عوامل پیش بینی کننده برای حل رفتار ضد اجتماعی در میان جوانان مراقبتی با دریافت خدمات مبتنی بر جامعه

Abstract Youth in the foster care system are more likely to be diagnosed with mental illness than those in the general population. Within this system, youth with antisocial behavior (e.g., aggressive, oppositional) are overrepresented. The challenges youth with antisocial behavior present to foster care systems make understanding the factors that predict remission in this population important for improving placement stability. Using Optimal Data Analysis (ODA), this study examines potential moderating effects of various individual, social, and strength variables on clinically significant decreases antisocial behavior in a sample of foster care youth over time. Results revealed positive improvements in youths' wellbeing to be the optimal predictor of resolution, followed by positive changes in family functioning and positive changes in adjustment to trauma (i.e., symptoms of PTSD). These results indicate that clinically significant decreases over time in antisocial behavior were associated with concurrent improvement in individual and environmental variables. Implications for service providers working with this population are discussed.

1. Introduction Not surprisingly, youth in the foster care system are more likely to experience psychological problems compared to those in the general population (Burns et al., 2004 and Landsverk and Garland, 1999). Youth with antisocial behavior are particularly overrepresented in the foster care population (Pilowsky & Wu, 2006). For example, White, Havalchak, Jackson, O'Brien, and Pecora (2007) found that 20% of youth in a foster care sample they studied had a diagnosis of Conduct Disorder, compared to only 7% in the general population. The presence of antisocial behavior presents a unique challenge to stakeholders in the foster care system, since it leads to more negative and severe long-term outcomes, including chronic deviant behavior such as theft, alcohol abuse, and sociopathy (Offord & Bennett, 1994). Further, these youth experience more challenges with placement in foster homes (Rolock, Koh, Cross, & Eblen-Manning, 2009) and have poorer placement stability (Barber, Delfabbro, & Cooper, 2001). Youth with antisocial behavior in the foster care population are also at an increased risk of stepping up in care, due to difficulty managing their behavior issues in a community setting (Hussey & Guo, 2005). Further, the most significant predictors of stepping up to higher levels of care include a history of criminal and/or delinquent behavior, elopement risk, and inappropriate sexual behavior, all which are included under the umbrella of antisocial behavior (Park, Jordan, Epstien, Mandell, & Lyons, 2009). Congregate care placement is particularly problematic for youth with antisocial behavior. Iatrogenic effects have been reported for youth in group-based care with disruptive behavior issues (Poulin et al., 2001 and Robst et al., 2011). When comparing congregate care to treatment foster care, Robst et al. (2011) found more negative effects following congregate care, including greater post-treatment felony charges and return to out-of-home and residential treatment placements. Negative effects have been found to be most robust for youth with initially low levels of delinquency (Poulin et al., 2001), highlighting the negative influence of group care on youth with antisocial behavior. The evidence that youth with antisocial behavior are at an increased risk of stepping up to a higher level of care and that the experience of this type of care is associated with poorer outcomes underlines the importance of community-based care options for this group of youth. With the significantly increased rates of mental health needs among youth in foster care, the child welfare system has been described as a “de facto public behavioral health care system” (Lyons & Rogers, 2004), prompting state child welfare agencies to seek to put systems and policies in place to appropriately match youth needs with the most effective treatments. In 1986, the Child and Adolescent Service System Program (CASSP) put forth a landmark proposal that set the stage for what would become the System of Care (SOC) model. The most consequential element of the SOC model holds that the community should be the centerpiece of any service system and should always be considered the treatment setting of choice (Stroul & Friedman, 1986, 1994). The SOC model also calls for services to be (a) delivered in the least restrictive environment, (b) individualized, (c) coordinated, (d) delivered as close to youths' home as possible, (e) involve all available adults in youths' lives, (f) recognize youth strengths, and (g) be culturally competent. The Wraparound approach to care is one of a few specific treatment modalities that is consistent with the SOC philosophy and has been found to be effective for youth with disruptive behavior disorders (Burchard, Bruns, & Burchard, 2002). Wraparound is a direct treatment application of the broad SOC model. Using existing community services and natural supports, the Wraparound system is a family-centered and child-focused intervention that capitalizes on youth strengths, creating an individualized, community-based treatment program that it is interagency coordinated and culturally competent (Burchard et al., 2002, Burchard et al., 1993 and VanDenBerg and Grealish, 1998). In 2002, the state of Illinois responded to the call to serve youth in their communities by developing a statewide community-based program designed to provide multi-modal services to at-risk youth in substitute care. The program was designed by the Illinois Department of Children and Family Services (DCFS) for children and adolescents who are capable of community functioning but were either at-risk of stepping up to specialized foster or residential care or were stepping down from these higher-level placements. The Illinois model, called community “SOC,” uses a Wraparound approach to treatment, which has been shown to be successful in the mental health, child welfare, and juvenile justice systems (Burchard et al., 2002). This approach is community-based and individualized; therefore, it is consistent with, but not identical to, an SOC approach to service planning and delivery. Prior research has found positive mental health outcomes for Illinois' community “SOC.” One study reported modest positive change in outcome trajectories on a composite measure clinical severity (Sieracki, Leon, Miller, & Lyons, 2008) and another stated that the SOC program is beneficial in preventing placement disruption in foster care youth (McClelland & Schneider, 2009). However, the limitation of this and other prior research in the behavioral health outcomes literature is that outcomes were studied using “main effects” predictors, and failed to test whether the outcome effects were moderated by other clinical variables, social context variables (e.g., caregiver issues) or individual strengths (e.g., psychological, educational). Further, prior research has used composite measures of emotional and behavioral problems and has not disaggregated youth by presenting problem type (e.g., diagnosis) to determine the potentially unique predictors of outcomes for these specific youth. Given the incidence of antisocial behavior in the foster care population, the unique risk these youth have of stepping up to higher levels of care, and the potentially chronic and severe negative outcomes associated with externalizing behavior, it is particularly appropriate to study the variables that specifically predict maintenance versus clinically significant decreases in antisocial behavior. The current study addresses these gaps in the literature by examining the potentially moderating effects of a range of individual, social, and strengths variables on decreases in antisocial behavior in a sample of foster care youth who are at risk of stepping up to residential care. In order to create a model for predicting clinically significant decreases in antisocial behavior among youth in foster care, Optimal Data Analysis (ODA) was used (Soltysik and Yarnold, 1993 and Yarnold and Soltysik, 2005). ODA is an exploratory, non-parametric data analysis method that maximizes the accuracy of the model created from the data sample. ODA's method of statistical analysis is best suited for the current study. The approach to the testing of multivariate interactions used by ODA allows for an unlimited number of variables to be tested to fit into the optimal predictive model. Traditional analyses, such ANOVA and regression, require the selection of specific predictors to be tested in a pre-described model. ODA permits the inclusion of an unlimited number of possible predictors without the specification of hypothetical interactions. Although some researchers argue that only those variables with supporting evidence in the literature should be included in the model of analysis, the techniques used by ODA are able to accommodate an unlimited number of variables without increasing the chance of error (Yarnold & Soltysik, 2005). By not placing restrictions on those variables included in the model, ODA allows variables not previously explored to be examined for involvement in mental health outcomes for youth in foster care. Additionally, ODA permits the creation of subgroups and the examination of moderators within the context of the model, rather than each variable needing to have a predictive effect for the entire group, as is the case in traditional models. For example, gender may moderate the effect of family functioning on the remission of antisocial behavior in youth in foster care. The methodology of ODA allows for the creation of a model that identifies the strongest predictors for each subgroup of the sample (Yarnold & Soltysik, 2005). Based on the previous literature with this clinical population (e.g., Coie and Dodge, 1998, Hinshaw and Lee, 2003 and Moffitt, 1993), a range of variables across the individual and his or her ecologies are suggested to predict outcome. However, it is important to note that the overwhelming majority of variables studied in the child and adolescent antisocial behavior literature have been main effects variables. The literature offers very little guidance on what will emerge from an exploratory statistical analysis designed specifically to unearth highly distinct moderations — many ODA studies unearth up to four or five total interactions. Therefore, the hypotheses below apply to the univariate ODA analyses that will be run and not to the final multivariate ODA results. With this caveat in mind, in the present study, we propose that being female, and the individual and social/ecological variables of having low reported danger to others, positive social behavior, high interpersonal strengths, and/or positive family functioning will predict decreases in antisocial behavior among youth.

3. Results 3.1. Descriptive statistics Descriptive statistics were computed for the sample of youth with antisocial behavior problems used in the analyses (see Table 1). Overall, 77 individual youth were included in the analyses. Youth ranged from 4 to 18 years old, with a mean age of 12.05 years (SD = 3.77), and males comprised more of the sample than females (63.4%). Services were received from 16 distinct agencies, with treatment periods averaging 297.42 days (SD = 125.05). Table 1. Descriptives and UniODA results for youth with antisocial behavior. Study variables % Time 1 M (SD) Time 2 M (SD) ODA time 1 predictors (p-value) ODA time 2 − time 1 difference score predictors (p-value) Age 12.05 (3.77) NA 0.229 NA Sex (male) 63.6 0.152 NA Treatment days 297.42 (125.05) NA 0.133 NA Psychosis 0.25 (0.59) 0.23 (0.51) 0.489a 0.351 Attention problems 1.55 (0.93) 1.32 (0.83) 1.00a < 0.001⁎⁎⁎ Depression 1.51 (0.84) 1.34 (0.64) 0.99a 0.001⁎⁎ Substance abuse 0.57 (0.91) 0.51 (0.74) < 0.001⁎⁎⁎ 0.160 Adjustment to trauma 1.62 (0.82) 1.27 (0.68) 0.015⁎ < 0.001⁎⁎⁎ Attachment 1.39 (0.86) 1.12 (0.85) 0.150 < 0.001⁎⁎⁎ Situational consistency of problems 1.79 (0.79) 1.47 (0.80) 1.00a < 0.001⁎⁎⁎ Temporal consistency of problems 2.07 (0.82) 1.76 (0.88) 1.00a < 0.001⁎⁎⁎ Danger to self 0.42 (0.70) 0.46 (0.72) 0.048⁎ 1.00a Elopement 0.73 (0.87) 0.64 (0.88) 0.029⁎ 0.067 Sexually abusive behavior 0.38 (0.71) 0.30 (0.59) 0.079 0.053 Social behavior 1.43 (0.85) 1.14 (0.84) 1.00a < 0.001⁎⁎⁎ Crime/delinquency 1.00 (0.97) 0.81 (0.99) 0.095 0.001⁎⁎ Intellectual functioning 0.57 (0.72) 0.61 (0.71) 0.126 0.024⁎ Physical functioning 0.34 (0.68) 0.25 (0.61) 0.592a 0.702a Family functioning 1.75 (1.00) 1.60 (0.90) 0.034⁎ < 0.001⁎⁎⁎ School functioning 2.00 (0.89) 1.64 (0.98) 0.045⁎ < 0.001⁎⁎⁎ Sexual development 0.66 (0.87) 0.49 (0.68) 1.00a 0.032⁎ Monitoring 0.99 (0.95) 0.86 (0.85) 0.087 0.045⁎ Treatment 1.03 (0.97) 1.03 (0.93) 0.062 0.109 Transportation 0.64 (0.67) 0.51 (0.62) 0.078 0.033⁎ Service permanence 1.29 (1.05) 1.12 (0.97) 1.00a 0.049⁎ Behavioral health 0.32 (0.57) 0.44 (0.73) 0.280a 0.242 Supervision 0.56 (0.79) 0.74 (0.93) 0.009⁎⁎ 0.008⁎⁎ Involvement with care 0.67 (0.78) 0.81 (0.97) 0.139 0.160 Knowledge 0.85 (0.69) 0.85 (0.78) 0.144 0.054 Organization 0.55 (0.68) 0.62 (0. 83) 0.059 0.933 Resources 0.81 (0.87) 0.74 (0.85) 0.039⁎ 0.016⁎ Residential stability 0.19 (0.43) 0.30 (0.62) 0.208 0.131 Safety 0.40 (0.62) 0.40 (0.68) 0.106 < 0.001 and Family strengths 1.55 (0.88) 1.54 (0.87) 0.202 0.001⁎⁎ Interpersonal strengths 1.48 (0.82) 1.38 (0.86) 0.271 0.001⁎⁎ Relationship permanence 1.78 (0.75) 1.52 (0.75) 1.00a 0.004⁎⁎ Educational strengths 1.50 (0.90) 1.34 (0.90) 0.059 0.012⁎ Vocational strengths 1.86 (1.03) 1.66 (0.99) 0.999a 0.168 Wellbeing 1.91 (0.65) 1.55 (0.79) 0.033⁎ < 0.001⁎⁎⁎ Spiritual strengths 1.39 (1.06) 1.21 (1.10) 0.921a 0.019⁎ Talents 1.56 (0.80) 1.28 (0.91) 0.288 < 0.001⁎⁎⁎ Inclusion 1.58 (0.90) 1.37 (0.93) 0.167 0.045⁎ The LOO approach insures the stability of the predictive model. CANS variables at time 1 and difference scores between time 2 and time 1 were entered into the ODA model as predictors of change in the CANS variable Antisocial Behavior. a Variable was not LOO (leave-one out) stable and, therefore, was not eligible to be entered into the overall classification tree model. ⁎ p < .05. ⁎⁎ p < .01. ⁎⁎⁎ p < .001. Table options The descriptive statistics for the CANS-MH composite scales (problem presentation, risk behavior, care intensity and organization, caregiver needs and strengths, and youth strengths) suggest that this sample's needs and strengths are consistent with other samples of child welfare youth being served in community settings (Lyons, 2004). However, the individual CANS-MH items comprising the various composite scales varied in their rated severity. For example, examining items from the problem presentation scale, antisocial behavior (M = 2.09, SD = 0.29) and adjustment to trauma (M = 1.62, SD = 0.82) were the highest rated items, with average scores nearing the moderate range of impairment across youth (i.e., a “2” rating on the CANS-MH item). This result is intuitive given that this sample was chosen due to their score on the antisocial behavior item of the CANS-MH and that, as a whole, the sample is a higher-risk group of youth in the child welfare system, where PTSD and externalizing behavior are likely to be prevalent and symptoms have been present for a relatively longer period of time. Regarding risk behaviors, temporal consistency (M = 2.07, SD = 0.82) and situational consistency (M = 1.79, SD = 0.79) were the items rated highest across the sample of youth, both items reaching moderate impairment. These findings are consistent with a population of foster care youth with complex needs, requiring multiple foster care placements. In terms of functioning challenges, absence of school (M = 2.00, SD = 0.89) and family (M = 1.75, SD = 1.00) strengths had the highest mean ratings across the sample. Again, this is consistent with a sample of youth who were referred because they were at-risk of stepping up to higher levels of care. Family and school are environmental contexts and the absence of support in these domains is among the most common reasons youth are referred to residential placements ( Stroul & Friedman, 1996). Regarding youth strengths, means were highest across the sample for absence of wellbeing (M = 1.91, SD = 0.65), vocation (M = 1.86, SD = 1.02), and relationship permanence strengths (M = 1.78, SD = 0.75). Absence of strength in wellbeing may be a psychological consequence of the trauma of being within the foster care system. Impairments in relationship permanence are an intuitive result of being a part of the foster care system, as children are often subject to not only removal from their biological parents, but also multiple foster home placements. In terms of the absence of vocational strengths, this sample had a mean age of approximately 12 years, so it is unlikely that these youth would have vocational strengths. 3.2. ODA results UniODA analyses were used to determine the optimal predictors of resolution of antisocial behavior (see Table 1). The identified optimal predictors established subgroups of youth predicted to experience resolution and those not predicted to experience resolution. Multivariate classification trees were created, first for the resolution group and then for the no resolution group, by using UniODA analyses for subsequent predictor variables, controlling for the optimal predictor, until variables no longer significantly predicted resolution. Many variables emerged as having high classification accuracy, both in the initial UniODA and subsequent analyses, however, a strategy was developed where all possible classification trees were created and that with the best overall classification accuracy was retained. Fig. 1 depicts the final ODA classification tree model for youth with clinically significant antisocial behavior at Time 1. Each rectangle signifies a decision point and arrows represent pathways of prediction. P values for each decision point are listed within the rectangles to show significance. The fractions and percentages included within the rectangles represent the number of correctly predicted individuals of the total number included in that category at that particular endpoint. The numbers listed next to the prediction pathway arrows specify the cutoff values for designation into classification categories. Dunn and Sidak adjusted per-comparison p adjustments ( Yarnold & Soltysik, 2004) were used to decrease Type I error. Only those decision points that met the Dunn and Sidak criteria were included. Predictors of outcome in a sample of youth with antisocial behavior: Optimal ... Fig. 1. Predictors of outcome in a sample of youth with antisocial behavior: Optimal Data Analysis (ODA) results. Figure options 3.2.1. Classification tree analysis Initial UniODA results indicated that predominately difference score variables emerged as significant predictors of resolution of antisocial behavior (see Table 1). Change in family functioning emerged as the optimal predictor of resolution. Positive change in family functioning (difference of “1”, “2”, or “3”) formed the group predicted to experience resolution (labeled node C) and those with a difference score less than “1” comprised the group predicted to not experience resolution. The subgroup predicted to experience resolution was not subjected to additional UniODA analyses because all youth in this subgroup were correctly classified (i.e., 100% of those with positive change in family functioning experienced resolution). An additional UniODA was run for the group predicted to not experience resolution from antisocial behavior (those with a family functioning difference score less than “1”). Change in adjustment to trauma entered the multivariate analyses next as it emerged as the next best predictor of resolution for this subgroup. Youth with positive change in their adjustment to trauma score (difference of “1”, “2”, or “3”) formed the subgroup predicted to experience resolution while those with no change or negative change in their adjustment to trauma score (difference of less than “1”) comprised the subgroup predicted to not experience resolution. Additional UniODA analyses for both groups did not reveal any variables that further classified the sample significantly. Those with no change or negative change in their adjustment to trauma score (difference of less than “1”) were predicted to not experience resolution with 82% accuracy (labeled group A) and those with positive change in their adjustment to trauma score (difference of “1”, “2”, or “3”) were accurately predicted to experience resolution of antisocial behavior on the CANS-MH in 75% of cases (labeled group B). Classification performance statistics were computed for the full CTA model for antisocial behavior, as well as the statistics for each of the resolution and no resolution group (see Table 2). The overall model was predicted with 89.3% accuracy. The mean sensitivity across classes was 90.8%, with a sensitivity of 87.2% for the resolution of antisocial behavior group and 94.4% for the group that did not experience resolution. The mean specificity across classes was similar, with a mean of 90.0% for the full CTA model. Specificity for the group that experienced resolution was 86.7% and 93.3% for the group whose antisocial behavior did not resolve. The overall classification tree predicted resolution 78.6% above chance, which is considered to be a “strong” effect strength according to parameters set forth by Yarnold and Soltysik (2005). Table 2. Classification performance summary for the classification tree model of resolution vs. no resolution of antisocial behavior (N = 75). Performance index Performance parameter Effect strength Overall classification accuracy 67/75 (89.3%) 78.6% Sensitivity (resolution) 41/47 (87.2%) 74.4% Sensitivity (no resolution) 34/36 (94.4%) 88.8% Mean sensitivity across classes 90.8% 81.6% Specificity (resolution) 39/45 (86.7%) 73.4% Specificity (no resolution) 28/30 (93.3%) 86.6% Mean specificity across classes 90.0% 80.0% Mean performance across classes 90.4% 80.8% Overall cross-classification table Predicted status No resolution Resolution Actual status No resolution 28 2 Resolution 6 39 Overall classification accuracy is the percentage of the total sample that is correctly classified by the overall tree model. Sensitivity is a predictive indicator of the percentage of the predicted classifications into a given category that were correct. Specificity is a descriptive index of the percentage of the actual members of a given category (i.e., those who experienced resolution of their antisocial behavior) that the classification tree correctly categorized. Effect strength is a standardized index of the performance of the model, defined as the percentage above chance that the model correctly predicts, on a 0–100 scale, where 0 is the performance expected by chance and 100 is perfect classification accuracy. The statistic is computed using the following formula: [(1 − {(100-model performance statistic) / (100/C)}) × 100%], where C is the number of response categories for the class variable (Yarnold, Soltysik, & Bennett, 1997, p. 1454). Effect strengths of 25% or less are considered weak, values between 25% and 50% are considered moderate, and those above 50% are considered strong (Yarnold & Soltysik, 2005).