Do Trajectories of Self- and Interpersonal Functioning Identify a Core Underlying Developmental Pathway for Personality Pathology in Late Adolescence and Early Adulthood?

Abstract

The Alternative Model for Personality Disorders (AMPD) is a dimensional model of personality disorder that describes difficulties in self- and interpersonal functioning as the common core of personality disorder and its hypothesized developmental pathway. However, empirical evidence in support of this developmental pathway is lacking, and the unique developmental relevance of self- and interpersonal functioning for the emergence of personality pathology independent from internalizing and externalizing psychopathology remains unknown. The aim of this pre-registered study (osf.io/xqc4u) was to leverage data from the 17-year prospective Preschool Depression Study (PDS) including 348 participants, over-sampled for depression, to evaluate the respective predictive utility of self and interpersonal functioning for the later development of personality pathology (in a subsample of 187 participants). Self- and interpersonal functioning were operationalized with self and interpersonal items from the Preschool Age Psychiatric Assessment (PAPA) and Child and Adolescent Psychiatric Assessment (CAPA). The final model included 7 timepoints (mean [SD] age 4.5[0.8] – 12.2 [1.1]) for self-functioning and 4 timepoints (age 8 onwards) for interpersonal functioning. Personality pathology in late adolescence was operationalized with the Borderline Personality Disorder (BPD) Features Scale for Children at timepoints 9-10 (mean [SD] age 16.3 [1.1] – 18.6 [1.2]; n = 187). When considered separately, bivariate growth curve modeling demonstrated significant intercepts and slopes for self-functioning (standardized intercept estimate = 0.206, standardized slope estimate = 0.357), and marginally significant slopes for interpersonal trajectories with BPD (standardized slope estimate = 0.233). When considered together, using a reduced model for trajectories, multivariate growth curve modeling showed predictive utility for self-functioning (standardized intercept estimate = 0.436, standardized slope estimate = 0.301) only, independent of internalizing and externalizing symptoms. Results point to the importance of incorporating explicit consideration of self-functioning in the conceptualization of personality, especially from a developmental point of view in service of prevention and early intervention for personality disorder.

General Scientific Summary

The results of this study highlight the predictive role of aspects of self-functioning for the development of personality disorder in late adolescence and early adulthood. It suggests that there may be a unique pathway related to self-functioning, independent of symptoms of internalizing and externalizing disorders, that could increase risk for the development of personality disorder.

Personality disorder is a lifespan disorder that typically onsets in adolescence and young adulthood (American Psychicatric Association, 2022; Newton-Howes et al., 2015; Sharp, 2022b). It affects up to 12.16% of the Western general population (Volkert et al., 2018) with rates in clinical populations estimated between 45-51% (Beckwith et al., 2014). Personality disorder is associated with significant functional impairment and poor long-term outcomes (Hastrup, Jennum, et al., 2019; Hastrup, Kongerslev, et al., 2019; Ostby et al., 2014), including high rates of suicide. For instance, a meta-analysis reported that the suicide rate among those with borderline personality disorder (BPD) is 45 times higher than the suicide rate in the general population and 20 times higher than the suicide rate among those with depression (Chesney et al., 2014). Despite its association with a high burden of disease, morbidity, and premature mortality (Bjorkenstam et al., 2015), research on the developmental precursors leading to personality disorder lags far behind that of other psychiatric disorders, impeding efforts towards prevention and early intervention (Biskin, 2015; Chanen & Nicol, 2021; Chanen et al., 2022).

Dimensional models of psychopathology, such as the Alternative Model for Personality Disorders (AMPD) in Section III of the Diagnostic and Statistical Manual for Mental Disorders (DSM-5; American Psychiatric Association, 2013) and the new International Classification of Diseases (ICD-11) system for diagnosing personality disorder, have been identified as a promising catalyst for the acceleration of research into the developmental precursors of personality disorder (Eggermont et al., 2023; Franssens et al., 2024; Sharp & Wall, 2017, 2021). Both systems define personality disorder, first, through the concept of Level of Personality Functioning (LPF; Criterion A of the AMPD), which refers to an individual’s self-functioning (identity and self direction) and interpersonal functioning (empathy and intimacy). The LPF is viewed as the core dimension of personality pathology shared by all manifestations or “flavors” of personality pathology and is evaluated across five levels, with 0 indicating healthy personality functioning and 4 indicating severely impaired self- and interpersonal functioning. The AMPD and ICD-11 regard the LPF as conditional to the diagnosis of personality disorder, but then requires (in the AMPD) or recommends (in the ICD-11) that clinicians also evaluate the manifestation of individuals’ personality dysfunction across five maladaptive trait domains (in the AMPD, Criterion B of the personality disorder diagnosis): Negative affectivity, Detachment, Antagonism, Disinhibition, and Psychoticism (or Anankastia in the ICD-11).

The expectation is that personality disorder does not emerge de novo, but that precursors of a developmental pathway should be observable in pre-adolescence and perhaps even in very young children, with continued predictive power as children age into adolescence (Cohen et al., 2005; DeClercq & Sharp, 2020; Sharp & DeClercq, 2020). Accordingly, a substantial literature supports the association between maladaptive traits (as assessed through temperament) and later personality disorder (see Bozzatello et al., 2019; De Clercq et al., 2014; Shiner, 2015; Skabeikyte & Barkauskiene, 2021; Stepp et al., 2016 for reviews). Much less research has been conducted to evaluate precursors of personality disorder in terms of LPF. Specifically with regard to precursors of maladaptive self-functioning, theories from diverging perspectives postulate that for personality disorder to onset in adolescence and/or young adulthood, anomalies in self-development must be observable already in pre-adolescence (Fonagy et al., 2002; Kernberg et al., 2000; Linehan, 1993). Indeed, developmental research (see Harter, 2012 for a review) has identified several precursors of self-development in concepts such as self-awareness, self-esteem, self-concept, self-regulation, self-attribution, self-consciousness, self-perception, autobiographical memory, and so on; however, we are aware of only three longitudinal studies that have evaluated the predictive utility of aspects of self-functioning for the development of later personality pathology – two of which evaluated self-control (Belsky et al., 2012; Hallquist et al., 2015), and one that evaluated initiative-taking (Boone et al., 2022). In addition, studies of the normative development of self describe identity synthesis as the culmination of self-development and the major developmental task of adolescence (Bogaerts et al., 2021; Branje, 2022; Erikson, 1959; Harter, 1999; Harter, 2012; McAdams, 2015a). Accordingly, we note that several cross-sectional studies have established evidence for identity diffusion as an index of maladaptive self-functioning in youth with personality disorder. For instance, Westen et al. (2011) replicated adult findings showing links between identity disturbance in 139 adolescent patients and BPD – specifically, lack of normative commitment, role absorption, identity incoherence, and identity inconsistency. More recently, Sharp, Vanwoerden, et al. (2021) evaluated age-varying associations between identity diffusion and borderline pathology in a community-based sample of over 2,000 adolescents and showed that throughout adolescence, identity diffusion is significantly associated with borderline personality features, and that this association strengthens with increasing age. Using measures of narrative identity, Lind et al. (2019) demonstrated a significant association between narrative incoherence and borderline features in adolescent inpatients. Similar cross-sectional findings of a link between self-reported narrative identity and borderline features have also been demonstrated in a community sample of adolescents (Balzen et al., 2024).

An even stronger literature base supports a role for interpersonal precursors of the development of personality disorder – most often in the context of family relationships—demonstrating insecure, disorganized attachment patterns among those with BPD (Khoury et al., 2020; Newman-Morris et al., 2020) and the exacerbating effect of invalidating, negative interactions with parents (Brumariu et al., 2020; Hessels et al., 2022; Musser et al., 2018; Steele et al., 2019); for reviews see Bozzatello et al. (2019), Skabeikyte and Barkauskiene (2021), and Stepp et al. (2016). While adaptive interpersonal functioning related to parents is no doubt essential for healthy personality development, the importance of peers is increasingly recognized as an additional determining factor (Runions et al., 2021). Importantly, research has shown that adolescents typically increase their valuation of peer relationships, develop greater psychological distance from parents, and renegotiate boundaries and responsibilities in family relationships (Fuligni & Eccles, 1993; Hallquist et al., 2015; Steinberg et al., 2006). Significant changes in the quality of children’s social interactions with peers and the role of peers in self-concept formation begin during preschool and continue through middle childhood and adolescence (Crone et al., 2022), and it has been shown that peer relationships during childhood and adolescence exacerbate trait-like vulnerabilities like negative affectivity in the prediction of later personality pathology (Vanwoerden et al., 2022). Research furthermore demonstrates that children who experience bullying—as victims or perpetrators—may develop BPD (Wertz et al., 2020). Victimization is prospectively linked to BPD and associated symptoms such as self-harm and suicidal ideation (Geoffroy et al., 2016; Klomek et al., 2015; Wolke et al., 2012). Elevated relational aggression during childhood may also precede the development of later personality disorder (Cavicchioli et al., 2024; Runions et al., 2021). For example, Nelson et al. (2014) found that preschool girls’ physical and relational aggression (as rated by teachers) was significantly associated with self-reported BPD symptoms ten years later. Research using the Pittsburgh Girls Study also indicated trajectories of peer problems from childhood through adolescence differentiating adolescent BPD from MDD (Beeney et al., 2021). In several other unique samples, middle childhood friendship exclusivity and relational aggression emerged as key drivers of personality pathology, even after controlling for current depressive symptoms (Crick et al., 2005; Franssens et al., 2023; Haltigan & Vaillancourt, 2016; Underwood et al., 2011; Vaillancourt et al., 2014).

Taken together, the above findings highlight the pivotal role of both self and interpersonal factors in the development and progression of personality disorders beginning as early as the preschool period. However, a few gaps remain. First, most studies assessing associations of self-constructs with personality disorder have thus far been cross-sectional, and while some prospective studies have been conducted for interpersonal factors, most either covered a brief stage of development, or excluded younger age ranges.

Second, the relative contributions of self- vs. interpersonal functioning have not been studied together. LPF is considered a unidimensional severity criterion, so some may argue that the evaluation of the differential impact of its component parts should not be undertaken. However, studies support unique variance ascribed to self and interpersonal functioning even in adults (see e.g. Morey et al., 2022 for a review), and specifically in adolescents when utilizing two-factor structures to model covariance (e.g. Wu et al., 2024). Therefore, we cannot assume the unidimensionality of LPF, especially during the early stages of development when it is well-known that psychological structures, including personality, are less crystalized (De Clercq et al., 2017; DeClercq & Sharp, 2020). For instance, developmental models of personality development based on two decades of empirical findings (e.g. McAdams, 2015a) suggest that different aspects of personality (dispositional traits, characteristic adaptations, and the narrative self) develop at different times and interact with each other over time to culminate in integrated personality functioning. Supporting these assumptions, neurocognitive research suggests unique but reciprocal pathways for self- vs. interpersonal processing in childhood and adolescence (Crone et al., 2022). Building on this basic science, we have suggested that it is the changes in self-development that explain the onset of personality disorder in adolescents (e.g. Sharp, 2020; Sharp et al., 2018; Sharp & Wall, 2017, 2021). Specifically, based on research showing the relative stability of maladaptive traits in childhood and adolescence, we have argued that dispositional traits and characteristic adaptations constitute continuous (and relatively stable) aspects of personality, but that qualitative maturational changes in neurocognition related to social- and meta-cognitive capacities during adolescence allow for the narrative (reflective) self to emerge, facilitating the full assessment of personality disorder in adolescence. If this is so, the respective predictive power of self variables should be significant such that it drives other aspects of personality dysfunction. As yet, though, we do not have studies that have tracked the progression of self-development over time as it relates to later personality pathology.

A final gap that this study wishes to address is the unique variance explained by self and interpersonal developmental pathways beginning in early childhood for the emergence of personality pathology when controlling for internalizing and externalizing psychopathology. This question is important to justify the need for a personality dysfunction concept that is distinguishable from internalizing-externalizing spectra given the known overlap between internalizing-externalizing spectra and traditional personality disorder (see Kotov et al., 2021; Tackett, 2006 for reviews).

In summary, the AMPD, which is a dimensional model of personality disorder, describes difficulties in self- and interpersonal functioning as the common core of personality disorder and its hypothesized developmental pathway. However, the translational innovation of this thesis depends on (1) empirical evidence in support of this developmental pathway; (2) determination of the components with unique developmental relevance at different stages of childhood for the development of personality pathology; and (3) determining the extent to which self- and interpersonal functioning, which also characterize internalizing and externalizing disorders, identifies a core developmental pathway to personality pathology. To this end, we leveraged data from the 17-year prospective Preschool Depression Study (PDS; Luby et al., 2009) to evaluate the respective predictive utility of self and interpersonal functioning, starting at around age 4 and spanning 7 time points, in relation to the development of personality pathology, operationalized as BPD, at ages 16 to 18. We use BPD as a general indicator for personality pathology, not to reify its existence, but because emerging data suggest that BPD may be a good indicator of general personality dysfunction as represented by the LPF (Clark et al., 2017; Sharp et al., 2015; Wright et al., 2016). Despite growing dissatisfaction with traditional categorical notions of personality disorder including BPD (Clark, 2007; Krueger et al., 2018; Livesley et al., 1985; Morey et al., 1985; Skodol et al., 2011; Tyrer & Alexander, 1979; Widiger & Frances, 1985; Widiger & Trull, 2007), these data show that BPD is the only traditional personality disorder that exclusively loads onto a general factor of personality dysfunction without additional variance explained in a specific factor, as compared with other personality disorders that are better represented by specific factors. Against this background, BPD offers a good proxy for general personal disorder and therefore serves as the general indicator for later personality dysfunction in our study. We hypothesized that (1) high levels of impairment in self- and interpersonal functioning over time will be associated with high levels of borderline pathology (as an index of general personality pathology) at follow-up; (2) the contribution of self-functioning to personality dysfunction will increase over time as children age into adolescence – with this we mean that self-functioning may be a stronger predictor of later personality functioning than interpersonal functioning; and (3) high levels of impairment in self- and interpersonal functioning over time will associate with personality pathology even when controlling for levels of internalizing and externalizing psychopathology.

Method

Participants and Procedures

The study protocol was approved by the appropriate institution’s human subjects review committee and was preregistered (osf.io/xqc4u; Boone et al., 2024). Data and code for the study are not publicly available but may be requested from the second author. We report how we determined our sample size, all data exclusions (if any), all manipulations, and all measures in the study.

This study used data from the 17-year longitudinal Preschool Depression Study (see Figure 1 for study design).

Figure 1. Preschool Depression Study Flowchart.

Initially, 306 children between the ages of 3 to 6 years (Timepoint 1) were recruited from daycare and primary care centers in the St. Louis region. The original cohort intentionally oversampled preschoolers with depressive symptoms (identified by the Preschool Feelings Checklist; Luby et al., 2002) and included smaller groups of healthy preschoolers and preschoolers with disruptive symptoms (Luby et al., 2009). Approximately 7 years later, 42 additional healthy comparison children were enrolled (Timepoint 6), bringing the total sample size to 348 participants.

All participants were invited to return for assessments approximately every one to two years for up to 10 total assessment waves. Among these 348 participants, 216 had previously completed MRI scanning during Timepoints 6-8 and were therefore invited to participate in Timepoints 9 and 10, when BPD features were assessed. Of these 216 invited participants, 187 completed at least one follow-up at Timepoint 9 or 10. Only six participants in the entire sample of 348 never experienced a depressive symptom at any wave. By the final assessment at Timepoint 10, 179 participants had met criteria for a lifetime diagnosis of major depressive disorder, and 169 had not.

Because Timepoint 8 included only 97 participants, it was collapsed with Timepoint 7 for analytic purposes (Figure 1). The 187 participants with BPD data (out of a planned N=216) did not differ significantly from the remaining 161 participants on age, sex, or race/ethnicity (see Table 1). Given that the total sample of 348 exceeds commonly recommended minimums for growth modeling (Curran et al., 2010), a growth curve analysis was deemed appropriate. Missing follow-up data were handled using established methods described in the Data Analytic Strategy section.

Table 1. Characteristics of Participants with Complete vs. Missing Data on BPD features.

	Complete (n=187)		Missing (n=161)		Complete vs. Missing
	%	N	%	N	χ2	p
Female sex	50.8	95	45.3	73	1.03	.309
Race					0.35	.838
White	54.0	101	51.5	83
Black	34.2	64	34.8	56
Other	11.8	22	13.7	22
	Mean	SD	Mean	SD	t	p
Baseline age	4.50	0.79	4.40	0.81	1.04	.301

Measures

Self-functioning

As per pre-registration, our aim was to derive a self-functioning sum score using items from the Preschool Age Psychiatric Assessment (PAPA; Egger & Angold, 2004), the Child and Adolescent Psychiatric Assessment (CAPA; Angold & Costello, 2000), and the Test of Self-Conscious Affect for Children (TOSCA-C; Tangney, 1990) shame scale.

The PAPA and CAPA are well-validated semi-structured diagnostic interviews assessing a broad range of child psychopathology, peer relationships, and family context (Angold & Costello, 2000; Egger et al., 2006). When children were ages 3.0-7.11, caregivers were interviewed using the PAPA. When children were ages 8.0-8.11, caregivers were interviewed using the CAPA, and from age 9.0 onwards, in addition to caregiver CAPA assessments, children were interviewed directly with the child version of the CAPA. When both child and parent reports were available, the highest (most severe) rating endorsed was used in analyses. Specifically, we selected 14 PAPA/CAPA items assessing self-hatred, loneliness, feeling unloved, feeling sorry for oneself, excessive guilt, hopelessness, helplessness (CAPA only), loss of affect (CAPA only), anger and resentfulness, touchy and easily annoyed, boredom, thoughts of death, grandiosity, and excessive bragging. These items were selected for their correspondence to AMPD-defined impairment in self-functioning, including poor identity definition, inflated or deflated self-esteem, difficulties with emotion regulation, poorly defined goals or standards of behavior, impaired self-reflection, and emptiness. A full list of questions can be seen in the online supplemental materials (Table S1). The PAPA and CAPA items were rated as either 0 (absent) or 2 (present) or as 0 (absent), 2 (subthreshold), or 3 (threshold). The PAPA also includes a rating of 1 for certain items, indicating normative behavior. Based on response rates across response options which showed low endorsement in the maximum response option (i.e. 3), for all items used, a score of 2 or 3 indicated the presence of psychopathology or worse functioning. All PAPA/CAPA items were dichotomized by assigning a value of 0 (absent) to ratings of 0 or 1 and a value of 2 (present) to ratings of 2 or 3. If both child-reported CAPA items and their corresponding parent-reported CAPA items were available, the maximum score was utilized; if only one was available, the available score was used. Subsequently, we calculated the mean score of all items. Internal consistency of PAPA self-functioning at baseline was assessed using McDonald’s omega, which is especially advantageous for scales with heterogeneous items or multidimensional structures (Dunn et al., 2014). McDonald’s omega was 0.71. A McDonald’s omega value of 0.70 or higher was considered acceptable, indicating adequate reliability for group-level comparisons.

In addition to the PAPA and CAPA self items, as per pre-registration, we also intended to use items from the TOSCA-C. The TOSCA-C is a child-report, scenario-based measure assessing propensity towards shame, guilt, externalization of blame, detachment, pride in self, and pride in behavior. When completing the TOSCA-C, children read scenarios and rate how likely it is that they would respond in each of several ways on a Likert scale from 1 (not at all true) to 5 (extremely true). This construct of shame in the TOSCA-C is closely tied to AMPD-defined impairment in self-functioning, which includes the propensity toward low self-esteem easily influenced by events or external evaluation. The shame scale of the TOSCA-C has previously demonstrated good internal consistency, convergent validity, and discriminant validity from the guilt scale (Watson et al., 2016).

Interpersonal functioning

An interpersonal functioning variable was created using interpersonal items from the PAPA/CAPA as well as items from the MacArthur Health and Behavior Questionnaire-Parent Version (HBQ-P; Essex et al., 2002). The HBQ-P is a well-validated parent-report questionnaire assessing child psychopathology, peer relationships, and functioning. Caregivers completed the version for younger children (HBQ-P 1.0) at each timepoint when the child was younger than 9 years old, and the version for older children (HBQ-P 2.1) at each timepoint when the child was 9 years old or older. Items were either rated on a Likert scale from 0 (never or not true) to 2 (often or very true) or on a Likert scale from 1 (not at all) to 4 (very much). Items were reverse-scored as needed so that higher ratings indicated more psychopathology or worse functioning.

To derive a mean score of interpersonal functioning, we selected 8 PAPA/CAPA items and an additional 9 CAPA items not included on the PAPA assessing relational and physical aggression, conflict, lack of empathy, lack of interest in people, and poor relationship quality with adults, peers, siblings and others (please see online supplemental materials [Table S1] for final item set). Similar to our approach for self-functioning, we utilized items from the child-reported CAPA with the corresponding item from the parent-reported CAPA based on item availability, using the maximum score if both were available at each wave. Subsequently, we calculated the mean score from the total of 17 items for the final analysis. Internal consistency of PAPA interpersonal functioning at baseline was assessed using McDonald’s omega, showing a value of 0.70.

We also selected 34 HBQ-P 2.1 (older child) items and 23 corresponding HBQ-P 1.0 (younger child) items. Selected HBQ items assessed relational and physical aggression towards peers and being rejected by peers. These PAPA/CAPA and HBQ items were selected for their correspondence to AMPD-defined impairment in interpersonal functioning, which includes limited empathy, difficulty estimating the impact of behavior on others, limited mutuality, distance from others, and/or conflictual relationships.

Personality Disorder

The dependent variable in this study is personality disorder, operationalized through the assessment of BPD features, in acknowledgment that BPD is viewed as an adequate proxy for general personality disorder (Clark & Ro, 2014; Sharp et al., 2015; Wright et al., 2016). Participants completed the Borderline Personality Disorder Features Scale for Children (BPFS-C; Crick et al., 2005), a 24-item self-report questionnaire assessing affective instability, negative relationships, identity problems, and impulsivity/recklessness at either Timepoint 9 or Timepoint 10 (or both). The BPFS-C was designed for children and adolescents age 9 and older and has demonstrated good construct validity, moderate stability over time, and internal consistency of α = .76 or greater (Crick et al., 2005). Items were rated on a 5-point Likert scale ranging from 1 (not true at all) to 5 (always true). If participants completed the BPFS-C at both Timepoint 9 and Timepoint 10, the highest score was used. This was done following Boone et al. (2022) to capture the most extreme manifestation of symptoms, as adolescents with BPD features may experience intense spikes in emotion dysregulation and/or impulsivity that are best captured by a maximum score. Peak symptom severity may better align with actual clinical risk and/or clinically meaningful symptoms.

Internalizing and Externalizing Psychopathology

Internalizing and externalizing scores were calculated using items from the PAPA/CAPA based on the internalizing and externalizing variables derived from this dataset for other publications (Gilbert et al., 2021; Gilbert et al., 2019). Internalizing items included core symptoms of generalized anxiety, separation anxiety, and post-traumatic stress disorders. Externalizing items included core symptoms of attention-deficit hyperactivity, oppositional defiant, and conduct disorders. We include a full list of selected items in the online supplemental materials (Table S2) demonstrating the face validity for internalizing and externalizing pathology as well their distinguishability from the items selected for self and peer functioning. Each symptom had possible values of 0 (absent) and 1 (present at subthreshold or threshold level), and then the scores were summed.

Data Analytic Strategy

Descriptive analyses

We conducted descriptive analyses to compute the mean and standard deviation (SD) for each variable at each timepoint, providing an overview of the distribution and variability of the data across the study period.

Two-factor confirmatory factor analyses

Before conducting main study analyses, we conducted a two-factor confirmatory factor analysis (CFA) to assess the factor structure of self- and interpersonal functioning at Timepoint 1. The two-factor approach helps determine whether a single overarching construct best explains the data or whether distinct but related factors provide a better fit. This approach was considered advantageous given the existing data supporting this structure for LPF data in both adults and adolescents (see Sharp & Wall, 2021; and Morey et al., 2022 for reviews). As this step was preliminary and not central to the primary aims, full two-factor CFA results are reported in the Supplemental Materials (Table S4).

Overview of the modeling approach

In our main analyses, we employed a stepwise modeling approach, beginning with univariate latent growth curve models to examine individual trajectories, followed by bivariate latent growth curve models to explore relationships between individual trajectories and later personality pathology (as indexed by borderline features), and finally multivariate latent growth curve models to capture more complex developmental patterns. To evaluate model adequacy and compare different specifications, we assessed several standard model fit indices, including the Comparative Fit Index (CFI), Tucker–Lewis Index (TLI), Bayesian Information Criterion (BIC), and Root Mean Square Error of Approximation (RMSEA).

Univariate Latent Growth Curve Models (LGCMs)

To model self- and interpersonal functioning over time, we first employed univariate latent growth curve models to estimate the slope, intercept, variance, and covariance for each measure of self- and interpersonal functioning. These models allowed us to evaluate how self- and interpersonal functioning changed across waves.

Bivariate LGCMs

Based on the best-fitting univariate models (PAPA/CAPA self-functioning and reduced PAPA/CAPA interpersonal functioning), we next explored the relationship between mean changes in either self- or interpersonal functioning (as indicated by the slope across waves) and their association with the maximum scores in our dependent variable (borderline features) at Timepoints 9 and 10. This was accomplished by first using bivariate latent growth curve models.

Multivariate LGCMs

We then extended our analysis to multivariate latent growth curve models to assess whether the mean changes in both self- and interpersonal functioning (PAPA/CAPA self-functioning and reduced PAPA/CAPA interpersonal functioning) predicted the maximum scores for later personality pathology (operationalized through BPD scores) at Timepoints 9 and 10. To control for potential confounding factors, we incorporated time-invariant covariates, including sex, race (with White as the reference group), and age at baseline. To address our third aim, we also controlled for baseline internalizing and externalizing symptoms in the bivariate and multivariate models.

Missing Data

Missing data were handled using the full information maximum likelihood method (Little & Rubin, 2019). Data were missing mainly due to participant attrition and there were no statistically significant differences in baseline sociodemographic characteristics (Vogel et al., 2023) – see also Table 1. Data were analyzed using R studio Version 2024.04.2+764. Analyses were conducted using the lavaan package in R (Rosseel, 2012).

Results

Descriptive Results

Descriptive statistics are presented in Table 2. Correlations between self-functioning and peer functioning variables at each timepoint and maximum personality disorder features (as indexed by BPD features) are available in supplemental Table S3.

Table 2. Descriptive Statistics of all Pre-registered Study Variables and Demographics by Wave.

Timepoints (N)	1 (N=305)	2 (N=277)	3 (N=262)	4 (N=233)	5 (N=262)	6 (N=277)	7/8 (N=163)	9/10 (N=187)
Maximum of PAPA/CAPA parent and child report self
Mean (SD)	0.168 (0.159)	0.150 (0.144)	0.166 (0.151)	0.152 (0.158)	0.140 (0.159)	0.103 (0.141)	0.092 (0.129)
Missing	3 (1.0%)	4 (1.4%)	3 (1.1%)	0 (0%)	0 (0%)	0 (0%)	1 (0.6%)
Maximum of PAPA/CAPA parent and child report interpersonal
Mean (SD)	0.279 (0.218)	0.252 (0.200)	0.256 (0.208)	0.380 (0.149)	0.405 (0.129)	0.380 (0.112)	0.404 (0.102)
Missing	3 (1.0%)	7 (2.5%)	7 (2.7%)	14 (6.0%)	26 (9.9%)	17 (6.1%)	1 (0.6%)
TOSCA-C shame
Mean (SD)				2.65 (0.681)	2.59 (0.681)	2.24 (0.660)
Missing				14 (6.0%)	33 (12.6%)	33 (11.9%)
HBQ
Mean (SD)	0.853 (0.495)	0.762 (0.450)	0.811 (0.452)	−0.0472 (0.947)	−0.650 (0.756)	−0.786 (0.720)	−1.01 (0.579)
Missing	22 (7.2%)	19 (6.9%)	16 (6.1%)	7 (3.0%)	8 (3.1%)	22 (7.9%)	1 (0.6%)
Sex
Male	157 (51.5%)	141 (50.9%)	134 (51.1%)	121 (51.9%)	141 (53.8%)	144 (52.0%)	85 (52.1%)
Female	148 (48.5%)	136 (49.1%)	128 (48.9%)	112 (48.1%)	121 (46.2%)	133 (48.0%)	78 (47.9%)
Age
Mean (SD)	3.96 (0.762)	4.99 (0.764)	6.01 (0.758)	8.52 (0.891)	9.67 (0.967)	10.5 (1.06)	11.7 (1.13)
Race
White	164 (53.8%)	154 (55.6%)	144 (55.0%)	134 (57.5%)	143 (54.6%)	145 (52.3%)	73 (44.8%)
Black	101 (33.1%)	89 (32.1%)	88 (33.6%)	71 (30.5%)	84 (32.1%)	97 (35.0%)	70 (42.9%)
Other	40 (13.1%)	34 (12.3%)	30 (11.5%)	28 (12.0%)	35 (13.4%)	35 (12.6%)	20 (12.3%)
Externalizing symptoms
Mean (SD)	7.22 (6.99)	5.87 (6.32)	5.58 (5.83)	5.68 (5.79)	5.72 (6.26)	3.98 (5.14)	2.86 (4.56)
Missing	3 (1.0%)	7 (2.5%)	9 (3.4%)	15 (6.4%)	28 (10.7%)	19 (6.9%)	2 (1.2%)
Internalizing symptoms
Mean (SD)	2.19 (2.70)	1.81 (2.42)	1.74 (2.58)	0.805 (1.67)	0.991 (1.91)	0.636 (1.43)	0.463 (1.27)
Missing	3 (1.0%)	7 (2.5%)	9 (3.4%)	12 (5.2%)	27 (10.3%)	16 (5.8%)	1 (0.6%)
Maximum BPFS-C score at Timepoint 9 and 10
Mean (SD)								59.6 (14.4)
Missing								161 (46.3%)

Note. PAPA = Preschool Age Psychiatric Assessment; CAPA = Child and Adolescent Psychiatric Assessment; TOSCA-C = Test of Self-Conscious Affect for Children; HBQ = MacArthur Health and Behavior Questionnaire-Parent Version.

Confirmatory Factor Analysis for Self- and Interpersonal Functioning

The two-factor CFA results support the utility of this factor structure for subsequent analyses. Although some fit indices (e.g., CFI = 0.861; TLI = 0.837) fall slightly below conventional cutoffs (e.g., > .90), other indices such as the RMSEA = 0.045 (90% CI [0.030, 0.059]) and SRMR = 0.062 indicate good model fit. Following Marsh et al. (2004), who caution against rigid adherence to cutoff values and emphasize the importance of considering model complexity, sample size, and theoretical justification, we interpret these indices collectively. Taken together, these indices indicate that the specified structure captures meaningful variance in the data (factor loadings can be found in supplemental Table S4). The interpersonal construct showed strong and significant factor loadings, indicating reliable measurement. The absence of a significant correlation between the self- and interpersonal factors (p = .163) suggests that these are distinct constructs, aligning with theoretical expectations of the distinctiveness of the developmental aspects of self- and interpersonal functioning. Given the model’s overall fit and the interpretability of the factor loadings, this structure provides a theoretically sound and empirically supported basis for subsequent analyses.

Overview of Model Fit Statistics

Evaluation of the models including CAPA, PAPA, TOSCA, and HBQ across all 7 waves demonstrated poor fit. Table 3 summarizes model fit indices for univariate, bivariate, and multivariate latent growth curve models, highlighting the progression from individual constructs to integrated models.

Table 3. Model Fit Statistics.

	χ2	df	p	CFI	TLI	RMSEA [90% CI]	SRMR	BIC
Univariate Models
TOSCA	12.19	1	< .001	0.92	0.77	0.20 [0.11, 0.30]	0.06
PAPA/CAPA self-functioning	46.54	22	.002	0.95	0.96	0.06 [0.03, 0.08]	0.08
PAPA/CAPA interpersonal functioning	317.26	38	< .001	0.38	0.41	0.14 [0.13, 0.16]	0.43	−1,187.18
PAPA/CAPA interpersonal functioning reduced model	12.27	4	.015	0.94	0.92	0.08 [0.03, 0.14]	0.046	−1,276.95
HBQ	488.05	31	< .001	0.00	−0.47	0.21 [0.19, 0.22]	0.30
Bivariate Models
PAPA/CAPA self-functioning and BPD	73.78	29	< .001	0.92	0.92	0.07 [0.05, 0.09]	0.09
PAPA/CAPA interpersonal functioning and BPD reduced model	20.47	7	.005	0.91	0.88	0.08 [0.04, 0.12]	0.07
Multivariate Models
self/interpersonal functioning and BPD reduced model	115.21	58	< .001	0.93	0.92	0.05 [0.04, 0.07]	0.07
self/interpersonal functioning and BPD reduced model with covariates	435.49	112	< .001	0.71	0.67	0.09 [0.08, 0.10]	0.16

Note. CFI = Comparative Fit Index; TLI = Tucker–Lewis Index; RMSEA = Root Mean Square Error of Approximation; SRMR = Standardized Root Mean Square Residual; BIC = Bayesian Information Criterion.

Among the univariate models, the PAPA/CAPA self-functioning model (CFI = 0.95, RMSEA = 0.06) demonstrated better model fit. As the full interpersonal model showed inadequate fit (CFI = 0.38, RMSEA = 0.14), we then focused on modeling interpersonal functioning using Timepoints 4-7. This reduced model demonstrated a better fit with fewer parameters, as evidenced by its lower BIC (−1,276.95) compared to the full model (−1,187.18). The reduced model offered a more parsimonious solution, reflected in higher CFI (0.94) and TLI (0.92) values. Thus, the reduced model was preferred due to its more efficient fit without overcomplicating the model. Therefore, we also focused on bivariate and multivariate models demonstrating good fit (PAPA/CAPA self-functioning using Timepoints 1-7 , PAPA/CAPA reduced interpersonal functioning using Timepoints 4-7 and later BPD using Timepoints 9-10). The TOSCA showed a high RMSEA (0.20) and low TLI (0.77), and the HBQ model performed worse (CFI = 0, RMSEA = 0.21).

Building on these results, bivariate models were developed by pairing the two well-fitting univariate models with later borderline scores. The PAPA/CAPA self-functioning and borderline features model (CFI = 0.92, RMSEA = 0.07) and the reduced PAPA/CAPA interpersonal functioning and borderline features model (CFI = 0.91, RMSEA = 0.08) both showed acceptable fit. These bivariate models were then integrated into a multivariate framework, resulting in the combined self/interpersonal functioning and borderline features reduced model, which exhibited strong fit (CFI = 0.93, RMSEA = 0.05). However, the inclusion of covariates in the multivariate model reduced the overall fit (CFI = 0.71, RMSEA = 0.09), reflecting the added complexity. On this basis, a decision was made to proceed with PAPA and CAPA items only to model self- and interpersonal functioning, respectively.

Univariate Latent Growth Curve Model Results

Table 4 summarizes the results of the univariate models. Model estimates indicated that PAPA/CAPA self-functioning scores on average decreased over time (slope raw estimate = −0.011, standardized estimate = −0.485, p <.001), indicating self-functioning in our sample improved over time. PAPA/CAPA interpersonal functioning scores on average increased over time (slope raw estimate = 0.004, standardized estimate = 0.125, p = .203), indicating interpersonal functioning in our sample worsened over time. For both PAPA/CAPA self- and interpersonal functioning, there were negative associations between initial status and functioning changes over time (see Table 4 and Figure 2).

Table 4. Univariate Latent Growth Curve Modeling of Self-functioning and Interpersonal Functioning.

	Self-functioning trajectory
	Raw estimate	Standardized estimate	Standard error	z	p
Latent variables
Intercept	0.171	1.450	0.008	21.474	<.001
Slope	−0.011	−0.485	0.002	−5.858	<.001
Variances
Intercept	0.014	1.000	0.002	6.149	<.001
Slope	0.001	1.000	0.000	4.283	<.001
Covariances
Intercept and Slope	−0.001	−0.510	0.000	−3.304	.001
	Interpersonal functioning trajectory
	Raw estimate	Standardized estimate	Standard error	z	p
Latent variables
Intercept	0.384	3.815	0.009	45.016	<.001
Slope	0.004	0.125	0.003	1.274	0.203
Variances
Intercept	0.010	1.000	0.002	4.136	<.001
Slope	0.001	1.000	0.001	1.830	.067
Covariances
Intercept and Slope	−0.002	−0.559	0.001	−2.258	.024

Figure 2. Trajectories of Self- and Interpersonal Functioning Observed Scores Over Time.

Bivariate Latent Growth Curve Model Results

Table 5 presents the relationship between PAPA/CAPA self-functioning and reduced PAPA/CAPA interpersonal functioning and later personality disorder features (as indexed by BPD scores) in separate, bivariate models. Initial self-functioning was positively associated with increased personality disorder severity (raw estimate = 1.839, standardized estimate = 0.206, p=.007), indicating that individuals with higher initial levels of impairment in self-functioning showed higher levels of later personality pathology. Slope in self-functioning also was positively associated with personality disorder features (raw estimate = 17.669, standardized estimate = 0.357, p <.05). While self-functioning was generally improving over time as suggested in the univariate LGCM findings, the results of the bivariate modeling suggest that individuals with less improvement (i.e. at a slower rate of negative slope) tended to have more severe personality disorder features at follow-up. In contrast, the initial level of interpersonal functioning did not significantly relate to personality disorder severity at follow-up (raw estimate = 0.043, standardized estimate =0.004, p = .831), suggesting it did not substantially associate with personality disorder severity. However, the slope in interpersonal functioning approached significance with a positive association to personality disorder features (raw estimate = 5.987, standardized estimate =0.233, p = .089), indicating a potential link that warrants further exploration.

Table 5. Bivariate Latent Growth Curve Modeling of Self-functioning and Interpersonal Functioning with BPFS-C as Dependent Variable.

	Self-functioning trajectory with BPFS-C as dependent variable
	Raw estimate	Standardized estimate	Standard error	z	p
Intercept	1.839	0.206	0.682	2.696	.007
Slope	17.669	0.357	7.862	2.247	.025
	Interpersonal functioning trajectory with BPFS-C as dependent variable
	Raw estimate	Standardized estimate	Standard error	z	p
Intercept	0.043	0.004	0.203	0.213	.831
Slope	5.987	0.233	3.517	1.702	.089

Multivariate Latent Growth Curve Model Results

As explained earlier when discussing model fit statistics, when all self- and interpersonal functioning measures were included in a single model to evaluate their relationships with later borderline features, the resulting model exhibited poor fit (see Table 3). Therefore, we focused on bivariate and multivariate models demonstrating good fit (PAPA/CAPA self-functioning, PAPA/CAPA reduced interpersonal functioning with later borderline features).

Table 6 summarizes the results of an evaluation of the associations of self- and interpersonal functioning in the same model. The results confirmed that initial self-functioning shows a positive association with personality disorder severity (raw estimate = 4.571, standardized estimate = 0.535, p < .001), indicating that individuals starting with higher levels of impaired self-functioning report more severe personality disorder symptoms at follow-up. Additionally, the slope in self-functioning over time was positively related to personality disorder features (raw estimate = 16.822, standardized estimate = 0.372, p =.005), again confirming that individuals with less improvement tended to have more severe personality disorder symptoms in late adolescence/early adulthood. Conversely, initial interpersonal functioning was negatively associated with personality disorder severity (raw estimate = −1.448; standardized estimate = −0.143, p = .001), suggesting that higher initial interpersonal functioning correlates with lower personality disorder severity. However, slope in interpersonal functioning did not significantly relate to personality disorder features (raw estimate = 2.002, standardized estimate =0.068, p = .616), indicating that level of interpersonal functioning over time does not substantially influence later symptom severity.

Table 6. Multivariate Latent Growth Curve Modeling of Self-functioning and Interpersonal Functioning with BPFS-C as Dependent Variable.

	Raw estimate	Standardized estimate	Standard error	z	p
Intercept of self-functioning	4.571	0.535	0.949	4.817	<.001
Slope of self-functioning	16.822	0.372	6.023	2.793	.005
Intercept of interpersonal functioning	−1.448	−0.143	0.434	−3.340	.001
Slope of interpersonal functioning	2.002	0.068	3.994	0.501	.616

We then examined the association between self- and interpersonal functioning and personality disorder features while accounting for internalizing and externalizing psychopathology, as well as other relevant covariates. As shown in Table 7, when covariates were included in the analysis, self-functioning remained significant, whereas the initial status of interpersonal functioning became nonsignificant (p = .06). This may suggest that the effect of interpersonal functioning on personality disorder severity may be washed out by the association between self- and interpersonal functioning. Overall, these findings underscore the intricate interplay between self- and interpersonal functioning in personality disorder, indicating that while self-functioning significantly affects personality disorder severity, the role of interpersonal functioning may not be as robust or consistent.

Table 7. Self- and Interpersonal Functioning Trajectory with BPFS-C as Dependent Variable, Adjusting for Covariates.

	Raw estimate	Standardized estimate	Standard error	z	p
Intercept of self functioning	3.626	0.436	1.278	2.838	0.005
Slope of self functioning	13.171	0.301	5.915	2.227	0.026
Intercept of interpersonal functioning	−1.865	−0.188	0.995	−1.875	0.061
Slope of interpersonal functioning	0.367	0.013	4.667	0.078	0.937
Sex	0.243	0.124	0.127	1.907	0.057
Age	−0.059	−0.126	0.031	−1.930	0.054
Black	0.209	0.101	0.165	1.269	0.205
Other race	0.093	0.032	0.253	0.367	0.714
Internalizing	0.033	0.087	0.028	1.162	0.245
Externalizing	0.010	0.068	0.016	0.626	0.531

Discussion

Dimensional models of developmental psychopathology, including personality pathology, have the potential to significantly improve clinical care in psychiatric services (Hopwood et al., 2018; Krueger et al., 2018). The publication of the AMPD in the DSM-5 facilitated a decade of research into its validity, reliability and clinical utility (Krueger & Hobbs, 2020; Morey et al., 2022; Sharp et al., in press; Sharp & Miller, 2022; Sharp & Wall, 2021; Waugh et al., 2017; Zimmermann et al., 2019); however, most of the research in this area has been psychometric in nature with a focus on clarifying the structure of personality pathology, and research with more direct applied impact – especially for prevention and early intervention – has lagged behind. In the present pre-registered study, we leveraged an existing prospective follow-up study to evaluate the unique developmental relevance of self- and interpersonal functioning for the emergence of personality pathology independent from internalizing and externalizing psychopathology. We focused on self- and interpersonal functioning because the AMPD (and ICD-11) identifies LPF (maladaptive self- and interpersonal functioning) as its entry criterion and conditional to the diagnosis of personality disorder. Based on research that identifies BPD as an adequate proxy for general personality dysfunction (Sharp et al., 2015; Wright et al., 2016), we used BPD features as our dependent variable. We hypothesized that (1) high levels of impairment in self- and interpersonal functioning over time will be associated with high levels of borderline pathology (as an index of general personality pathology) at follow-up; (2) the contribution of self-functioning to personality dysfunction will increase over time as children age into adolescence such that self-functioning may be a stronger predictor of personality functioning than interpersonal functioning; and (3) high levels of impairment in self- and interpersonal functioning over time will associate with personality pathology even when controlling for levels of internalizing and externalizing psychopathology.

Results demonstrated that while initial levels of self- (ages 3-4) and interpersonal functioning (age 8) associated with personality disorder severity in late adolescence/young adulthood (ages 16-18), change in self-functioning (between ages 3-15), but not interpersonal functioning (age 8 onwards), was associated with later personality disorder features. Specifically, our data showed that self-functioning improved over time in children from ages 3 to 15, and that a slower rate of improvement in self-functioning (indicated by a less negative slope) was associated with higher levels of personality pathology in late adolescence/young adulthood. These findings held even when controlling for internalizing and externalizing psychopathology – which in this sample were higher than what is typically found in community samples given that the sample was recruited to be enriched with higher levels of depression and externalizing problems.

That self-functioning improves from childhood through adolescence is consistent with research demonstrating that as executive functioning and meta-cognitive capacities improve, age-related improvements are observed in children’s ability to self-reflect and the ability to translate information gleaned from self-reflection into appropriate behavioral adjustments (Crone et al., 2022; Lyons & Zelazo, 2011). Our results suggest that patterns of divergence from this normative improvement in self-functioning from early childhood to early adolescence are associated with higher risk for personality pathology later in development. In interpreting these findings, we emphasize that our measure of self-functioning did not include identity diffusion, which typically onsets in adolescence, increases until mid-to late adolescence and then decreases (Bogaerts et al., 2021; Eggermont et al., 2023; Sharp, Vanwoerden, et al., 2021). Identity diffusion is not typically assessed in pre-adolescent children, because identity consolidation and integration require certain meta-cognitive capacities that only come online in adolescence (Crone et al., 2022). The unique contribution of this study therefore lies in the identification of aspects of self-functioning that can be assessed in young children that associate with later personality pathology in order to identify early childhood indicators of impairment in LPF. The importance of identifying these risk factors in early childhood is underscored by the greater malleability of these features during this developmental period (Luby, et al., 2020; McLaughlin, et al., 2012). The CAPA and PAPA items used in the current study that allowed for the assessment of self-functioning in children as young as age 3 included items related to self-hatred, loneliness, feeling unloved, feeling sorry for oneself, excessive guilt, hopelessness, helplessness, undifferentiated affect, anger and resentfulness, being touchy and easily annoyed, boredom, thoughts of death, grandiosity, and excessive bragging (see online supplemental material for a full description of items). We are not aware of any other study that has evaluated these risk factors prior to age 10; however, we note one study that evaluated aspects of self-functioning in relation to personality pathology in 10-year-olds. Franssens et al. (2024) used psychological network modeling to examine the longitudinal associations between borderline-related traits in 718 youth across four timepoints spanning six years (ages 10-15) and showed that lack of self-confidence strongly affected other borderline-related traits across time, such as anxiousness and stress coping, thereby identifying self-confidence as a potential core aspect of personality that steers the development of other traits. The clinical implication of such findings, including our own, is that these aspects of self-functioning can be monitored and scaffolded developmentally in children and adolescents to ensure optimal personality development and protection against the onset of personality disorder. Interventions at earlier points in development focused on aspects of self may therefore hold promise.

That the above features of childhood and adolescent self-functioning appear to predict personality pathology in late adolescence and early adulthood even when controlling for internalizing and externalizing psychopathology is noteworthy in the context of discussions about the value of the LPF construct, especially its self components, in the context of personality disorder. Personality-psychopathology spectrum approaches (see Clark, 2005; Kotov et al., 2021; Shiner, 2009; Tackett, 2006 for reviews), represented best most recently by the Hierarchical Taxonomy of Psychopathology (HiTOP; Kotov et al., 2021) describe personality traits as derivatives of basic temperamental features that interact with the environment to ultimately result in personality profiles that differentiate one individual from another (Rothbart, 2007; Shiner, 2015). An important finding that unified the temperament and basic personality literature in this regard has been the finding that the same dimensional structure of psychopathology – broadly speaking, the internalizing-externalizing spectra – explains covariance among temperamental, personality, and personality disorder traits. There has been a debate around the extent to which LPF is represented in these spectra (e.g. Widiger et al., 2019), so a finding demonstrating independence of aspects of LPF (as shown here) contributes to the conclusion that aspects of LPF, especially self-functioning, may have some specificity for the prediction of later personality pathology that are not accounted for by the internalizing-externalizing spectra.

Consistent with our hypothesis that self-functioning may be more associated with later personality pathology than interpersonal functioning, our findings demonstrated that the predictive role of interpersonal functioning was less robust than that of self-functioning in the current study. Evidence in support of the role of initial levels of interpersonal functioning was only evident in the multivariate model, while evidence in support of the role of both initial levels and change in levels of self-functioning were evident across all models. The lack of predictive power for interpersonal functioning should be interpreted against the background of the fact that we were not able to model interpersonal functioning across all waves. Specifically, while self-functioning was modeled from ages 3-15, interpersonal functioning was modeled only from ages 8-15. It is possible that very early difficulties in interpersonal functioning and change in interpersonal functioning including preschool age ranges may carry more predictive power than the time period modeled here. It is not clear why the multivariate model with earlier waves of interpersonal variables did not converge. The fact that the HBQ items were not included and may provide better indices of interpersonal functioning is a possibility. Of course, it is possible that interpersonal functioning is a non-specific marker of all psychopathology rather than core to personality pathology, but these are empirical questions leaving a gap for future research with improved methods to further evaluate the role of interpersonal functioning vis-à-vis self-functioning in the emergence and timing of onset of personality pathology.

Taken together, however, our findings do seem to provide preliminary support for the prominent role ascribed to self-functioning in the development of personality (McAdams, 2015a, 2015b, 2015c; McAdams & Olson, 2010) and personality disorder (Sharp, 2020; Sharp et al., in press; Sharp & Boone, in press; Sharp et al., 2018; Sharp & Wall, 2017, 2021) and provide an alternative viewpoint compared to other dimensional models of psychopathology like HiTOP (e.g. Kotov et al., 2021) and approaches that propose to rename personality disorder as interpersonal disorders (e.g. Wright et al., 2022). Specifically, with regard to HiTOP, questions have arisen about whether maladaptive traits, which are well represented by traditional internalizing-externalizing-psychotic spectra, are enough to represent all aspects of personality. Demonstrating that precursors of LPF, especially in the self-domain, predict personality pathology independent from internalizing and externalizing symptoms suggests that traits may not be enough. The assessment of self-functioning, in our view, offers the additional, more process-oriented components of personality that have to do with intrapsychic functioning – a reflective self that dynamically makes sense of self and others in the context of important relationships and interactions (Sharp, 2022a; Sharp & Vanwoerden, 2022; Sharp & Wall, 2018, 2021).

With regard to approaches that promote the renaming of personality disorder as interpersonal disorders, our findings suggest that such renaming would diminish the role of the aspect of personality functioning that denotes “personhood” – namely the construction of an integrated sense of self that builds over time through childhood into adolescence. While interpersonal emphasis as represented in Wright and colleagues’ (2022) proposal includes self-functioning, it suggests that interpersonal functioning is the core concept by which personality functioning is defined, while we maintain that it is in the representation of self that the core and distinguishing features of personality disorder lie. One may argue that this is simply a semantic differentiation, but we contend that semantics matter, and that naming a disorder should ideally closely reflect what it represents – in this case, self and interpersonal functioning and not just interpersonal functioning.

There are several additional limitations in the current study that should be taken into account in the interpretation of the findings presented here. Importantly, we were not able to model TOSCA and HBQ items as originally planned and preregistered. While PAPA and CAPA items provided adequate approximation of self- and interpersonal functioning, TOSCA items (available from age 8 onwards) and HBQ items (available across all waves) would have provided additional indices of self- and interpersonal functioning. In addition, while we approximated LPF as best we could by leveraging the available data of an existing prospective study, it is not LPF proper. While LPF has been effectively evaluated in adolescents (Barkauskiene et al., 2022; Cosgun et al., 2021; Weekers et al., 2021; Wu et al., 2024), careful consideration of the pre-adolescent operationalization of LPF needs to take place if longitudinal studies are to model LPF proper. It may be that developmentally appropriate precursors of self-functioning such as those studied here are indeed the best option; however, others have advocated in the past for the possibility of the evaluation of identity diffusion and disorganization in pre-adolescents (Kernberg, 1990; Kernberg & Chazan, 1998; Kernberg et al., 2000). A further limitation is the fact that only half of the sample had follow-up data on personality disorder symptoms. While we dealt with this statistically, and no significant differences were shown in demographic variables for those who were included vs. excluded, this still may have introduced bias in the study.

Despite these limitations, the current study is the first, to our knowledge, that identifies indicators of self-functioning in children as early as the preschool period and trajectories of self-functioning into adolescence that associate with features of personality disorder in late adolescence and early adulthood. At present, there are very few interventions specifically designed for the prevention and early intervention of personality pathology. Researchers have advocated for pre-adolescent intervention at the level of traits (e.g. Franssens et al., 2024; Hutsebaut & Aleva, 2021; Perepletchikova & Goodman, 2014; Sharp et al., in press; Sharp, Kerr, et al., 2021). Here, we add to this the possibility of becoming more intentional in the assessment of self-functioning in children and adolescents, with a focus as early as the preschool period, and importantly, the development and evaluation of interventions that scaffold optimal development of self. Such interventions may focus directly on self-development in children, but may also utilize the caregiving (and broader social) environment as an important mediator for the development of self. Indeed, the theories in which the LPF construct is grounded explain the development of LPF as primarily attachment-based (Bender et al., 2011; Blatt & Lerner, 1983; Fonagy, 1989; Fonagy et al., 2002; Kernberg, 1967; Kernberg, 1984; Livesley, 2003; Masterson, 1988). For instance, object relations theory (Kernberg 1967, 1984) suggests that basic personality structures, including self and identity, are shaped by interactions with significant others throughout development. Similarly, mentalization-based theory (Fonagy, 1991; Fonagy et al., 2002) identifies the early caregiving environment as an essential laboratory for the development of reflective capacity that undergirds self-development. Consistent with these theories, the findings of this study provide initial evidence for the importance of finding ways to explicitly integrate a consideration of self-development to support youth in their personality development.