Individual Growth Curve Analysis Illuminates Stability and Change inPersonality Disorder Features: Title and subTitle BreakThe Longitudinal Study of Personality Disorders FREE

Personality disorders (PDs) are assumed to be stable over time. Theprevailing diagnostic nomenclature (DSM-IV,¹ DSM-III,² and DSM-R-III³ ) for these prevalentdisorders^{4 - 5} embraces this assumptionin asserting that PDs are “enduring patterns” of behavior thatare “inflexible,” and “stable over time.”¹ However, data supporting this core assumption, drawnfrom properly designed longitudinal studies, are essentially nonexistent,and the long-term stability of PDs remains essentially undocumented terrain.

Initial studies of the stability of PDs generally used 2-wave test-retestresearch designs.⁶ Such studies are typicallyconducted across short time spans and examine the stability of individualdifferences by examining correlation coefficients and comparing group averagesof PD features at the 2 time points. However, the inadequacy of the test-retestdesign for illuminating stability or change has long been noted in the longitudinalresearch methodology literature.^{7 - 10} Twowaves of longitudinal data represent an extremely limited design for investigatingchange because (1) the amount of change between the first and second occasionsof measurement cannot tell us anything about the shape of each person’sindividual growth trajectory between those times, and (2) estimates of truechange are difficult to obtain from the observed 2-wave data.^{8 ,10}

Alternatively, the methodological superiority of the prospective multiwavelongitudinal design for the study of stability and change has long been welldocumented.^{7 - 12} Inaddition to using a multiwave design, a proper longitudinal study involvesthe collection of data across sensibly spaced intervals, wherein the variableof interest is continuous in nature, can be equated across occasions of measurement,and remains construct valid for the entire study period.¹⁰ Moreover,for interview-based assessments, blinded assessment is a critical requirementsuch that the same subject is never seen by the same interviewer more thanonce to ensure against “halo effects,” which diminish validityand elevate stability estimates artifactually.^{13 - 14} Finally,the most informative extraction of meaning from longitudinal data for theinvestigation of stability and change takes the individual growth curve (orpath) as the critical unit of analysis.^{7 ,10 - 11}

The Longitudinal Study of Personality Disorders (LSPD),^{4 ,15} begunin 1990, is a prospective multiwave longitudinal study of personality pathology,normal personality, and temperament sponsored by the National Institute ofMental Health, Washington, DC. A major goal of the LSPD is the life-span studyof the stability of PD symptomatology. The LSPD subjects were drawn initiallyfrom a nonclinical population, thus avoiding the usual confounds attendingthe study of hospital/clinic PD cases (eg, treatment/time confounds, Berkson’sbias, and selection of extreme cases). Herein, we describe an analysis ofindividual growth trajectories of PD features from the LSPD.^{4 ,15} Ouranalyses transcend the previous LSPD¹⁵ reportby examining stability and change in a more informative manner and by illustratingthe power, richness, and utility of the individual growth curve (IGC) approach(also known as hierarchical linear modeling,^{16 - 17} multilevelmodeling,¹⁸ covariance components models,¹⁹ or random-coefficient regression models²⁰ )for longitudinal data analysis in psychopathology.

In the previous LSPD report,¹⁵ the stabilityof individual differences in the PDs over time was examined using the between-wavecorrelation approach, and individual differences were found to be relativelystable during the 4-year span. However, this approach provided only thin 2-wavesnapshots of the entire spectrum of interindividual differences in growth.Average levels of PD symptomatology, assessed using repeated-measures multivariateanalysis of variance (MANOVA),¹⁵ provided someevidence of PD feature reduction over time (albeit with small effect sizes).However, the previous MANOVA approach (1) could not accommodate the existingunequal spacing of PD assessments across study subjects, (2) did not disentangleimportant aspects of individual change such as initial levels of symptomatologyand rate of change as typically implemented, and (3) implausibly assumed comparablegrowth across all subjects, wherein heterogeneity of growth is more likely.²¹ As such, although the previously reported correlationaland MANOVA approaches represented a reasonable first pass through the LSPD¹⁵ data, they did not tap the true richness therein.An analysis of individual growth trajectories^{10 - 11} isrequired at this point to represent this richness and to improve our understandingof the stability or change in personality disorders over time.

The 258 subjects in the LSPD¹⁵ were drawnfrom a population consisting of 2000 first-year undergraduate students.⁴ Subjects were assigned to a possible PD (PPD) groupor a no PD (NPD) group according to the International Personality DisorderExamination DSM-III-R–Screen (IPDE-S)⁴ (response rate, 84.2%). The PPD subjects metthe diagnostic threshold for at least 1 specific DSM-III-R PD, whereas NPD subjects (1) did not meet the DSM-III-R–defined threshold for diagnosis and (2) had fewer than 10 PDfeatures across all disorders. Extensive detail concerning the initial subjectselection procedure and sampling is given elsewhere.⁴ The258 subjects consisted of 121 men (46.9%) and 137 women (53.1%). The PPD groupincluded 134 (68 men and 66 women); the NPD group, 124 (53 men and 71 women).All subjects gave voluntary written informed consent and received an honorariumof $50 at each wave. Of the initial 258 subjects, 250 completed all 3 assessmentwaves and are included in this analysis. Five PPD and 3 NPD subjects did notcomplete all 3 waves, for a final sample of 64 men (49.6%) and 65 women (50.4%)in the PPD group and 53 men (43.8%) and 68 women (56.2%) in the NPD group.Race and ethnicity in the final sample were as follows: 9 (3.6%) African American;12 (4.8%) Latin or Hispanic; 180 (72.0%) white; 43 (17.2%) Asian-Pacific Islander;2 (0.8%) Native American; and 4 (1.6%) other. Mean age at study entry was18.89 years (SD, 0.51 years). Additional sample characteristics are summarizedin Table 1.

The LSPD has a prospective multiwave longitudinal design, with subjectsinitially undergoing evaluation at 3 time points (ie, first, second, and fourthyears in college). Although not required for application of individual growthmodeling,¹⁰ the LSPD data are balanced, inthat all subjects have 3 waves of data, and time structured, in that everyoneundergoes repeated assessment on the same 3-wave schedule,¹⁰ althoughthe time between assessments varies from case to case. Interview assessmentswere conducted by experienced PhD or advanced MSW clinicians. Finally, asthe LSPD is a naturalistic prospective study, subjects were free to seek psychologicaltreatment of their own accord.

The IPDE-S is a 250-item self-administered true-false PD screening inventorydeveloped by Armand W. Loranger, PhD. The diagnostic efficiency and psychometricproperties of the IPDE-S in a 2-stage screen application were described previously.⁴

The IPDE is the well-known semistructured interview that assesses DSM and ICD-10 PD features^{22 - 24} and was used in theWorld Health Organization/Alcohol, Drug Abuse, and Mental Health Administation–sponsoredInternational Pilot Study of Personality Disorders.²⁴ The DSM-III-R criteria were assessed in this study. Clinicallyexperienced interviewers received training in IPDE administration and scoringby Dr Loranger and were supervised throughout the project by one of us (M.F.L.)who was blind to the subjects’ identity, putative PD status, and allprevious assessment information. The interrater reliability for IPDE assessmentswas excellent at all 3 waves, ranging from 0.84 to 0.92 for all PD dimensions.The interviewers were blind to the putative PD group status of the subjectsand to all prior LSPD PD assessment data, and subjects never underwent assessmentby the same interviewer more than once.

The Structured Clinical Interview for DSM-III-R–Non-PatientVersion²⁵ is a semistructured DSM-III-R Axis I clinical interview for use with nonpatients. The interviewwas administered first, followed by the IPDE.

We used IGC analysis to investigate change in PD features over time.This method of analyzing within-subject change was popularized by Rogosa andcolleagues^{7 ,11} and representsthe most powerful way to assess change in a continuous dimension over timewithin subjects.^{10 - 11 ,16 - 17} Thedependent variable used in these analyses was the number of PD features ratedas present on the IPDE, which yielded continuous dimensional scores for the11 DSM-III-R Axis II PDs, the 3 clusters (A, B, andC), and total PD features. Dimensional measures of PD ensured the greatestsensitivity to the investigation of stability and change (qualitative diagnoseswould not be appropriate for a study of change in this framework).

The IGC approach hypothesizes that, for each individual, the continuousoutcome variable (eg, the number of PD features) is a specified function oftime, called the individual growth trajectory, plus error. This trajectoryis often specified as a simple linear function of time, in which case it contains2 important unknown individual growth parameters—an intercept and aslope—that determine the shape of individual true growth over time.The individual intercept parameter represents the net elevation of the trajectoryover time, ie, the true mean level of the PD features for the individual atthe onset of the study (or, alternatively, whenever the origin of the timescale has been defined). The individual slope parameter represents the rateof change over time and is the within-person rate of change in PD featuresover time in the present study. Despite the obvious theoretical importanceof individual slope in studies of change, it has not been widely examinedin research on psychopathology. Once an individual growth trajectory has beenspecified (at level 1) to represent the individual change over time, a level2 model can be specified to describe the investigators’ hypotheses aboutthe way that the individual growth parameters contained in the level 1 modelare related to between-subjects factors (eg, subject sex and diagnostic group).

The IGC approach has several advantages. First, interindividual variabilityin assessment intervals can be tolerated in specifying the individual trajectories,unlike repeated-measures MANOVA. Second, although estimation of an individualgrowth trajectory and its precision generally requires more than 2 observationson the individual over time, the IGC modeling approach is highly tolerantof missing data, permitting subjects who have incomplete data to participatein the analyses. Third, using modern software, the IGC modeling approach simultaneouslyuses data on all individuals at every time point to concurrently investigatewithin- and between-individual change, with concomitant improvements in precisionand power. Fourth, when lengthy multiwave data are available, IGC modelingpermits the flexible specification and rich investigation of nonlinear individualchange over time.

In our analyses, the hypothesized levels 1 and 2 statistical modelswere fitted simultaneously to the LSPD data using full-maximum likelihoodestimation and the HLM-5 computer program.²⁶ Weconducted our analyses sequentially. First, we conducted a set of unconditionalgrowth analyses¹⁰ in which we posited a linearindividual change trajectory at level 1, but did not attempt to predict interindividualvariation in the growth parameters by between-subject factors. Such unconditionalanalyses are useful for partitioning the outcome variation into variance componentsthat describe the net variation in slope and intercept across individuals.Second, we conducted a set of conditional analyses in which we examined systematicinterindividual differences in intercept and slope as a function of 3 between-subjectpredictors, namely study group (PPD vs NPD), subject sex, and age in yearsat entry into the study. These predictors yield fixed effects in the predictionof the slope and intercept values retained from the level 1 analysis. Thefitting of the level 2 model also yields estimates of residual variance thatdescribe remaining interindividual variability in the individual slopes andintercepts (as well as their covariance) after accounting for the hypothesizedfixed effects, giving rise to the variance components (ie, σ²₀, σ²₁, σ₀₁) (Table 2).

In addition, supplementary analyses were conducted to determine whetherstatistical interactions among the group, sex, and age at entry variableswere required as predictors in the level 2 model or whether inclusion of additionallevel 2 predictors (ie, Axis I disorder and treatment) would substantiallyimprove the model fit for a particular PD dimension. Improvement in modelfit was assessed by comparison of deviance statistics. Fixed-effects and variancecomponents were tested for statistical significance using the provided z statistics (2-tailed). Sample comparisons of simple proportionswere performed using the χ² test.

As reported previously,¹⁵ the lifetime DSM-III-R Axis I diagnoses (Table 3) of the study subjects are for definite and probable disorders.Eighty-one (62.8%) of the PPD subjects received an Axis I diagnosis comparedwith 32 (26.4%) of the NPD subjects (χ²₁ = 33.30; P<.001). Forty-one PPD subjects (31.8%) vs 21 NPD subjects(17.4%) reported a history of treatment by wave 3 (χ²₁= 6.97; P<.01). Finally, as of wave3, 16% of the sample received a probable or definite diagnosis for at least1 Axis II PD (or PD not otherwise specified).

The PD features of each of the 250 study subjects were assessed 3 timesduring the 4-year study period. The average age of study subjects at the assessmentwaves were 18.89 years (SD, 0.51 years) at wave 1, 19.83 years (SD, 0.54years) at wave 2, and 21.70 years (SD, 0.56 years) at wave 3. The time betweenassessments for each subject was calculated in years, using each subject’sdate of birth and exact assessment date, and then centered on age at entryinto the study for each study subject (with age at entry being included asa predictor at level 2). The mean time from entry into the study (wave 1)to wave 2 was 0.95 years (SD, 0.14 years); from wave 1 to wave 3, 2.82 years(SD, 0.23 years). Centering the assessment intervals on age at entry and includingage at entry as a predictor at level 2 accounts for each subject’s uniquechronological age when he or she began the study and causes the individuallevel 1 intercepts to represent the true value of the wave 1 assessments asthe subjects’ initial status.

The IGC modeling analyses were performed separately for the outcomevariables of total PD features, cluster PD features, and each individual DSM PD. The heterogeneity in the individual growth trajectoriesis impressive, and they are plotted, using an exploratory ordinary least squaresapproach, for total PD features in Figure 1.

First, an unconditional growth model (ie, containing no level 2 predictors)was fitted for all PD dimensions, providing estimates of the average elevationand rate-of-change parameters and their natural variation across all subjectson entry into the study. The results of these fits are given in Table 4 and Table 5, whichcontain estimates of the fixed-effects and variance components for the unconditionalgrowth trajectories for each of the various PD dimensions. The estimated averageelevation of the individual growth trajectories on entry into the study (intercepts)differed significantly from 0 for all PD dimensions (P<.001;all large effects). In addition, each intercept contained significant variability(σ²₀) (P<.001) availablefor prediction at level 2 in subsequent conditional models.

The estimated average rates of change also differed significantly from0 for all of the PD dimensions, except for the paranoid PD dimension, indicatingthat much change over time was evident in the PD data (typically medium orlarge effects). In addition, all of the variance components associated withrate of change (σ²₁), with the exception of clusterC disorders (compulsive, avoidant, passive-aggressive, and dependent PDs)and paranoid PD, were statistically significant, suggesting that there weresignificant amounts of variation in change that could potentially be predictedin subsequent level 2 models. Finally, the estimated slopes from the unconditionalgrowth analyses support interesting conclusions with respect to the rate atwhich PDs actually change with time. Specifically, we estimate that totalPD features decrease by 1.4 PD features with each passing year. At the levelof clusters, annual rates of change over the study period were as follows:0.35 cluster A features per year, 0.65 cluster B features per year, and 0.40cluster C features per year.

Next, we distinguished the individual growth trajectories by the subjects'group, sex, and age at study entry. We introduced level 2 predictors to explainany between-person variation in the individual elevation and rate-of-changeparameters. The 2 primary between-subjects factors of interest were the groupmembership (PPD vs NPD) and subject sex. In addition, we included each subject’sage at entry into the study as a predictor at level 2 to account for interindividualvariation in change associated with actual age (ie, developmental level).For all 15 models fitted, intermodel comparisons of goodness-of-fit (deviance)statistics disclosed that the inclusion of group, sex, and age at study entryas level 2 predictors significantly improved the fit beyond that achievedin the unconditional growth models (schizoid PD, P < .05;all others, P < .01). The results ofthe conditional analyses are presented in Table6 and Table 7, which includesestimates of the fixed-effects and variance components associated with eachlevel 2 predictor (study group, sex, and age at entry), the approximate P value for testing that these effects are 0 in the population,and an estimate of the effect size correlation coefficient (r).²⁷ Table5 also contains estimates of the variance components from the level2 models, which were also tested for statistical significance.

With respect to elevation of the individual growth trajectories, groupwas a statistically significant predictor of individual elevation parametersin PD features for all PD dimensions across time (all P < .001, except schizotypal PD [P < .002]and schizoid PD [P < .07]), with effectsizes ranging from 0.12 to 0.44 (median, 0.30), indicating medium effects.Sex was less substantially predictive of elevation parameters; although statisticallysignificant (P≤.05) for 8 of 15 curves, most effectsizes were 0.18 or less (median, 0.14), indicating small effects. The PPDstatus corresponded to higher elevation for all PD dimensions, and male subjectsdisplayed higher elevation, except for histrionic and dependent PD. Age atentry into the LSPD predicted little variation in PD growth curve elevations,except for narcissistic PD (P < .09).All of the variance components estimates for elevation (σ²₀) indicated that there remained significant variation in elevationthat could be modeled beyond the 3 predictors that we had selected.

The critical growth parameter for investigating stability and changein PD features over time is the individual slope parameter, as it directlyindexes the rate and direction of individual change over time. In the level2 prediction of slope, group membership was significantly predictive of therate of change in PD features for total PD features; cluster A, B, and C dimensions;and the paranoid, borderline, narcissistic, histrionic, avoidant, obsessive-compulsive,and dependent PDs (all P≤.05). For these PD dimensions,the median effect size was 0.26, indicating a medium-sized effect, and thedirection of the fixed effects shows that PPD subjects were showing higherrates of change (ie, PD feature declines). Group status was less predictiveof rate of change for schizoid (P < .43),schizotypal (P < .06), passive-aggressive(P < .24), and antisocial (P < .10) PDs. Inspection of the fixed effects for sexindicate that it was less predictive of slope and, therefore, essentiallyunrelated to change in PD features over time, attaining statistical significanceonly for narcissistic PD (P < .03).Overall, the effect sizes for sex in relation to the rate of change were quitesmall (median effect size, r = 0.07). Ageat entry into the LSPD was minimally associated with the rate-of-change parameters(Table 4). Overall, study group statuswas the factor most strongly associated at level 2 with rate of change inthe PDs. For many PD dimensions, namely clusters A and B and schizoid, schizotypal,antisocial, borderline, narcissistic, and histrionic PD, the variance componentestimates for rate of change (σ²₁) indicated therewas also additional significant variation in elevation that could still bemodeled beyond the 3 predictors that we had selected.

Finally, in sensitivity analyses suggested by the arithmetic and distributionalproperties of the dependent variable (as a count of features), we refittedall of our unconditional and conditional models by replacing the existingoutcome by its square root. This yielded a pattern of results completely consistentwith those reported herein for the untransformed PD variables.

The change observed in PD features over the study period is clearlyshown in fitted growth trajectories recovered from the IGC analysis for thetotal PD features index presented in Figure 2. In this figure, we plot fitted growth trajectories for prototypicalmembers of each group and sex. Prototypical PPD subjects display marked changeover time.

We also investigated whether statistical interactions among the level2 predictors had any impact on the prediction of the PD outcome variables.In 1 set of supplementary IGC analyses, we included a group × sexinteraction term as an additional predictor at level 2, along with the group,sex, and entry age main effects, and we refitted all of the hypothesized statisticalmodels. For 14 of the 15 models fitted (ie, total PD, 3 clusters, and 11 PDs),inclusion of the group × sex interaction did not result inimproved fit over and above corresponding models containing only the maineffects for group, sex, and age at entry. For the antisocial PD outcome, however,inclusion of the group × sex interaction resulted in an improvementin model fit (χ²₂ = 9.31; P < .01), and the interaction was predictive of elevation(P < .003) but not slope (P = .29). In addition, we refitted all hypothesized modelsand included age at entry × sex and age at entry × groupinteractions, none of which improved model fit over corresponding main effectsmodels.

We also considered whether other important level 2 variables might predictelevation and/or rate of change in the PD dimensions. One such hypothesizedvariable was the presence of any Axis I disorder for study subjects beforeor during the study period, as it is reasonable to suspect that stabilityof PD features could be influenced by an Axis I disorder in evidence beforeor during the study period. Indeed, 45.2% of the total sample had some formof lifetime (or current) Axis I disorder as of wave 3. In 12 of the 15 fittedPD models, including the presence of an Axis I disorder as a predictor atlevel 2 provided an improvement in model fit beyond the fit achieved withonly the main effects of group, sex, and entry age (P < .05).However, the presence of an Axis I disorder was related significantly onlyto the elevation parameter values (for 9 PD dimensions) and not to any ofthe slope parameters (ie, change).

Another potential level 2 predictor of interest was the presence ofsome form of psychological/psychiatric treatment in subjects before or duringthe study period. In this sample, 24.8% of the subjects reported receivingtreatment at some point during their lives, before or during the study. Whenincluded as a level 2 predictor, treatment exposure provided an improvementin model fit in 10 of 15 cases (all P < .05).The impact of treatment, however, was limited entirely to elevation (intercept)values in 7 of those 10 models (P ≤ .05)(total PD, cluster A and B totals, and paranoid, schizotypal, borderline,and narcissistic PDs) and had no statistically significant relation to rateof change (ie, slope) for any of the 15 PD dimensions.

The current diagnostic nomenclature (DSM-IV)clearly asserts that PDs are “enduring patterns” of behavior thatare “inflexible,” “stable” over time, and “oflong duration.”¹ However, empirical datasupporting such assertions are virtually nonexistent. The LSPD^{4 ,17} wasbegun 14 years ago, in part to investigate this core DSM assumption. In a previous LSPD report,¹⁵ itwas concluded that “PD features display relatively high levels of individualdifference stability and appreciable mean level stability, with some changeoccurring over time,” although “the changes were relatively small.”However, that analysis had limited resolving power for issues of stabilityand change, as it was inadequately sensitive to the heterogeneity of individualgrowth trajectories and the unique spacing of assessments for each subjectin the LSPD. The present study, which used the more powerful IGC analysisapproach, was able to characterize the nature and amount of change observedin PD dimensions in the LSPD subjects more precisely. Specifically, althougha previous MANOVA-based analysis¹⁵ hinted atsome change in PD feature levels over time (with small effect sizes), it didnot speak to the rate of change as reflected in the true growth (change) ofeach subject apart from their initial level of PD features. Our IGC analysisclarified that indeed there was considerable change noted in PD features overtime. Group membership (PPD vs NPD) emerged as an important factor that predictednot only the elevation of the individual growth trajectories (at study entry)as a medium-sized effect, but also showed a comparable medium-sized effectin predicting the rates of individual change in PD features over time. Clearevidence of statistically significant individual change was observed for nearlyall PD dimensions studied, and this change was typically and uniformly inthe direction of decreasing PD features over time. In sum, PD features donot appear to be as inflexible and enduring as that suggested by the DSM criteria.^{1 - 3}

What could account for this pattern of change in PD features over time?Variance components estimated in our level 2 analyses showed that additionalvariability in the PD rate-of-change parameters remains that could be predictedby factors beyond our hypothesized predictors of group, sex, and entry age.The presence of an Axis I disorder in a study subject or receipt of treatmentbefore or during the study period had little relationship with rates of changefor the PDs. Fortunately, the IGC modeling approach will allow for the inclusionof additional predictors (eg, personality and temperament) in future effortsto dissect individual change in PD features over time,^{10 - 11 ,28} andsuch future work will be model guided.²⁹

Several caveats should be considered with these data. The present sampleis more homogenous in age, educational achievement, and social class thanthe US population at large, and it consists of young adults. All of thesefeatures may differentially affect the study results. We note adjustment touniversity life across the college years (particularly the freshman-year transition)may have played a role in the changes we observed.¹⁵ Also,LSPD subjects were selected from a population of first-year university studentsthat may have been censored for some of the most severely PD-affected individuals.However, 16% of the LSPD sample received a diagnosis of an Axis II conditionby the end of the study period using the highly conservative IPDE and a ratethat accords well with community studies,^{5 ,30} and45.2% had a lifetime (or current) Axis I disorder by the end of college. Unfortunately,no other multiwave longitudinal study of PDs includes proper methodologicalsafeguards (eg, blinded assessments, no treatment/time confounds) with whichto compare these results (compare Shea et al^{31 (p2037)}) Additional longitudinal studies of PDs using clinical samplesare welcome.

William James³² claimed: “by theage 30, the character has set like plaster, and will never soften again,”which appears true for some aspects of normal personality, but not all.^{33 - 36} However,the DSM criteria notwithstanding, clinical experiencesuggests that some PD features may diminish over time or be generally lessstable, and our results support this impression. Continued life-span studyof the LSPD subjects will allow us to specify the long-term change or stabilityfor personality pathology. With planned waves 4 and 5 data collections, morecomplex functions (eg, quadratic and cubic) for representing nonlinear individualchange will become estimable.¹⁰ We expect thatadditional waves of data will shed further light on this fascinating issueand should provide a more refined appreciation of the natural history of personalitydisorders.

Correspondence: Mark F. Lenzenweger, PhD,Department of Psychology, State University of New York at Binghamton, ScienceIV, Binghamton, NY 13902-6000 (mlenzen@binghamton.edu).

Submitted for Publication: September 17, 2003;final revision received February 25, 2004; accepted March 16, 2004.

Funding/Support: This study was supported inpart by grant MH-45448 from the National Institute of Mental Health, Washington,DC (Dr Lenzenweger).

Acknowledgment: We thank Armand W. Loranger,PhD, for providing training and consultation on the use of the InternationalPersonality Disorder Examination and Lauren Korfine, PhD, for project coordinationin the early phase of the study.