Trauma-Focused Cognitive-Behavioral Therapy for Children Sustained Impact of Treatment 6 and 12 Months Later

Abstract

This study presents the findings from 6- and 12-month follow-up assessments of 158 children ages 4–11 years who had experienced sexual abuse and who had been treated with Trauma-Focused Cognitive-Behavioral Therapy (TF-CBT) with or without the inclusion of the trauma narrative (TN) treatment module and in 8 or 16 treatment sessions. Follow-up results indicated that the overall significant improvements across 14 outcome measures that had been reported at posttreatment were sustained 6 and 12 months after treatment and on two of these measures (child self-reported anxiety and parental emotional distress) there were additional improvements at the 12-month follow-up. Higher levels of child internalizing and depressive symptoms at pretreatment were predictive of the small minority of children who continued to meet full criteria for posttraumatic stress disorder at the 12-month follow-up. These results are discussed in the context of the extant TF-CBT treatment literature.

Trauma-focused Cognitive-Behavioral Therapy (TF-CBT) is an evidence-based treatment for child trauma. In a recent meta-analysis of psychosocial treatments for children and adolescents exposed to traumatic events, TF-CBT was the only treatment meeting the “well-established criteria” threshold (Silverman et al., 2008). TF-CBT consists of the following three treatment modules or phases, each provided to children and parents: (1) skills-building components to enhance children’s affective, behavioral, biological, and cognitive self-regulation and parenting interventions to enhance caregiver coping, behavior management skills, and support of the child; gradual exposure to the child’s trauma reminders is included throughout these components; (2) trauma narrative (TN) during which children describe and cognitively process their personal trauma experiences; and (3) treatment closure including conjoint caregiver–child sessions and safety planning. Typically, one third of TF-CBT treatment is dedicated to each of these modules. The major purpose of the TN is to provide more intensive exposure work. This builds on the desensitization process that had been initiated with the skills-based components (Cohen, Mannarino, & Deblinger, 2006).

A two-site, randomized study was conducted to examine the differential effects of TF-CBT provided with or without the TN during 8 or 16 sessions to address the questions of (1) whether and for which children the TN is beneficial; and (2) the optimal length of TF-CBT treatment. Within the two TN conditions, the proportionality of the three TF-CBT phases was also varied. In the 8 session TN condition, normal TF-CBT proportionality was maintained (3 sessions for skills; 3 for TN; 2 for treatment closure) but in the 16-session TN condition, this was altered with disproportionate time spent on the TN (3 sessions for skills; 11 for TN, 2 for treatment closure). The rationale for adding 8 TN sessions and no other TF-CBT components to the 16-session TN condition was to increase the probability that treatment differences related to this condition could be interpreted as being the result of more intensive exposure work and not other factors. We hypothesized that (1) all four conditions would significantly improve posttraumatic stress disorder (PTSD) symptoms since a previous study documented that TF-CBT skills alone can effectively do so (CATS Consortium, 2010); (2) the 16 session conditions (YES TN and No TN) would be more effective than the 8 session conditions (Yes TN and No TN) for improving PTSD since 8 sessions would be of insufficient duration; and (3) the 8-session Yes TN condition would be superior to the 8-session No TN condition and possibly to the other conditions for improving children’s fear and anxiety, since this condition provided all TF-CBT phases in optimal proportions.

The original study included 179 children ages 4–11 years who had experienced childhood sexual abuse (CSA). Significant posttreatment improvements were found across all four treatment conditions for 14 outcome measures, with no differences between conditions with respect to child PTSD hyperarousal, internalizing behaviors, depressive symptoms, sexualized behaviors, shame, body safety skills, and parental depression (Deblinger, Mannarino, Cohen, Runyon, & Steer, 2011). The 8-session TN condition was more effective and efficient for reducing parents’ abuse-specific distress and children’s general anxiety than the 8-session No TN condition while both TN conditions were more effective in reducing abuse-related fear than the No TN conditions. Also, the 16-session conditions were more effective at improving PTSD reexperiencing and avoidance than the 8-session conditions. Finally, the 16-session No TN condition was more effective at improving parenting practices than both TN conditions while both No TN conditions were more effective in reducing externalizing behavior problems than both TN conditions.

Our primary hypothesis for the current follow-up study was that all of the improvements attained at posttreatment would be maintained at 6- and 12-month follow-ups. A secondary hypothesis was that the above differences among conditions would be maintained for the duration of the follow-up. We also examined predictors of nonresponse to treatment, that is, whether background or clinical characteristics of the children and their caregivers might identify the children who continued to meet DSM-IV-TR (Diagnostic and Statistical Manual of Mental Disorders (Fourth edition, text revision) diagnostic criteria for PTSD 12 months after treatment. Our previous study for 8- to 14-year-olds found that more severe pretreatment depression predicted later PTSD for children randomly assigned to the client-centered therapy condition but not for those assigned to the TF-CBT condition (Deblinger, Mannarino, Cohen, & Steer, 2006). However, a recent meta-analysis demonstrated that child maltreatment predicts worse treatment outcomes for youth and adults with depression (Nanni, Uher & Danese, 2012). Therefore, we were particularly interested to see whether pretreatment depressive symptoms predicted poorer treatment response to TF-CBT in younger children.

Method

Initial Sample

Detailed descriptions about the randomized controlled design, treatment procedures, and the specific posttreatment outcomes for the 14 self-report and clinical rating scales that were employed are described by Deblinger and colleagues (2011). Briefly, 210 children (ages 4–11) and their parents or caregivers were initially randomly assigned to 8 No TN, 8 Yes TN, 16 No TN, and 16 Yes TN conditions. The term parent is used here to describe the parent or primary caregiver who participated in the treatment and provided the principal information about the child. The children were recruited from two sites (Pittsburgh, PA, and Stratford, NJ). The assessment and treatment protocols were consistent across sites, and the study was approved by each site’s respective Institutional Review Board. The majority of children (84%) were living with one or both biological parents. Although children in foster care were included in the study and 15% were living in formal or informal foster homes, only two children were living with nonrelative foster parents. The rest were living in kinship foster placements (i.e., with grandparents, aunts, or other family members) that remained stable throughout the duration of the study. The informant changed during the course of the study in less than 1% of the cases.

To be eligible for the study, children must have experienced contact sexual abuse, which was confirmed by NJ’s Division of Youth and Family Services, PA’s Department of Children, Youth, and Families, a law enforcement official, or a professional with recognized expertise in conducting evaluations of CSA. Children had at least five PTSD symptoms, including at least one avoidance, reexperiencing, and hyperarousal symptom. With respect to exclusion criteria, children with significant developmental disabilities (IQ < 70) were excluded. A child or a parent could not have a serious medical or mental health illness (i.e., psychosis) that would interfere with his or her participation in treatment but no children were excluded for these reasons.

Sixteen (8%) children and their parents never returned after being assigned to treatment, and 15 (7%) children and their parents left after attending only one or two sessions. These 31 (15%) children and their parents were defined as dropouts. A meta-analysis of cognitive behavioral therapy (CBT) studies by Barkham, Shapiro, Hardy, and Rees (1999) found that three sessions are required for benefits to occur. Additionally, we used this cut-off for our previous multi-site study (Cohen, Deblinger, Mannarino, & Steer, 2004) after conducting intention to treat analyses that documented that there were no statistical differences between those completing three sessions and those who completed more sessions. Therefore, we chose to base our present analyses on those with three or more sessions as we had done in our previous study. Thus, 179 (85%) children attended at least three TF-CBT sessions, and these completers composed the sample that Deblinger and colleagues (2011) analyzed in their study. However, there were only 158 children and 144 parents who were assessed at the end of treatment on the principal outcome measures. The difference between the number of parents and children was attributable to 17 (8%) siblings being included in the study. Siblings were randomized to the same condition and the results did not differ as a function of sibling inclusion. Deblinger and colleagues (2011) reported that the missing data in their study were found to be “missing at random” and thus did not appear to bias the reported results.

Follow-Up Sample

Because we were primarily interested in ascertaining whether any changes in the 14 outcome measures had occurred between the end of therapy and the follow-up evaluations at 6 and 12 months, the current sample was restricted to the 158 children and 144 parents who had completed at least structured diagnostic interviews at the end of treatment. This sample was composed of 98 (62%) girls and 60 (38%) boys whose mean age at admission had been 7.60 (SD = 2.07) years. The children represented 103 (65%) Caucasians, 23 (15%) African Americans, 10 (6%) Hispanic-Americans, and 22 (14%) with other ethnic origins. Oral–genital contact, penile penetration, or both had been experienced by 96 (61%) of the children, and 66 (42%) of the perpetrators were related or unrelated adults with the remaining perpetrators being older children or teens. Of the 144 parents, the modal parent was the child’s biological mother (N = 121, 84%) who was either currently married or cohabiting with a partner (N = 80, 56%) and who was employed either full-time or part-time (N = 86, 60%). Fifty-seven (40%) parents reported that they had also experienced contact CSA.

Outcome Measures

Except where noted, measures were collected for all children and raw total scores were used for all instruments. The rationale for including a wide variety of outcome measures (as opposed to focusing only on one outcome such as PTSD) was that even young traumatized children, and particularly those who experience sexual abuse, are known to develop difficulties in diverse domains (e.g., Cohen, Mannarino & Deblinger, 2010), and that it is important to obtain information from multiple reporters (e.g., parents and children). Most parents did not want schools to know that their children were participating in this project so it was not feasible to obtain teacher report measures.

The Schedule for Affective Disorders and Schizophrenia for School-Age Children-Present and Lifetime Version (K-SADS; Kaufman et al., 1997), a semistructured interview, was administered independently to all children and caregivers in the study as a “gold standard” interviewer-administered assessment of DSM-IV-TR PTSD symptoms and diagnostic status. Consensus ratings (caregiver and child) for each item were used, using the “or” criteria; that is, if the caregiver or child endorsed an item, it was considered to be present as this is considered to yield the most reliable PTSD diagnosis for prepubertal youth (Scheeringa, 2008). The number of KSADS symptoms representing Reexperiencing, Avoidance, and Hyperarousal DSM-IV-TR symptom clusters, respectively, was summed to construct a total score representing each cluster. The intra-class correlations for the Reexperiencing, Avoidance, and Hyperarousal total scores in the initial sample were, respectively, .92, .85, and .84. A higher score represents more severe PTSD symptoms.

Parent Report Measures

The Beck Depression Inventory-II (BDI-II; Beck, Steer, & Brown, 1996) is a self-report instrument that was used to measure the severity of parental depression. The BDI-II has generally been found to have high internal consistency. The coefficient α of the BDI-II total scores in the initial sample was .92. A higher score indicates more severe depression.

The Child Behavior Checklist (CBCL; Achenbach, 1991) is a 120-item parental rating scale that assesses children’s externalizing and internalizing behavioral problems. These subscales display good convergent and discriminant validities (Seligman, Ollendick, Langley, & Baldacci, 2004). Internal consistency α for the internalizing scale was .88 and for the externalizing scale was .92. A higher score indicates more severe behavior problems. It should be noted that T scores were not used in the data analysis for the CBCL. As children were being evaluated over a year, changes in T scores might have been spurious simply because subjects potentially were being compared with a different normative group.

The Child Sexual Behavior Inventory (CSBI; Friedrich et al., 1992) is a 42-item parent rating inventory that assesses children’s sexual behavior problems on a 4-point Likert-type scale from 0 = never to 3 = at least once a week. A sample question is “Kisses other children they do not know well.” The test–retest reliability after 1 month has been found to be .80. Internal consistency coefficient α for the initial sample was .84. A higher score indicates the presence of more sexual behavior problems.

The Parent Emotional Reaction Questionnaire (PERQ; Cohen & Mannarino, 1996) is a parent self-report instrument to assess parental distress related to their child being sexually abused. It is rated on a 5-point Likert-type scale (1 = never; 5 = always). A sample question is “I have felt upset about my child being abused.” The 2-week test–retest reliability was found to be .90. Internal consistency was .87 and coefficient α in the initial sample was .90. A higher score indicates greater parental distress.

The Parent Practices Questionnaire (PPQ; Strayhorn & Weidman, 1988) is a 35-question self-report parent measure that assesses parents’ positive parenting interactions with their children on either a 7- or 5-point Likert-type scale (0 = never to 5 = many times each day). A sample question is “How often do you physically punish your child, for example by spanking?” A modified version of the PPQ was used in this study (Deblinger, Stauffer, & Steer, 2001). Internal consistency is .78. The coefficient α for the PPQ total scores in the initial sample was .82. A higher score indicates more optimal parenting practices.

Child Report Measures

The Children’s Depression Inventory (CDI; Kovacs, 1992) is a well-validated and reliable 27-item self-report instrument used to assess children’s depression for children ages 7 and older (Seligman et al., 2004). Thus, the CDI was only administered to subjects aged 7 or older in this study. The coefficient α for the CDI total scores in the initial sample was .87. A higher score indicates more severe depressive symptoms.

The Fear Thermometer (Fear; Hersen & Bellack, 1988) consists of a pictorial representation of a thermometer to assess children’s level of fear or discomfort. Children were asked to mark the segment that best described their current level of fear/discomfort. Good test–retest coefficients and concurrent validities have been established for the Fear measure (Hersen & Bellack, 1988). A higher score indicates greater fear.

The Multidimensional Anxiety scale for children (MASC; March, Parker, Sullivan, Stallings, & Conners, 1997) is a 39-item self-report measure with well-established construct validity that was used to assess global anxiety in this study for children ages 7 and older. The coefficient α of the MASC total scores in the initial sample was .90. A higher score indicates greater anxiety.

The Shame Questionnaire (Shame; Feiring, Taska, & Lewis, 1999) is an 8-item self-report instrument used to assess children’s feelings of shame about being abused. The measure has demonstrated excellent test–retest reliability and good construct and predictive validity. The coefficient α for the Shame total scores in the initial sample was .65. A higher score indicates greater shame.

The What If Situations Test (WIST; Sarno & Wurtele, 1997) is a brief interview using vignettes that was used to assess children’s abilities to recognize and respond effectively to hypothetical abusive situations. The coefficient α for the WIST Skill scores in the initial sample was .84. Higher scores indicate greater safety skills.

Procedures

Detailed descriptions about the TF-CBT treatment are described by Deblinger and colleagues (2011). Therapy was provided in 90-min sessions, most of which were divided into 45-min individual sessions for the child and caregiver, respectively. Some sessions included 30 min of conjoint parent–child time. All four therapists provided all four treatment conditions. The TN component was only provided to those who had been randomly assigned to the two Yes TN conditions. Although gradual exposure was provided during the skills phase to all children, only the children assigned to the two TN conditions were actively encouraged to develop and cognitively process a detailed narrative about their sexual abuse and related experiences which they reviewed with the therapist and caregiver. Children and caregivers assigned to the No TN conditions focused more sessions on psychoeducation about sexual abuse and skills-building exercises. In all conditions, both children and parents received TF-CBT components involving psychoeducation about sexual abuse and skills (i.e., relaxation, affective modulation, cognitive coping, and body safety training) as well as parenting skills training. Parents and children assigned to the 16 No TN condition were given more opportunities to review psychoeducational material and engage in skill repetition and practice. As described above, the parents and children in the 16 TN condition participated in 11 TN sessions that focused on reviewing the details of their traumatic experiences and their associated thoughts, feelings, and sensations through the expansion of their narrative as well as other exposure activities (e.g., writing a poem or letter related to the sexual abuse).

A project coordinator at each site had provided a detailed explanation of the study to the parents and children. After the parent and child read and signed their respective consent and assent forms, the project coordinator completed the screening questions and administered the assessment battery, if appropriate. Children under the age of 7 years were administered only the instruments appropriate for their age; this primarily explains why fewer children in the present sample had depression and anxiety outcome data. Following the initial assessment, each child was randomly assigned to one of the four treatment conditions. The families were told that they would be paid $25 for completing the initial evaluation, $25 for a 4-week evaluation, and $50 for an 8-week evaluation, 16 week evaluation, and 6- and 12-month evaluations. Because the project coordinator was blind to the treatment assignment, the children and parents were told of the treatment assignment by the therapist during the first treatment session. All siblings participating in the study were assigned to the same condition. The therapists had graduate degrees in psychology, clinical social work, or a related field and had at least 3 years of clinical experience. The project coordinator at each site also conducted all of the 6- and 12-month follow-up evaluations. Follow-up assessments were conducted as close as possible to the actual 6- or 12-month anniversary of the treatment completion date but always within 2 weeks of that date.

Data Analysis

A mixed-model approach was used to compare the adjusted posttreatment, 6-month, and 12-month scores for the 14 outcome measures. The overall design was assumed to reflect a three-factor, repeated-measures analysis of covariance (MM-RANCOVA) in which the main effects were for length of treatment (8 sessions or 16 sessions), inclusion of the TN (No, Yes), and time (posttreatment, 6- and 12-month follow-ups). The covariate for the posttreatment, 6-month, and 12-month scores was the respective pretreatment score for the outcome measure score under analysis. The children and parents were again considered to represent a random effect. Because an MM-RANCOVA uses a restricted maximum likelihood (REML) technique to generate parameter estimates from all of the available data, unbiased estimates of the variance–covariance matrix can be also assessed by taking into account the loss of degrees of freedom attributable to the fixed effects in the model. Therefore, the present study tested to determine if the variance–covariance matrices for each of the 14 outcome measures corresponded to unstructured, compound symmetry, or first-order autoregressive effects (Little, Milliken, Stroup, & Wolfinger, 1996). Based on the values of the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) indices for each variable’s matrix, it was consistently found that the assumption of compound symmetry afforded the lowest AIC and BIC values across all of the 14 outcome measures while the parameter estimates for each variance–covariance term were significantly different from 0 according to the Wald statistics.

The SAS Multiple Imputation and Multiple Imputation Analysis procedures were also used to estimate the missing 6- and 12-month follow-up outcome scores, and 10 maximum-likelihood complete sets of data were again generated for each of the 14 outcome measures based on the initial number of respondents with posttreatment scores. These imputation techniques are the same as those used by Deblinger and colleagues (2011). Once again, all of the parameter estimates for the multiple imputation data sets corresponded to those that had been found without imputing missing data. Consequently, only the MM-RANCOVA results for non-imputed data for each outcome measure are described. Hedges’ g (Hedges & Olkin, 1985) was used to estimate the effects sizes of the adjusted mean differences for significant mean differences.

A power analysis based on the use of a MM-RANCOVA controlling for a pretest measure would yield a power of 80% for detecting a significant difference for a medium effect size f (Cohen, 1992) of .30 for the main effects of number of sessions (8 vs. 16), use of TN (No, Yes), and time (posttreatment, 6 months, and 12 months) along with the main effects’ interactions for an α of .0037, two-tailed test. The .0037 level represented a Bonferroni adjustment of .05/14 to control the familywise error rate of performing 14 separate MM-RANCOVAs.

To determine whether any pretreatment variables significantly predicted meeting DSM-IV-TR PTSD diagnostic criteria at 12 months, six selected background characteristics and the 14 pretreatment scores of the children and parents were regressed on child PTSD status at 12 months (0 = No, 1 = Yes). Even though it might be expected that the 14 symptom outcome measures (e.g., anxiety and depression) would be positively correlated with each other, the number of children completing all instruments (ranging from 70 to 140) was too small to conduct a multiple regression approach in which all 20 predictors were simultaneously regressed on meeting PTSD diagnostic criteria. Therefore, a separate logistic regression analysis was used for each predictor, and a Bonferroni adjustment of α (.05/20) was used to control for the familywise error rate of performing 20 individual analyses.

Results

Correlates of Completing Follow-Up Evaluations

Before conducting the MM-RANCOVAs, it was determined (a) whether the children and parents who completed 6- and 12-month follow-up evaluations were different from those who had not completed 6- and 12-month follow-up evaluations and (b) whether other variables might need to be controlled for in the MM-RANCOVAs, besides the pretreatment outcome measure scores. The selected background and clinical characteristics were sex, being Caucasian, age, treatment site, length of treatment, and inclusion of the TN (0 = No, 1 = Yes). Using a Bonferroni adjustment of α divided by 20 to control for the familywise error rate from calculating 20 correlations for each of the 6- and 12-month follow-up completion variables, there was only one significant correlation for age, r = −.28, p < .01 as the child’s age was inversely related to completing a 12-month follow-up evaluation. This correlation was < .30. Therefore, age was not employed as another covariate in the MM-RANCOVAs because Allison (1995) found that covariates with correlations < .30 did not have any significant or meaningful clinical impact upon outcome. Furthermore, Deblinger and colleagues (2011) had previously found that age was neither a significant covariate nor yielded significant interactions in their MM-ANCOVAs with the 14 outcome measures. Given the general lack of significant and meaningful correlations (r > .30), it was concluded that there were no important differences between the children and parents who did and did not complete 6- and 12-month follow-up evaluations.

The numbers of children or parents completing either the 6- or 12-month follow-up evaluations seldom varied by more than five subjects within each of the four treatment conditions. However, it should not be assumed that the same children and their parents completed both the 6- and 12-month evaluations. The odds ratio for completing both sets was 9.51 (95% Confidence interval [CI]: [4.34, 20.81]), but only 92 (58%) of the 158 children completed both 6- and 12-month follow-up evaluations, and 31 (20%) children did not complete either follow-up evaluation. However, 15 children (9%) who had not completed a 6-month evaluation completed a 12-month evaluation, whereas 20 (13%) children who had completed a 6-month evaluation failed to complete a 12-month evaluation, McNemar Test, p = .50.

Posttreatment and Follow-Up Comparisons

Table 1 presents the adjusted means (M_adj) and standard errors (SE) for the adjusted means as calculated by the MM-RANCOVAs that were performed for the 14 outcome measures. Using a Bonferroni adjustment of α divided by 14 to control for the familywise error rate, there were only two significant differences with respect to time, and no significant interactions of time with the main effects of length of therapy or the inclusion of the TN component. The MM-RANCOVA results in Table 1 thus indicate that the overall significant improvements across the 14 outcome measures that had been reported by Deblinger and colleagues (2011) at posttreatment were sustained 6 and 12 months after treatment. However, the adjusted mean score for parental distress at posttreatment was higher than the adjusted mean score for parental distress at 12 months, and this difference represents a medium effect size (Hedges’ g = .60). This difference indicates that the parents’ distress related to their child being sexually abused had continued to decrease 12 months later. Likewise, children’s adjusted mean anxiety scores at posttreatment were higher than the adjusted mean anxiety scores at 12 months, and this adjusted mean difference also reflects a medium effect size (Hedges’ g = .61). These results suggest that the children’s levels of general anxiety had continued to decrease at 12 months.

Table 1.

Mixed-Model Repeated Analyses of Covariance for Outcome Measures

			Time
			Posttest		6 Months		12 Months
Length of therapy	Trauma narrative	N	M adj	SE	M adj	SE	M adj	SE
K-SADS-Reexperiencing
8 Sessions	No	40	1.76	0.22	1.37	0.26	1.03	0.25
8 Sessions	Yes	39	1.86	0.22	1.44	0.25	1.25	0.25
16 Sessions	No	35	1.02	0.24	0.92	0.26	0.71	0.26
16 Sessions	Yes	44	1.41	0.21	1.62	0.24	1.46	0.25
Total			1.51	0.11	1.34	0.13	1.11	0.13
K-SADS-Avoidance
8 Sessions	No	40	1.77	0.21	1.68	0.24	1.36	0.23
8 Sessions	Yes	39	1.68	0.21	0.93	0.24	0.84	0.24
16 Sessions	No	35	1.09	0.22	1.00	0.24	0.74	0.25
16 Sessions	Yes	44	1.35	0.20	1.36	0.22	1.38	0.23
Total			1.47	0.10	1.24	0.12	1.08	0.12
K-SADS-Hypervigilance
8 Sessions	No	40	1.49	0.22	1.35	0.25	0.90	0.25
8 Sessions	Yes	39	1.26	0.22	1.20	0.25	1.06	0.25
16 Sessions	No	35	0.78	0.23	0.80	0.25	0.60	0.26
16 Sessions	Yes	44	1.33	0.21	1.22	0.23	1.56	0.24
Total			1.24	0.11	1.14	0.12	1.03	0.13
CBCL-Internalizing
8 Sessions	No	40	7.10	1.03	6.32	1.16	5.63	1.14
8 Sessions	Yes	38	6.75	1.04	7.17	1.13	6.73	1.11
16 Sessions	No	35	4.01	1.08	5.99	1.15	5.25	1.19
16 Sessions	Yes	43	6.64	0.98	7.69	1.06	7.61	1.12
Total			6.13	0.51	6.79	0.56	6.31	0.57
CBCL-Externalizing
8 Sessions	No	40	8.64	1.03	8.41	1.22	6.40	1.19
8 Sessions	Yes	38	9.97	1.04	10.13	1.18	9.53	1.15
16 Sessions	No	35	6.08	1.10	7.66	1.19	8.06	1.25
16 Sessions	Yes	43	10.35	0.99	10.34	1.10	10.34	1.19
Total			8.76	0.52	9.14	0.59	8.58	0.60
CSBI-Child sexual behavior
8 Sessions	No	39	3.85	0.69	1.81	0.81	2.00	0.79
8 Sessions	Yes	38	3.45	0.70	3.80	0.78	3.44	0.76
16 Sessions	No	35	1.68	0.73	2.58	0.79	2.39	0.82
16 Sessions	Yes	43	3.13	0.66	2.43	0.73	2.10	0.78
Total			3.03	0.35	2.66	0.39	2.48	0.39
PERQ-Parental distress
8 Sessions	No	38	32.25	1.68	32.34	1.88	27.68	1.83
8 Sessions	Yes	36	27.77	1.69	25.38	1.82	24.59	1.79
16 Sessions	No	30	27.19	1.86	25.14	1.98	25.59	1.96
16 Sessions	Yes	36	30.64	1.71	30.96	1.81	27.93	1.85
Total			29.46	0.87	28.45	0.93	26.45	1.28
Time: F(2,198) = 9.08** MadjDiff. Posttest > 12 months
PPQ-Parenting practices
8 Sessions	No	37	151.86	1.38	149.17	1.63	152.86	1.60
8 Sessions	Yes	36	151.12	1.40	151.45	1.60	150.57	1.58
16 Sessions	No	31	156.07	1.51	154.51	1.71	154.63	1.68
16 Sessions	Yes	36	149.26	1.40	149.54	1.56	148.24	1.63
Total			152.08	0.71	151.17	0.81	151.58	0.81
BDI-II-Parental depression
8 Sessions	No	38	8.05	1.16	9.57	1.34	6.20	1.33
8 Sessions	Yes	37	6.31	1.17	6.48	1.31	6.99	1.28
16 Sessions	No	31	5.50	1.28	4.98	1.39	4.20	1.39
16 Sessions	Yes	36	8.60	1.20	9.99	1.31	9.60	1.40
Total			7.12	0.60	7.75	0.67	6.75	0.67
CDI-Child depression
8 Sessions	No	28	6.51	1.00	3.78	1.27	4.12	1.21
8 Sessions	Yes	26	6.20	1.03	5.65	1.32	4.49	1.32
16 Sessions	No	29	6.09	0.97	3.32	1.04	4.61	1.08
16 Sessions	Yes	29	4.61	0.96	4.03	I.I0	4.90	I.I9
Total			5.85	0.49	4.20	0.59	4.53	0.60
MASC-Child anxiety
8 Sessions	No	28	48.86	2.95	45.89	3.73	34.74	3.50
8 Sessions	Yes	26	36.78	3.06	34.24	3.80	31.96	3.78
16 Sessions	No	28	36.51	2.88	37.64	3.08	35.81	3.13
16 Sessions	Yes	29	41.98	2.83	42.84	3.17	37.81	3.34
Total			41.03	1.45	40.15	1.71	35.08	1.71
Time. F(2,I44) = 7.35* MadjDiff. Posttest > 12 months
Child shame
8 Sessions	No	35	3.24	0.48	1.97	0.60	1.50	0.57
8 Sessions	Yes	27	2.21	0.55	2.20	0.66	2.08	0.64
16 Sessions	No	33	1.85	0.49	1.53	0.54	1.41	0.55
16 Sessions	Yes	29	2.93	0.52	1.57	0.56	2.48	0.62
Total			2.56	0.25	1.82	0.29	1.87	0.30
Child fear
8 Sessions	No	40	3.02	0.22	1.87	0.27	1.94	0.26
8 Sessions	Yes	38	2.14	0.23	2.10	0.26	1.69	0.26
16 Sessions	No	36	2.66	0.24	2.18	0.26	2.20	0.27
16 Sessions	Yes	42	1.97	0.22	2.35	0.24	2.14	0.26
Total			2.45	0.11	2.12	0.13	1.99	0.13
WIST-Child safety skills
8 Sessions	No	41	19.44	0.61	20.30	0.74	19.21	0.71
8 Sessions	Yes	39	19.34	0.62	18.73	0.73	18.57	0.71
16 Sessions	No	36	19.86	0.65	20.58	0.72	19.72	0.73
16 Sessions	Yes	44	19.07	0.59	18.83	0.66	18.68	0.71
Total			19.42	0.31	19.61	0.36	19.05	0.36

Note. BDI-II = Beck Depression Inventory-II, CBCL = Child Behavior Checklist, CDI = Children’s Depression Inventory, Fear = Fear Thermometer, K-SADS = Schedule for Affective Disorders and Schizophrenia for School Age Children, MASC = Multidimensional Anxiety Scale for Children, PERQ = Parent Emotional Reaction Questionnaire, PPQ = Parent Practices Questionnaire, WIST = What If Situations Test Safety Skills subscale, TN = trauma narrative, M_adj = adjusted mean, M_adjDiff = adjusted mean difference, SE = standard error of the adjusted mean

^*p < .05, Bonferroni adjusted (α/14).

^**p < .01

Predicting PTSD at 12 Months

Although the MM-RANCOVAs found that the posttreatment gains for 12 outcome measures had been sustained and two outcome measures had improved at the 12-month follow-up, there were 12 (11%) of the 113 children who completed 12-month PTSD assessments who continued to meet DSM-IV-TR diagnostic criteria for PTSD. Five of these 12 children had not met PTSD diagnostic criteria at the end of treatment, but seven children who had met PTSD diagnostic criteria at the end of treatment continued to meet such criteria at 12 months. In contrast, 80 (71%) of these 113 children had met such criteria at admission to the study; the 60% decrease in the rate of meeting PTSD diagnostic criteria at the end of 12 months was thus significant, McNemar Test, p < .001.

To ascertain whether sex (0 = Male, 1 = Female), being Caucasian (0 = No, 1 = Yes), age (years), treatment site (0 = PA, 1 = NJ), length of treatment (0 = 8 sessions, 1 = 16 sessions), inclusion of the TN (0 = No, 1 = Yes) and the pretreatment scores of the 14 outcome measures predicted meeting PTSD diagnostic criteria at 12 months, meeting PTSD diagnostic criteria at the 12-month follow-up (0 = No, 1 = Yes) was separately regressed on each of the aforementioned 20 variables. (Since there were only 12 [11%] children who met PTSD criteria at the 12-month follow-up, this number of children was too small to permit a multiple logistic regression approach, especially when the number of children who had been administered the self-report depression and anxiety scales decreased to 70 as Table 2 shows.) Table 2 displays the logistic regression results for each of the 20 variables. Again, using a Bonferroni adjustment of α divided by 20 to control for the familywise error rate, there were only two significant predictors; pretreatment internalizing behavior problems and pretreatment depression raw total scores. For every 1-point increase in internalizing raw total scores at pretreatment, there was a 9% increase in the odds of meeting diagnostic criteria for PTSD at 12 months. The increase in the odds of meeting PTSD diagnostic criteria for depression raw total scores was approximately twice the odds for the Internalizing scores. For every 1-point increase in the depression raw total scores at pretreatment, the odds for meeting criteria for PTSD at 12 months rose by 17%.

Table 2.

Logistic Regression Analyses for Predicting PTSD at 12 Months With Selected Background Characteristics and Admission Outcome Measure Scores

Variable	N	B	SE	Wald (1)	Odds ratio	95% CI
Sex (0 = Male, 1 = Female)	113	0.26	0.61	0.18	1.30	[.39, 4.29]
Caucasian (0 = No, 1 = Yes)	113	0.14	0.65	0.05	1.16	[.33, 4.10]
Age (years)	113	0.08	0.15	0.29	1.08	[.81, 1.45]
Site (0 = PA, 1 = NJ)	113	1.12	0.65	2.99	3.05	[.86, 10.80]
Length of therapy (0 = 8 Sessions, 1 = 16 Sessions)	113	0.83	0.64	1.67	2.30	[.65, 8.12]
Use of trauma narrative (0 = No, 1 = Yes)	113	1.16	0.70	2.77	3.18	[.81, 12.45]
K-SADS-Reexperiencing	113	0.78	0.35	4.96	2.18	[1.10, 4.32]
K-SADS-Avoidance	113	0.58	0.26	5.08	1.78	[1.08, 2.95]
K-SADS-Hypervigilance	113	0.83	0.32	6.85	2.29	[1.23, 4.26]
CBCL-Internalizing	113	0.09	0.03	9.73*	1.09	[1.03, 1.16]
CBCL-Externalizing	113	0.06	0.03	4.03	1.06	[1.00, 1.13]
CSBI-Child Sexual Behavior	112	0.03	0.03	1.01	1.03	[.97, 1.10]
PERQ-Parental Distress	106	0.02	0.03	0.73	1.02	[.97, 1.08]
PPQ-Parenting Practices	105	0.01	0.02	0.06	1.01	[.96, 1.05]
BDI-II-Parental Depression	105	0.02	0.03	0.57	1.02	[.97, 1.08]
CDI-Child Depression	70	0.16	0.05	9.70*	1.17	[1.06, 1.29]
MASC-Child Anxiety	70	0.02	0.02	0.97	1.02	[.98, 1.05]
Child Shame	87	0.09	0.08	1.22	1.09	[.93, 1.28]
Child Fear	113	0.15	0.21	0.53	1.16	[.77, 1.76]
WIST-Child safety skills	113	0.00	0.05	0.00	1.00	[.91, 1.11]

Note. BDI-II = Beck Depression Inventory-II; CBCL = Child Behavior Checklist; CDI = Children’s Depression Inventory; Fear = Fear Thermometer; K-SADS = Schedule for Affective Disorders and Schizophrenia for School Age Children; MASC = Multidimensional Anxiety scale for children; PERQ = Parent Emotional Reaction Questionnaire; PPQ = Parent Practices Questionnaire; WIST = What If Situations Test Safety Skills subscale, CI = confidence interval.

^*p < .05, Bonferroni adjusted (α/20).

Because the same children’s and parents’ scores were being compared over time, the outcome measures’ raw scores sufficed for the MM-RANCOVA analyses. However, raw scores are less useful for clinical screening purposes, so internalizing and depression raw total scores were converted into normalized T scores based on the normative tables provided in their respective manuals for various sex and age group combinations. The mean pretreatment internalizing T scores for the 12 children who did and 101 children who did not meet diagnostic criteria for PTSD at 12 months were, respectively, 71.33 (SD = 12.71) and 59.93 (SD = 12.18), t(111) = 3.05, p < .01, Hedges’ g = .57. Likewise, the mean pretreatment depression T scores for the 8 children who did and 62 children who did not meet symptom criteria for PTSD at 12 months, were, respectively, 65.75 (SD = 15.07) and 49.26 (SD = 9.69), Welch’s t(8) = 3.02, p < .05, Hedges’ g = 1.01. The effect size for the mean difference of the Internalizing T scores was moderate whereas the effect size for the mean difference of the depression T scores was large. Most importantly, the mean internalizing and depression T scores were, respectively, two and one standard deviations (SD = 10) above the mean T score of 50 for children in general. The children who entered TF-CBT treatment being rated by their parents as displaying much above average internalizing behavior problems or who described themselves as having above average symptoms of depression were more likely to meet symptom criteria for PTSD 12 months after treatment than the children who entered TF-CBT treatment being rated by their parents as displaying only above average internalizing behavior problems or who described themselves as having average symptoms of depression. Consequently, the optimal TF-CBT treatment for children who have experienced sexual abuse needs not only to address PTSD but also to identify and treat children with symptoms of clinical depression or who are rated by their parents as displaying high levels of internalized distress.

Discussion

Consistent with the primary hypothesis for this study, the findings from our 6- and 12-month follow-up assessments clearly demonstrate that TF-CBT treatment gains had been sustained. This was true regardless of whether children and parents had 8 or 16 sessions of treatment or whether or not the children were actively encouraged to develop a TN and process it with the therapist. Moreover, these results are consistent with earlier studies which have demonstrated that TF-CBT treatment gains are sustained for 1 to 2 years after the treatment ends (Cohen & Mannarino, 1997; Deblinger et al., 2006; Deblinger, Steer, & Lippmann, 1999).

Although all four groups continued to improve, the differences between the four conditions at posttreatment were not sustained at the 6- and 12-month follow-ups. The latter findings were not consistent with our hypothesis that posttreatment differences would continue throughout the follow-up period. The most likely explanations are that over time, the commonalities across the four conditions outweighed the differences and the power to detect differences may have been limited by reduced sample size due to drop outs during the follow-up period. It is important to emphasize that through the process of gradual exposure (Cohen et al., 2010), therapists talked about sexual abuse with children and caregivers during every treatment component, including the children receiving the No TN conditions. In contrast to older youth, many of the young children in this study had limited capacities to develop or process narratives in depth, further minimizing differences between Yes TN and No TN conditions.

There were two dependent variables in the current study that continued to decline during the posttreatment period. First, parental emotional distress was significantly lower at the 12-month follow-up compared with posttreatment and reflected a medium effect size. Although it is not clear why this occurred, it would make sense that parental distress would decline further as parents observed that their children were continuing to do well and that the sexual abuse was not likely to have a long-term negative impact.

The other dependent variable that continued to decline posttreatment was children’s self-report of anxiety. Specifically, children’s anxiety was significantly lower at the 12-month follow-up than at posttreatment, and this decline also reflected a medium effect size. This likely reflects ongoing consolidation of skills that children learned during TF-CBT, resulting in greater self-confidence and lower anxiety.

Although the TF-CBT initial treatment gains were impressive (Deblinger et al., 2011) and were sustained at the two follow-up assessments, regardless of treatment condition, 12 children (11%) continued to meet full criteria for PTSD at the 12-month follow-up. Children’s internalizing behavior problems and self-reported depressive symptoms predicted their 12-month PTSD status. Specifically, children whose internalizing behavior problems were much above average (2 SDs) or whose self-reported depression scores were above average (1 SD) at pretreatment were significantly more likely to meet full symptom criteria for PTSD at the 12-month follow-up.

Regarding the first finding, it is likely that very high levels of internalizing behavior problems reflect a general level of internal distress whether of biological or environmental etiology which may make these young children more resistant to trauma interventions in general. With respect to the second finding, high levels of comorbid depressive symptoms also appeared to have a negative impact on children’s response to TF-CBT. It is important to note that these self-reported depressive symptoms were only obtained from children ages 7–11 years; if the younger children had also been included (with an age-appropriate measure of depression), the findings might have been different. Interestingly, in a previous TF-CBT outcome study (Deblinger et al., 2006), pretreatment depression was positively related to the total number of PTSD symptoms at posttreatment and the 12-month follow-up but only for those children assigned to the client-centered therapy treatment condition. However, comorbid depressive symptoms did not predict posttreatment or 12-month follow-up PTSD outcomes for those assigned to the standard 12-session TF-CBT protocol utilized in that investigation. These discrepant findings in terms of predictors of PTSD outcomes may reflect differences between these two investigations in terms of PTSD treatment protocols and the different age ranges of the subject samples.

Though there were no interactive effects by treatment condition found in the current investigation, it is possible that this finding reflects the mix of different types of TF-CBT packages and/or the much younger age range of the population in this study (i.e., 4–11 years versus 8 to 14 years in the previous study). While TF-CBT has been found to produce significant reductions in depressive and internalizing symptoms in the current and previous studies, the current findings suggest that TF-CBT may require some modifications and tailoring for a small percentage of young children exposed to trauma with the highest levels of depression. Given the overlap of current evidence-based treatments for childhood depression with the TF-CBT skills components, how to modify treatment for these children is a critical question that we have recently begun to explore with child depression treatment experts. Perhaps such modifications would place relatively greater emphasis on cognitive processing and/or interpersonal interactions (D. Brent, personal communication, July 28, 2011).

The study was limited by the relatively small numbers of children in each condition (barely enough power to detect differences among conditions for most outcomes) and the lack of self-report instruments to assess depressive and anxiety outcomes in younger children. Additionally, the study did not use the newer alternative algorithm for assessing PTSD (Scheeringa, 2008) that emerged as this study was being designed and conducted. Also, the fact that most children were living in stable home settings differentiated this cohort from children in other foster settings or residential care who experience multiple placement disruptions and lack of a consistent supportive adult presence. Finally, although allowing us to answer some questions, the disproportionate application of the TN and processing in the 16-session Yes TN condition (which was not consistent with usual TF-CBT implementation) did not allow us to evaluate other critical questions, such as whether 16 sessions of proportionate TF-CBT would have been superior to the 8-session Yes TN condition or whether these two TN conditions would have been superior to the No TN conditions for other outcomes. Given these study limitations, in our view, the lack of significant differences found on follow-up between the narrative and no narrative conditions should not deter clinicians from eliciting trauma narratives when implementing TF-CBT. Trauma narratives often reveal idiosyncratic dysfunctional abuse-related beliefs that if undetected might undermine children’s psychosocial development. Thus, it seems critically important to address and correct such beliefs in order to minimize the risk of long-term psychosocial difficulties that are so common among survivors of child sexual abuse.

The practice implications of our findings are worth noting. This study provides additional data to the existing strong empirical evidence supporting the durability of TF-CBT for maintaining positive treatment effects over time. Taken in the context of previous TF-CBT studies, these findings emphasize that although there are some initial benefits to tailoring TF-CBT according to children’s presenting symptoms, over time these differences are less important than the overall effectiveness of providing TF-CBT for improving a wide range of trauma symptoms in young children who have experienced sexual abuse.