Health Psychology

Identification of Distinct Depressive SymptomTrajectories in Women Following Surgery for BreastCancerLaura B. Dunn, Bruce A. Cooper, John Neuhaus, Claudia West, Steven Paul, BradleyAouizerat, Gary Abrams, Janet Edrington, Debby Hamolsky, and Christine MiaskowskiOnline First Publication, July 4, 2011. doi: 10.1037/a0024366

CITATIONDunn, L. B., Cooper, B. A., Neuhaus, J., West, C., Paul, S., Aouizerat, B., Abrams, G.,Edrington, J., Hamolsky, D., & Miaskowski, C. (2011, July 4). Identification of DistinctDepressive Symptom Trajectories in Women Following Surgery for Breast Cancer. HealthPsychology. Advance online publication. doi: 10.1037/a0024366

Identification of Distinct Depressive Symptom Trajectories in WomenFollowing Surgery for Breast Cancer

Laura B. Dunn, Bruce A. Cooper, John Neuhaus, Claudia West, Steven Paul, Bradley Aouizerat,Gary Abrams, Janet Edrington, Debby Hamolsky, and Christine Miaskowski

University of California, San Francisco

Objective: Depressive symptoms, common in breast cancer patients, may increase, decrease, or remainstable over the course of treatment. Most longitudinal studies have reported mean symptom scores thattend to obscure interindividual heterogeneity in the symptom experience. The identification of differenttrajectories of depressive symptoms may help identify patients who require an intervention. This studyaimed to identify distinct subgroups of breast cancer patients with different trajectories of depressivesymptoms in the first six months after surgery. Method: Among 398 patients with breast cancer, growthmixture modeling was used to identify latent classes of patients with distinct depressive symptomprofiles. These profiles were identified based on Center for Epidemiological Studies�Depression(CES-D) scale scores completed just prior to surgery, and 1, 2, 3, 4, 5, and 6 months after surgery.Results: Four latent classes of breast cancer patients with distinct depressive symptom trajectories wereidentified: Low Decelerating (38.9%), Intermediate (45.2%), Late Accelerating (11.3%), and Parabolic(4.5%) classes. Patients in the Intermediate class were younger, on average, than those in the LowDecelerating class. The Intermediate, Late Accelerating, and Parabolic classes had higher mean baselineanxiety scores compared to the Low Decelerating class. Conclusions: Breast cancer patients experiencedifferent trajectories of depressive symptoms after surgery. Of note, over 60% of these women wereclassified into one of three distinct subgroups with clinically significant levels of depressive symptoms.Identification of phenotypic and genotypic predictors of these depressive symptom trajectories aftercancer treatment warrants additional investigation.

Keywords: cancer, breast, depression, anxiety, growth mixture modeling

Women with breast cancer are more likely than women withoutbreast cancer to experience depressive symptoms (Den Oudsten,Van Heck, Van der Steeg, Roukema, & De Vries, 2009). Depres-sive symptoms have adverse effects on quality of life (Bower,2008; Fann et al., 2008), may impact treatment by reducing ad-herence (DiMatteo, Lepper, & Croghan, 2000) and affecting other

health behaviors, and may increase patients’ perception of pain andfatigue (Fann et al., 2008; Gaston-Johansson, Ohly, Fall-Dickson,Nanda, & Kennedy, 1999).

Despite these negative effects, depressive symptoms frequentlygo undetected and untreated (Greenberg, 2004; Pirl, 2004). There-fore, improvement in the identification of breast cancer patientswith depressive symptoms is needed (Bower, 2008). Most of thestudies that attempted to identify predictors of depressive symp-toms were cross-sectional. While some longitudinal studies haveprovided a more complete picture of how depressive symptomschange over time, they had methodological limitations. Specifi-cally, most of these studies reported mean symptom scores for theentire sample. This approach tends to obscure the heterogeneity indepressive symptoms. In fact, estimates of the prevalence of per-sistent depressive symptoms in women with breast cancer rangefrom 12% to 25% (Burgess et al., 2005; Deshields, Tibbs, Fan, &Taylor, 2006). Thus, in addition to reporting mean changes insymptom scores, there is a need to examine longitudinal data forwithin-subject change, as well as symptom trajectories. This ap-proach may permit identification of women at increased risk forworse depressive symptoms during and after treatment (Hensel-mans et al., 2010).

Newer methods of longitudinal data analysis like growth mix-ture modeling (GMM) (Muthen & Muthen, 2000) can identifylatent classes of patients with similar symptom trajectories. Todate, only four studies used GMM to identify latent classes ofoncology patients with distinct symptom trajectories (Donovan,

Laura B. Dunn, John Neuhaus, and Gary Abrams, School of Medicine,University of California, San Francisco; Bruce A. Cooper, Claudia West,Steven Paul, Janet Edrington, Debby Hamolsky, and Christine Mias-kowski, School of Nursing, University of California, San Francisco; Brad-ley Aouizerat, School of Nursing and Institute for Human Genetics, Uni-versity of California, San Francisco.

This study was funded by grants from the National Cancer Institute(CA107091 and CA118658). Dr. Bradley Aouizerat is funded through theNational Institutes of Health Roadmap for Medical Research Grant (KL2RR624130). Dr. Laura B. Dunn received funding from the Mount ZionHealth Fund. Dr. Christine Miaskowski is an American Cancer SocietyClinical Research Professor. This project was supported by NIH/NCRRUCSF-CTSI Grant Number UL1 RR024131. Its contents are solely theresponsibility of the authors and do not necessarily represent the officialviews of the NIH.

Correspondence concerning this article should be addressed to ChristineMiaskowski, RN, PhD, FAAN, Professor and Associate Dean, Universityof California, 2 Koret Way, Box 0610, San Francisco, CA 94143-0610.E-mail: chris.miaskowski@nursing.ucsf.edu

Health Psychology © 2011 American Psychological Association2011, Vol. ●●, No. ●, 000–000 0278-6133/11/$12.00 DOI: 10.1037/a0024366

1

Small, Andrykowski, Munster, & Jacobsen, 2007; Helgeson, Sny-der, & Seltman, 2004; Henselmans et al., 2010; Rose et al., 2009).Two studies examined predictors of distress in breast cancer pa-tients over one year (Henselmans et al., 2010) and over four years(Helgeson et al., 2004). Precedent for using GMM to examinedepressive symptoms over time comes from studies that identifiedsubgroups with distinct trajectories of depressive symptoms in thegeneral population (Carragher, Adamson, Bunting, & McCann,2009; Colman, Ploubidis, Wadsworth, Jones, & Croudace, 2007),in patients with major depressive disorders (Hunter, Muthen,Cook, & Leuchter, 2010), in family caregivers of dementia patients(Taylor, Ezell, Kuchibhatla, Ostbye, & Clipp, 2008), in perinatalpatients (Mora et al., 2009), and in postcoronary artery bypasssurgical patients (Murphy et al., 2008).

Depressive and anxiety disorders are closely related and anxietyis common among breast cancer patients (Stark et al., 2002). Thesesymptoms are often subsumed under the concept of “distress” inthe psycho-oncology literature (Jacobsen et al., 2005). However,depressive and anxiety symptoms may occur independently as wellas jointly. One of the aims of our larger study was to identifyinterrelationships among a variety of symptoms that are commonin breast cancer patients. Therefore, both anxiety and depressionwere evaluated using valid and reliable measures. For the purposesof this paper, we evaluated the relationship of trait and stateanxiety to the identified depressive symptom trajectories.

Given the paucity of research that attempts to account for theheterogeneity in depressive symptoms in patients with breast can-cer (Fann et al., 2008; Helgeson et al., 2004), the purposes of thisstudy were: to identify distinct latent classes of breast cancerpatients based on self-reported depressive symptoms from justprior to surgery through six months after surgery, and to evaluatefor differences in demographic and clinical characteristics andunderlying trait anxiety among these latent classes. In addition, thelevels and trajectories of anxiety symptoms among the differentlatent classes were evaluated. Based on the literature (Compas etal., 1999; Helgeson et al., 2004; Kroenke et al., 2004), we hypoth-esized that younger age and higher levels of trait and state anxietywould be associated with the subgroups of patients with higherlevels of depressive symptoms over time.

Method

Patients and Settings

This longitudinal study is part of a larger study that evaluatedneuropathic pain and lymphedema in a sample of women whounderwent breast cancer surgery. Patients were recruited fromBreast Care Centers located in a Comprehensive Cancer Center,two public hospitals, and four community practices. Patients wereeligible if they were: a woman � 18 years old who would undergobreast cancer surgery on one breast; able to read, write, andunderstand English; and provided written informed consent. Pa-tients were excluded if they were: having bilateral breast cancersurgery and/or had distant metastasis at the time of diagnosis. Atotal of 516 patients were approached, and 410 enrolled in thestudy (response rate 79.4%). The major reasons for refusal were:too busy, overwhelmed with the cancer diagnosis, or insufficienttime available to complete the baseline assessment prior to sur-gery.

Instruments

At enrollment, demographic information was obtained (age,gender, marital status, education, ethnicity, employment status,living situation, and financial status). At each subsequent assess-ment, patients provided information on current treatments forbreast cancer. Medical records were reviewed to obtain informa-tion on stage of disease, surgical procedure, neoadjuvant treatment,and reconstructive surgery.

The Karnofsky Performance Status (KPS) scale is widely usedto evaluate functional status in patients with cancer and has wellestablished validity and reliability (Karnofsky, 1977; Karnofsky,Abelmann, & Craver, 1948). Patients rated their functional statususing the KPS scale that ranged from 30 (I feel severely disabledand need to be hospitalized) to 100 (I feel normal; I have nocomplaints or symptoms).

The Self-Administered Comorbidity Questionnaire (SCQ),which was developed to assess comorbidity in clinical and healthservice research settings (Sangha, Stucki, Liang, Fossel, & Katz,2003), consists of 13 common medical conditions, including de-pression, described in plain language (i.e., no prior medical knowl-edge needed). Patients were asked to indicate if they currently hadthe condition (“yes/no”), and if “yes,” to indicate whether theyreceived treatment for it (“yes/no”) and whether it limited theiractivities (“yes/no”). Patients could also add two additional con-ditions not listed on the instrument. Each condition yields a max-imum of three points; therefore, the maximum score totals 45points if the open-ended items are used and 39 points if only the 13closed-ended items are used. SCQ-13 scores are reported in thispaper. The SCQ has well-established validity and reliability andhas been used in studies of patients with a variety of chronicconditions (Brunner et al., 2008; Sangha et al., 2003).

The Center for Epidemiologic Studies-Depression (CES-D)scale consists of 20 items representing the major symptoms in theclinical syndrome of depression. Scores can range from 0 to 60,with scores of � 16 indicating the need for clinical evaluation formajor depression. The CES-D has well-established concurrent andconstruct validity (Carpenter et al., 1998; Radloff, 1977; Sheehan,Fifield, Reisine, & Tennen, 1995). In the current study, the Cron-bach’s alpha was 0.90.

The Spielberger State-Trait Anxiety Inventories (STAI-T andSTAI-S) consist of 20 items, each rated from 1 to 4. Total scoresfor each scale can range from 20 to 80, with higher scores indi-cating greater anxiety. The STAI-T measures a person’s predispo-sition to anxiety and estimates how a person generally feels. TheSTAI-S measures an individual’s transitory emotional response,with items assessing worry, nervousness, tension, and apprehen-sion related to how a person feels “right now”. Scores � 31.8and � 32.2 suggest high levels of trait and state anxiety, respec-tively (Fletcher et al., 2008; Spielberger, 1983). Both inventorieshave well-established criterion and construct validity and internalconsistency (Bieling, Antony, & Swinson, 1998; Kennedy,Schwab, Morris, & Beldia, 2001; Spielberger, 1983). In the currentstudy, the Cronbach’s alphas for the STAI-T and STAI-S were0.88 and 0.95, respectively.

Study Procedures

The Committee on Human Research at the University ofCalifornia, San Francisco and each study site approved the

2 DUNN ET AL.

study. During the patient’s preoperative visit, a staff memberexplained the study to the patient. For those women willing toparticipate, the staff member introduced the patient to theresearch nurse, who met with the women, determined eligibilityand obtained written informed consent prior to surgery. Afterproviding consent, the patient completed the baseline studyquestionnaires (Assessment 0). Patients were contacted twoweeks after surgery to schedule the first postsurgical appoint-ment. The research nurse met with the patients in their home,the Clinical Research Center, or the clinic at 1, 2, 3, 4, 5, and6 months after surgery. During each study visit, the womencompleted the study instruments.

Statistical Analyses

Descriptive statistics and frequency distributions were calcu-lated for the sample characteristics and symptom severity scores.

Unconditional growth mixture modeling (GMM) with robustmaximum likelihood estimation was carried out with Mplus Ver-sion 5.21 (Muthen & Muthen, 2009) to identify latent classes (i.e.,subgroups of patients) with distinct depressive symptom trajecto-ries over the six months of the study. GMM is an extension oflatent growth curve analysis that extends the estimation of a singlegrowth curve—represented as latent variables (i.e., intercept andslope coefficients) and variance components for them—to theestimation of a new latent categorical variable that identifies latentgrowth curves for two or more classes (Jung & Wickrama, 2008;Kreuter & Muthen, 2008; Mo & Bodner, 2007; Muthen & Muthen,2000; Muthen, 2001; Muthen, Collins, & Sayer, 2001; Muthen &Kaplan, 2004).

The number of latent growth classes that best fit the datawere identified using the guidelines recommended by Jung andWickrama (2008); Muthen (2001); Nylund, Asparaouhov, &Muthen (2007), and Tofighi and Enders (2008). First, a modelwith two latent growth classes was fit to the data, then a modelwith three latent growth classes was fit, and the procedure wasrepeated until the final iteration of the model was not supported.Model fit for the GMM was assessed statistically by identifyingthe model with the lowest Bayesian Information Criterion(BIC), and by testing the “K” versus “K-1” class models todetermine whether a model with K classes fit the data betterthan a model with K-1 classes with the parametric bootstrappedlikelihood ratio test (BLRT; Jung & Wickrama, 2008; Nylund,Asparouhov, & Muthen, 2007; Tofighi & Enders, 2008). Inaddition, we examined the Vuong-Lo-Mendell-Rubin Likeli-hood Ratio Test (VLMR) for the “K” versus “K-1” class mod-els. The VLMR test has been shown to be anticonservative inidentifying the “correct” number of classes in some mixturemodels. However, when the VLMR test is nonsignificant, itdoes provide evidence that the K-class model is not better thanthe K-1-class model (Nylund et al., 2007).

Intercepts and linear and quadratic slopes for each latent classwere estimated for each model. Intercept and linear slope vari-ances were estimated for each class and were allowed to differacross classes. Given the relatively small sample size, the within-class quadratic slope variance was fixed at zero, because estima-tion failed when quadratic slopes were free to vary even for thesimplest (K � 2) model. Even for the initial two-class model, itwas necessary to fix the linear slope variance to zero for our largest

class. Without setting the slope variance to zero, the model couldnot be estimated due to a nonpositive definite covariance matrixfor the class. The change trajectory for this large class was flat withminimal within-class variance, similar to what would be expectedin a latent class growth model (Nagin & Tremblay, 2001; Roeder,Lynch, & Nagin, 1999).

Mixture models are known to produce solutions at local max-ima, so each model was fit with random starts to be sure that thesolution for the model with the maximum log likelihood valueswas replicated (Muthen & Muthen, 1998-2008). Entropy (i.e., theproportion of latent vs. predicted class membership) was estimatedfor each solution, with � .80 being preferred. Better-fitting modelsshould produce higher entropy values (Celeux & Soromenho,1996; Muthen & Muthen, 1998-2008). In addition to evaluatingthe fit indices, the best fitting model was visually inspected byplotting observed against model-predicted values to determinewhether the predicted trajectories followed the empirical trajecto-ries for the classes, and to evaluate whether the predicted plots“made sense” theoretically and clinically (Muthen & Kaplan,2004).

After identifying the latent class solution that best fit our data,differences among the predicted classes were examined for impor-tant covariates and concurrent outcomes outside our models. Al-though this approach is not the ideal strategy (Delucchi, Matzger,& Weisner, 2004; Kreuter & Muthen, 2008; Muthen, 2002; Petraset al., 2008), it was judged to be more prudent given our relativelysmall sample. Missing data were accommodated by Mplus Version5.21 through the use of Full Information Maximum Likelihood andthe use of the Expectation-Maximization algorithm. This methodassumes that any missing data are ignorable (i.e., missing atrandom; Muthen, 2002; Schafer & Graham, 2002).

Analyses of variance (ANOVA) and chi-square analyses wereused to assess for differences in demographic and clinical charac-teristics and symptom severity scores among the GMM latentclasses. Linear mixed effect model analyses were used to test fordifferences in State anxiety scores at baseline and over time amongthe latent classes, using the SPSS MIXED module. Post hoccontrasts were done to evaluate for differences among the GMMclasses at baseline (i.e., intercept) as well as for GMM class � timeinteractions (i.e., Does the change over time in state anxiety varyacross the different GMM classes?). Post hoc contrasts were doneusing the Bonferroni procedure to control the overall familywisealpha level of the six pairwise contrasts for the four GMM classesat 0.05. For any of the six pairwise contrasts, a p value of � 0.008(.05/6) was deemed statistically significant. Data were analyzedusing SPSS Version 18.0 (SPSS, 2010) and Mplus Version 5.21(Muthen & Muthen, 2008).

Results

Results of GMM Analysis

Four distinct classes of depressive symptom trajectories wereidentified using GMM (see Figure 1). As shown in Table 1, afour-class solution provided the best model fit. The four-classmodel was selected because of its improvement over the three-class model (i.e., the 4-class BIC � 17010.84, which is smallerthan the 3-class BIC � 17026.16; BLRT �2, p � .01, entropy �.71), with each class maintaining a reasonable size and interpret-

3DISTINCT DEPRESSIVE SYMPTOM TRAJECTORIES

ability (Jung & Wickrama, 2008). In addition, the VLMR testshowed that the 5-class solution did not fit better than the 4-classsolution.

The parameter estimates for the four latent classes are listed inTable 2. The classes were named based on the overall shape of thetrajectory. The largest percentage of patients was classified into theIntermediate class (n � 180, 45.2%). This class had a mean totalCES-D score prior to surgery that was just above the clinicallysignificant CES-D cutpoint of 16 (i.e., M � 17.1), with a gradualincrease in the mean score over the course of the study (see Table1). The second largest class was called the Low Decelerating class(n � 155, 38.9%). Prior to surgery, this subgroup’s mean CES-Dscore was 6.8, with mean symptom scores decreasing slightly overthe course of the study. Two smaller classes contained 15.8% ofthe patients. The Late Accelerating class (n � 45, 11.3%) had amean CES-D score that was elevated prior to surgery (i.e., M �24.1). This mean score decreased below the cutpoint of 16 atapproximately one to two months after surgery, followed by anoverall increase over the fifth and sixth months after surgery. Thefourth class, called the Parabolic group (n � 18, 4.5%), had a meanCES-D score (mean 13.8) that was lower than the mean CES-Dscore of the Intermediate group prior to surgery. However, thisclass’s CES-D scores increased steeply over the first three monthsafter surgery, peaked at approximately three months, and thendecreased to presurgical levels at six months postsurgery.

Patient Characteristics

As summarized in Table 3, most patients were Caucasian andwell-educated. Approximately 60% were married or partnered.Nearly one-quarter lived alone. Approximately 22% (n � 87) ofthe women endorsed currently having depression on the SCQ.

Differences in Demographic and ClinicalCharacteristics Among the Four Latent Classes

As shown in Table 3, statistically significant differences werefound among the four latent classes in age, education, and KPSScore. No significant differences were found among the four latentclasses in comorbidities (SCQ score), ethnicity, marital status,working for pay, or living alone.

Post hoc contrasts for significant overall findings revealed thatpatients in the Intermediate class were statistically significantlyyounger than those in the Low Decelerating class. In terms ofeducation, the post hoc contrasts were not significant at the pre-determined threshold. In terms of KPS scores, those in the Inter-mediate class had lower scores than those in the Low Deceleratingclass. The subgroups did not differ in terms of stage at diagnosis.The only treatment variables that distinguished among the groupswere having an axillary lymph node dissection, and having post-operative adjuvant chemotherapy. Post hoc contrasts revealed thata higher percentage of women in the Intermediate class had anaxillary lymph node dissection (n � 82, 45.6%) compared to thosein the Late Accelerating (n � 10, 22.2%) class. Post hoc analysesfor adjuvant chemotherapy following surgery did not reveal sig-nificant differences among the latent classes.

Statistically significant differences were found among the GMMlatent classes in the percentage of women who endorsed depres-sion on the SCQ (�2 � 8.94, p � .03). Across the four GMMclasses, 16.8% of the Low Decelerating class (i.e., 26 of 155women), 23.9% of the Intermediate class (i.e., 43 of 180 women),35.6% of the Late Accelerating class (16 of 45 women), and 11.1%of the Parabolic class (two of 18 women) reported depression. Posthoc contrasts revealed that a higher percentage of women in theIntermediate class than in the Low Decelerating class reported acurrent problem with depression.

Figure 1. Observed and estimated mean CES-D trajectories for patients in each of the latent classes, as wellas the mean CES-D scores for the total sample.

4 DUNN ET AL.

Differences in Trait and State Anxiety Scores Amongthe Four Latent Classes

As Table 3 shows, the Intermediate and Late Acceleratingclasses had higher mean trait anxiety (STAI-T) scores prior tosurgery compared to the Low Decelerating class (both p � .001).Figure 2 illustrates the differences at baseline and over time inmean state anxiety (STAI-S) scores among the four classes. Atbaseline, the Low Decelerating class had a statistically signifi-cantly lower mean STAI-S score compared to the other three latentclasses (all p � .001). The post hoc contrasts for the class by timeinteractions for changes in STAI-S scores over time showed nodifferences among the four latent classes in the trajectory ofanxiety symptoms.

Discussion

Only two studies were found that used GMM to identify latentclasses of oncology patients based on changes in depressive symp-toms over time (Helgeson et al., 2004; Henselmans et al., 2010).

This paper adds to this nascent literature and provides additionalevidence that four distinct subgroups of patients exhibit consider-able heterogeneity in their experience of depressive symptomsduring treatment for breast cancer (Deshields et al., 2006).

The patient subgroups in this study are almost identical to thosereported by Henselmans and colleagues (2010) who evaluated 171breast cancer patients, assessed at five timepoints (from afterdiagnosis to approximately one year after diagnosis). Using abroad measure of psychological distress (i.e., General HealthQuestionnaire [GHQ]), four distress trajectories, namely no dis-tress (36%), distress during the active treatment phase only (33%),distress in the reentry and survivorship phase (15%), and chronicdistress (15%) were identified. Lower levels of neuroticism (ten-dency to experience negative affect), higher levels of mastery(sense of control over one’s life) and optimism, and fewer physicalcomplaints from adjuvant treatment distinguished the no distressgroup from the other three groups. Consistent with their previouswork, in the multivariate analyses, mastery was the only uniquepredictor of group membership (Helgeson et al., 2004).

Table 1Fit Indices for the GMM Class Solutions for 398 Breast Cancer Patients

GMM LL AIC BIC Entropy BLRT VLMR (df)

1-Classa �8523.34 17171.67 17215.52 n/a n/a n/a2-Class �8495.90 17029.80 17105.54 .64 112.16�� 112.16� (6)3-Class �8441.24 16930.48 17026.16 .69 90.82�� 90.82� (6)4-Classb �8415.62 16891.24 17010.84 .71 51.24� 51.24� (6)5-Classc �8395.19 16862.37 17005.88 .70 40.87� 40.87ns (6)

Note. GMM � Growth mixture model; LL � loglikelihood; AIC � Akaike Information Criterion; BIC � Bayesian Information Criterion; BLRT �parametric bootstrapped likelihood ratio test for K-1 (H0) vs K classes; VLMR � Vuong-Lo-Mendell-Rubin likelihood ratio test for K-1 (H0) vs K classes;ns � not significant.a Latent growth curve with linear and quadratic components; Chi2 � 77.186, 24 df, p � .00005, CFI � .94, RMSEA � .075. b As noted in bold font,the 4-class model was selected as the best fitting model. c While a five-class solution had the smallest BIC, three other indicators led us to reject the 5-classsolution in favor of the 4-class solution. Entropy decreased for the 5-class solution, indicating a worse fit; the best loglikelihood value was not replicatedin 65 out of 77 bootstrap draws for the BLRT, making that test result unreliable; and the VLMR was not significant, indicating that too many classes hadbeen extracted.� p � .05. �� p � .00005.

Table 2Parameter Estimates for Latent Classes From 7 Assessmentsa

Parameter estimatesb

Low Deceleratingclass (n � 155)

Intermediate class(n � 180)

Late Acceleratingclass (n � 45)

Parabolic class(n � 18)

Mean (SE) Mean (SE) Mean (SE) Mean (SE)

MeansIntercept 6.88��� (0.76) 16.38��� (1.20) 22.06��� (2.37) 13.30��� (1.96)Linear Slope �0.25 (0.35) 0.21 (0.52) �5.98��� (1.25) 11.20��� (1.27)Quadratic Slope �0.07 (0.05) �0.11 (0.08) 0.97��� (0.20) �1.96��� (0.18)

Variancesc

Intercept 3.67� (1.41) 66.32��� (13.06) 75.49�� (22.10) 27.94�� (8.36)Linear Slope 0d 1.55��� (0.34) 4.41��� (0.99) 2.12�� (0.63)I with S Covariance 0d �5.35��� (1.45) �5.35��� (1.45) �5.35��� (1.45)

Note. SE � Standard Error.a Trajectory group sizes are for classification of individuals based on their most likely latent class probabilities. b Growth mixture model estimates wereobtained with robust maximum likelihood estimation. c Variances for quadratic slopes were fixed at zero, and covariances between intercepts and slopeswere held equal across classes, to aid in model convergence due to the relatively small sample sizes. Residual variances for the indicators at each assessmentwere held to be equal across classes, to allow comparisons of the growth parameters. d Fixed at zero.� p � .01. �� p � .001. ��� p � .0005.

5DISTINCT DEPRESSIVE SYMPTOM TRAJECTORIES

Heterogeneity in patients’ experiences with depressive symp-toms was demonstrated in another study of women with breastcancer who underwent radiation therapy (RT) (Deshields et al.,2006). In this longitudinal study, five distinct subgroups of womenwere identified based on their CES-D scores (i.e., Never De-pressed, Recover, Become Depressed, Stay Depressed, Vacillate).The only demographic characteristic that distinguished among the

groups was the number of children at home, with the BecomeDepressed group having more children at home and the Vacillategroup having significantly fewer children at home than the NeverDepressed group. In addition, at the completion of RT, all of thegroups had significantly higher Spielberger State Anxiety scorescompared to the Never Depressed group. While findings from theDeshields et al. study suggest that patients differed in their expe-

Table 3Demographic and Clinical Characteristics of Total Sample (n � 398) and of Four Latent Classes

Characteristic

Total sample(n � 398)

Low Deceleratingclass (1)

n � 155 (38.9%)

Intermediateclass (2)n � 180(45.2%)

Late Acceleratingclass (3) n � 45

(11.3%)

Parabolicclass (4)n � 18(4.5%)

Omnibus statistics and Post hoccomparisons�Mean (SD) Mean (SD) Mean (SD) Mean (SD) Mean (SD)

Age (years) 54.9 (11.6) 57.3 (11.0) 53.0 (11.9) 55.6 (11.0) 52.4 (10.7) F(3, 394) � 4.41; p � .005 1 � 2Education (years) 15.7 (2.7) 15.8 (2.5) 15.9 (2.8) 14.5 (2.2) 15.8 (3.4) F(3, 389) � 3.51; p � .015KPS Score 93.2 (10.3) 95.5 (8.7) 91.1 (11.1) 92.7 (11.5) 95.6 (7.0) F(3, 387) � 5.31; p � .001 1 � 2SCQ-13 Score 4.3 (2.8) 4.0 (2.5) 4.6 (3.1) 4.4 (3.1) 3.7 (2.4) F(3, 393) � 1.42; p � .237CES-D Total Score 13.7 (9.8) 6.8 (4.7) 17.1 (8.6) 24.1 (11.1) 13.8 (8.0) F(3, 377) � 78.66; p � .001 1 � 2, 3, 4;

2 � 3; 3 � 4Trait Anxiety Score 35.3 (9.0) 30.6 (6.3) 38.6 (9.2) 38.8 (9.6) 35.5 (8.0) F(3, 373) � 28.12, p � .001 1 � 2, 3State Anxiety Score 41.8 (13.5) 35.0 (11.2) 45.2 (12.6) 50.4 (14.2) 45.0 (13.5) F(3, 379) � 26.91, p � .001 1 � 2, 3, 4

n (%) n (%) n (%) n (%)

Ethnicity (% White) 255 (64.4%) 107 (69.5%) 112 (62.6%) 26 (57.8%) 10 (55.6%) �2 � 21.98; p � .23Marital status (%

married/partnered) 213 (54.1%) 96 (62.3%) 88 (49.2%) 24 (54.5%) 5 (29.4%) �2 � 21.53; p � .12Works for pay (% yes) 189 (47.8%) 78 (50.3%) 83 (46.6%) 19 (42.2%) 9 (52.9%) �2 � 1.23; p � .75Lives alone (% yes) 95 (24.2%) 34 (22.1%) 41 (23.0%) 13 (29.5%) 7 (41.2%) �2 � 3.87; p � .28Stage of disease at

diagnosis0 64 (16.9%) 23 (15.5%) 28 (16.3%) 11 (25.6%) 2 (12.5%) �2 � 25.86; p � .21I 143 (37.7%) 66 (44.6%) 51 (29.7%) 19 (44.2%) 7 (43.8%)IIA 99 (26.1%) 35 (23.6%) 49 (28.5%) 11 (25.6%) 4 (25.0%)IIB 39 (10.3%) 12 (8.1%) 23 (13.4%) 1 (2.3%) 3 (18.8%)IIIA 18 (4.7%) 5 (3.4%) 13 (7.6%) 0 (0%) 0 (0%)IIIB 6 (1.6%) 2 (1.4%) 4 (2.3%) 0 (0%) 0 (0%)IIIC 8 (2.1%) 3 (2.0%) 4 (2.3%) 1 (2.3%) 0 (0%)IV 2 (0.5%) 2 (1.4%) 0 (0%) 0 (0%) 0 (0%)

Surgical treatmentBreast-conserving 318 (79.9%) 127 (81.9%) 142 (78.9%) 35 (77.8%) 14 (77.8%) �2 � 0.69; p � .88Mastectomy 80 (20.1%) 28 (18.1%) 38 (21.1%) 10 (22.2%) 4 (22.2%)Sentinel node biopsy

(% yes) 328 (82.4%) 133 (85.8%) 144 (80.0%) 36 (80.0%) 15 (83.3%) �2 � 2.15; p � .54Axillary lymph node

dissection (% yes) 149 (37.5%) 52 (33.8%) 82 (45.6%) 10 (22.2%) 5 (27.8%) �2 � 11.10; p � .01 2 � 3Breast reconstruction

at time of surgery(% yes) 86 (21.7%) 28 (18.2%) 41 (22.8%) 11 (24.4%) 6 (33.3%) �2 � 2.88; p � .41

Neoadjuvantchemotherapy (%yes) 79 (19.9%) 26 (16.9%) 44 (24.4%) 6 (13.3%) 3 (16.7%) �2 � 4.55; p � .21

Post-operativetreatment��

Adjuvantchemotherapy 133 (33.4%) 42 (27.1%) 73 (40.6%) 12 (26.7%) 6 (33.3%) �2 � 7.83; p � .05

Radiation�� 224 (56.3%) 93 (60.0%) 95 (52.8%) 26 (57.8%) 10 (55.6%) �2 � 1.81; p � .61

Note. KPS � Karnofsky Performance Status; SCQ � Self-Administered Comorbidity Questionnaire; CES-D � Center for Epidemiological Studies –Depression.� Bonferroni post hoc pairwise comparisons with � � .05: Numbers refer to latent classes, e.g., for CES-D: 1 � 2,3,4 represents Low Decelerating classhad a lower mean CES-D score than the Intermediate Decelerating, Late Accelerating, or Parabolic classes. Only significant post hoc contrasts areshown. �� Post-operative chemotherapy and radiation were coded dichotomously yes/no if the patient had received these treatments at any point duringthe course of the study.

6 DUNN ET AL.

rience of depressive symptoms at the completion of and at threeand six months after RT, the use of categorical data did not allowfor the identification of underlying or latent classes of patients withdistinct symptom trajectories. Variations within and between sub-groups were minimized by the use of dichotomous categorizations.In contrast, in the present study, the use of GMM enabled theclassification of individuals into four groups or latent classes basedon similarities in scores over time.

The largest latent class in our study (45%) had depressivesymptom scores that on average were just above the clinicallysignificant CES-D cutpoint of 16. This Intermediate class mayrepresent a group of patients with subsyndromal depression. Inaddition, this group exhibited significantly higher baseline trait andstate anxiety scores and higher anxiety scores over time, comparedto the Low Decelerating group. This group also had a lowerfunctional status score than the Low Decelerating group.

The Late Accelerating and Parabolic classes together accountedfor 16% of the sample. These two groups had distinct symptomtrajectories that warrant additional investigation to determine un-derlying causes for these changes in depressive symptoms overtime. These groups may represent patients in need of intervention.Our findings are consistent with those of Helgeson and colleagues(2004), who identified a subgroup of women with breast cancerwith elevated levels of distress for up to four years after diagnosis.Moreover, they found two subgroups with elevated distress levels,who comprised over 30% of their sample.

As predicted, the distinct latent classes differed from one an-other in terms of age, with the Low Decelerating group beingsignificantly older than the Intermediate group. This finding isconsistent with previous reports that found that, on average, oldercancer patients, including those with breast cancer, have lowerlevels of depressive symptoms and better overall health-relatedquality of life compared to younger patients (Helgeson et al., 2004;

Kroenke et al., 2004; Parker, Baile, de Moor, & Cohen, 2003).Various explanations are offered for why younger adults are morelikely to have elevated depressive symptoms in the context ofcancer (Compas et al., 1999; Kroenke et al., 2004). Proposedfactors include differences in the types of treatments received, theseriousness of side effects (e.g., abrupt menopause, infertility,sexual dysfunction), and adaptiveness of coping mechanisms used(Compas et al., 1999; Mosher & Danoff-Burg, 2005). However, inat least one study of older women with breast cancer (Ganz et al.,2003), the older age groups had higher study refusal rates. There-fore, it is possible that those older patients who agree to participatein psychosocial or symptom-related research, particularly studiesrequiring multiple, fairly lengthy assessments over a prolongedperiod of time, may be relatively healthier and less distressed thantheir nonparticipating counterparts.

The majority of the patients in the study had relatively highfunctional status scores. The lower functional status score reportedby participants in the Intermediate class compared to the LowDecelerating group, while statistically significant and clinicallymeaningful (effect size, d � 0.43), is not easily explained andwarrants investigation in future studies.

The finding that state anxiety distinguished the Low Decelerat-ing class from the other three classes at baseline through sixmonths of follow-up highlights the important relationship betweenthese two symptoms. Comorbid depressive and anxiety symptoms,even at subsyndromal levels, are associated with impaired func-tioning and QOL (Das-Munshi et al., 2008; Merikangas et al.,2003). Investigators are actively pursuing common etiologicalfactors for these two symptoms (Goldberg, Krueger, Andrews, &Hobbs, 2009), including genetic (e.g., polymorphisms in the sero-tonin transporter; Caspi et al., 2003) and personality-mediatedfactors (e.g., neuroticism; Gonda et al., 2009; Kendler, Kuhn, &Prescott, 2004; Munafo, Clark, Roberts, & Johnstone, 2006). Fur-

Figure 2. Changes over time in mean State Anxiety scores for patients in each of the latent classes. Post hoccontrasts for intercept 1 � 2, 3, and 4 all p � .0001. No differences among the CES-D groups on the slopes forState Anxiety.

7DISTINCT DEPRESSIVE SYMPTOM TRAJECTORIES

ther understanding of the temporal and etiological relationshipsbetween anxiety and depression in cancer patients and the predis-posing factors for these symptoms (independently and jointly)warrant additional investigation.

This study has several clinical implications. First, the findingthat nearly half of the patients (Intermediate class, 45%) hadslightly elevated or subsyndromal levels of depressive symptomssuggests that identification of these patients may be as important asidentifying those patients with clearly elevated symptoms. Giventhat most screening instruments (e.g., Distress Thermometer; Ja-cobsen et al., 2005) use fewer items to screen for depression, it isunclear whether these patients would have been identified withsuch brief screeners. Thus, this substantial subgroup, may remainunidentified, yet may need referral or intervention. Little researchexists on the effects of subsyndromal depressive symptoms ontreatment outcomes, functional status, and QOL in cancer patients.Future research needs to clarify the prevalence, correlates, andimplications of subsyndromal levels of depressive as well asanxiety symptoms.

Second, it was notable that a large subgroup (39%) experiencedno to minimal depressive symptoms over the course of this study.This finding is consistent with previous reports that a substantialproportion of women with breast cancer exhibit very low levels ofdistress (Deshields et al., 2006; Helgeson et al., 2004; Henselmanset al., 2010). As these researchers noted, patients in this subgroupmay be demonstrating a resilient pattern of response to stressfulevents.

This resilient pattern may be more common than previouslythought. In suggesting that resilience is common in adults whoexperience significant trauma, including a life-threatening illness,Bonanno (2004) postulated four distinct profiles of response (i.e.,resilience, recovery, and chronic and delayed disruptions in func-tioning). Given the consistency of these findings, investigatorsshould seek to identify factors that may protect individuals fromthe deleterious effects of the various stressful aspects of cancerdiagnosis and treatment. From both clinical and research perspec-tives, distress interventions should be designed and tailored forindividuals at highest risk. In contrast, resilient individuals maynot need an intervention or may need a different type of interven-tion to maintain their resilience.

Third, these trajectories may represent underlying traits, which,at times of increased stress, predispose individuals to differenttrajectories of psychological symptoms. Moreover, other studieshave reported on the influence of putative predisposing,personality-related factors (e.g., optimism, neuroticism, trait anx-iety) in relation to depressive symptoms in women with breastcancer (Den Oudsten et al., 2009; Henselmans et al., 2010). Fur-ther work is needed to understand the degree to which traitsinfluence psychological distress (both depression and anxiety)over time in cancer patients, and to develop and test interventionsto improve coping skills in those predisposed to greater psycho-logical distress by virtue of their underlying traits.

Finally, further work is needed to understand the longitudinalrelationships among depressive symptoms and other prevalentsymptoms in cancer patients, particularly fatigue, pain, and sleepdisturbance. Research on the underlying neurobiology of depres-sion in cancer suggests that these symptoms may have a commonunderlying basis (Raison & Miller, 2003).

Several limitations must be acknowledged. While informationon depressive and anxiety symptoms were obtained through validself-report measures, future studies need to include a clinicalevaluation of previous and concurrent psychiatric comorbidities.The fact that the major reasons for refusal were being too over-whelmed with their cancer treatment or too busy may have led toan underestimation or overestimation of depressive symptoms inthis sample. It is possible that the four latent classes may reflectsome unique characteristics of this sample.

Finally, while other studies have used GMM to identify distinctlatent classes of patients with and without cancer based on self-reports of depressive symptoms (Carragher et al., 2009; Colman etal., 2007; Helgeson et al., 2004; Henselmans et al., 2010; Hunteret al., 2010), the findings from this study must be interpreted withcaution until they are replicated in future studies. Ideally futurestudies should be done with sample sizes that are large enough toallow for confirmatory analyses of both the number and trajecto-ries of the latent classes, as well as the phenotypic and genotypiccharacteristics that are unique to each class.

Additional research is needed to learn whether these distinctlatent classes can be replicated. In addition to demographic andclinical variables, investigators should examine coping styles, per-sonality traits, and other preexisting individual characteristics aspotential mediators and moderators of latent class membership.Research with other seriously or chronically ill populations wouldassist investigators and clinicians to understand how individualdifferences manifest in response to illness. If these latent classeswith distinct symptom trajectories are reproduced, these findingswould strengthen the notion that underlying traits are essential tounderstanding the incidence and course of distress or resilience inindividuals affected by chronic and serious illnesses like cancer.

References

Bieling, P. J., Antony, M. M., & Swinson, R. P. (1998). The State-TraitAnxiety Inventory, Trait version: Structure and content re-examined.Behaviour Research and Therapy, 36, 777–788.

Bonanno, G. A. (2004). Loss, trauma, and human resilience: Have weunderestimated the human capacity to thrive after extremely aversiveevents? American Psychologist, 59, 20–28.

Bower, J. E. (2008). Behavioral symptoms in patients with breast cancerand survivors. Journal of Clinical Oncology, 26, 768–777.

Brunner, F., Bachmann, L. M., Weber, U., Kessels, A. G., Perez, R. S.,Marinus, J., & Kissling, R. (2008). Complex regional pain syndrome1–the Swiss cohort study. BMC Musculoskeletal Disorders, 9, 92.

Burgess, C., Cornelius, V., Love, S., Graham, J., Richards, M., & Ramirez,A. (2005). Depression and anxiety in women with early breast cancer:Five year observational cohort study. British Medical Journal, 330, 702.

Carpenter, J. S., Andrykowski, M. A., Wilson, J., Hall, L. A., Rayens,M. K., Sachs, B., & Cunningham, L. L. (1998). Psychometrics for twoshort forms of the Center for Epidemiologic Studies-Depression Scale.Issues in Mental Health Nursing, 19, 481–494.

Carragher, N., Adamson, G., Bunting, B., & McCann, S. (2009). Subtypesof depression in a nationally representative sample. Journal of AffectiveDisorders, 113, 88–99.

Caspi, A., Sugden, K., Moffitt, T. E., Taylor, A., Craig, I. W., Harrington,H., . . . Poulton, R. (2003). Influence of life stress on depression: Mod-eration by a polymorphism in the 5-HTT gene. Science, 301, 386–389.

Celeux, G., & Soromenho, G. (1996). An entropy criterion for assessingthe number of clusters in a mixture model. Journal of Classification, 13,195–212.

8 DUNN ET AL.

Colman, I., Ploubidis, G. B., Wadsworth, M. E., Jones, P. B., & Croudace,T. J. (2007). A longitudinal typology of symptoms of depression andanxiety over the life course. Biological Psychiatry, 62, 1265–1271.

Compas, B. E., Stoll, M. F., Thomsen, A. H., Oppedisano, G., Epping-Jordan, J. E., & Krag, D. N. (1999). Adjustment to breast cancer:Age-related differences in coping and emotional distress. Breast CancerResearch and Treatment, 54, 195–203.

Das-Munshi, J., Goldberg, D., Bebbington, P. E., Bhugra, D. K., Brugha,T. S., Dewey, M. E., . . . Prince, M. (2008). Public Health significance ofmixed anxiety and depression: Beyond current classification. BritishJournal of Psychiatry, 192, 171–177.

Delucchi, K. L., Matzger, H., & Weisner, C. (2004). Dependent andproblem drinking over 5 years: A latent class growth analysis. Drug andAlcohol Dependence, 74, 235–244.

Den Oudsten, B. L., Van Heck, G. L., Van der Steeg, A. F., Roukema,J. A., & De Vries, J. (2009). Predictors of depressive symptoms 12months after surgical treatment of early-stage breast cancer. Psychoon-cology, 18, 1230–1237.

Deshields, T., Tibbs, T., Fan, M. Y., & Taylor, M. (2006). Differences inpatterns of depression after treatment for breast cancer. Psychooncology,15, 398–406.

DiMatteo, M. R., Lepper, H. S., & Croghan, T. W. (2000). Depression isa risk factor for noncompliance with medical treatment: Meta-analysis ofthe effects of anxiety and depression on patient adherence. Archives ofInternal Medicine, 160, 2101–2107.

Donovan, K. A., Small, B. J., Andrykowski, M. A., Munster, P., &Jacobsen, P. B. (2007). Utility of a cognitive-behavioral model to predictfatigue following breast cancer treatment. Health Psychology, 26, 464–472.

Fann, J. R., Thomas-Rich, A. M., Katon, W. J., Cowley, D., Pepping, M.,McGregor, B. A., & Gralow, J. (2008). Major depression after breastcancer: A review of epidemiology and treatment. General HospitalPsychiatry, 30, 112–126.

Fletcher, B. S., Paul, S. M., Dodd, M. J., Schumacher, K., West, C.,Cooper, B., . . . Miaskowski, C. (2008). Prevalence, severity, and impactof symptoms on female family caregivers of patients at the initiation ofradiation therapy for prostate cancer. Journal of Clinical Oncology, 26,599–605.

Ganz, P. A., Guadagnoli, E., Landrum, M. B., Lash, T. L., Rakowski, W.,& Silliman, R. A. (2003). Breast cancer in older women: Quality of lifeand psychosocial adjustment in the 15 months after diagnosis. Journal ofClinical Oncology, 21, 4027–4033.

Gaston-Johansson, F., Ohly, K. V., Fall-Dickson, J. M., Nanda, J. P., &Kennedy, M. J. (1999). Pain, psychological distress, health status, andcoping in patients with breast cancer scheduled for autotransplantation.Oncology Nursing Forum, 26, 1337–1345.

Goldberg, D. P., Krueger, R. F., Andrews, G., & Hobbs, M. J. (2009).Emotional disorders: Cluster 4 of the proposed meta-structure forDSM-V and ICD-11. Psychological Medicine, 39, 2043–2059.

Gonda, X., Fountoulakis, K. N., Juhasz, G., Rihmer, Z., Lazary, J., Laszik,A., . . . Bagdy, G. (2009). Association of the s allele of the 5-HTTLPRwith neuroticism-related traits and temperaments in a psychiatricallyhealthy population. European Archives of Psychiatry and Clinical Neu-roscience, 259, 106–113.

Greenberg, D. B. (2004). Barriers to the treatment of depression in cancerpatients. Journal of the National Cancer Institute Monographs, 32,127–135.

Helgeson, V. S., Snyder, P., & Seltman, H. (2004). Psychological andphysical adjustment to breast cancer over 4 years: Identifying distincttrajectories of change. Health Psychology, 23, 3–15.

Henselmans, I., Helgeson, V. S., Seltman, H., de Vries, J., Sanderman, R.,& Ranchor, A. V. (2010). Identification and prediction of distresstrajectories in the first year after a breast cancer diagnosis. HealthPsychology, 29, 160–168.

Hunter, A. M., Muthen, B. O., Cook, I. A., & Leuchter, A. F. (2010).Antidepressant response trajectories and quantitative electroencephalog-raphy (QEEG) biomarkers in major depressive disorder. Journal ofPsychiatric Research, 44, 90–98.

Jacobsen, P. B., Donovan, K. A., Trask, P. C., Fleishman, S. B., Zabora, J.,Baker, F., & Holland, J. C. (2005). Screening for psychologic distress inambulatory cancer patients. Cancer, 103, 1494–1502.

Jung, T., & Wickrama, K. A. S. (2008). An introduction to latent classgrowth analysis and growth mixture modeling. Social and PersonalityPsychology Compass, 2, 302–317.

Karnofsky, D. (1977). Performance Scale. New York: Plenum Press.Karnofsky, D., Abelmann, W., & Craver, L. (1948). The use of nitrogen

mustard in the palliative treatment of cancer. Cancer, 1, 634–656.Kendler, K. S., Kuhn, J., & Prescott, C. A. (2004). The interrelationship of

neuroticism, sex, and stressful life events in the prediction of episodes ofmajor depression. American Journal of Psychiatry, 161, 631–636.

Kennedy, B. L., Schwab, J. J., Morris, R. L., & Beldia, G. (2001).Assessment of state and trait anxiety in subjects with anxiety anddepressive disorders. Psychiatric Quarterly, 72, 263–276.

Kreuter, F., & Muthen, B. O. (2008). Analyzing criminal trajectory pro-files: Bridging multilevel and group-based approaches using growthmixture modeling. Journal of Quantitative Criminology, 24, 1–31.

Kroenke, C. H., Rosner, B., Chen, W. Y., Kawachi, I., Colditz, G. A., &Holmes, M. D. (2004). Functional impact of breast cancer by age atdiagnosis. Journal of Clinical Oncology, 22, 1849–1856.

Merikangas, K. R., Zhang, H., Avenevoli, S., Acharyya, S., Neuenschwan-der, M., Angst, J., & Zurich Cohort Study. (2003). Longitudinal trajec-tories of depression and anxiety in a prospective community study: TheZurich Cohort Study. Archives of General Psychiatry, 60, 993–1000.

Mo, W., & Bodner, T. E. (2007). Growth mixture modeling: Identifyingand predicting unobserved subpopulations with longitudinal data. Orga-nizational Research Methods, 10, 635–656.

Mora, P. A., Bennett, I. M., Elo, I. T., Mathew, L., Coyne, J. C., &Culhane, J. F. (2009). Distinct trajectories of perinatal depressive symp-tomatology: Evidence from growth mixture modeling. American Jour-nal of Epidemiology, 169, 24–32.

Mosher, C. E., & Danoff-Burg, S. (2005). A review of age differences inpsychological adjustment to breast cancer. Journal of PsychosocialOncology, 23, 101–114.

Munafo, M. R., Clark, T. G., Roberts, K. H., & Johnstone, E. C. (2006).Neuroticism mediates the association of the serotonin transporter genewith lifetime major depression. Neuropsychobiology, 53, 1–8.

Murphy, B. M., Elliott, P. C., Higgins, R. O., Le Grande, M. R., Worcester,M. U., Goble, A. J., & Tatoulis, J. (2008). Anxiety and depression aftercoronary artery bypass graft surgery: Most get better, some get worse.European Journal of Cardiovascular Prevention and Rehabilitation, 15,434–440.

Muthen, B., & Muthen, L. K. (2000). Integrating person-centered andvariable-centered analyses: Growth mixture modeling with latent trajec-tory classes. Alcohol Clinical and Experimental Research, 24, 882–891.

Muthen, B. O. (2001). Latent variable mixture modeling. In G. A. Mar-coulides & R. E. Schumacker (Eds.), New Developments and Techniquesin Structural Equation Modeling (pp. 1–33). Mahwah, NJ: Erlbaum.

Muthen, B. O. (2002). Beyond SEM: General latent variable modeling.Behaviormetrika, 29, 81–117.

Muthen, B. O., Collins, L. M., & Sayer, A. G. (2001). Second-generationstructural equation modeling with a combination of categorical andcontinuous latent variables: New opportunities for latent class-latentgrowth modeling. In New methods for the analysis of change. Decade ofbehavior. (Vol. 1, pp. 291–322). American Psychological Association(APA).

Muthen, B. O., & Kaplan, D. W. (2004). Latent variable analysis: Growthmixture modeling and related techniques for longitudinal data. In D.

9DISTINCT DEPRESSIVE SYMPTOM TRAJECTORIES

Kaplan (Ed.), The sage handbook of quantitative methodology for thesocial sciences (pp. 345–368). Newbury Park, CA: Sage.

Muthen, L. K., & Muthen, B. O. (1998–2008). Mplus User’s Guide (5thEdition). Los Angeles: Muthen & Muthen.

Muthen, L. K., & Muthen, B. O. (2008). Mplus (Version 5.1). Los Angeles:Muthen & Muthen.

Muthen, L. K., & Muthen, B. O. (2009). Mplus (Version 5.21). LosAngeles: Muthen & Muthen.

Nagin, D. S., & Tremblay, R. E. (2001). Analyzing developmental trajec-tories of distinct but related behaviors: A group-based method. Psycho-logical Methods, 6, 18–34.

Nylund, K. L., Asparouhov, T., & Muthen, B. O. (2007). Deciding on thenumber of classes in latent class analysis and growth mixture modeling:A Monte Carlo simulation study. Structural Equation Modeling, 14,535–569.

Parker, P. A., Baile, W. F., de Moor, C., & Cohen, L. (2003). Psychosocialand demographic predictors of quality of life in a large sample of cancerpatients. Psychooncology, 12, 183–193.

Petras, H., Kellam, S. G., Brown, C. H., Muthen, B. O., Ialongo, N. S., &Poduska, J. M. (2008). Developmental epidemiological courses leadingto antisocial personality disorder and violent and criminal behavior:Effects by young adulthood of a universal preventive intervention infirst- and second-grade classrooms. Drug and Alcohol Dependence, 95,S45–59.

Pirl, W. F. (2004). Evidence report on the occurrence, assessment, andtreatment of depression in cancer patients. Journal of the NationalCancer Institute Monographs, 32, 32–39.

Radloff, L. S. (1977). The CES-D scale: A self-report depression scale forresearch in the general population. Applied Psycholical Measurement, 1,385–401.

Raison, C. L., & Miller, A. H. (2003). Depression in cancer: New developmentsregarding diagnosis and treatment. Biological Psychiatry, 54, 283–294.

Roeder, K., Lynch, K. G., & Nagin, D. S. (1999). Modeling uncertainty inlatent class membership: A case study in criminology. Journal of theAmerican Statistical Association, 94, 766–776.

Rose, J. H., Kypriotakis, G., Bowman, K. F., Einstadter, D., O’Toole, E. E.,Mechekano, R., & Dawson, N. V. (2009). Patterns of adaptation inpatients living long term with advanced cancer. Cancer, 115(18 Suppl),4298–4310.

Sangha, O., Stucki, G., Liang, M. H., Fossel, A. H., & Katz, J. N. (2003).The Self-Administered Comorbidity Questionnaire: A new method toassess comorbidity for clinical and health services research. Arthritis andRheumatism, 49, 156–163.

Schafer, J. L., & Graham, J. W. (2002). Missing data: Our view of the stateof the art. Psychological Methods, 7, 147–177.

Sheehan, T. J., Fifield, J., Reisine, S., & Tennen, H. (1995). The measure-ment structure of the Center for Epidemiologic Studies DepressionScale. Journal of Personality Assessment, 64, 507–521.

Spielberger, C. (1983). Manual for the State-Trait Anxiety Inventory (FormY). Palo Alto, CA: Consulting Psychologists Press.

SPSS. (2010). SPSS for Windows (Version 18). Chicago: SPSS, Inc.Stark, D., Kiely, M., Smith, A., Velikova, G., House, A., & Selby, P.

(2002). Anxiety disorders in cancer patients: Their nature, associations,and relation to quality of life. Journal of Clinical Oncology, 20, 3137–3148.

Taylor, D. H., Jr., Ezell, M., Kuchibhatla, M., Ostbye, T., & Clipp, E. C.(2008). Identifying trajectories of depressive symptoms for womencaring for their husbands with dementia. Journal of the American Ge-riatrics Society, 56, 322–327.

Tofighi, D., & Enders, C. K. (2008). Identifying the correct number ofclasses in growth mixture models. In G. R. Hancock & K. M. Samuelsen(Eds.), Advances in latent variable mixture models (pp. 317–341). Char-lotte, NC: Information Age Publishing.

10 DUNN ET AL.