Long-term stability of perceived childhood trauma in individuals with borderline personality disorder

ABSTRACT

Background: The Childhood Trauma Questionnaire (CTQ) is a widely used tool for the retrospective assessment of childhood trauma (CT). However, evidence on its long-term stability remains limited, especially in diagnostically complex populations such as individuals with borderline personality disorder (BPD). Given the disorder’s characteristic emotional instability and altered self-perception, assessing the stability of CTQ scale scores in this group is particularly important.

Methods: In a longitudinal design, CTQ and the Borderline Symptom List-23 (BSL-23) scores were collected at two timepoints (T0: n = 417; T1: n = 195) from individuals with current or past BPD. The interval between assessments ranged from 4–11 years (mean 7.72 years).

Results: CTQ total scores showed excellent internal consistency (McDonald’s ω = .95) and high long-term stability (intraclass correlation coefficient = .82), regardless of assessment interval. Subscale stabilities ranged from moderate to high (Sexual Abuse = .86; Physical Abuse = .81; Emotional Neglect = .77; Physical Neglect = .76; Emotional Abuse = .75). Changes in CTQ subscale scores showed small to moderate positive associations with changes in BPD symptom severity (β = .23 –.33), indicating that increases in perceived childhood trauma were accompanied by increases in symptom severity, except for Physical Neglect (β = .16, p = .078).

Conclusions: The CTQ scales demonstrated strong long-term stability over up to 11 years in a heterogeneous BPD sample. Individual changes in perceived childhood trauma were associated with corresponding changes in BPD symptom severity. These results support the CTQ’s utility as a retrospective measure of childhood trauma.

HIGHLIGHTS

• Reported childhood trauma remains highly stable over several years in individuals with current/past borderline personality disorder.

• Changes in perceived childhood trauma were linked to changes in borderline personality disorder symptom severity, underscoring the clinical relevance of retrospective trauma assessments.

• Findings support the Childhood Trauma Questionnaire as a retrospective measure of childhood trauma for both research and clinical practice.

1. Introduction

Traumatic childhood experiences comprise adverse interpersonal experiences occurring before the age of 18. Such childhood trauma (CT) can be described along the dimensions of Physical or Emotional Neglect as well as Physical, Sexual, and/or Emotional Abuse (Bernstein et al., 2003). CT represents the most significant singular risk factor for mental disorders and is also associated with greater symptom severity and poorer clinical outcomes (Duarte et al., 2020; Nelson et al., 2017; Aas et al., 2023; Sahle et al., 2022; Daníelsdóttir et al., 2024; Baldwin et al., 2023). CT is likewise considered a risk factor for the development of personality disorders, especially for borderline personality disorder (BPD) (Riemann et al., 2024).

The biosocial model of BPD, originally proposed by Linehan (1993), conceptualises the disorder as the result of a transactional process between biological vulnerability and an invalidating social environment. At its core, BPD is characterised by pervasive emotion dysregulation, affective instability, impulsivity, unstable interpersonal relationships, and recurrent self-harm or suicidal behaviour. Heightened emotional sensitivity, intense affective responses, and a prolonged return to baseline are exacerbated by environments that ignore, punish, or trivialise emotional expression – particularly during early developmental stages (Linehan, 1993). Within this framework, invalidating and potentially traumatic early environments are considered central environmental risk factors. Individuals growing up in such environments often fail to develop effective strategies for managing intense emotional states, increasing the likelihood of impulsive behaviour, unstable interpersonal relationships, and chronic emotional distress (Linehan, 1987). Developmental extensions further emphasise dynamic transactions between child temperament and caregiver responses, underscoring the relevance of early adverse experiences for the emergence of BPD symptoms (Crowell et al., 2009).

In this context, early traumatic experiences are of central relevance. CT represents a key environmental factor that interacts with BPD symptomatology (Riemann et al., 2024) such as difficulties in emotion regulation (van Dijke et al., 2011; Turniansky et al., 2019; Rosenstein et al., 2018) and dissociation (Schulze et al., 2024; Spitzer et al., 2000; Watson et al., 2006). Individuals with BPD report higher levels of CT across all trauma dimensions compared to psychologically healthy controls (Schmitz et al., 2021).

The Childhood Trauma Questionnaire (CTQ; Bernstein et al., 2003) is one of the most commonly used instruments for the retrospective assessment of CT (Georgieva et al., 2021). Although it is widely used in both clinical practice and research to retrospectively assess CT, there is still a limited number of studies that have systematically examined its psychometric properties – specifically its long-term stability (Aloba et al., 2020; Cay et al., 2022; Hagborg et al., 2022; Zhang et al., 2024). Understanding the stability of trauma reports in individuals with BPD is therefore critical for both theoretical models and clinical assessment. This gap is particularly relevant given ongoing concerns about the accuracy of retrospective self-reports. Retrospective self-reports of CT may be influenced by memory biases (Hardt & Rutter, 2004), which can lead to both over- and underreporting, thereby highlighting potential discrepancies between subjective recollections and objective records (Brown et al., 1998; Baldwin et al., 2019; Danese and Widom 2020, 2023; Widom, 2019). Beyond memory bias per se, the accuracy of retrospective self-reports may also be shaped by how traumatic experiences are perceived, encoded, and evaluated. Traumatic memories may be incompletely or distortedly represented in memory and subject to dysfunctional appraisals (Ehlers & Clark, 2000). Additional factors that may contribute to memory fragmentation or altered reporting include dissociative symptoms (Bedard-Gilligan & Zoellner, 2012), individuals’ motivations to disclose, withhold, or fabricate information about maltreatment (Coleman et al., 2024), as well as depressive symptoms (Spinhoven et al., 2012).

A conceptual limitation concerns the terminology frequently used in the CTQ literature. The term test-retest reliability is often applied to studies with extended time intervals between assessments. However, from a psychometric standpoint, such intervals, exceeding two months (see e.g. Gnambs 2014), are generally considered inappropriate for estimating true test-retest reliability, as the assumption that the underlying construct remains stable becomes increasingly implausible. Accordingly, we refer to stability rather than reliability throughout this study when addressing long-term associations. Notably, research has shown that reliability (i.e. resistance to measurement error) and stability (i.e. consistency of the underlying construct over time) are conceptually and empirically distinct; for instance, test-retest reliability and stability coefficients can diverge markedly across different timeframes and constructs (Chmielewski & Watson, 2009).

To date, only a handful of studies have investigated the long-term stability of CTQ scores across different clinical and non-clinical populations. For example, Cay et al., (2022) reported high test–retest reliability for the CTQ total score over a 20-month period (intraclass correlation coefficient [ICC] > .81) in patients with schizophrenia, bipolar disorder, and healthy controls (Cay et al., 2022), indicating good to excellent stability. Bader et al., (2009) reported moderate to high ICCs between .74 and .94 over an average of 13 months in a German-speaking clinical sample (Bader et al., 2009). Hagborg et al., (2022) found moderate stabilities over a one-year interval in adolescents, with coefficients exceeding .61 across all subscales (Hagborg et al., 2022), which can be interpreted as moderate stability. Similarly, Zhang et al., (2024) observed reliability in adolescents over a one-year period, with total score ICCs above .60 and subscale ICCs above .40 (Zhang et al., 2024), suggesting moderate reliability for the total score and low to moderate reliability for some subscales. Even longer intervals have been examined: in a 10-year follow-up of individuals with bipolar disorder, Hosang et al., (2023) reported substantial agreement on most CTQ subscales, with kappa values up to .85 for the total score, which can be interpreted as almost perfect agreement (Landis & Koch, 1977), but lower stability for physical and emotional neglect (κ = .41 and κ = .43, respectively) (Hosang et al., 2023; Landis & Koch, 1977), reflecting moderate agreement (Landis & Koch, 1977). Supplemental Table S1 provides more information of previous studies examining the long-term stability of CTQ scores. To date, however, no studies have specifically examined the long-term stability of the CTQ in individuals with symptoms or a diagnosis of BPD (Zhang et al., 2024).

In contrast to the growing but still fragmented body of evidence on long-term stability, the internal consistency of the CTQ has been more widely reported and is generally considered acceptable to good. Multiple studies across diverse samples – ranging from healthy controls to individuals with PTSD, depression, or substance use disorders – have found high internal consistency for the total scale (e.g. Cronbach’s α > .85) and for most subscales (Jiang et al., 2018; Peng et al., 2023; Xu et al., 2023; Wingenfeld et al., 2010; He et al., 2019). However, there is a consistent pattern of lower reliability for the Physical Neglect subscale, with some studies reporting Cronbach’s alpha values well below .70 (Bader et al., 2009; Klinitzke et al., 2012; Schulz et al., 2014).

Although a few studies have examined the long-term stability of the CTQ, none have included participants diagnosed with BPD. Given that BPD is characterised by affective instability, dissociation, and shifts in self-perception (Herzog et al., 2022), it is particularly important to determine whether retrospective reports of childhood trauma remain stable over time in this population. Retrospective reports of childhood trauma are influenced by current psychological states (Bedard-Gilligan & Zoellner, 2012; Coleman et al., 2024; Spinhoven et al., 2012) and associated memory biases (Hardt & Rutter, 2004). Because fluctuations in BPD symptom severity involve affective, dissociative, and depressive processes that shape trauma appraisal and recall (Ehlers & Clark, 2000), changes in symptoms may be closely linked to changes in perceived childhood trauma over time.

The present study aims to address this gap by systematically examining the stability of the CTQ over an extended time frame in a BPD sample. The objectives of the study are a psychometric investigation of the CTQ in patients with a current or past BPD diagnosis over a follow-up period of 4–11 years, with a particular focus on the stability of the scale scores and correlates of changes in CTQ scores. Based on previous findings, we hypothesised that (1) perceived CT shows high long-term rank-order stability, such that individuals largely maintain their relative position within the sample over time, and (2) that within-person changes in perceived CT are closely linked to concurrent changes in BPD symptom severity over time, suggesting that symptom fluctuations may influence the appraisal of retrospectively reported CT.

2. Methods

This study was part of a larger research cohort established by KFO 256, a Clinical Research Unit supported by the German Research Foundation (DFG) to examine the mechanisms underlying disturbed emotion processing in BPD (Schmahl et al., 2014). Additionally, data collection was extended within a project titled Symptom Course of Borderline Personality Disorder: A Follow-up Study.

2.1. Recruitment and enrolment

2.1.1. Baseline-Assessment, T0

Recruitment took place across Germany using advertisements in social media, newspapers, radio, and podcasts, along with flyers distributed to general practitioners, psychiatrists, psychotherapists, self-help organisations, and psychiatric inpatient and outpatient facilities. This strategy enabled the assessment of 417 individuals with BPD between 2012 and 2018. Inclusion criteria were based on the International Personality Disorder Examination (IPDE) (Loranger et al., 1997), which was administered by trained psychologists. Participants were classified as having current BPD if they met five or more DSM-IV BPD criteria, or as having remitted BPD if they met no more than three criteria and reported no self-injurious behaviour in the past two years. Exclusion criteria included current substance use disorder, psychotic disorder, or bipolar I disorder. Prior to participation, all individuals provided written informed consent and completed sociodemographic questionnaires as well as clinical self-report measures.

2.1.2. Follow-up, T1

All patients who participated in the first or second funding period of KFO 256 and had either (1) a DSM-IV diagnosis of current BPD or (2) BPD in remission – defined with a maximum of three BPD criteria with no self-injury in the past two years – were re-invited (in the years 2022–2024). A total of n = 196 participants who took part at T0 provided informed consent for the follow-up assessment and completed both interview and questionnaire measures.

2.2. Assessments

Participants were asked to provide basic sociodemographic information, including age and gender. Age was assessed as a continuous variable in years. Gender was recorded as a categorical variable (T0: female, male; T1: female, male or other).

The German version of the Childhood Trauma Questionnaire (CTQ) (Klinitzke et al., 2012) was used to assess self-reported traumatic childhood experiences. The German version of the CTQ comprises 28 items, measuring five dimensions of childhood adversity: Physical, Sexual, and Emotional Abuse, as well as Physical and Emotional Neglect. Each dimension consists of five items rated on a 5-point Likert scale ranging from ‘never true’ to ‘very often true,’ yielding subscale scores between 5 and 25. A total score can be calculated, ranging from 25 to 125. Additionally, a set of three items captures the construct of minimisation/denial (Bernstein et al., 2003). The German version of the CTQ was validated in a representative sample of the general population (Klinitzke et al., 2012). Confirmatory factor analysis (CFA) supported the original five-factor structure (Emotional Abuse, Physical Abuse, Sexual Abuse, Emotional Neglect, Physical Neglect), though with only sufficient model fit. Internal consistencies (Cronbach’s α) for the subscales were as follows: Emotional Abuse α = .88, Physical Abuse α = .83, Sexual Abuse α = .96, Emotional Neglect α = .89, and Physical Neglect α = .64. Test-retest reliability (over a 4-week interval) ranged from r = .79 (Physical Abuse) to r = .86 (Sexual Abuse). Evidence for construct validity was provided by significant associations with depression, anxiety, and life satisfaction (Klinitzke et al., 2012).

The International Personality Disorder Examination (IPDE) (Loranger et al., 1997) is a semi-structured diagnostic interview developed by the World Health Organization for the assessment of personality disorders according to ICD-10 and DSM-IV criteria. The ICD-10 module comprises 59 items assessing the presence of personality disorder traits across all major ICD-10 personality disorder categories. Each item is rated on a three-point scale (0 = not present, 1 = accentuated, 2 = criterion level). The IPDE has demonstrated good interrater reliability (kappa coefficients typically > .70) and test-retest reliability (ICC > 0.86), as well as satisfactory validity across patients with BPD (Carcone et al., 2015). Interviews are conducted by trained clinicians or psychologists following standardised guidelines.

The Borderline Symptom List-23 (BSL-23) (Bohus et al., 2009) was used to assess the severity of borderline-specific symptoms experienced during the past week. The 23 items are rated on a 5-point Likert scale ranging from 0 to 4, with higher mean scores reflecting greater symptom severity. The instrument demonstrates high test-retest reliability over a one-week interval (r = .82) (Bohus et al., 2009).

2.3. Data analysis

Data analyses were conducted in R (version 4.4.1). First, to explore whether missingness in CTQ scores at follow-up (T1) was systematically related to baseline characteristics, we conducted a logistic regression analysis. A binary outcome variable was created indicating whether the CTQ total score at T1 was missing (1 = missing, 0 = observed). Predictor variables included age, gender, baseline CTQ total score (T0), and BSL mean score at T0. Second, descriptive statistics were computed for demographic and clinical characteristics at baseline (T0) and follow-up (T1), including distributions of age, gender, and BPD diagnoses. In addition, the CTQ Minimization/Denial scale was reported descriptively. Minimisation scores are considered to reflect denial of adverse experiences and may be influenced by cultural factors, particularly with regard to items assessing physical neglect (Şar et al., 2021). The literature suggests that denial of trauma should not be conflated with dissociative symptomatology: the two appear to reflect different psychological and neurobiological response patterns, with dissociation not reducible to low or biased reporting (Mutluer et al., 2018). Minimisation was calculated by summing the number of maximum endorsements (score = 5) across the three minimisation items (e.g. ‘I had the perfect childhood’), yielding a total score ranging from 0 to 3. Importantly, minimisation/denial status has not been shown to significantly moderate associations between CTQ scores and clinical variables (MacDonald et al., 2015), supporting its descriptive use in the present study.

The CTQ was subjected to CFA using the lavaan package (version 0.6.19) separately at T0 and T1 to test the established five-factor structure with missing data being handled using full information maximum likelihood (FIML). All models were estimated using maximum likelihood estimation with robust standard errors and scaled test statistics (MLR). Model fit was evaluated using multiple indices, including the comparative fit index (CFI), the root mean square error of approximation (RMSEA), and the standardised root mean square residual (SRMR). We used established cutoff criteria, with acceptable model fit defined as CFI ≥ .90 (Byrne 1994), RMSEA ≤ .08 (Browne & Cudeck, 1992), and SRMR ≤ .08 (Hu & Bentler, 1999). Internal consistency was assessed using McDonald’s omega (ω), calculated via the compRelSEM function from the semTools package. This approach estimates reliability based on the parameters of the confirmatory factor model, ensuring consistency between the measurement model and reliability estimation (Savalei & Reise, 2019).

We evaluated longitudinal measurement invariance of the CTQ’s five-factor structure across baseline (T0) and follow-up (T1) using a series of CFAs. Because the analysis of missing data (see below) indicated that age and gender significantly predicted missingness in CTQ scores at follow-up, these variables were included as auxiliary variables using the cfa.auxiliary function from the semTools package. This approach incorporates auxiliary variables into the estimation without specifying direct effects on the latent structure, thereby improving the robustness and efficiency of parameter estimation. To account for shared method variance, we specified correlated residuals (i.e. correlated uniquenesses) between identical items measured at T0 and T1. Longitudinal measurement invariance was assessed in three increasingly restrictive models. First, a configural model was fitted in which all factor loadings and item intercepts were freely estimated across time points. This model served to establish that the same general factor structure held across T0 and T1. Second, a metric invariance model constrained corresponding factor loadings to be equal over time, testing whether items were related to their respective latent constructs in a consistent way across measurement occasions. Third, a scalar invariance model imposed equality constraints on both factor loadings and item intercepts, allowing us to evaluate whether participants with the same latent trait levels would provide equivalent item responses over time, a prerequisite for meaningful interpretation of mean-level change. Model fit was assessed using χ², degrees of freedom, CFI, RMSEA, and SRMR, as well as information criteria (AIC, BIC). We compared nested models based on changes in fit indices, using established guidelines: a decrease in CFI ≤ .010 (i.e. ΔCFI ≥ –.010) was considered negligible (Cheung & Rensvold, 2002), and increases in RMSEA (ΔRMSEA ≤ .015) or SRMR (ΔSRMR ≤ .010) were considered acceptable (Chen 2007).

To evaluate the stability of CTQ subscale scores across time, we included correlations between latent factors at T0 and T1 in the longitudinal CFA model, providing estimates of rank-order stability based on latent variables. In addition, mean-level differences across time were captured by freely estimating latent means at T1, while fixing the means at T0 to zero, thereby allowing the model to test for systematic changes in latent factor levels over time. As a supplementary analysis (see Supplemental Material), we also estimated intraclass correlations (ICCs) for CTQ total and subscale scores using random-intercept mixed-effects models based on all available observations across T0 and T1, allowing the inclusion of participants with data at only one measurement occasion. This approach provides variance-component ICCs that are robust to incomplete longitudinal data. Alluvial plots were used to visualise transitions between CTQ severity categories from T0 to T1 and were therefore based on participants with observed data at both time points for the respective scale. To examine whether the time gap between assessments influenced change in CTQ scores, regression analyses of the CTQ total score were conducted in participants with data at both time points, as change scores are only defined for paired observations.

Correlation analyses were performed to explore manifest associations between CTQ scores and measures of symptom severity (BSL-23) at both time points. To investigate longitudinal associations between CTQ scores and borderline symptom severity (BSL-23), bivariate latent change score (LCS) models were estimated (Kievit et al., 2018). The models capture both within-construct change (self-feedback effects) and cross-domain influences across time. Specifically, the models included CTQ scores (total or each subscale) and BSL-23 total scores at baseline (T0) and follow-up (T1), allowing latent change factors (dCTQ, dBSL) to be regressed on their respective T0 values (self-feedback) as well as on the other construct’s T0 value (cross-lagged paths). Residual covariances were estimated between both baseline values and latent change scores. Missing data were handled using FIML and robust standard errors were computed using the MLR estimator. To control the false discovery rate (Benjamini & Hochberg, 1995), p-values were adjusted across the 18 pre-specified hypothesis tests from the bivariate LCS models (CTQ_T0 → ΔBSL, BSL_T0 → ΔCTQ, and ΔCTQ–ΔBSL covariance for each CTQ scale: total, EA, PA, SA, EN, PN). Both unadjusted and FDR-adjusted p-values are reported.

3. Results

3.1. Missing data

The logistic regression using standardised continuous predictors for missing data at T1 showed that older age was associated with significantly lower odds of missing CTQ data (OR = 0.667, 95% CI [0.520, 0.848], p = .001). Gender significantly predicted missingness, with females being more likely to have missing data at T1 compared to males (OR = 2.114, 95% CI [1.066, 4.359], p = .036). Higher CTQ total scores at baseline also predicted greater odds of missing data (OR = 1.337, 95% CI [1.064, 1.692], p = .014). BSL symptom severity at T0 did not significantly predict missingness (OR = 0.905, 95% CI [0.722, 1.132], p = .383).

3.2. Descriptive statistics

Participants who had consented to the study (T0: n = 417; T1: n = 196) but had missing data in the CTQ were excluded for the descriptive statistics only. Therefore at T0, data on n = 360 participants (88% female) were reported, with a mean age of 28.87 years (SD = 8.13, range = 15–55 years). At baseline (T0), 296 participants met criteria for current BPD (84%), while 56 participants were classified as remitted (16%). At follow-up (T1), 100 participants met criteria for current BPD (51%), and 94 participants were classified as remitted (49%). Descriptive analyses for both time points are presented in Table 1.

Table 1.

Sample description for CTQ, its subscales and the BSL-23 for T0 and T1.

	T0		T1
	n	M (SD)	n	M (SD)
CTQ (total)	360	61.06 (19.56)	195	60.41 (18.50)
CTQ EA	360	16.47 (5.59)	195	16.13 (5.26)
CTQ PA	360	8.82 (5.11)	195	8.48 (4.58)
CTQ SA	360	8.66 (5.78)	195	9.02 (5.52)
CTQ EN	360	17.07 (5.50)	195	16.72 (5.44)
CTQ PN	360	10.04 (3.93)	195	10.06 (3.87)
CTQ Min	357	0.62 (0.54)	195	0.54 (0.50)
BSL-23	355	1.67 (0.91)	194	1.32 (0.94)

Note: Some data were missing for certain participants; therefore, the sample size is reported for each variable. CTQ was quantified as sum score and BSL-23 as mean score. CTQ = Childhood Trauma Questionnaire, BSL-23 = Borderline Symptom List-23, EA = Emotional Abuse, PA = Physical Abuse, SA = Sexual Abuse, EN = Emotional Neglect, PN = Physical Neglect, Min = Minimisation/Denial score. There are six grades of symptom severity for the BSL-23 mean score: none or low: 0–0.28, mild: 0.28–1.07, moderate: 1.07–1.87, high: 1.87–2.67, very high: 2.67–3.47 and extremely high: 3.47–4 (Kleindienst et al., 2020). Severity categories for the CTQ subscales were defined according to the cut-off scores proposed by Bernstein and Fink (1998). For Emotional Abuse, scores of 5–8 indicate none/minimal, 9–12 low to moderate, 13–15 moderate to severe, and ≥16 severe to extreme exposure. For Physical Abuse, scores of 5–7 indicate none/minimal, 8–9 low to moderate, 10–12 moderate to severe, and ≥13 severe to extreme exposure. For Sexual Abuse, scores of 5 indicate none/minimal, 6–7 low to moderate, 8–12 moderate to severe, and ≥13 severe to extreme exposure. For Emotional Neglect, scores of 5–9 indicate none/minimal, 10–14 low to moderate, 15–17 moderate to severe, and ≥18 severe to extreme exposure. For Physical Neglect, scores of 5–7 indicate none/minimal, 8–9 low to moderate, 10–12 moderate to severe, and ≥13 severe to extreme exposure. Şar et al., (2021) reported CTQ Min scores of 0.2 (SD = 1.1) in psychiatric patients, 0.3 (SD = 0.6) in patients with dissociative disorder and 0.9 (SD = 1.1) in non-clinical participants.

3.3. Stability of childhood trauma questionnaire

At baseline (T0), the trauma subscale with the highest percentage of participants with extreme exposure was Emotional Abuse (n = 211, 59%), followed closely by Emotional Neglect (n = 191, 53%). Fewer participants reported extreme levels of Physical Neglect (n = 83, 23%), Sexual Abuse (n = 70, 19%), and Physical Abuse (n = 69, 19%). At follow-up (T1), the trauma subscales with the highest percentage of participants with extreme exposure were Emotional Abuse (n = 107, 55%) and Emotional Neglect (n = 94, 48%), followed by Sexual Abuse (n = 40, 21%), Physical Neglect (n = 40, 21%), and Physical Abuse (n = 33, 17%).

The CTQ 25-item version demonstrated excellent internal consistency at both timepoints for Emotional Abuse (ω_T0 = .87 and ω_T1 = .86), Physical Abuse (ω_T0 = .90 and ω_T1 = .89), Sexual Abuse (ω_T0 = .97 and ω_T1 = .95) and Emotional Neglect (ω_T0 = .91 and ω_T1 = .92) also showed high consistency, whereas Physical Neglect showed poor values (ω_T0 = .57 and ω_T1 = .62).

The CFA results for both T0 (scaled values: CFI = .910, TLI = .899, RMSEA = .071, SRMR = .060) and T1 (scaled values: CFI = .891, TLI = .876, RMSEA = .079, SRMR = .073) indicated an acceptable model fit for the five-factor structure. While the CFI and TLI values are close to or just at the conventional cutoff of .90, the RMSEA (between .071 and .079) and SRMR (between .06 and .07) suggest that the model provided an acceptable representation of the data at both time points. These indices suggest that while the model captures the overall structure reasonably well, some degree of misfit remains. Inspection of modification indices (MIs) revealed substantial localised misfit. For instance, at both T0 and T1, high MIs suggested correlated residuals between items within the same subscales (e.g. in the Physical Abuse subscale, MI at T0 = 73.03; MI at T1 = 104.98), and among Emotional Neglect items (e.g. MI at T0 = 47.31; MI at T1 = 38.69), indicating possible item redundancy or shared method variance. Notably, some reverse-coded items showed large MIs for cross-loadings onto Emotional Neglect or Sexual Abuse factors, raising concerns about item ambiguity and potential factor overlap. Additionally, the latent variable covariance matrix at T0 included very high inter-factor correlations, especially between Emotional Neglect and Physical Neglect (r = 0.972), which may reflect poor discriminant validity between these constructs. Standardised residual correlation matrices revealed several item pairs with residual correlations > .10, indicating remaining unexplained covariance beyond what was accounted for by the latent structure. Taken together, while the hypothesised five-factor model demonstrates adequate global fit, the results suggest notable localised strain, particularly within the Physical Neglect subscale and across items with conceptual or linguistic overlap.

The configural model provided acceptable baseline fit to the data (χ²(1100) = 2265.80, p < .001; scaled CFI = .906; scaled RMSEA = .048; SRMR = .074). Imposing equality constraints on factor loadings (metric invariance) did not significantly degrade model fit (χ²(1120) = 2292.62, p < .001; scaled CFI = .905; RMSEA = .048; SRMR = .075; ΔCFI = −.001; ΔRMSEA = .000; ΔSRMR =  + .001), all well within recommended cutoffs (ΔCFI ≤ .010; ΔRMSEA ≤ .015; ΔSRMR ≤ .010). Further constraining item intercepts (scalar invariance) resulted in a statistically significant, yet substantively small, reduction in fit (χ²(1145) = 2323.98, p < .001; scaled CFI = .905; RMSEA = .048; SRMR = .075; ΔCFI = .000; ΔRMSEA = .000; ΔSRMR = .000). Given that ΔCFI and ΔRMSEA remained within acceptable limits, these results support full scalar invariance of the five-factor CTQ model across time (T0 and T1).

In the scalar-invariant longitudinal CFA, all CTQ domains showed very high latent retest stability across the two measurement occasions. Latent correlations between the same factors at T0 and T1 were uniformly large, ranging from r = .81 to .88, indicating strong rank-order consistency of trauma-related constructs over time. At the item level, retest correlations were more heterogeneous, ranging from small to large (r = −.14 to .78), suggesting that while individual item responses fluctuate, the underlying latent trauma dimensions remain highly stable. The CTQ subscales were also strongly intercorrelated at both time points, reflecting a substantial shared trauma component. Correlations among latent factors at T0 ranged from r = .31 to .97, and at T1 from r = .25 to .94. In particular, Emotional Neglect and Physical Neglect were almost perfectly correlated at both occasions (T0: r = .97; T1: r = .94), indicating limited discriminant validity between these two neglect dimensions in this sample. For the mean-level change, only Sexual Abuse showed a significant increase over time (ΔM = 0.17, SE = 0.05, z = 3.34, p = .001), whereas no significant mean-level changes were observed for Emotional Abuse (ΔM = 0.07, p = .248), Physical Abuse (ΔM = 0.02, p = .588), Emotional Neglect (ΔM = −0.04, p = .388), or Physical Neglect (ΔM = 0.03, p = .287). Supplemental Table S2 provides standardised factorloadings, Table S3 factor correlations for the scalar-invariant model of the CTQ at T0 and T1 and Figure S1 estimated ICCs for CTQ total and subscale scores.

3.4. Associations between BPD symptom severity and changes in CTQ scores

Strong manifest intercorrelations among subscale scores were also observed, especially between Emotional Abuse and Emotional Neglect (T0: r = 0.69; T1: r = 0.67), and between Emotional Neglect and Physical Neglect (T0: r = 0.65; T1: r = 0.62). Associations between BSL-23 and individual CTQ subscales were generally small to moderate (e.g. BSL-23 and Emotional Abuse at T0: r = .21, T1: r = 0.30). Only the total CTQ score was significantly correlated with the BSL-23 at T0 (r = .16), whereas at T1 no CTQ–BSL-23 correlations reached significance.

The bivariate latent change score models revealed that higher baseline CTQ total scores were weakly associated with subsequent increases in symptom severity over time (β =  0.154, p  =  .042, p_FDR= .084), whereas the reverse path, from baseline symptoms to CTQ change, was not significant. Subscale analyses showed some divergence (see Table 2): Emotional Abuse (β =  0.169, p  =  .022, p_FDR= .049) and Physical Neglect (β =  0.162, p =  .021, p_FDR= .049) similarly predicted increases in symptoms, whereas only Physical Abuse showed a significant reverse effect, with higher baseline symptoms predicting reductions in CTQ scores (β =  −0.186, p =  .001, p_FDR= .010). While directional effects were generally weak or absent across subscales, change scores for CTQ and BSL were significantly associated for all subscales (β =  .23–.33, ps ≤ .021, ps_FDR ≤ .049), except for Physical Neglect (β = .16, p = .079, p_FDR= .142), suggesting that changes in symptom burden and trauma perception tend to co-occur over time. Figure 1 shows the LCS model for the CTQ total.

Table 2.

Standardised estimates (β) and p-values from Bivariate Latent Change Score models for CTQ subscales and symptom change.

Scale	CTQ_T0 → ΔBSL β [CI]	CTQ_T0 → ΔBSL (p)	BSL_T0 → ΔCTQ β [CI]	BSL_T0 → ΔCTQ (p)	ΔCTQ ↔ ΔBSL β [CI]	ΔCTQ ↔ ΔBSL (p)
Total	.154 [.010, .200]	.042	−.071 [−.209, .066]	.326	.327 [.177, .477]	<.001
EA	.169 [.030, .308]	.022	−.034 [−.169, .101]	.625	.233 [.091, .374]	.004
PA	.089 [−.062, .239]	.248	−.186 [−.297, –.076]	.001	.282 [.107, .457]	.020
SA	.043 [−.124, .210]	.613	−.063 [−.179, .052]	.286	.258 [.100, .415]	.003
EN	.100 [−.036, .235]	.156	.069 [−.069, .206]	.325	.257 [.110, .404]	.002
PN	.162 [.029, .296]	.021	−.065 [−.218, .088]	.412	.162 [−.022, .346]	.079

Note: EA = Emotional Abuse, PA = Physical Abuse, SA = Sexual Abuse, EN = Emotional Neglect, PN = Physical Neglect, CI = 95% confidence interval.

Figure 1.

Structural equation model of the bivariate latent change score analysis examining the longitudinal associations between childhood trauma (CTQ total score) and borderline symptomatology (BSL-23). Latent change factors (ΔCTQ, ΔBSL) were defined as the residualized difference between T1 and T0 scores. Paths represent standardised estimates; dashed arrows represent fixed parameters; solid arrows represent freely estimated paths. The model includes self-feedback paths, cross-domain coupling, and covariances between baseline levels and change scores.

4. Discussion

The aim of this study was to examine the psychometric properties of the CTQ, particularly its long-term stability in patients with BPD. The patients were assessed over a period ranging from 4–11 years. We hypothesised (1) that the CTQ scale scores are highly consistent over time, indicating that the subjective perception of childhood trauma is relatively stable, and (2) that individual changes in perceived CT are closely linked to changes in BPD symptom severity over time.

4.1. CT in patients with BPD

The most frequently reported traumatic experiences in the present sample were Emotional Abuse (59% at T0) and Emotional Neglect (53% at T0). These findings are consistent with previous research identifying Emotional Abuse and Emotional Neglect as among the most commonly reported forms of CT in individuals with BPD (Igarashi et al., 2010). Moreover, both Emotional Abuse and Emotional Neglect have been identified as significant predictors in the development of BPD (Fossati et al., 2016; Wu et al., 2022), and are frequently found to be associated with the disorder (Lobbestael et al., 2010; Zanarini et al., 1997). These results are in line with theoretical models of BPD development, such as those proposed by Linehan (1993) and Crowell et al., (2009). Specifically, childhood experiences of Emotional Abuse and Emotional Neglect may reflect an invalidating environment – a central component in Linehan’s biosocial theory of BPD. Similarly, in the extended developmental model by Crowell et al. (2009), emotionally neglectful and abusive early caregiving experiences are conceptualised as key environmental risk factors that interact with biological vulnerabilities, ultimately contributing to the emergence of BPD symptomatology. However, this assumption was only partially supported in the present study, as only Emotional Abuse and Physical Neglect at baseline (T0) were associated with subsequent changes in symptom severity over time. This may suggest that symptom trajectories in BPD are not predetermined by early trauma alone, but are likely influenced by multiple additional factors across development. Future research should therefore consider a broader range of potential influences on symptom change over time.

4.2. Long-term stability of the CTQ

The ICCs observed for the total score and subscales of the CTQ suggest moderate stability for Emotional Neglect, Physical Neglect and Emotional Abuse, and high stability for the total score, Physical Abuse and Sexual Abuse. These findings are consistent with previous research demonstrating similarly robust stability indices across both clinical and non-clinical samples (Badenes-Ribera et al., 2024; Xiang et al., 2021; Georgieva et al., 2021; Aizpurua et al., 2024). For example, Xiang et al., (2021) reported ICCs exceeding .75 over an 11-month period, with the exception of the Emotional Neglect subscale, which showed a notably lower value (ICC = .54). Furthermore, our supplementary analyses did not reveal any significant influence of the time interval between baseline and follow-up assessments on absolute changes in, or autoregressive effects of, CTQ total scores. This suggests that the observed long-term stability is not merely a function of shorter or longer time intervals between assessments.

The lower ICCs of the subscales Emotional Neglect, Physical Neglect and Emotional Abuse, as well as the relatively high intercorrelations between Emotional Abuse and Emotional Neglect, as well as Emotional Neglect and Physical Neglect, could be an indication of a lower discriminatory power of the individual constructs. Emotional Neglect and Emotional Abuse encompass subtle, subjective, and context-dependent experiences – such as lack of warmth, emotional rejection, or insufficient support (Bernstein et al., 2003) – that are not easily distinguishable, either in recollection or in perception by respondents. This complexity may contribute to measurement variability and is further compounded by evidence linking these subscales to current psychological states, including depressive symptoms (Dehn & Beblo, 2019), neuroticism (Fujimura et al., 2023; Ogle et al., 2017), and mood (Miranda & Kihlstrom, 2005). Such associations suggest that reports of emotional trauma may be influenced not only by past experiences but also by present affective conditions (Danese & Widom, 2020). Moreover, developmental and temporal factors appear to shape how individuals report traumatic experiences. Research indicates that younger individuals tend to distinguish more clearly between emotional and physical forms of trauma, potentially due to the recency of events and greater contextual specificity. In contrast, adults recalling more distant childhood experiences often report emotional and physical trauma as co-occurring, which may reflect a more integrated or generalised memory structure (Paivio & Cramer, 2004). This shift over time could be due to memory degradation or a broader interpretive framework for understanding early trauma.

4.3. Associations between BPD symptom severity and changes in CTQ scores

While correlations between CTQ and BSL-23 scores were generally small to moderate, they suggest a meaningful association between childhood trauma reports and symptom severity. In the full sample, small to moderate cross-sectional correlations were found between BSL-23 and CTQ subscales, though only the total CTQ score showed a significant association with BSL-23 in the longitudinal subsample at baseline, and no significant correlations emerged at follow-up. These comparatively weaker associations may, to some extent, reflect the reduced symptom severity among participants willing to engage in long-term follow-up. When examining change over time, latent change score models revealed that Emotional Abuse and Physical Neglect significantly predicted increases in BPD symptom severity. Notably, while reverse effects were generally absent, higher baseline BPD symptom severity predicted a decrease in self-reported Physical Abuse over time, hinting at a possible retrospective reappraisal of adverse experiences in the context of symptom change. This finding may tentatively point to shifts in how past abusive experiences are appraised or emotionally weighted over time, rather than to changes in the underlying autobiographical memory (Leer et al., 2014; Quirk et al., 2010). Such reappraisal processes may occur in parallel with symptom change and reduced emotional distress.

Importantly, change scores for the CTQ and BSL-23 were significantly correlated for all subscales, with the exception of Physical Neglect. This pattern supports the notion that fluctuations in trauma reporting co-occur with changes in symptom burden. Together, these findings partially support the hypothesis that CTQ stability is affected by shifts in BPD symptom severity, especially in terms of parallel change, while providing limited evidence for a direct influence of symptom severity on retrospective trauma reports. One potential mechanism underlying these parallel changes is that fluctuations in emotional states associated with symptom worsening (e.g. increased negative affect, distress, or affective instability) may influence the accessibility and appraisal of autobiographical memories (Faul & LaBar, 2023), thereby shaping retrospective reports of childhood trauma. Consistent with prior research (Yang et al., 2018; Jiang 2024; Shen et al., 2023), moderate associations between CTQ and BSL-23 were more pronounced in specific subscales such as Emotional Abuse. However, heterogeneous findings across studies likely reflect differences in sample composition and methodology. Future research with more homogeneous clinical samples and refined temporal resolution is needed to clarify how fluctuations in psychological state shape the stability of self-reported childhood trauma.

4.4. Internal consistency of the CTQ

The confirmatory factor analysis supported the five-factor structure with acceptable global fit, although localised misfit and redundancy within certain subscales, especially Physical Neglect. warrant further investigation. The presence of high modification indices and substantial residual correlations among item pairs suggests that some items may not uniquely reflect their intended constructs. For example, the repeated indications of cross-loadings and correlated errors within the neglect subscales raise concerns about item redundancy, conceptual overlap, or even translation-related ambiguities, if applicable. The very high correlation between Emotional Neglect and Physical Neglect may indicate poor discriminant validity between these constructs. Factor-analytic investigations frequently reveal overlap between the constructs of Emotional Abuse, Emotional Neglect and Physical Neglect, while the Sexual Abuse and the Physical Abuse subscales tends to emerge as distinct and clearly defined factors (Scher et al., 2001). Nevertheless, it is worth noting that the Emotional Neglect and Emotional Abuse subscales of the CTQ have been shown to be more sensitive in detecting emotional trauma than interview-based assessments, highlighting their potential utility despite psychometric limitations (Spinhoven et al., 2014). Beyond measurement considerations, emerging network-analytic evidence suggests that emotional abuse may function as a central hub in borderline personality disorder, linking other forms of childhood trauma to core BPD characteristics (Schulze et al., 2022; Macchia et al., 2025).

The Physical Neglect subscale, in particular, remains problematic. Aizpurua et al., (2024) reported inadequate internal consistency (α = .57), a finding that aligns with previous research highlighting conceptual and structural limitations inherent in the original design of this subscale (Aizpurua et al., 2024). Supporting this, multiple studies have confirmed the five-factor model of the CTQ but reported poor fit for the Physical Neglect factor (Schulz et al., 2014; Karos et al., 2014). For instance, Schulz et al., (2014) found an α of just .40 for this subscale in a large German population sample, whereas all other subscales demonstrated strong internal consistencies ranging from .80 to .92. This pattern suggests that Physical Neglect may reflect a heterogeneous bundle of experiences rather than a coherent latent dimension, which reduces its psychometric precision and limits the interpretability of findings involving this subscale. Future assessments might benefit from a broader and more differentiated set of indicators to better capture the construct of physical neglect.

4.5. Limitations

At both assessment points, mean BSL-23 scores in the present sample fell within the moderate range (0.7–1.7; see (Kleindienst et al., 2020) for severity classification). In contrast, previous studies employing the BSL-23 in samples of individuals with BPD have reported higher levels of symptom severity, with mean scores typically falling in the high range (1.8–2.2; Bohus et al., 2009; Kleindienst et al., 2020; Nicastro et al., 2016) or even in the very high range, as evidenced by Abel et al. (2025), who reported a mean BSL-23 score of 2.9 (Abel et al., 2025). These discrepancies may suggest that participants in the current study exhibited relatively lower symptom severity, which could potentially influence the observed stability of CTQ scores. Future research should consider incorporating clinician-administered interviews to more accurately capture the extent of symptom severity and its potential impact on retrospective self-report measures such as the CTQ.

The present analyses did not account for potential confounding factors such as engagement in psychotherapy during the follow-up period, intervening life events, or exposure to new traumatic experiences. These factors may affect clinical symptomatology, particularly intrusive memories (Astill Wright et al., 2021); however, existing evidence suggests that pharmacological and psychotherapeutic interventions primarily modify the emotional appraisal and affective salience of traumatic memories rather than erasing the memory of the traumatic event itself (Leer et al., 2014; Quirk et al., 2010). Importantly, the current study focused on the psychometric stability of the CTQ, which appeared to be high despite the lack of control for these potential confounding factors.

This study was embedded in a larger research cohort established by KFO 256, in which multiple additional self-report questionnaires (e.g. Barratt Impulsiveness Scale Version 11, (Patton et al., 1995); Difficulties in Emotion Regulation Scale (Gratz & Roemer, 2004); HEXACO Personality Inventory-Revised (Ashton and Lee 2009); Personality Inventory for DSM-5 (Krueger et al., 2012)) and structured or semi-structured clinical interviews (e.g. Structured Clinical Interview for DSM-IV (Beesdo-Baum et al., 2019) and Semi-Structured Interview for Personality Functioning DSM-5 (Zettl et al., 2019)) were administered during the assessment period. The extensive multi-instrument assessment may have introduced methodological biases, including question order or context effects that influence responses to subsequent measures (Oldendick, 2008), survey conditioning effects due to repeated participation (Duan et al., 2007), and mere-measurement effects, whereby the act of assessment itself can alter perceptions or responses (Long et al., 2025). In addition, symptom-focused assessments may have increased negative affect, thereby enhancing the accessibility of negative autobiographical material via mood-congruent recall and potentially influencing retrospective trauma reports (Faul & LaBar, 2023). Consistent with evidence that trauma-related cues are more readily activated and processed in individuals with PTSD (Ehring & Ehlers, 2011; Sündermann et al., 2013), symptom- or trauma-proximal questions may have functioned as cognitive primes. Future studies should explicitly examine and, where possible, control for such assessment-related effects.

Moreover, due to the availability of only two measurement points, our design does not allow for a statistical separation of measurement error (reliability) and true change in the construct (stability). Future studies employing at least three time points are needed to disentangle these components more effectively.

Despite these limitations, the overall psychometric performance of the CTQ remains robust. The acceptable model fit indices and consistently high stability coefficients across most subscales support the conclusion that the CTQ is aconsistent tool over a long-term time interval for assessing a broad range of CT. Continued refinement of specific subscales –particularly Physical Neglect– may further enhance the instrument’s validity and clinical utility.

5. Conclusion

The present findings support the temporal stability of CTQ scores, with a very long-time interval between the survey dates and even within a diagnostically heterogeneous clinical population characterised by frequent fluctuations in symptom severity, such as individuals with BPD. This underlines the CTQ’s robustness as a retrospective measure of CT and contributes to the broader discussion regarding the stability of self-report questionnaires in both clinical and research settings.

While CTQ scores showed overall long-term stability, the observed associations with symptom change suggest that retrospective trauma reports may, to some extent, fluctuate alongside changes in psychological state. These findings indicate that the CTQ remains a generally consistent instrument, even in clinical populations with varying symptom trajectories. Thus, the CTQ can be considered a valuable tool in longitudinal research and clinical assessment, while also warranting careful interpretation in the context of symptom dynamics.

Funding Statement

This work was supported by German Research Foundation (DFG) [grant number 190034061]; Endorsed by German Center for Mental Health (DZPG).

Disclosure statement

No potential conflict of interest was reported by the author(s).

Declaration of generative AI and AI-assisted technologies in the writing process

During the preparation of this work the authors used ChatGPT (version GPT-4, accessed via https://chat.openai.com) developed by OpenAI solely in order to language refinement. No content was generated or substantively written by the tool, and all ideas, analyses, and conclusions are the authors’ own. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.

Data availability statement

The participants of this study did not give written consent for their data to be shared publicly, so due to the sensitive nature of the research supporting data is not available.

Supplemental Material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/20008066.2026.2646755.