. (For example, in the CLPS, all subjects were “treatment seeking” at time of recruitment. In the first 25- to 36-month interval, 64% of BPD subjects were in psychotherapy and 68% received medication consults; Bender et al., 2001, 2006.) Treatment utilization was even higher for BPD patients in the MSAD study, as all were recruited as inpatients. At 6-year follow-up, three-quarters of borderline subjects in the MSAD study were still in psychotherapy and taking medication, though 70% of patients in treatment were already diagnostically remitted (Zanarini, Frankenburg, Hennen, Reich, & Silk, 2006; Zanarini, Frankenburg, Hennen, & Silk, 2004). Through the 10-year follow-up, a large number of borderline patients continued to receive outpatient treatment for prolonged periods of time, despite diagnostic and symptomatic remission rates as high as 88% and 93% (Horz, Zanarini, Frankenburg, Reich, & Fitzmaurice, 2010). Compared to the CLPS and MSAD studies, only 16.1% of our subjects had any outpatient treatment over a 10-year follow-up (Soloff & Chiappetta, 2018a), which may be attributed to the unavailability of long-term treatment in the public health sector as well as unaffordability of private care for most of our subjects.

The role of interpersonal relationships in facilitating good psychosocial outcome was equivocal from the patients’ perspective. Among self-help measures that helped, avoidance of intimate relationships was cited by 20.7% of subjects. In contrast, only 12.7% credited interpersonal relationships with actually facilitating good psychosocial outcomes. Subjects who avoided intimate relationships were not entirely socially isolated, but reported some contacts with friends, family, and mental health resources. Social isolation in BPD patients was a risk factor for suicidal outcome in the landmark 27-year follow-up study of Paris and Zweig-Frank (2001), suggesting the need for long-term follow-up of subjects who defensively avoid relationships.

In our overall sample of 150 BPD subjects, 11.3% reported same-sex relationships. This exceeds prevalence estimates for the general U.S. population, which range from 2.3% to 7.7% for “gay, lesbian, and bisexual populations,” depending greatly on definitions (identity vs. behavior), method of ascertainment, age of sample, and even year of the survey (Centers for Disease Control and Prevention, 2018; Herbenik et al., 2017). Among the 290 BPD inpatients in the MSAD study, same-sex relationships were reported in 36.6% (Reich & Zanarini, 2008). The MSAD study concluded that, “same gender attraction and/or intimate relationship choice may be an important interpersonal issue for approximately one-third of both men and women with BPD.” Differences in prevalence rates between studies are attributable to sample characteristics, that is, McLean Hospital inpatients (MSAD) compared to our balanced inpatient, outpatient, and non-patient community sample.

A secondary intent of our study was the characterization of clinical differences between the best and worst outcome subjects at baseline. These data may guide early identification of subjects at greatest risk. Using baseline and time-varying follow-up data, we previously reported that full-time employment, social and vocational adjustment at follow-up, and the absence of MDD, SUD, or anxiety disorders at follow-up increased the likelihood of good psychosocial outcomes in BPD subjects followed 2 to 31 years (mean 9.9 years) (Soloff & Chiappetta, 2018b). Poor outcome subjects are significantly more impaired at baseline, with older age, lower GAS scores, and more borderline psychopathology, especially affective symptoms. They have greater degrees of neuroticism, which reflects a temperamental vulnerability to negative affectivity. A high prevalence of depressive and anxious symptoms is reflected in both objective and subjective assessments. Compared to good outcome subjects, more poor outcome subjects have past histories of psychiatric hospitalization and medication use. More poor outcome subjects live in poverty. Social deficits associated with poverty include restricted educational and vocational opportunity. These poor-prognosis patients may be best served by vocational rehabilitation models of care (Links, 1993).

CLINICAL IMPLICATIONS

High rates of diagnostic remission in longitudinal studies have generated an undue optimism about the good prognosis of patients with BPD. This perspective must be tempered by the very low rates of good psychosocial outcome or recovery. Remission of diagnostic criteria does not predict good psychosocial outcome in our studies. High rates of sustained treatment utilization, including psychotherapies, years after diagnostic and symptomatic remission, support the validity of this conclusion.

Symptomatic and diagnostic recurrences are frequent over time, even after periods of sustained remission. For example, in the MSAD study, rates of diagnostic recurrence increased to 36% over 14 years of follow-up (Zanarini, Frankenburg, Reich, & Fitzmaurice, 2012). Symptoms of borderline psychopathology waxed and waned over a 16-year prospective follow-up (Zanarini, Frankenburg, Reich, & Fitzmaurice, 2016). Clinical narrative reports in our study clearly reflect symptomatic instability, even among good outcome subjects.

Patients with strong internal motivation report benefits from self-help measures more often than professional interventions. Psychoeducation, physical self-care, and structured activities related to educational and vocational goals are cited as most helpful. In treatment settings, these interventions are given little attention in favor of formal psychotherapies, yet they are useful to self-motivated patients and are quite cost-effective.

Alcohol and substance use disorders were problematic for both good and poor outcome subjects, and generally did not receive the specialized attention they required. Similarly, the high prevalence of MDD, SUD, and anxiety disorders in both groups suggests a need for treatment focused on these psychiatric comorbidities, independent of BPD. Medication use tends to become cumulative and prolonged in patients with BPD (Zanarini, Frankenburg, Reich, Harned, & Fitzmaurice, 2015). While aggressive pharmacotherapy is clearly needed for comorbid psychiatric disorders, medication use for targeted borderline symptoms is generally recommended for brief time periods. (Our subjects were very clear about the success or failure of medication treatment efforts.)

CONCLUSIONS

Clinical narratives (qualitative findings) differ from results of structured interviews (quantitative findings) primarily in the degree to which symptoms are experienced and reported as problematic by the subject. The formal structure and setting of diagnostic interviewing constrains complete reporting, especially in regard to the relative importance of specific symptoms and comorbidities in achieving psychosocial functioning from the patient’s perspective. That is, clinical narratives identify and magnify symptoms and comorbidities most important to the patient. (In this regard, clinical narratives more closely resemble data derived in therapy settings.) The gap between rates of remission and recovery in BPD is explained in part by clinical narrative data, augmenting and amplifying data derived by structured interviewing.

LIMITATIONS

Our subjects had lower SES than patients in the CLPS (37% H-R IV, V) or MSAD studies (3.4 on a 5-point H-R scale), which limits comparison between studies (Gunderson, et al., 2000; Zanarini, Frankenburg, Hennen, Reich, & Silk, 2005). Low SES contributes to limited educational and vocational opportunities, key predictors of psychosocial outcome, as well as access to private or long-term psychiatric treatment.

The DIB/DIB-R interviews have a 2-year time frame. For subjects who have had inter-current episodes of MDD, SUD, anxiety disorders, suicidal behaviors, or hospitalizations outside the 2-year time frame, the follow-up DIB/DIB-R scores may minimize vulnerability. These interval events may have a profound impact on psychosocial outcomes. Similarly, avoidance of all interpersonal relationships in the 2-year time frame, a very common coping mechanism among our subjects, results in lower (improved) DIB/DIB-R scores, though reflecting true borderline psychopathology.

Clinical narrative reporting incurs the risk of subjects’ exaggerating good and minimizing poor outcomes. Adverse events such as suicide attempts, hospital admissions, drug and alcohol abuse, impulsive, aggressive, and antisocial behaviors may be minimized in narrative reports, though documented in structured follow-up interviews. Similarly, causal mechanisms cannot be determined from subjective data. Good outcome subjects may well credit their own internal motivation for their better outcomes.