The introduction of personality disorders (PDs) in the third edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-III) (1) led to a substantial increase in empirical research and clinical interest (2). Yet, the advent of specific diagnostic criteria and a multiaxial approach that differentiated PDs (i.e., Axis II) from clinical syndromes (i.e., Axis I) set the stage for an ongoing controversy about the conceptualization and diagnosis of PDs. While PDs were defined as discrete, distinct categories, the shortcomings of such a categorical classification model became quickly apparent (3–5), and a shift to a more dimensional model, in which PDs are perceived as extreme variants of normal personality dimensions, became inevitable (6, 7). With the upcoming 11th edition of the International Classification of Diseases (ICD-11) (8), the conceptualization of PDs is finally in transition, acknowledging PDs to be only “relatively” stable (9–11). For over decades, however, temporal stability consisted in one of the major distinguishing features between Axis I and Axis II disorders with the stability of PDs being substantially higher than for other mental disorders. Yet cumulative findings slowly appeared to question the stability of PDs, by suggesting considerable improvement over time (12, 13). Thus, against the common assumption that PDs are “enduring,” “inflexible,” and “stable” the categorical stability of PDs has found to be not much higher than the stability of other mental disorders (14). Indeed, the Collaborative Longitudinal Study of PDs (CLPS) (15), which investigated the stability of schizotypal, borderline, avoidant, and obsessive-compulsive PDs over time, found that fewer than half of PD patients still met the criteria for a diagnosis after 2 years (16). With regard to borderline PD (BPD), 85% of the original sample had remitted after 10 years (17).

Nevertheless, as outlined in Morey and Hopwood’s narrative review (18), temporal stability is a complex notion and has to be examined with respect to several factors. First, estimates tend to vary as a function of the type of stability being assessed. In the present study, the focus relies on the two types of stability that have been studied most frequently, namely mean-level and rank-order stability. Mean-level stability refers to the degree to which the average level of a PD or a PD trait in a given sample changes over time. Rank-order stability, on the other hand, refers to the consistency of an individual’s relative ordering compared to others in a given sample, capturing, thus, the extent to which interindividual differences persist over time (18). Rank-order stability is high if the participants in a given sample maintain their ordering with regard to a specific PD or PD trait relative to each other over time, even if the sample as a whole increases or decreases with regard to that PD or PD trait. As such, rank-order changes are independent of mean-level changes (19). Second, estimates depend in part on the type of PD construct being assessed (i.e., categories or traits), suggesting higher stability for dimensional traits rather than for distinct categories (20–22). In their narrative review, Grilo and McGlashan (21) reported that the rank-order stability for meeting any PD diagnosis is fair to moderate, while individual PD diagnoses often exhibit lower stability. In contrast, dimensional scores tend to show slightly higher stability estimates. Durbin and Klein (20) confirmed these findings by showing that rank-order stability was low to fair for categorical PD diagnoses over a 10-year follow-up in depressed outpatients, while rank-order stability for dimensional PD traits was fair to moderate. According to Grilo et al. (23), mean-level stability, when assessed dimensionally, is generally lower than rank-order stability, which indicates that symptoms tend to decrease on average, but the rank-ordering of individuals within a defined sample remains roughly the same. Third, estimates may be affected by the assessment method being used to measure PDs. Self-report questionnaires tend to show a relatively higher stability than clinical interviews (20, 24). For instance, the findings from Samuel et al. (22) for dimensional ratings showed significantly greater rank-order and mean-level stability for self-report questionnaires compared to clinical interviews. Findings regarding categorical PD diagnoses, in contrast, indicated comparable rank-order and mean-level stability. Finally, Morey and Hopwood (18) outlined how the clinical status and age range of a given sample are critical factors affecting PD stability estimates over time. Studies investigating the course of PDs, however, seem to focus mainly on adult samples, and studies on children and adolescents are scarce. This paucity of research has been in part due to the widespread reluctance to diagnose PDs in youth (25, 26) and to the belief that personality in adolescence is inconstant and characterized by emotional outbursts and impulsive behavior (27, 28). Existing literature, however, clearly states that PDs can be validly and reliably diagnosed among juveniles (27, 28) and that the stability of PDs in adolescence is found to be comparable to the stability in adulthood (29, 30).

Given the apparent number of developmental tasks [e.g., achieving emotional independence from parents, developing close relationships with peers, preparing for a professional occupation (31)], the transition from adolescence to adulthood seems to be a salient period for investigating the stability of PDs (18, 32). To the best of our knowledge, however, only two studies have explicitly investigated the stability of PDs from adolescence to early adulthood. The Children in the Community (CIC) study investigated the stability of PD traits in a community sample ranging in age from 9 to 28 (33). Findings show that mean PD traits were highest in adolescence and declined linearly to adulthood, although effect sizes were small. Rank-order stability was found to be low to moderate, and cluster C traits seemed to be less stable than cluster A and B traits (34). Similarly, Bornovalova et al. (35), who investigated the stability and heritability of BPD in a community sample, showed a significant mean-level decline from age 14 to 24, although rank-order stability was high. A third study, namely the study from Chanen et al. (36), investigated the 2-year stability of PDs in older adolescent outpatients, aged 15–18 years, and found that 74% of those diagnosed with a PD at baseline still met the criteria for a PD at follow-up. Regarding dimensional ratings, both rank-order and mean-level stability ranged from low (PD NOS) to moderate (borderline, histrionic, and schizotypal) to high (antisocial and schizoid) (36).

Given the apparent role of developmental influences on the etiology of PDs, studies about the stability of PDs in high-risk samples are surprisingly lacking. The aim of the present study was therefore to examine the prevalence of PDs and their stability over a 10-year period from adolescence to adulthood in adolescents placed in residential care and juvenile-justice institutions. Due to multiple risk factors – such as childhood adversities (37), unfavorable parenting practices, low socioeconomic status, parental mental disorders (38), early mental-health problems (e.g., ADHD, oppositional defiant disorders, and attachment disorders), symptoms of depression and anxiety (39), substance use (40), self-harming behavior (41), psychopathic traits, and youth delinquency (42) – adolescents in residential care and juvenile-justice institutions are particularly at risk of developing a PD, and PD prevalence rates among them are high, ranging from 18 to 40% across studies (43–45). To account for conceptual and methodological factors, both categorical and dimensional mean-level and rank-order stability were investigated.

The aim of the current study was to examine the prevalence rates as well as the mean-level and rank-order stability of PDs over a 10-year follow-up in adolescents placed in residential care and juvenile-justice institutions. Both the stability of PD categories and the stability of dimensional PD traits were analyzed from adolescence to adulthood. The present findings indicated high PD prevalence rates in young adults with a history of child welfare and juvenile-justice placements, while PD diagnoses and PD traits exhibited only low to moderate stability over the 10-year follow-up.

At least three findings have to be discussed in more detail. First, PD prevalence rates substantially increased from adolescence to adulthood in this high-risk sample. While the normative course of BPD during adolescence is described as an increase of BPD pathology from puberty to young adulthood (57), most previous findings indicate a general decline in PDs and PD traits beginning in young adulthood (17). On the other hand, the prevalence rates of any PD as well as of specific PDs are consistent with the existing literature; the prevalence rates of PDs in institutionalized youth and young adults with a history of out-of-home care have been found to range between 18 and 40% across studies (43–45). A recent meta-analysis on mental disorders in incarcerated youth, which included 30 studies of 8,000 participants, indicated that antisocial and borderline PDs were relatively common in both males and females, while the prevalence of narcissistic and schizotypal PDs was comparably low (58). The current study seems to confirm this pattern, as antisocial and borderline PDs were among the most frequently diagnosed disorders, both at baseline and at follow-up. An increase in PD diagnoses from adolescence to adulthood in this sample, may, thus, be explained by the fact that many adolescents in residential care and juvenile-justice institutions have experienced severe childhood adversities (e.g., child abuse and neglect), which are shown to significantly contribute to the development of PDs (59, 60). For instance, the meta-analysis by Porter et al. (37) found that patients with borderline PD were over 13 times more likely to report childhood adversity than non-clinical controls. In addition, participants in this high-risk sample were likely to have experienced a range of other critical risk factors, such as unfavorable parenting practices, low socioeconomic status, childhood psychopathology, including high substance use, self-harming behavior, and youth delinquency, which have also been shown to be significantly associated with the development of PDs over time (38–42). Given the multifaceted nature of problems faced by juveniles in child welfare care and juvenile-justice institutions, the institutions often lack the professional and financial means to detect personality problems at an early stage, leading to delays in diagnoses and appropriate treatment. Delaying appropriate diagnoses, in turn, carries clinical risk, as evidence is accumulating that many of the harms associated with PDs occur early in the course of the disorder (61), and delay tends to lead toward greater impairments and poorer outcomes (62).

Second, on the categorical level, the mean-level stability of any PD was only moderate, and the mean-level stabilities of specific PDs were low to moderate, except for schizoid PD (high). The concordance between baseline and follow-up assessments (i.e., Cohen’s κ) was low, both for any PD and for individual PDs, except for schizoid PD (moderate). The rank-order stability (i.e., tetrachoric correlation (r_tet) of any PD category was moderate. For individual diagnoses, the rank-order stability ranged from low (i.e., borderline, avoidant, PD NOS, passive-aggressive PDs) to moderate (i.e., paranoid, narcissistic, antisocial, obsessive-compulsive, depressive PDs) to high (schizoid, dependent PDs). Regarding categorical mean-level stability, Chanen et al. (36) found a higher proportion of enduring cases (74%) compared to our findings (47%), which may be due to the shorter follow-up interval (2 years), the clinical status of participants (outpatients), and the narrower age range (15–18 years old) in their study. Indeed, the explorative age-sensitive analyses in the Supplementary Material revealed a higher categorical mean-level stability for the participants who were 15–18 years old than for the participants who were 12–14 years old, although the stability still seems to be lower than that found by Chanen et al. (36). Categorical mean-level stabilities for individual PDs, however, were similar to those found by Chanen et al. (36). As such, participants may have changed specific PDs (from one PD category to another category) but did not discard the general diagnosis of a PD over time. Noteworthy, however, is that 24 (20.9%) participants first developed a PD in young adulthood. As the explorative age-sensitive analyses revealed, older adolescents (15–18 years) were more likely to meet a PD diagnosis first at follow-up than younger adolescents (12–14 years). This suggests that the onset of a PD indeed lies in later adolescence and that some of the present sample had not yet passed the critical age. Another explanation might be that PDs in (young) adolescence are more difficult to detect (63). In addition, older adolescents with a PD diagnosis between 15 and 18 years may have already had longer and more stable patterns of personality pathology, which, therefore, may be more predictive of unfavorable long-term outcomes. Nevertheless, a total of 12 (10.4%) participants improved from baseline to follow-up and no longer met the criteria for a PD in adulthood. While this could have been due to several factors (e.g., treatment or spontaneous remission), it is also possible that these participants no longer met the diagnosis of a PD but still exhibited PD symptoms. This, in turn, is a major concern of the categorical classification system, as it is based on an arbitrary diagnostic threshold that can be easily met (PD diagnosis) or not met (no PD diagnosis) by an increase or decrease in a single criterion.

Regarding categorical rank-order stability, the poor concordances between the baseline and follow-up assessments (i.e., Cohen’s κ) for any PD and for individual PD diagnoses are consistent with those found by Chanen et al. (36). Findings regarding rank-order stability measured with tetrachoric correlations (r_tet) are difficult to compare across studies, since Cohen’s κ remains the most common statistical measure for assessing the rank-order stability of categorical data. Overall, rank-order stability nevertheless seemed to be higher for specific PD diagnoses (i.e., paranoid, narcissistic, avoidant, dependent, PD NOS, and passive-aggressive PDs) than mean-level stability for these PD diagnoses, which suggests that even if the specific diagnoses did not remain the same over time, the rank ordering of participants with such a disorder appeared to be more or less the same. Both the rank-order stability and the mean-level stability of borderline PD were particularly weak, which indicates that on average, neither the category nor the rank ordering of participants with a borderline PD remained the same over time. While this may seem somewhat surprising, it is consistent with the narrative review from Bondurant et al. (64), which suggests that there is only little diagnostic borderline PD stability in adolescence. Interestingly, both Cohen’s κ and tetrachoric correlation coefficients (r_tet) were considerably higher for older adolescents at baseline (15–18 years) compared to younger adolescents (12–14 years old) at baseline (see Supplementary Table 2), which suggests that diagnoses in early adolescence should be treated with caution.

Third, on the dimensional level, PD scores significantly increased for most of the disorders, except for schizotypal, avoidant, narcissistic, borderline, dependent, and passive-aggressive traits. Histrionic traits significantly decreased from baseline to follow-up. Effect sizes were generally low, except for antisocial and obsessive-compulsive traits. In contrast to our findings, Johnson et al. (34) found a significant mean-level decline in dimensional ratings from adolescence to adulthood, and Chanen et al. (36) found neither a significant increase nor a decrease in PD traits, except for paranoid (increase), antisocial (increase), and depressive PDs (decrease). One explanation is that the study by Johnson et al. (34) was conducted in a community-based sample, while the study by Chanen et al. (36) was conducted with older adolescent outpatients. The overall low to moderate dimensional rank-order stability in the present study was, however, consistent with the rank-order stability found in the studies by Johnson et al. (34) and Chanen et al. (36). This indicates that although mean-level PD traits tended to increase among adolescents in residential care and juvenile-justice institutions through adulthood, their individual rank ordering seemed to be less stable, emphasizing interindividual differences among participants. The additional explorative age-sensitive analyses revealed higher dimensional mean-level and rank-order stability estimates regarding older participants (15–18 years old) than younger participants (12–14 years old). On the one hand, this highlights the presence of PD traits in early adolescence but on the other hand, suggests that PD diagnoses before the age of 15 should be interpreted with caution.

Three main findings can be drawn from the current study. First, the prevalence rates of PDs in young adults with a history of child welfare and juvenile-justice placements are high. Second, most categorical PD diagnoses and dimensional PD traits increased from adolescence to adulthood in our sample. Third, overall, the findings indicate low to moderate stability of PDs and PD traits from adolescence to adulthood, although the extent of stability differed according to the PD construct (i.e., categorical diagnoses or dimensional traits), the type of stability (i.e., mean-level or rank-order stability) and the specific PD and PD trait being assessed. As a result, the current findings are in accordance with the growing evidence, that PDs are not that stable. This in turn, emphasizes the current shift to a more dimensional model and highlights the use of the upcoming ICD-11 that acknowledges PDs as only “relatively” stable.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving humans participants were reviewed and approved by the Ethics Committees on Research Involving Humans at the University of Basel and the University of Lausanne (Switzerland) as well as the Institutional Review Board at the University of Ulm (Germany). The follow-up study procedure was approved by the Ethics Committee Northwestern and Central Switzerland. Written informed consent to participate in this study was provided by the participants and the participants’ legal guardian/next of kin, if participants were under 18 years old.

Dd’H, MSt, CB, and KS contributed to conceiving and designing the present manuscript. Dd’H wrote the first draft of the manuscript and analyzed the data. Dd’H, DB, SS, and CB collected the data. MSt supervised the data analyses. CB, MB, NJ, MSc, JF, and KS commented on an earlier draft of the article and supervised the entire process. All authors read and approved the final manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.