Participants
The present study examined narrative coherence of autobiographical memories in a subsample of BPD participants and healthy controls from a larger cohort reported elsewhere (36). Twenty-six women aged 18–45 years who met diagnostic criteria for BPD according the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) were included in the study. Twenty-eight controls were matched on age and sex, as well as parental education to assure that BPD participants and controls did not differ in socioeconomic status of the family of origin (missing data on parental education for two BPD participants). Parent education was coded in one of five categories (1 = no education; 2 = trained worker; 3 = skilled worker; 4 = theoretical education of lower academic level; 5 = academic education). When data on parental education was available for both parents, the mean of the parental education categories was reported. We furthermore recorded the educational status for each subject to control for inter-individual variability in educational level coded in one of six categories (1 = no education or primary school; 2 = skilled worker; 3 = university-preparatory school; 4 = short academic education; 5 = academic education equivalent to a bachelor’s degree; 6 = academic education equivalent to a master’s degree). The matching and control variables are displayed in Table 1.
Table 1
Matching variables, control measures, symptom scale, and psychopatology scales for patients and controls.
| BPD subjects (N = 26) | Controls (N = 28) |
---|
| Control variables |
Age | 28.67 ± 7.21 | 28.88 ± 8.77 |
Parental education | 2.45 ± 0.96 (−2) | 2.71 ± 0.79 |
Subject education | 2.52 ± 1.68 (−1) | 3.85 ± 1.32 |
WAIS VCI | 91.15 ± 10.4 | 103.32 ± 6.2 |
Memory word count | 75.04 ± 34.9 | 84.63 ± 29.3 |
CTQ | 60 (44–71) (−3) | 28 (25–33) (−1) |
| Symptom scale |
ZAN-BPD | 12.6 ± 6.4 | - |
| Psychiatric diagnostic comorbidityn(%) |
Major Depressive Disorder | 3 (11,5) | 0 (0.0) |
Bipolar II Disorder | 0 (0.0) | 0 (0.0) |
Agoraphobia | 6 (23.1) | 0 (0.0) |
Panic Disorder without Agoraphobia | 7 (26.9) | 0 (0.0) |
Social Phobia | 7 (26.9) | 0 (0.0) |
General Anxiety Disorder | 2 (7.7) | 0 (0.0) |
Obsessive-Compulsive Disorder | 2 (7.7) | 0 (0.0) |
Posttraumatic Stress Disorder | 6 (23.1) | 0 (0.0) |
| Personality disorders n(%) |
Avoidant | 9 (34.6) | 0 (0.0) |
Dependent | 2 (7.7) | 0 (0.0) |
Obsessive-Compulsive | 6 (23.1) | 0 (0.0) |
Paranoid | 5 (19.2) | 0 (0.0) |
Schizotypal | 0 (0.0) | 0 (0.0) |
Schizoid | 0 (0.0) | 0 (0.0) |
Histrionic | 2 (7.7) | 0 (0.0) |
Narcissistic | 2 (7.7) | 0 (0.0) |
Antisocial | 2 (7.7) | 0 (0.0) |
Data is shown with mean ± standard deviations except the CTQ score, which is displayed with medians and lower and upper quartiles because the variable significantly deviated from the normal distribution. Note that missing data for a variable is displayed with a superscript encapsulated in hyphens (−n) in the descriptive statistic cells for the patients and controls, respectively. Missing data is also described in the method section. WAIS VCI = Wechsler Adult Intelligence Scale Verbal Comprehension Index; BIS-11 = Barrat Impulsiveness Scale; CTQ = Childhood Trauma Questionnaire; BPAQ = Buss-Perry Aggression Questionnaire; ZAN-BPD = Zanarini Rating Scale for Borderline Personality Disorder. |
Exclusion criteria for all participants included DSM-IV lifetime psychotic disorder, bipolar I disorder or substance use disorder, history of significant head trauma, severe chronic physical or neurological illness such as seizure disorder, encephalitis or stroke or a VCI score below 70 points.
BPD participants were recruited from Psychiatric Clinic Roskilde, an outpatient clinic in Region Zealand Psychiatry, Denmark, specialized in treating BPD, and controls were recruited through local advertisement. All participants received thorough written and verbal information about the project and were required to provide written informed consent before inclusion in the study. The study was approved by the Regional Ethics Committee for Science Ethics of Zealand. Recruitment procedures are described in more detail elsewhere (35).
Assessments
Diagnostic assessment. BPD participants and controls were screened for psychiatric disorders by trained clinicians with the Mini International Neuropsychiatric Interview (36), and for personality disorders with the Structured Clinical Interview for DSM-IV Axis II Disorders (37). All subjects in the BPD cohort fulfilled the criteria of BPD.
Borderline symptoms. Severity of borderline symptoms within the last two weeks was assessed in BPD participants with the nine-item Zanarini Rating Scale for Borderline Personality Disorder (ZAN-BPD), which is a clinician-administered semi-structured interview (38). The items, each of which reflect the core symptoms associated with BPD, were scored on a scale ranging from 0–4 based on frequency and severity of BPD symptoms. The ZAN-BPD total score is presented in Table 1.
Childhood trauma. We assessed childhood trauma experiences with the validated Danish version of the Childhood Trauma Questionnaire (CTQ) (39). The CTQ is a standardized retrospective 28-item self-report inventory that measures the occurrence and severity of experienced childhood trauma including emotional, physical and sexual abuse, emotional and physical neglect scored on a 5-point Likert scale (40). We used the total score in the present study. Three BPD participants and one person in the control group had missing data on the CTQ.
Verbal comprehension index. We have previously shown that the BPD participants in the present study displayed lower verbal comprehension relative to controls (35, 41) and the VCI scale has been shown to be positively associated with reading comprehension and spelling (42). Because our participants were asked to write down their autobiographical memories, individual differences in verbal comprehension abilities could affect how the autobiographical memories was communicated on paper. Therefore, we assessed individual differences in verbal knowledge and reasoning with the Similarities, Vocabulary and Information subscales from the Wechsler Adult Intelligence Scale, Fourth edition (43), which were combined to estimate the Verbal Comprehension Index (VCI).
Autobiographical Memory
Participants were asked to recall autobiographical memories of specific events from their lives. Each participant was asked to recall a total of six autobiographical memories, two of which had to be self-guiding, socially shared, and directive in theme, respectively. The participant was requested to recall either a self-guiding, socially shared or directive autobiographical memory by cue-card with one of the following prompts: Try recalling an event that you think says something about your identity (Self-guiding autobiographical memory), Try recalling an event that you often share with other people (Socially shared autobiographical memory), and Try recalling an event that you think of in order to solve current or future problems (Directive autobiographical memory).
Participants were presented with one cue-card at a time in a randomized order. The response time was recorded from when a cue-card was shown until the participant verbally indicated that a memory had been recalled. If the participant had not responded after three minutes, the trial was registered as an omission. Participants were instructed to write down the memories on paper with no time restraint and were instructed to communicate each event as if they told it for the first time to a new friend with whom they felt comfortable. Eight BPD participants had a total of 15 omissions, including 6 directive, 5 social and 4 self-guided memories, while we registered a total of 4 omissions across four controls including 3 directive memories and 1 social memory.
Narrative coherence was coded using an adapted version of the coding system developed by Baerger & McAdams (1999), which has previously been used in adult (21) and adolescent (44) clinical samples. Narrative coherence was operationalized using four subscales: Orientation, Structure, Affect and Integration. To ensure rater-reliability, we simplified the original scale, which ranged from 1–7 points (9), to a 5-point scale with 1 representing minimum coherence, and 5 maximum coherence for each of the four subscales. A similar simplified adaption of the coding procedure has been successfully employed before (21, 44). A global score for total narrative coherence was calculated based on the average of the four subscales. The Orientation subscale measures how satisfactorily the story provides the reader with adequate context to comprehend the story. Context includes critical background information leading up to the beginning of the story, including sufficient information about the main characters, and when and where the story takes place. The Structure subscale evaluates the logical flow of the story and whether events and reactions appear causally and temporally connected. A story scoring high in structure should also display the structural elements of an episode system described in detail elsewhere (9). The Affect subscale measures how coherently affective words or phrasings are incorporated to make evaluative points, giving a clear indication of the feelings experienced by the persons described in the autobiographical memories. The Integration subscale pertains to how skillfully the narrator communicates the point of telling the story. Autobiographical memories high in integration convey the significance of the experience to the narrator and how it relates to the narrators’ overall life. Descriptive statistics for narrative coherence scores are displayed in Table 2.
Table 2
Descriptive statistics for outcome measures for the patient and control group.
| BPD subjects (N = 26) | Controls (N = 28) |
---|
| Outcome measures |
Episodic specificity | 1.02 ± 0.75 | 1.11 ± 0.73 |
Orientation subscale | 2.64 ± 0.69 | 3.23 ± 0.70 |
Structure subscale | 2.71 ± 0.72 | 3.35 ± 0.74 |
Affect subscale | 2.30 ± 0.70 | 2.18 ± 0.55 |
Integration subscale | 2.37 ± 0.57 | 2.72 ± 0.56 |
Data is shown with group mean ± standard deviations. The displayed data for the narrative subscales and episodic specificity score is based on each subject's mean scores from the six autobiographical memories. |
Memories were considered specific if they were situated in a specific time and place, typically not exceeding the course of a day. Episodic specificity was scored on a scale ranging from 0–3. A score of 0 points indicated a generalized memory with no episodic details, a 1-point score indicated an autobiographical memory with few and fragmented episodic details, a memory that scored 2 points included some references to specific and associated episodic details, and a memory that scored 3 points was characterized by being rich in episodic details, and firmly situated in time and place. The episodic specificity score is presented in Table 2.
Inter-rater reliability. Three blinded coders (G.B, M.S.J and L.S.P) rated the Episodic specificity score, and the Orientation, Structure, Emotion and Integration subscale scores, respectively, based on anonymized data from 30 randomly selected participants. Inter-rater reliability was estimated from the mean score of the six autobiographical memories for each participant for the Episodic specificity, Orientation, Structure, Emotion and Integration scores, respectively. Inter-rater reliability was estimated using intraclass correlation coefficient (ICC) using a two-way random model with absolute agreement for single measures. Results showed a moderate to good reliability for the Episodic specificity score (ICC = 0.880), and the Orientation (ICC = 0.655), Structure (ICC = 0.766), Emotion (ICC = 0.703) and Integration (ICC = 0.785) subscale scores. Because the inter-rater reliability was satisfactory, we used the first author's ratings for the statistical analyses of all participants.
Statistical analysis
Statistical analyses were conducted in SPSS 25. A p value below .05 was considered significant. Group differences for the matching and control variables were tested with two-tailed t tests. Group differences in omissions were tested with the non-parametric Mann-Whitney U test.
To assure that group differences were not mediated by potential confounders all ANCOVA and multiple linear regression models were controlled for age, subject education, parent education, VCI score and number of words used in the autobiographical stories. We used a two-way repeated measures ANCOVA to test our primary hypothesis that the autobiographical memories in BPD participants had lower narrative coherence compared to controls and to test for group differences in the social, self and directive memory functions. Group was entered as the between-subject factor while the narrative subscales were entered as the first within-subject factor and the memory functions were entered as the second within-subject factor in the ANCOVA model. To test group differences in episodic specificity, we likewise used repeated measures ANCOVA, in which group was entered as the between-subject factor while episodic specificity for the social, self and directive scores were entered as within-subject factor. If significant results were observed, multiple linear regression were used to test the directionality of the observed group differences. We further explored if significant group differences were coupled to self-reported childhood adversity. Initially, the CTQ score was entered as a control variable to see if group differences persisted when controlling for childhood trauma. Subsequently, we included an interaction term for group by CTQ score to examine if the respective significant outcome measure showed a different relationship with experienced childhood adversity in BPD participants relative to controls. If the interaction term was not significant, we explored whether the control variable was associated with the significant outcome measure across BPD participants and controls. Lastly, we explored whether the significant outcome measure was associated with symptom severity in BPD participants entering the ZAN-BPD score as the predictor of interest. For the covariates used in the multiple linear regression models, missing values were replaced by the mean of the respective variable to limit unwarranted exclusion of subjects. All multiple linear regression models were visually inspected to ensure normal distribution of the residuals and all model covariates fulfilled criteria of noncollinearity with Tolerance > .3.
Continuous variables were determined to have a significantly non-normal distribution if either of the respective standard errors for the skewness or kurtosis were above or below Z ± 1.96 (two-sided p < .05) in the BPD participants and/or controls. Non-normally distributed variables were normalized using the Rankit transformation, and the normalized values were used in the statistical analyses. The CTQ score was significantly non-normally distributed (p < .05) and was successfully normalized with Rankit transformation. The remaining variables all appeared normally distributed.