Behat,. Res. Ther. Vol. 33, No. 3, pp. 271-281, 1995
Pergamon
0005-7967(94)00053-0
CHILD,
Copyright ~S 1995 Elsevier Science Ltd Printed in Great Britain. All rights reserved 0005-7967/95 $9.50 + 0.00
PARENT AND FAMILY DYSFUNCTION PREDICTORS OF OUTCOME IN COGNITIVE-BEHAVIORAL TREATMENT OF ANTISOCIAL
AS
CHILDREN
ALAN E. KAZDIN Department of Psychology, Yale University, P.O, Box 208205, New Haven, CT 06520-8205, U.S.A. (Received 16 June 1994)
Summary--The present study examined factors that predicted favorable treatment outcomes among clinically referred conduct problem children (N = 105, ages 7-13) who received cognitive-behavioral treatment. Three domains (severity and breadth of child impairment, parent stress and psychopathology and family dysfunction) assessed at pretreatment were predicted to affect treatment outcome. The results only partially supported the prediction. Less dysfunction in each of the domains predicted who responded favorably to treatment on parent ratings of deviance and prosocial functioning but not on teacher ratings of these outcomes. The findings have implications for identifying youths who respond to available treatments. The results also underscore fundamental questions about the assessment of treatment effects and the criteria for evaluating outcome.
INTRODUCTION
Aggressive and antisocial behaviors are among the most frequent bases for referring children and adolescents to psychotherapy (Gilbert, 1957; Kazdin, Siegel & Bass, 1990). Youth referred for treatment raise special challenges because they often evince multiple conduct problems (e.g. aggression, stealing, lying, truancy, firesetting), meet criteria for multiple psychiatric diagnoses (e.g. conduct disorder with attention deficit disorder or depression), and experience other characteristics (e.g. peer rejection, school delays) that impede adaptive functioning in everyday life. A number of promising treatments (e.g. variations of cognitive behavioral treatment, parent training, and family therapy) have emerged based on evidence that they can achieve significant changes over the course of treatment and follow-up (at least up to one year) in controlled trials with clinical samples (see Dumas, 1989; Kazdin, 1993; Pepler & Rubin, 1991). Even so, the qualifier "promising" remains warranted because clinically significant change and long-term follow-up remain to be firmly established. Our own work has focused on problem-solving skills training (PSST) and parent management training (PMT). We have found that PSST alone and in combination with PMT are more effective than client-centered, relationship-based treatment and a nonspecific treatment control condition among inpatient and outpatient youth (Kazdin, Bass, Siegel & Thomas, 1989; Kazdin, Esveldt-Dawson, French & Unis, 1987a, b). Although treated youth improve significantly, most of them remain outside of the "normal" (normative) range of functioning at home or at school. Although efforts are needed to bolster treatment effects, it would also be helpful to identify youth for whom treatment is effective. From a conceptual standpoint, it would be important to understand what factors moderate outcome, whether they vary among different treatments, or indeed are not influenced by treatment. From a clinical perspective, identification of cases likely to respond or fail to respond would improve triage of youth, so that available treatments can be administered where they are likely to produce favorable results and novel approaches, including new treatments, variants of existing treatments and multiple-treatment combinations, can be applied for youth less likely to respond. Yet within child and adolescent psychotherapy research, rarely have studies examined child, parent, or family factors that influence treatment outcome (Kazdin, Bass, Ayers & Rodgers, 1990). A model of research is to identify a particular attribute (e.g. subtype of dysfunction, age, sex) and to examine whether that characteristic interacts with treatment in producing a particular outcome (see Smith & Sechrest, 1991). A complementary approach, less often used in outcome 271
272
Alan E. Kazdin
research, is to consider multiple variables as risk factors that may accumulate or contribute to outcome. In a risk-factor model, several influences may operate, none of which is necessarily strong in its influence or a necessary condition for a particular outcome. In the case of conduct disorder youth, multiple characteristics influence the onset and course of dysfunction (Kazdin, 1987; Rutter & Giller, 1983). Severity and breadth of child impairment, parental stress and psychopathology, and poor family relations and family dysfunction are among the many factors that predict onset of conduct disorder and poor long-term prognosis. These factors alone or in combination are also likely to influence treatment outcome. The reason is that impairment of the child, parent and family has a pervasive impact on participation in treatment (e.g. attendance to the sessions, adherence to treatment prescriptions, support of treatment goals and practices at home). Also, the impact of a given treatment is likely to be diminished when the scope and severity of dysfunction of the child, parent and family are extensive. The present study evaluated whether characteristics of conduct problem youth and their families affect treatment outcome. Three domains of risk were examined including severity of child dysfunction, parent stress and psychopathology and family dysfunction. We predicted that more favorable status in each domain at pretreatment would portend more positive treatment outcomes in functioning at home and at school. We evaluated whether responsiveness to treatment among youth and families who received PSST alone or PSST in combination with PMT varied as a function of child, parent and family dysfunction. Our prior work has suggested that both treatments lead to change among clinically referred youth, as noted above, and more recently that the combined treatment (PSST + PMT) leads to greater change than PSST alone (Kazdin, Siegel & Bass, 1992). In the present study we were interested in predictors of positive responsiveness to treatment in relation to normative levels of functioning. To that end, we used as outcome criteria, standardized parent and teacher rating scales to assess child dysfunction (Achenbach, 1991a, b). These measures provide excellent normative data and permit evaluation of the extent to which treatment places youth within the "normal" range of functioning. A prior study (Kazdin et al., 1992) reported treatment outcomes for different interventions but did not evaluate factors related to responsiveness to treatment. We drew all cases from that prior study (n = 53) who received PSST and PSST+ PMT; for the present study new cases (n = 52) were run who received the same interventions to increase the overall sample size. METHOD Participants Child characteristics. The study was conducted at an outpatient treatment service (Child Conduct Clinic) for children referred for aggressive, antisocial and oppositional behavior. The outpatient service is affiliated with a larger child psychiatry service that is the initial point of contact and triage for treatment. The participants included 105 children (30 girls and 75 boys) who ranged in age from 7-13 (M = 10.3). As mentioned previously, 53 of the cases (50.5%) were drawn from a prior study; 52 additional cases (49.5%) were run for the present study. Sixty-six (62.9%) of the children were Caucasian; 31 (29.5%) were African-American, 7 (6.7%) were Hispanic; and 1 (0.9%) was Asian-American. Diagnoses of the children, based on DSM-III-R criteria (American Psychiatric Association, 1987), were obtained from a structured interview administered at intake. Reliability of Axis I diagnosis, from observation of the diagnostic interview by an independent observer for the new sample of cases, yielded high agreement (kappa = 0.93 across all diagnoses). Principal Axis I diagnoses for the children included conduct disorder (43.3%), oppositional-defiant disorder (39.4%), various other disorders including attention deficit hyperactivity disorder, anxiety disorder or major depression (12.5%), or no diagnosable Axis I disorder (4.8%). Most children (66.7%) met criteria for more than one disorder (M = 2.1 disorders). Parent and family characteristics. The primary caretaker of the child included biological mothers (89.5%), step, foster or adoptive mothers (7.6%), or other relative or guardian (2.9%). Forty-two (39.0%) of the children came from one-parent families. Social class, based on the Hoilingshead and Redlich (1958) index, yielded the following breakdown from lower to higher socioeconomic class: Classes V (23%), IV (34%), III (26%), II (11%) and I (6%). Estimated monthly income for families
Cognitive-behavioraltreatment of antisocial children
273
ranged from 0 to $500 to >$2500 (Mdn. range = $1000-$1500); 21.6% of the families received social assistance.
A ssessmen t Child, parent and family domains assessed at pretreatment were predicted to influence responsiveness to treatment. The domains included: severity and breadth of child dysfunction; parent stress and psychopathology; and family dysfunction. Measures within these domains were used as predictors of child functioning at home and at school at the end of treatment. Measures drew on varied assessment formats (interviews, questionnaires) and sources of information (e.g. parents, teachers, children). The measures and their administration have been described elsewhere and hence are only highlighted here (see Kazdin et al., 1992). Child dysfunction. We predicted that severity of child impairment at pretreatment would influence treatment outcome. Five measures were used to represent this domain and encompassed both severity of antisocial symptoms and psychopathology more generally. From the diagnostic interview, we assessed the total number of conduct disorder symptoms (from DSM-III-R) to measure breadth of antisocial behaviors and the total number of other (nonconduct disorder) symptoms present across DSM-III-R diagnoses, to provide a broader measure of overall dysfunction. From the Risk Factor Interview administered to parents [RFI; Kazdin, Mazurick & Bass (1993)], we used the scale measuring child history of antisocial and delinquent behavior. The scale includes 18 items about the child's history of dysfunction in this area. Two child self-report measures were also used. Children were interviewed directly using the Self-Report Delinquency Checklist [SRD; Elliott, Dunford & Huizinga (1987)], which measures frequency of delinquent acts within the past year. The total score of delinquent acts was used from the scale. Also, children completed the Children's Action Tendency Scale [CATS; Deluty (1979)] which measures aggression, assertiveness and submissiveness. From this scale, the total aggression score was used. Parent stress and dysfunction. Higher stress and greater parental dysfunction were predicted to be associated with poorer child outcomes at posttreatment. This domain was presented by four measures. Parents completed the Parenting Stress Index [PSI; Abidin (1983)]. From the PSI, two scores were examined, namely, total perceived stress and life events. These scales reflect parent evaluations of several areas as sources of stress and specific events that have occurred, respectively. To measure depression, parents completed the Beck Depression Inventory [BDI; Beck et al. (1961)]. Also, parents completed the Hopkins Symptom Checklist [SCL-90; Derogatis & Cleary (1977)] to assess a broader range of symptoms (e.g. somatization, depression, anxiety). Total scores from the BDI and SCL-90 were used. Family dysfunction. We predicted that less family dysfunction, in relation to child-rearing practices and interpersonal relations in the home, would predict more favorable treatment outcomes. Four scores obtained from two measures represented this domain. From the RFI parent interview, mentioned earlier, the adverse child-rearing practices scale was used. This includes 29 items that cover a range of child-rearing practices associated with conduct disorder (e.g. poor parent monitoring and supervision of the child, use of harsh and inconsistent punishment). Parents also completed the Family Environment Scale [FES; Moos, Insel & Humphrey (1974)] to assess interpersonal relationships and organizational structure of the family. The scales (comprising 90 true-false items) load on three broad dimensions: relationship (e.g. cohesion, expressiveness, conflict), personal growth (e.g. independence, achievement orientation, active-recreational orientation), and system maintenance (organization and control). Treatment outcome assessment The purpose of the study was to examine predictors of positive therapy outcomes, i.e. significant improvement among children. Treatment outcome is multifaced and no single measure or assessment modality provides a complete picture of therapeutic change. To test the predictions, parent and teacher ratings were selected as outcome criteria because: (1) normative data are available for the measures and hence provide a developmentally informed basis to evaluate the level of functioning of youth relative to nonreferred peers; B R T 33/3--£
274
Alan E. Kazdin (2) the measures permit evaluation of deviance and prosocial functioning and adjustment at home and at school, all of which are critically relevant outcomes, given their relation to long-term prognosis; and (3) these are the most frequently used outcome criteria for child therapy (Kazdin, Bass, Ayers & Rodgers, 1990).
Parent and teacher ratings of child functioning were administered immediately before and after treatment. Posttreatment measures served as the outcome criteria to evaluate predictors, noted previously. Parents completed the Child Behavior Checklist [CBCL; Achenbach (1991a)]. This measure includes 118 items, each rated on a 0-2 point scale, that comprise multiple behavior problem scales. The total behavior problem score includes all items reflecting overall severity of dysfunction. The CBCL also includes a total social competence score (comprised of subscales assessing child participation in activities, interactions with others and school performance) which was used to evaluate prosocial functioning. To evaluate school performance, the children's teachers completed the Child Behavior ChecklistTeacher Report Form [CBCL-TRF; Achenbach (1991b)]. The measure parallels the structure of the parent version of the scale. The total behavior problems scale was used to evaluate child dysfunction at school. The adaptive functioning summary score, a composite of four subscales (working hard, behaving appropriately, learning, happy), was used to evaluate positive adjustment at school. Treatment administration
After intake assessment, children and families began treatment. Families were assigned randomly to receive PSST or PSST + PMT. Thus, in all cases, youth received PSST but for approximately half of the cases, parents received PMT. In PSST, children were seen individually for 20-25 sessions; each session was approx. 1 hr long and scheduled weekly. PSST used a variety of cognitivebehavioral techniques that develop and apply problem-solving skills to interpersonal situations in everyday life. For PMT, parents were seen for approximately 16 sessions (approx. 1-2 hr each and scheduled weekly). PMT focused on developing adaptive child-parent interaction and altering child behavior at home and at school. Over the course of therapy, the child or parent(s) were routinely brought into the other's sessions as part of treatment. Parents and children were seen and participated in each treatment condition, although the extent and nature of their participation varied as a function of the different conditions. Treatments were provided individually to each family. Each of the treatments spanned approx. 7-8 months. Twelve full-time, masters degree level clinicians (10 female, 2 male; 11 Caucasian, 1 African-American; ages 24-56) served as therapists. Details of the treatment and the parent and child participation in both PSST and PMT are provided elsewhere (Kazdin et al., 1992). RESULTS Preliminary analyses Pretreatment status. Table 1 provides the scores for subject demographic variables, pretest measures (predictors) and posttreatment outcomes (criteria) for cases assigned to the two treatment conditions (PSST and PSST + PMT). At pretreatment, multivariate analyses (Hotelling's T 2) were used to compare the two treatment conditions on subject and sociodemographic characteristics (Table 1), as well as the three domains (child, parent and family dysfunction) predicted to relate to outcome. The treatment groups did not differ on characteristics or predictors at pretreatment. Change over the course of treatment. The focus of the study was on predictors of posttreatment outcomes. Of initial interest is whether youth changed over the course of treatment. For the parent- and teacher-completed CBCLs, within-group t tests were computed from pre to posttreatment. For PSST cases, improvements from pre to posttreatment were evident for parent ratings of total behavior problems [t (50) = 5.02, P < 0.001] and social competence [t (50) = 4.64, P <0.001] and for teacher ratings of total behavior problems [t(49)= 4.18, P < 0.001] and adaptive functioning [t (48) = 3.09, P < 0.01]. For PSST + PMT cases, improvements also were evident in parent ratings of total behavior problems [t (52) = 6.89, P < 0.001] and social competence [t(52)=4.15, P <0.001] and teacher ratings of total behavior problems [t(50)=4.51,
C o g n i t i v e - b e h a v i o r a l treatment o f antisocial children
275
Table I. Means and standard deviations (or proportions) for the total sample (N = 105) and separately for problem-solving skills training (PSST) and problem-solving skills training + parent management training (PSST + PMT) Total sample Domains/measures
M (%)
SD
10.26 71.40 100.50 34.87 39.05 37.10 62.9/29.5/7.7 74.28
1.49
PSST M
PSST + PMT M
10.55 75.00 97.38 35.35 40.38 40.38 59.6/34.6/5.8 73.08
9.97 69.92 103.55 34.40 37.74 33.96 45.6/48.5/4.9 75.47
Family characteristics Child age Child sex (% male) WlSC-R full scale Mother age Single-parent family % Minority (%) (White/black, all other Head household bioparent (%)
Socioeconomic disadl,antage Hollingshead class Income level On public assistance (%) Child d),s[~metion No. Conduct dis sympt No. Other symptoms Child antisoc hist Self-Rep delinq CATS-Aggression CBCL Tot beh prob CBCL-social competence CBCL T R F - T o t beh prob C B C L - T R F - a d a p t i v e funct
Parental stress and dy,~[~mction PSI total score Life events Beck depress inv SCL-90
Family .[unctioning Adverse child rearing FES-relationships FES personal development FES system maintenance
16.98 6.33
45.68 3.64 21.60
17.80 1.61
46.86 3.41 22.00
44.55 3.86 21.15
3.51 23.94 19.88 10.91 6.18 71.04 37.45 66.27 36.44
2.48 7.96 1.79 8.79 3.99 9.11 8.06 8.73 8.19
3.69 23.85 19.92 10.96 6.79 71.33 35.76 65.47 36.16
3.33 24.04 19.84 10.87 5.58 70.75 39.08 67.08 36.73
265.37 9.49 8.89 57.89
49.25 8.41 8.50 55.01
261.75 I 1.27 9.69 62.65
268.92 7.74 8.09 53.21
52.24 7.61 27.30 0.00
6.53 5.01 6.36 2.41
52.65 7.29 26.85 -0.06
51.80 7.92 27.74 0.06
Notes. With the exception of income level, higher scores on each of the socioeconomic status and stress measures reflect greater dysfunction, level of stress, or symptoms. For the socioeconomic measures, Hollingshead and Redlich class reflects points that can be used to convert to one of 5 classes. Higher point totals refer to lower educational/occupational status.
P < 0.001], and adaptive functioning [t (50) = 2.23, P < 0.05]. For both treatment conditions, the changes reflect reductions of emotional and behavioral problems and improvements in prosocial functioning. The outcome differences between groups were not of direct interest in this study. Yet, differential outcomes, if evident, are of interest in relation to predicting change. Analyses of covariance of CBCL measures at post (using pretreatment as the covariate) indicated that youth who received PSST + PMT were significantly more improved on parent ratings of total behavior problems [F(1,101) = 4.27, P < 0.05] than youth who received PSST. However, on three other outcome scores (parent ratings of social competence, teacher ratings of total behavior problems or adaptive functioning) no differences were evident (each F < 1). Also, at posttreatment a greater proportion of youth who received PSST + PMT fell within the normal range on parent total behavior problem scores than PSST youth (45.3 and 26.9%, respectively, X 2( 1, N = 105) = 3.83, P = 0.05). Chi square tests on the proportion of youth in the normal range on other measures (parent-rated social competence; teacher-rated behavior problems and adaptive functioning) yielded no differences between treatments. Overall, the results indicated that cases who received PSST and PSST + PMT were no different at pretreatment, changed over the course of treatment, and were similar in their posttreatment status on three of four parent and teacher scale scores. On the fourth scale (parent CBCL, total behavior problem), the combined treatment led to a more favorable outcome. These results are in keeping with prior outcome results. Predicting treatment
outcome
Data reduction and analyses. Three domains of risk were predicted to influence treatment outcome. To evaluate each domain and to reduce redundancy in the analyses, correlations were
276
Alan E. Kazdin
computed between measures within a given domain. The purpose was to determine if the measures were highly correlated and hence redundant. For this evaluation, we defined highly correlated as r i> 0.85 (and hence shared variance of >f72%). No correlation met this criterion. Similarly, correlations of measures between domains did not meet this criterion. Consequently, none of the measures was deleted from the domains described previously.* Outcome criteria. The purpose of the study was to test whether child dysfunction, parent stress and psychopathology, and family dysfunction predicted treatment outcome. Outcome was evaluated by examining whether performance at posttreatment fell within the normative range on measures of functioning at home and at school. As is well known, parent and teacher evaluations of child dysfunction correlate only in the low-to-moderate range (Achenbach, McConaghy & Howell, 1987; Kazdin, 1994). In the present sample, with a range more restricted than in broader normative samples, the correlations were even lower than usual. For parent and teacher ratings of total behavior problems, the correlation at pretreatment was low (r = 0.06, NS). Similarly, parent and teacher ratings of prosocial behavior (social competence and adaptive functioning) yielded a low correlation (r = 0.16, NS). Thus, parent and teacher data were evaluated as separate outcomes. We were interested in improvements in total behavior problems and prosocial functioning. Within raters, these domains were correlated in the low to moderate range at pretreatment (for parent ratings, r - 0.31, P < 0.02 and for teacher ratings, r - 0.57, P < 0.001). The correlations indicate that level of deviance is negatively correlated with prosocial functioning. For examining treatment outcome, we evaluated parent CBCL and teacher CBCL separately. However, for a given rater (e.g. parent), improvements in behavior problems and prosocial competence were considered together, because within raters these were moderately correlated. That is, for a case to be considered as having responded well to treatment, we examined status of symptoms (total behavior problem) and social functioning (social competence or adaptive functioning). As a criterion, we relied on scale scores that discriminate clinical and nonclinical range for both total behavioral problems and prosocial functioning (Achenbach, 1991 a, b). Multivariate analyses of variance indicated no differences between treatment conditions (PSST + PMT) and domains of interest at pretreatment nor an interaction between treatment condition in child, parent and family domains at pretreatment on posttreatment (outcome). All cases received PSST and approximately one-half of the cases also received PMT. Treatment condition (receipt of PMT) was included in all analysis (0, 1, dummy variable). Child functioning at home. On the parent CBCL, children were considered to have responded well to treatment if they fell within the nonclinic range at posttreatment on both measures of symptoms and prosocial functioning. On the parent CBCL, the criteria for falling within the normative (nonclinic) range of functioning are reflected in total behavior problem scores (T ~<60) and social competence scores (T/> 39). Children who fell within the normative range on both of these measures at posttreatment were considered to have responded well to treatment (favorable responders) and were compared to those who did not meet this criterion (others). Two groups of youth were identified based on the above parent CBCL criteria. The means and standard deviations for each of the domains of interest are presented in Table 2. We expected that child, parent, and family dysfunction would predict treatment outcome. Multivariate analyses (Hotellings T 2) compared to those who entered the range of normative functioning (favorable responders) on both measures of total behavior problems and prosocial functioning with those who did not. The results revealed significant differences between favorable responders and others for child [T2(5,93)=4.09, P <0.01], parent [T2(4,98)=4.01, P <0.01], and family domains of dysfunction [T2(4,95)= 3.15, P < 0.05]. These results support the prediction that severity and breadth of child dysfunction, parent stress and psychopathology and family dysfunction influence treatment outcome. To examine the contribution of individual pretest measures on outcome, discriminant analyses were completed to classify favorable responders and others on the parent measure. As an initial screen for variables to include in the analysis, predictors were selected if group differences (between favorable responders and others) were significant at P ~<0.10. This alpha level was adopted because *Correlations referred to here are available from the author.
277
Cognitive-behavioral treatment o f antisocial children Table 2. Child, parent and family dysfunction at pretreatment for cases that responded well ( + response) or less well (other) to treatment separately for parent and teacher outcome measures Parent CBCL
Domains/measures
+ Response (n = 27)
(n = 78)
2.59 19.29 19.54 9.67 7.19
3.83 25.57 20.00 11.35 5.83
237.63 6.44 5.26 29.30 49.46 10,00 29.19 0.59
Teacher C B C L
Other
+ Response
Other
(n = 36)
(n = 66)
2.27 c 3.74 ~ 1.13 < I - 1.53
3.19 23.42 19.85 7.97 6.17
3.58 24.32 19.92 12.38 6.27
274.97 10.54 10.14 67.78
3.58 c 2.22 a 2.64 c 3.28 ~
263.67 8.08 8.42 53.25
266.74 10.33 9.23 61.20
< I 1.28 <1 < I
53.18 6.78 26.64 -0.21
2.58 c 2.98 ~ 1.81 ~ 1.49
50.76 7.83 27.97 -0.31
52.89 7.59 26.92 0.15
1.54 < 1 < I <1
t test
t test
Child dyffunction No. Conduct dis sympt No. Other s y m p t o m s
Child antisoc hist Self-rep delinq CATS-aggression Parental stress and PSI total score
< I < I <1 2.49 ¢ < 1
dy.~function
Life events Beck depress inv SCL-90
Family functioning Adverse ch rear FES-relation F E S pers devel F E S syst main
Note: +Response refers to a favorable response to treatment, as defined by functioning within the "'normal" range at posttreatment on total behavior problem and prosocial functioning measures. ap ~< 0.10, bp ~< 0,05, cP ~< 0.0l.
the prediction was that favorable response to treatment is likely to be influenced by multiple risk factors, some of which may be small in their individual contribution, but contribute to risk nevertheless. The t tests are also presented in Table 2. Of the variables listed in Table 2, 9 variables differentiated favorable responders from other cases and were entered into a discriminant analysis. The purpose was to identify individual predictors and the extent to which more favorable responders and less favorable responders could be correctly identified. Treatment condition (receipt of PMT or not) was also entered into the analysis, as noted previously. The discriminant analysis used the direct method so that each measure that differentiated favorable responders and others was included. The results yielded a significant discriminant function [Wilks' lambda = 0.81, X 2(10) = 19.26, P < 0.05]. The canonical correlation, a measure of the degree of association between discriminant scores and group membership, was 0.43. Table 3 presents the variables included in the function, their standardized canonical discriminant function coefficients, and their correlation with group status. The correlations convey that influences from each domain (child severity of dysfunction, parent stress and dysfunction and family relations) were related to responsiveness. Treatment (PMT) also contributed to outcome, as expected based on prior analyses. With this discriminant function (Table 3), 76.9% of the youth who responded very Table 3. Discriminant analyses of child, parent and family measures predictive of outcome separately for parent and teacher CBCL with standardized discriminant function (SDF) coefficients and correlations (r)
Parent CBCL Domains/measures
SDF
Teacher C B C L r
SDF
r
0.98
0.98
0.21
0.22
Child dysfunction No. Conduct dis sympt No. Other s y m p t o m s
0.11 0.43
0.48 0.73
Child antisoc hist Self-rep delinq
CATS-aggression Parental stress and PSI total score
dysJunction
Life events Beck depress inv SCL-90
0.34 0.03 -0.16 0.19
0.71 0.46 0.51 0,67
Family ,/unctioning Adverse ch rear FES-relation F E S pers devel FES syst main Treatment (PMT or no P M T ) *Non-zero number, but beyond two decimals.
-0.00" -0.32 0.00"
0.54 0.65 -0.36
-0.37
-0.33
278
Alan E. Kazdin
favorably to treatment and 69.3% of those who responded less favorably could be correctly classified, yielding an overall classification accuracy of 71.3%. Child functioning at school. The teacher-completed CBCL was evaluated in a parallel fashion. Children were considered to have responded well to treatment if they fell in the nonclinic range at posttreatment on both measures of symptoms and prosocial functioning. Those whose scores fell within the normative range on total behavior problems (T ~<60) and adaptive functioning scales (T >/39) at treatment outcome were considered to have responded well to treatment; those who were within the clinic range and hence in both domains at outcome were considered to have responded well to treatment. The means and standard deviations for favorable responders and others for each of the domains of interest are presented in Table 3. Multivariate analyses of variance of pretreatment measures separately by domain revealed no differences between those who responded well to treatment and those who responded less well for child [T2(5,90) = 1.95, P < 0.10], parent [T 2 < 1, NS], and family domains (T 2 < 1, NS). For treatment outcome, as measured by school performance, the prediction that different domains of functioning influence outcome was not supported. As with prior analyses, we were interested in examining the contribution of individual pretest measures on outcome using discriminant analyses to classify favorable responders and others on the teacher CBCL. Only the multivariate analysis from the child domain approached significance; measures within this domain were examined. As before, a t-test comparison of favorable responders and others was used as a screen to include predictors in the discriminant analysis. Self-reported delinquency (SRD) by the child was the only pretreatment variable that differentiated favorable responders and others. This measure and treatment conditions were entered into a discriminant analysis. The results yielded a significant discriminant function [Wilks' lambda = 0.94, Z 2(2) = 6.22, P < 0.05] with a canonical correlation of 0.25. Table 3 presents the two variables included in the function, their standardized canonical discriminant function coefficients, and their correlation with group status. With this discriminant function (Table 3), 66.7% of the youth who responded well to treatment but only 50.0% of those who responded less well could be correctly classified (overall classification accuracy of 55.8%). Although the discriminant function attained significance, quite clearly the predictions about domains that influenced outcome were not supported on teacher measures of child functioning at school. Subject and demographic factors were examined in relation to the prediction of treatment outcome, although no hypotheses were advanced for these factors. However, outcome status was not significantly related to child age, sex, parent marital status, socioeconomic status and income. These variables did not improve prediction when added to the discriminant functions, noted previously. DISCUSSION Child, parent, and family dysfunction were predicted to influence outcome in cognitivebehavioral treatment among youth referred to antisocial and aggressive behavior. The findings partially supported the prediction. Child severity and breadth of dysfunction, parental stress and psychopathology and family dysfunction were related to levels of deviant behavior and prosocial functioning among youth at the end of treatment. However, these results were obtained only for child functioning at home (parent ratings) and not for child functioning at school (teacher ratings). Child outcome at school was related to (predicted by) severity of initial dysfunction (self-report delinquency). Yet, other measures of child dysfunction and none of the measures of parent and family dysfunction were related to child treatment outcome, as measured by school functioning. The findings provide some support for the view that child, parent and family dysfunction influence treatment outcome. In the treatment of aggressive and antisocial youth, these findings have important implications for providing treatment. It is likely that many families could profit from existing treatments and identifying such families would have direct benefit. Apart from replication of the findings, much more work is needed before practical benefits could be realized. First, it would be important to identify levels of impairment in child, parent and family dysfunction and the relation of these levels to the likely benefits (outcomes) of treatment. A more fine-grained analysis of predictors and outcomes is needed to establish the benefits of the findings. Second, the
Cognitive-behavioral treatment of antisocial children
279
ways in which child, parent and family dysfunction moderate or mediate outcome would be very helpful to understand as well. The ways in which these domains affect outcome, once understood, might serve as a basis for intervention. That is, one might alter those facets associated with the domains that deter therapeutic change and in that way make favorable treatment outcomes more likely. Clearly, further work is needed to describe and to explain the interrelations among these predictors and ways in which they influence outcome. The study raises issues about the evaluation of treatment outcome that warrant comment. Mentioned previously was the sparse literature on predictors of treatment outcome in the context of therapy trials with children and adolescents. The present results underscore some of the challenges. Well documented in the broader literature is the low-to-moderate correlation between parent and teacher ratings of a given domain of child functioning (e.g. deviant behavior at home and at school). In a clinic sample (with a restricted range), the relation may be quite low, as was evident in the present study. If measures of child functioning at home and at school show little correlation, it is reasonable to expect that the predictors associated with outcome in one setting might not overlap with those associated with outcome in another setting. Consequently, support for predictions obtained with one outcome measure but not another may not be surprising. Although one would want a set of predictors that is robust across outcomes, this may be difficult to obtain in light of the relations among outcome measures. A related assessment issue is the role of method factors in the present results. Treatment outcome, as measured by parent ratings of child functioning, was reliably predicted by child, parent, and family dysfunction at pretreatment. These results may have been obtained because of common method variance across predictors and outcome. Although the predictors included parent- and child-completed measures, only those measures completed by parent significantly predicted treatment outcome. The common rater (parent) across predictors and outcome might explain the results. This does not explain all of the findings including the different degrees of contribution among parent-completed measures to child outcome. Also, for child functioning at school, the only predictor of outcome in this study was severity of deviance on one of the child-completed measures. Thus, predictors did not all reflect a common rater between pretreatment measures and treatment outcome. Even so, common method factors remain a plausible explanation for the pattern of support in the present study. The use of multiple measures is widely recognized to be important because of the many domains of functioning relevant to child adjustment and functioning and the inherent limits of single methods, sources of information and samples of performance. The little or no relation among some of the outcomes of interest raises challenges in evaluating treatment efficacy and differential responsiveness to treatment. One solution might be to look for composite treatment measures, i.e. ways of combining multiple criteria. Yet, it would be important to identify domains of correlated outcomes rather than merely combining measures in a summary fashion. Perhaps across a large number of relevant outcomes, we can identify a small set of outcome domains (latent variables) and use these domains, rather than individual measures, as outcome criteria. Although such combinations are easily derived statistically, the challenge is to obtain normative data on such combinations to facilitate their interpretation from a developmental perspective. Another line of work might be to prioritize outcomes more analytically among cases. Among youth referred to treatment for conduct disorder, some function poorly at home or at school or both. Perhaps, predictors of outcome ought to be examined after first identifying more fine-grained problem profiles at pretreatment. This raises important issues regarding the delineation and diagnosis of subtypes of youth with conduct problems and the possibility that treatment outcomes among different types would be predicted by different factors. Although this is feasible in principle, how to subtype such youth in general, and specifically in relation to predictors of treatment response, is in the early stages of development. A number of limitations place constraints on the present results. First, parent and teacher ratings, even if on well standardized measures, have their limitations. Indeed, any single assessment source (archival records, direct behavioral assessment) is limited. A broader outcome battery is needed. The measures were restricted in the present report to permit evaluation of pre-established outcome criteria (normal range) of functioning for symptoms and prosocial behavior. In this regard, parent and teacher ratings are among the very few modalities with sufficient normative data by age and
280
Alan E. Kazdin
sex of the child. At the same time, even such extensive data as that collected on the Child Behavior Checklist are not free from empirical challenge; the scores and cutoffs regarded as indicative of the normal range are not uniform across all samples, ethnic groups, and geographical locations (e.g. Sandberg, Meyer-Bahlberg & Yager, 1991). Developmentally based normative data on outcomes of interest in treatment as a general area of work is critically important to provide the means to better interpret therapeutic change. Second, the analyses were restricted to treatment outcome assessed at the end of treatment (posttreatment). The effects of treatment and treatment outcomes can vary at posttreatment and later follow-up. Quite possibly the degree to which outcome could be predicted and the profile of predictors vary at different points in time. Finally, the study focused on prediction of clinically significant outcomes, i.e. return of youth to normative (nonclinic) range of functioning. Although this is a desirable criterion to achieve, as any single criterion, it can be challenged. "Success" in treating conduct disorder youth, at least given current progress, might be measured in many ways such as maintaining youth so they do not become worse (i.e. altering likelihood of untoward prognosis), reliably improving functioning in everyday life, even if only slightly; improving the quality of parent and family life; and retaining children in the home rather than in foster care, day or residential treatment or juvenile detention. Clinically significant change is a goal toward which to strive but a failure to achieve the goal is not necessarily trivial from the standpoint of the individual case or society at large. As noted at the outset of the paper, very few studies examine the nontreatment variables that influence outcome in child and adolescent psychotherapy. This stands in sharp contrast to the adult psychotherapy literature which has examined a variety of client, therapist and within-treatment (process) factors that influence outcome (see Bergin & Garfield, 1994). The results of the present study illustrate some of the problems in outcome assessment upon which such research depends. Much further work is needed both in developing outcome domains to guide research and in evaluating predictors of outcome. Clinically severe antisocial behavior among children is known to be fairly resistant to treatment. Yet many youth do improve with treatment, and this has been well documented. Who these children are and the contexts from which they come have yet to be elaborated. The present study is an initial attempt to identify characteristics of youth and their families likely to influence outcome. Acknowledgements---Completion of this research was supported by a Research Scientist Award (MH00353) and a grant
(MH35408) from the National Institute of Mental Health. Staffat the Yale Child Conduct Clinic have contributed to the present work including Lisa Stoddard, Jennifer Mazurick, Todd Siegel, Susan Breton, Susan Bullerdick, and Elif Tongul. Correspondence concerning this paper should be directed to: Alan E. Kazdin, Department of Psychology,Yale University, P.O. Box 208205, New Haven, CT 06520-8205, U.S.A. REFERENCES Abidin R. R. (1983). Parenting Stress Index--manual. Charlottesville, VA: Pediatric PsychologyPress. Achenbach T. M. (1991a). Manual for the Child Behavior Checklist~4-18 and 1991 Profile. Burlington, VT: University of Vermont, Department of Psychiatry. Achenbach T. M. (1991b). Manual for the Teacher's Report Form and 1991 Profile. Burlington, VT: Universityof Vermont, Department of Psychiatry. Achenbach T. M., McConaughy S. H. & Howell C. T. (1987). Child/adolescent behavioral and emotional problems: implications of cross-informant correlations for situational specificity. Psychological Bulletin, 101, 213-232. American Psychiatric Association. (1987). Diagnostic and statistical manual of mental disorders (3rd edn Revised). Washington, DC: American Psychiatric Association. Beck A. T., Ward C. H., Mendelson M., Mock J. & Erbaugh J. (1961). An inventory for measuring depression. Archives of General Psychiatry, 4, 53-63. Bergin A. E. & Garfield S. L. (Eds) (1994). Handbook of psychotherapy and behavior change (4th edn). New York: Wiley. Deluty R. H. (1979). Children's Action Tendency scale: a self-report measure of aggressiveness, assertiveness, and submissiveness in children. Journal of Consulting and Clinical Psychology, 47, 1061-1071. Derogatis L. R. & Cleary P. A. (1977). Confirmation of the dimensional structure of the SCL-90: a study in construct validation. Journal of Clinical Psychology, 33, 981-989. Dumas J. E. (1989). Treating antisocial behavior in children: child and family approaches. Clinical Psychology Review, 9, 197-222. Elliott D. S., Dunford F. W. & Huizinga D. (1987). The identification and prediction of career offenders utilizing self-reported and officialdata. In Burchard J. D. & Burehard S. N. (Eds), Preventing delinquent behavior (pp. 90-121). Newbury Park, CA: Sage. Gilbert G. M. (1957). A survey of"referral problems" in metropolitan child guidancecenters. Journal of Clinical Psychology, 13, 37-42.
Cognitive-behavioral treatment of antisocial children
281
Hollingshead A. B. & Redlich F. C. (1958). Social class and mental illness. New York: Wiley. Kazdin A. E. (1987). Conduct disorders in childhood and adolescence. Newbury Park, CA: Sage. Kazdin A. E. (1993). Treatment of conduct disorder: progress and directions in psychotherapy research. Development and Psychopathology, 5, 277-310. Kazdin A. E. (1994). Informant variability. In Reynolds W. & Johnson H. (Eds), Handbook of depression in children and adolescents (pp. 249-271). New York: Plenum. Kazdin A. E., Esveldt-Dawson K., French N. H. & Unis A. S. (1987a). Problem-solving skills training and relationship therapy in the treatment of antisocial child behavior. Journal of Consulting and Clinical Psychology, 55, 76-85. Kazdin A. E., Esveldt-Dawson K., French N. H. & Unis A. S. (1987b). The effects of parent management training and problem-solving skills training combined in the treatment of antisocial child behavior. Journal of the American Academy of Child and Adolescent Psychiatry, 26, 416-424. Kazdin A. E., Bass D., Siegel T. & Thomas C. (1989). Cognitive-behavioral treatment and relationship therapy in the treatment of children referred for antisocial behavior. Journal of Consulting and Clinical Psychology, 57, 522-535. Kazdin A. E., Siegel T. C. & Bass D. (1990). Drawing upon clinical practice to inform research on child and adolescent psychotherapy: a survey of practitioners. Professional Psychology: Research and Practice, 21, 189-198. Kazdin A. E., Bass D., Ayers W. A. & Rodgers A. (1990). Empirical and clinical focus of child and adolescent psychotherapy research. Journal of Consulting and Clinical Psychology, 58, 729-740. Kazdin A. E., Siegel T. & Bass D. (1992). Cognitive problem-solving skills training and parent management training in the treatment of antisocial behavior in children. Journal of Consulting and Clinical Psychology, 60, 733-747. Kazdin A. E., Mazurick J. L. & Bass D. (1993). Risk for attrition in treatment of antisocial children and families. Journal of Clinical Child Psychology, 22, 2-16. Moos R. H., Insel P. M. & Humphrey B. (1974). Family, work, andgroup environment scales. Palo Alto, CA: Consulting Psychologists Press. Pepler D. J. & Rubin K. H. (Eds) (1991). The development and treatment of childhood aggression. Hillsdale, N J: Erlbaum. Rutter M. & Giller H. (1983). Juvenile delinquency: Trends and perspectives. New York: Penguin Books. Sandberg D. E., Meyer-Bahlberg H. F. L. & Yager T. J. (1991). The Child Behavior Checklist nonclinical standardization samples: should they be utilized as norms? Journal of the American Academy of Child and Adolescent Psychiatry, 30, 124-134. Smith B. & Sechrest L. (1991). Treatment of aptitude x treatment interactions. Journal of Consulting and Clinical Psychology, 59, 233-244.