International Journal of Mental Health & PsychiatryISSN: 2471-4372

All submissions of the EM system will be redirected to Online Manuscript Submission System. Authors are requested to submit articles directly to Online Manuscript Submission System of respective journal.

Research Article, Int J Ment Health Psychiatry Vol: 4 Issue: 1

Longitudinal Trajectories of Personality Disorders: A Growth Mixture Modeling Analysis

Siny Tsang1*, Stephanie Kasen2 and Alan S. Brown1,2

1Department of Epidemiology, Columbia University Mailman School of Public Health, USA

2Department of Psychiatry, Columbia University Medical Center, New York State Psychiatric Institute, USA

*Corresponding Author : Siny Tsang
Department of Epidemiology, Mailman School of Public Health, Columbia University, New York, NY 10032, USA
E-mail: [email protected]

Received: February 19, 2018 Accepted: March 12, 2018 Published: March 20, 2018

Citation: Tsang S, Kasen S, Brown AS (2018) Longitudinal Trajectories of Personality Disorders: A Growth Mixture Modeling Analysis. Int J Ment Health Psychiatry 4:1. doi: 10.4172/2471-4372.1000159


Among clinical and community samples, personality disorders (PDs) have been consistently reported to have negative effects throughout the lifespan. Although longitudinal studies have reported a general decline in PD symptoms, it is unclear whether there are subgroups of individuals with different PD trajectories from early adolescence to adulthood. This study aimed to examine this question in a large representative community sample and identify early childhood predictors of these varying trajectories. Longitudinal data collected across 20 years from participants in the Children in the Community (CIC) Study, a representative nonclinical cohort, were used. Trajectories of each PD cluster across four waves of assessments were estimated using growth mixture modeling. The extent to which early childhood risk factors were associated with latent class membership was examined. Two distinct trajectories were each identified for cluster A, B, and C PD symptoms from early adolescence into mid-adulthood. Most participants followed a trajectory of decreasing PD symptoms whereas a small proportion showed increases in PD symptoms over time. Individuals with depressive symptoms or a singleparent household in early childhood were more likely to be in the latent class with increasing PD symptoms across time. There is considerable heterogeneity in PD symptom trajectories over time among a large non-clinical sample in the community. Increases in PD symptoms in early adolescence may indicate elevated PD symptoms later in life. Findings highlight the need for early intervention and treatment to guide this subgroup of adolescents towards normative development and a reduction of PD symptoms.

Keywords: Personality disorder; Trajectory analysis; Growth mixture modeling; Children in the community study


Personality disorders (PDs) are characterized by serious disturbances in ways of thinking, behaving, and interacting with others. Among youths, such disturbance may adversely affect the development of cognitive and social competence and adequate emotional controls that are expected by adulthood [1]. Nonetheless, few studies have examined developmental trajectories for PDs from early adolescence into adulthood; moreover, even less is known about the heterogeneity of PD trajectories (i.e., change in level and shape) in non-clinical samples. The purpose of the current study is twofold: The first is to examine whether there exists subgroups of individuals in the ongoing trajectories of traditional PD symptom clusters A, B, and C (the schizotypal, dramatic-erratic, and anxiety symptom clusters, respectively) from adolescence into the adult years; the second is to test whether certain recognized childhood risk factors are associated with differences in level and shape of PD trajectories. We use data from the Children in the Community (CIC) study, gathered prospectively at multiple time points over more than two decades, to examine potential change. Thus, our findings are based on data that allow a life-span approach to the study of PD, extending knowledge gained from studies based on retrospective data on earlier risk or from studies designed as short-term baseline-outcome assessments of prior risk and PD.

Consequences, stability, and antecedents of adult PD

PDs are associated with poorer quality of life and dysfunction, including lower career and economic status, and poorer health status, parenting skills, and overall quality of life, in adult clinical [2] and community samples [3-5].

The Collaborative Longitudinal Personality Disorders Study (CLPS) [6,7] examined the stability of schizotypal, borderline, avoidant, and obsessive-compulsive personality disorders among adult patients who were seeking or had been in mental health treatment across a two-year period using four semi-structured interviews. Patients in each PD group showed substantial decreases in the average proportion of diagnostic criteria met, suggesting a decrease in symptom severity over time [8]. More recent work examining PD symptoms of the CLPS participants over a ten-year period also indicated a significant reduction in the average PD symptoms across time, though the PD constructs showed increased stability in later years [9].

The Longitudinal Study of Personality Disorders (LSPD) assessed PD symptoms in a non-clinical undergraduate sample at three time points over four years [10]. Consistent with the CLPS findings [8,9], a substantial decline was observed in the overall PD and cluster-level PD features. However, higher rates of decline were observed among individuals in the “possible” personality disorder group (met DSMIII- R diagnostic threshold for at least one PD) relative to those with no PD [11].

Although typically not measured until adulthood, personality disturbances have been shown to have their roots in earlier years [12-17]. For example, conduct problems in childhood are associated with personality disorders in adolescence [14] and adulthood [18]. Children who showed borderline pathology in childhood are more likely to develop PD in early adulthood [19]. Studies have also reported links between difficulties in temperament during childhood and cluster A [20], cluster B [21,22], and cluster C [23] PDs. Nonetheless, very few studies have examined the stability and change of PD with multiple assessment time points.

Early childhood risk factors

Several demographic risk factors in early childhood have been associated with PD symptoms in adolescence and early adulthood. Children from families of low socioeconomic status (SES) have higher risks for mental disorders than their peer counterparts in less disadvantaged families [24-27]. Low parental SES was associated with increased risk for offspring PD in adolescence [28], and low family SES was shown to be consistently related to elevated symptoms of schizotypal and borderline PD for over 25 years [29].

Associations between parental age and adverse outcomes in offspring have been reported, with children born to younger mothers having worse behavioral outcomes [30], and lower levels of SES, educational attainment, and health status [31,32]. A recent study showed that the offspring of teenage mothers have the highest risk for any mental disorders [33]. Previous research using the CIC data has indicated that unwanted pregnancy, single parent household and parental conflict were associated with elevated PD symptoms in adolescent and young adult offspring [34].

Additionally, comorbidity between PD and Axis I disorders suggest that depressive symptoms or depressive disorders in early childhood may be linked to PD in adolescence and adulthood [35]. Borderline personality disorder symptoms in adolescence have been associated with depressive symptoms in early adolescence [16]. In another community sample, depressive and irritable symptoms in early and middle childhood were associated with increased odds of PDs in adolescence [14,34].Childhood physical, sexual, and emotional abuse have been consistently associated with increased risk for PDs in adolescence or adulthood [15,34,36-40]. Those forms of childhood maltreatment are indicative of poor parenting and often occur together [1]. Accordingly, it is plausible that earlier risks may alter PD symptom trajectories among certain subgroups.

The Children in the Community (CIC) study

The CIC study is a prospective 30-year study of adult outcomes of childhood risk in a large, randomly selected community sample of families [41,42]. Repeated assessments of Axis I disorder symptoms and Axis II PD symptoms were collected for at multiple time points from early adolescence into adulthood. Consistent with the previously described multi-wave studies over lengthy periods [8,11,43], and with shorter-term studies, there was a substantial decline in overall PD traits from adolescence to early adulthood [10,44-46]. Nevertheless, changes in PD traits reported in the aforementioned studies reflect the average changes in symptoms for the study sample as a whole, which may be a poor representation of potential patterns of changes in PD trajectories for certain subgroups of individuals [29]. Despite the overall trend of decreasing PD symptoms over time, heterogeneity in individual PD trajectories has been noted [11,29]. For instance, although most CIC participants showed an overall decline in PD symptoms, a substantial minority (21%) of the participants exhibited increases in PD symptoms within a nine-year period [43].

To date, whether there are subgroups of individuals who show different patterns of PD traits over time has only been examined in a recent LSPD study [47]. Three distinct trajectories were identified for participants in the “possible” personality disorder group and those in the no personality disorder group. The trajectories showed differences in baseline levels of PD symptoms and rate of change in PD symptoms across the four-year assessment period. Those results suggest that the general decline in PD symptoms may not approximate the trajectories of certain participants, and that the heterogeneity in PD trajectories may consist of different patterns of changes in PD symptomology. Considering the modest sample size and relatively short follow-up period of the LSPD study, it remains unclear whether varied trajectories of PD symptoms are present in a larger community sample. In addition, a much longer follow-up period is needed to explore whether the heterogeneity of PD symptoms extends from adolescence throughout adulthood.

Materials and Methods


This study used longitudinal data collected from participants in the CIC Study. The participants were a cohort of children (Age: M = 5.8, SD = 5.6, range = 1 to 10) whose families were randomly selected from 100 neighborhoods in two upstate New York counties in 1975 [41,48]. About 800 mothers and one randomly sampled child from each family have participated in four follow-up interviews since 1975. The sociodemographic characteristics of these families (91% Caucasians; 51% Catholics, 40% Protestants) were generally representative of the sampled region and families in the northeastern United States [29,41]. The CIC study was approved by the Institutional Review Boards at Columbia University and the New York State Psychiatric Institute. Written informed consent or assent was obtained from all participants after the interview procedures were fully explained. Details regarding the CIC study methodology have been reported previously [41,48].

The current study focuses on the respondents who participated in the four follow-up interviews (W2 in 1983, W3 in 1985-1986, W4 in 1991-1991, and W5 in 2001-2004) in which Axis I mental disorders and Axis II PDs were assessed as part of the CIC protocol.


Personality disorder (PD) symptoms: At the time of the CIC’s W2 interview, no single instruments were available for the assessment of PD in childhood or adolescence because PD before adulthood was not recognized until the DSM-III-R. An instrument to assess PD was developed by adapting items from existing, validated scales or by writing age-appropriate items when necessary [49]. Algorithms were developed by a team of clinical researchers to combine the items in correspondence to the DSM-III-R diagnostic criteria [44], and were later updated with the publication of DSM-IV [43,50]. The PD scales have excellent internal consistency (Cronbach’s alpha: 0.81-0.92), with good construct validity, test-retest reliability, and predictive utility across an 11-year interval [49].

In this study, PD symptoms assessed from the participant’s clinical interviews at W2 to W5 were utilized. PD scales grouped at the traditional DSM cluster levels were used to maximize the amount of variance in each PD scale for the detection of subgroups of individuals with different trajectories over time: cluster A (paranoid, schizoid, and schizotypal PD), cluster B (antisocial, borderline, histrionic, and narcissistic PD), and cluster C (avoidant, dependent, and obsessive-compulsive PD). All symptom scales were transformed into percentage of maximum possible scores (POMP) [51], ranging from 0 to 100 percentage units. Table 1 presents the descriptive statistics of the symptoms scales at each wave.

Early childhood risk factors: Family socioeconomic status (SES), operationalized as a standardized sum of measures of both parents’ education levels, family income, paternal occupational status, and maternal occupation status (if employed), was obtained from the mother’s report at W1. Past studies have suggested that the combination of those components is a better indicator of family SES than the individual measures [29,52,53].

Mother’s age at participant’s birth was computed as the difference between mother’s reported age and the sample child’s age at W1. Unwanted pregnancy was assessed with two questions, “While you were pregnant with (sample child), how did you feel about having the baby?” and “While you were pregnant, how did (sample child’s) father feel about your having this baby?” at the W1 interview. The participant’s mother responded to both questions on a 5-point Likerttype scale (1=very happy to 5=very unhappy). Responses in both questions were summed to indicate the degree to which the pregnancy was unwanted, with higher values indicating the pregnancy was less desired [54].

Family structure was reported by mothers at W1. This information was used to create two dichotomous variables reflecting single parent households, and whether parents had separated/divorced by the time the sample child turned one. Participants’ depression symptoms at W1 were reported by the mother on the following: depressive or irritable symptoms: complaining or irritable, doesn’t fall asleep easily, isn’t happy on waking, complains a long time, and hurts self often. Due to the low frequencies of endorsement on these items, these items were subsequently dichotomized to indicate the presence of depression symptoms.

History of child abuse (sexual and/or physical) was obtained from official records for each participant. Considering the low incidence rates of child abuse, reports of sexual and physical abuse were combined to create a dichotomous variable.

Data analysis

Growth mixture modeling (GMM) was used to estimate the trajectory and identify possible subgroups with different trajectories, of each PD cluster across the four waves of assessment. GMM is an extension of a latent growth curve model (LGCM), which estimates the average level (latent intercept) and rate of change (latent slope), and individual heterogeneity around the mean trajectory (variances of the latent intercept and slope). GMM allows for differences in the shape (i.e., level and rate) of change in subgroups, or latent classes, within the data [55-58]. For each PD cluster, GMMs were tested for up to four latent classes to examine whether the changes in PD clusters over time can be described with one or more distinct trajectories. A one-class GMM in which one growth curve is used to describe the change over time for all individuals is essentially a LGCM. Age at each time point was included as time-varying covariates given the large age variability in the sample. All analyses were performed in Mplus Version 7.4 [59], and missing data was handled with Full Information Maximum Likelihood estimation.

Model fit was compared using the Bayesian Information Criterion [60], with lower BIC indicating better model fit. The best latent class solution for each PD cluster was determined by having the lowest BIC value relative to other models and a conceptually meaningful model. Entropy, a measure of classification accuracy, was also reported for each GMM. A model with entropy closer to 1 suggests higher classification accuracy [61,62].

Once the optimal number of latent classes (lowest BIC with conceptually meaningful trajectories) was determined, we examined whether the early childhood risk factors differentially predict latent class membership. Associations between childhood risk factors and latent class memberships were examined using the three-step method (R3STEP) in Mplus. This method takes into the account the posterior probabilities of the latent class membership as well as the classification uncertainty rate in the evaluation of the link between the latent classes and covariates [63,64].


Descriptive statistics

Overall, the average symptoms decreased over time for cluster A, B, and C PDs (Table 1). Similar patterns were observed for most individual PDs, with the exception of paranoid, antisocial, and obsessive-compulsive PDs. Descriptive statistics of the early childhood risk factors are shown in Table 2.

  Wave 2
n = 748
(Mean age = 13.7 ± 2.6)
Wave 3
n = 729
(Mean age = 16.1 ± 2.7)
Wave 4
n = 716
(Mean age = 22.0  ± 2.7)
Wave 5
N = 678
(Mean age = 33.2 ± 2.9)
Cluster A PD 27.3 8.8 24.3 7.7 21.9 7.8 19.6 8.6
Paranoid 37.0 15.0 39.8 16.0 32.4 14.5 26.9 15.0
Schizoid 25.7 10.8 25.7 10.5 22.0 9.8 20.7 9.6
Schizotypal 19.6 9.2 16.8 7.7 15.7 7.7 14.7 8.4
Cluster B PD 28.3 10.5 26.9 10.0 26.1 10.2 20.7 10.4
Antisocial 17.7 12.0 18.7 12.7 19.9 13.2 16.3 13.2
Borderline 26.0 11.9 25.0 10.5 23.4 11.1 18.9 11.2
Histrionic 38.7 16.2 35.6 14.6 33.9 14.9 27.6 14.7
Narcissistic 41.8 15.7 37.0 16.4 34.9 15.9 25.0 15.1
Cluster C PD 38.1 9.3 35.2 9.2 33.5 10.0 31.5 10.4
Avoidant 34.7 12.2 31.6 11.3 30.2 12.3 29.8 13.4
Dependent 34.8 11.2 31.2 12.0 27.9 12.9 23.0 12.7
Obsessive compulsive 53.5 14.1 52.8 13.5 54.1 13.7 54.2 11.2

Table 1: Descriptive statistics of psychiatric disorder symptom scales across interview waves.

  Mean SD
Mother’s age at participant’s birth 25.6 5.8
Unwanted pregnancy 3.45 1.59
Family socioeconomic status 9.95 0.97
  N %
Single parent household 71 7.0%
Parents separated by age 1 208 20.2%
Child abuse (sexual or physical abuse) 60 8.9%
Depression symptoms at W1 33 3.4%

Table 2: Descriptive statistics of early childhood risk factors.

Growth mixture modeling

Results of the GMM analyses consistently showed a two-class solution to be the best fit for cluster A, B, and C PDs (Supplemental Table 1). As shown in Table 3, the average posterior probabilities of class assignment were very high (all > .80, except the class 2 assignment for cluster C PDs). The results for the two-class solution are presented in Supplemental Figures 1-3 for cluster A, B, and C PDs, respectively. Similar results were obtained when only individuals with no missing data (N = 545) across the four interview waves were included (Supplemental Tables 2 and 3).

  Any Cluster A PD Any Cluster B PD Any Cluster C PD
  Class 1 (Decreasing) Class 2 (Increasing) Class 1 (Decreasing) Class 2 (Increasing) Class 1 (Decreasing) Class 2 (Increasing)
Intercept mean (SE) 30.60 (0.44) 23.3 (2.10) 34.20 (0.55) 22.14 (2.79) 40.99 (0.58) 39.45 (2.90)
Intercept variance (SE) 56.09 (8.11) 56.09 (8.11) 127.42 (12.62) 127.42 (12.62) 93.36 (11.83) 93.36 (11.83)
Slope mean (SE) -0.37 (0.02) 0.54 (12.6) -0.43 (0.02) 0.82 (0.18) -0.36 (0.03) 0.29 (0.13)
Slope variance (SE) 0.05 (0.02) 0.05 (0.02) 0.13 (0.03) 0.13 (0.03) 0.13 (0.03) 0.13 (0.03)
Correlation between intercept and slope -1.20 (0.32) -1.20 (0.32) -3.02 (0.50) -3.02 (0.50) -2.82 (0.54) -2.82 (0.54)
Proportion of sample 95.5% 4.5% 96.7% 3.3% 91.3% 8.7%
Average posterior probability for Class 1 98.4% 1.6% 98.8% 1.2% 95.3% 4.7%
Average posterior probability for Class 2 19.6% 80.4% 12.3% 87.7% 21.1% 78.9%

Table 3: Estimated statistics of 2-class growth mixture models for any cluster A, B, and C psychiatric disorders.

The trajectories of the two-class solution were relatively similar for cluster A, B, and C PDs. Class 1, which showed a declining trajectory over time (henceforth referred to as “decreasing”), consists of the majority of the sample (95.5% for cluster A, 96.7% for cluster B, and 91.3% for cluster C PDs). The average posterior probabilities were .98, .99, and .95 for cluster A, B, and C PDs, respectively. Class 2, which showed an increase in PD symptoms overtime, includes a very small proportion of the sample (4.5% for cluster A, 3.3% for cluster B, and 8.7% for cluster C). The average posterior probabilities were .80, .88, and .79 for cluster A, B, and C PDs, respectively.

Several features of these results are worth highlighting. First, individuals in both classes have similar levels of cluster A PD symptoms at the earliest assessment age (10 years old). Over time, individuals in these classes show two distinct trajectories, one showing decrease in cluster A PD symptoms, whereas the other group showing increases in cluster A PD symptoms over time. Second, the group with decreasing cluster B PD symptoms shows a slight increase in symptoms during adolescence before a steady decline in symptoms through adulthood, whereas the group with increasing cluster B PD symptoms displays increases in symptom levels from early adolescence through midadulthood. Third, although the two classes for cluster C PD symptoms have similar trends to those for cluster A PD symptoms, the average symptom levels for cluster C PDs were higher than those for cluster A PDs across the assessment ages. Specifically, results suggested that individuals endorsed more cluster C PD symptoms than cluster A PD symptoms.

Direct effects of early childhood risk factors on latent class membership

Next, the extent to which latent class membership was associated with early childhood risk factors was investigated. The early childhood risk factors included were family SES, mother’s age at the child’s birth, unwanted birth, single parent household, parents broke up by the time the child turned one, history of child abuse, and depressive symptoms at W1. Single parent household (OR = 14.08) and childhood depressive symptoms at W1 (OR = 20.91) were associated with increased odds of being in the increasing class for cluster A PDs (Table 4). The presence of depressive symptoms at W1 was also associated with increased odds of being in the increasing class for cluster B PDs (Table 5). None of the early childhood risk factors were statistically significantly associated with latent class membership for cluster C PDs (Table 6).

  Decreasing PD Increasing PD Reference = Decreasing
  M (SD) / N (%) M (SD) / N (%) Est (SE) p OR 95% CI
Intercept     -5.99 (5.26) 0.26    
Mother’s age at participant’s birth 25.9 (5.8) 25.2 (5.1) 0.05 (0.05) 0.36 1.05 0.95, 1.17
Unwanted pregnancy 3.4 (1.5) 3.7 (1.7) 0.04 (0.31) 0.90 1.04 0.56, 1.91
Family socioeconomic status 10.0 (1.0) 9.8 (1.3) 0.05 (0.53) 0.92 1.06 0.37, 2.99
Single parent household 35 (4.6%) 4 (13.8%) 2.65 (1.13) 0.02 14.08 1.53, 129.76
Parents separated by age 1 164 (21.4%) 6 (20.7%) -0.50 (1.08) 0.64 0.61 0.07, 5.01
History of abuse 55 (8.6%) 4 (20.0%) 1.25 (1.14) 0.28 3.48 0.37, 32.70
Depression symptoms at Wave 1 15 (2.1%) 2 (8.7%) 3.04 (1.30) 0.02 20.91 1.63, 267.72

Table 4: Associations between early childhood risk factors and latent class membership for cluster A PDs.

  Decreasing PD Increasing PD Reference = Decreasing
  M (SD) / N (%) M (SD) / N (%) Est (SE) p OR 95% CI
Intercept     -4.24 (6.91) 0.54    
Mother’s age at participant’s birth 25.9 (5.7) 26.1 (6.5) 0.12 (0.09) 0.20 1.12 0.94, 1.35
Unwanted pregnancy 3.4 (1.5) 3.7 (1.2) 0.04 (0.27) 0.90 1.04 0.61, 1.76
Family socioeconomic status 10.0 (1.0) 3.7 (1.2) -0.35 (0.59) 0.55 0.70 0.22, 2.24
Single parent household 36 (4.7%) 3 (15.8%) 3.10 (1.70) 0.07 22.24 0.80, 619.01
Parents separated by age 1 166 (21.4%) 4 (21.1%) -0.82 (1.76) 0.64 0.44 0.01, 13.91
History of abuse 55 (8.5%) 4 (28.6%) 2.00 (1.20) 0.10 7.43 0.70, 78.48
Depression symptoms at Wave 1 15 (2.1%) 2 (11.1%) 3.67 (1.50) 0.01 39.21 2.08, 740.29

Table 5: Associations between early childhood risk factors and latent class membership for cluster B PDs.

  Decreasing PD Increasing PD Reference = Decreasing
  M (SD) / N (%) M (SD) / N (%) Est (SE) p OR 95% CI
Intercept     -1.80 (2.63) 0.49    
Mother’s age at participant’s birth 25.9 (5.7) 25.6 (5.8) 0.01 (0.04) 0.75 1.01 0.94, 1.10
Unwanted pregnancy 3.4 (1.5) 3.5 (1.7) 0.04 (0.15) 0.78 1.04 0.78, 1.40
Family socioeconomic status 10.0 (1.0) 9.8 (1.1) -0.11 (0.27) 0.69 0.90 0.53, 1.53
Single parent household 36 (4.8%) 3 (7%) 1.18 (0.86) 0.17 3.25 0.60, 17.47
Parents separated by age 1 159 (21.2%) 11 (25.6%) -0.17 (0.67) 0.80 0.85 0.23, 3.13
History of abuse 52 (8.4%) 7 (17.9%) 0.75 (0.65) 0.25 2.11 0.59, 7.58
Depression symptoms at Wave 1 14 (2%) 2 (5%) 0.49 (1.58) 0.76 1.64 0.07, 36.12

Table 6: Associations between early childhood risk factors and latent class membership for cluster C PDs.


The present study identified two distinct trajectories of cluster A, cluster B, and cluster C PD symptoms across 20 years in a communitybased longitudinal study. The majority of the sample showed declines in cluster A, B and C PD symptoms from early adolescence to adulthood. This overall decline in PD symptoms is consistent with findings among psychiatric patients [8,9,65,66] and community samples [10,11,43]. It is possible that the general decline in PD symptoms for most participants reflect the process of maturation and increased ability to control impulses as individuals transition from early adolescence into adulthood [43]. Closer examination of the trajectory of the decreasing class in cluster B PD symptoms revealed a slight increase in symptoms during adolescence (between the ages of 13 and 18) before the overall decline in symptoms after adolescence. This increase in cluster B PD symptoms during adolescence may be analogous to the peak in delinquent and criminal behavior observed during adolescence [67,68], which may be attributed to the attention seeking and impulsive behaviors common and normative among teenagers. As young individuals mature with age, they gain social competence [42], become better at controlling their impulses, and consider the impact of their behavior on long term goals [69,70], which could have resulted in the decline in cluster B PD symptoms post-adolescence. There is similar evidence linking psychosocial maturity with a decline in antisocial behavior as adolescents transition into adulthood [70].

In contrast, a very small proportion of individuals showed increases in cluster A, B, and C PD symptoms across 20 years. A previous CIC study reported that 21% of the participants showed increases in PD symptoms from early adolescence to early adulthood [43]. The current findings further extended the trajectory duration into mid-adulthood, and more importantly, showed heterogeneity in the trajectories of all three PD clusters. Even though these groups of participants did not show elevated PD symptoms at W2, the steady increase in PD symptoms across time suggests that symptomatology may become more apparent in mid- to late-adolescence, and continue throughout adulthood. These individuals may have deficiencies in developing self-control, and the perspective and temperament associated with maturity. Over time, these deficiencies may culminate in personality pathologies, which may be further exacerbated by interactions with their environment. The social difficulties experienced by young individuals with elevated PD symptoms may compromise the successful development of psychosocial maturity, especially as social demands typically increase with age during this period of life.

Our findings showed that depressive symptoms in childhood may be an important variable differentiating individual in the decreasing class versus the increasing class for cluster A or cluster B PD symptoms. Participants who showed early signs of depressive symptoms had increased odds of belonging to the increasing PD class for cluster A and cluster B PD symptoms as they transitioned from adolescence to adulthood, which is consistent with previous research indicating that maladaptive traits during childhood are related to PD traits in later years [14,15,71-73]. Previous CIC studies have linked depressive symptoms in adolescence to PDs in young adulthood [74,75]. The current study extends such findings, showing that the PD symptom trajectories may be different for individuals who showed depressive or irritable symptoms in early childhood. It is possible that depressive symptoms evident at young ages may interfere with normative development, such as social control, cognitive development, and selfesteem, which in turn may lead to social deficiencies and delayed or unsuccessful development of social competence [75].

In addition, being raised in a single parent household was associated with increased odds of being in the increasing class for cluster A PD. Although that result should be interpreted with caution due to the small number of single-parent households, it is consistent with previous reports of negative associations between single-parent status and increased offspring PD symptoms [34]. Recent studies also found that family adversity (e.g., poverty, single parent household, difficult life circumstances) is associated with increasing borderline personality disorder symptoms [76] and maladaptive behavior (e.g., antisocial PD and marijuana use) in adulthood [77].

Clinical implications

Taken together, the current study showed that there is considerable heterogeneity in PD symptom trajectories over time among a large non-clinical sample in the community. The majority of the individuals showed declines in PD symptoms over time, suggesting that the decrease in PD symptoms may reflect the normative process of psychosocial development as individuals mature into adulthood. In contrast, a small proportion of individuals did not exhibit reductions in PD symptoms, but instead, showed increases in PD symptoms as they transitioned from early adolescence into mid-adulthood. This relatively small proportion of individuals with increasing PD symptoms is in line with the low PD prevalence estimates in our previous CIC study [49] and is comparable to estimates reported in other epidemiological studies [78-82].

Our findings suggest that the developmental process for a relatively small group of individuals may deviate from the normative process of maturation, in part due to an increasing trajectory of PD symptoms into adulthood. Of note, these individuals did not show elevated PD symptoms in early adolescence. One possible reason may be that the clinical measures used are not sensitive enough to detect PD symptoms in late childhood and early adolescence that may be maladaptive in adulthood. An alternative explanation may be that certain PD symptoms are not present and thus not measurable until late adolescence. Nonetheless, our results highlight the need to identify youths who show an increase in PD symptoms during adolescence, and to implement early interventions (e.g., improving social skills) that may reduce PD symptoms as they transition into adulthood.

Considering that PDs in adolescence are associated with poorer quality of life in adulthood [3], quality of life may be worse for individuals who follow an increasing PD symptom trajectory than those who showed declining PD symptoms over time. Further work is needed to examine whether quality of life in mid-adulthood differs between individuals who exhibited increases versus decreases in PD symptoms from adolescence to adulthood. In addition, more work is warranted to investigate the extent to which co-occurring axis I disorders (e.g., mood disorders, anxiety disorder, disruptive disorders, substance use disorders) may affect the trajectories of PD symptoms over time.

Strengths and limitations

Several strengths of the current study are noteworthy. PD symptoms were assessed prospectively in all interviews, thus minimizing measurement error due to retrospective reports and over/ under-reporting of PD symptoms during adolescence. Moreover, the follow-up period was far longer than in previous studies; we were able to examine the trajectories of PD across approximately 40 years, from early adolescence throughout adulthood.

This study had several limitations. First, although the estimates of PD symptoms are not artificially inflated from this randomly selected community sample [83], the latent classes that evidenced increasing symptoms over time represented only less than 10% of the sample. Findings regarding the increasing PD latent class and direct effects of latent class membership should be interpreted in light of the limited sample proportion. Second, although the demographic characteristics of the participants were representative of the sampled region at the time the study began, it is unclear whether our findings are generalizable to the larger population in other geographic areas and/or other racial/ethnic groups. Third, the use of official records of childhood abuse may have underestimated the actual incidences of abuse. Thus, the current null findings of abuse should be interpreted with caution. Fourth, cluster-level PD symptoms were used to attain sufficient power, as such, whether similar trajectories are observed for individual PD within each cluster was not investigated. Finally, it remains unclear whether the current findings will generalize to clinical samples, which are more likely to show more PD symptoms than the current community sample.


In summary, the present study identified two distinct trajectories of clusters A, B, and C PD symptoms from early adolescence to midadulthood among a large non-clinical community sample. The majority of the sample showed a decreasing trend of PD symptoms over time, whereas a small subset of individuals evidenced an increasing trend of PD symptoms across the same duration. Early signs of depressive and irritable symptoms and single-parent household status may be risk factors for increases in PD symptoms over time. Our findings suggest that increased PD symptoms in early adolescence may be a warning signal for elevated PD symptoms in later years, thus highlighting the need for early intervention and treatment implementation as a means of fostering normative development and a reduction of reducing adolescent PD symptoms as youth transition into adulthood.

Conflict of interest

On behalf of all authors, the corresponding author states that there is no conflict of interest.


Siny Tsang is supported by the research training grant 5-T32-MH 13043 from the National Institute of Mental Health.


Track Your Manuscript