Using CANS as a Placement Decision Support Algorithm to Predict Outcome of Youth in Child Welfare Placed in Residential Treatment Northwestern University Feinberg School of Medicine Mental Health Services & Policy Program Brian Chor, B.S., Gary McClelland, Ph.D., & Neil Jordan, Ph.D. 6/29/2009 Residential Treatment & Placement Most expensive form of mental health treatment Least restrictive setting Per diem ≈ $100 ~ $600; Annual cost /child ≈ $50,000 ~ $75,000 In Illinois in 1995, $350 million/$450 million budget invested in residential treatment To protect children and provide services in least restrictive setting available; efficacy of community-based treatment Misconstrued as “fail to proceed”, last resort placement Absence of standard placement criteria Inconsistency and fallibility of judgment Non-clinical criteria (e.g., bed availability, location) Introduction → Hypothesis → Methods → Results → Discussion Multi-Disciplinary Team Enhances traditionally individualized clinical decisions Pooled, multi-disciplinary expertise Family and youth participation Shared responsibility; consensual decisions Develops and sustains consistent and accountable decisions Limited evidence of efficacy Group better than individual decision-making? Group incurs same limitations of individual decisionmaking ↑ Group members ≠ ↑ Consistent decisions Feasibility and complexity of multi-disciplinary process Introduction → Hypothesis → Methods → Results → Discussion Decision Support Algorithm Articulates and clarifies logic of placement criteria best served by a level of care Level of Care = f (Criteria) Standardizes criteria, restores clinical focus Enhances quality, consistency, and predictability of decisions Aids, not replaces, human clinical judgment Empirically predicts differential anticipated outcomes Children deemed not eligible should not benefit Versatility Mental health Substance abuse Nursing care Hospitalization Introduction → Hypothesis → Methods → Results → Discussion Synthesis & Hypothesis High cost Policy of least restrictive setting Lack of standardized placement criteria Limited empirical research on placement decisionmaking Needs-based and evidence-based practices Understanding 2 models of placement decisions through the language of clinical outcome is necessary Hypothesis: Concordant placement decisions between the 2 models predictive of greater clinical improvement in subsequent residential treatment than discordant decisions: Multidisciplinary Predicted Clinical Team Improvement Concordant Residential Residential Greater Discordant Less Restrictive Residential Less Decision Algorithm Introduction → Hypothesis → Methods → Results → Discussion Sample 544 wards of Illinois DCFS whose residential placements were determined by the multidisciplinary Child and Youth Investment Teams (CAYIT); CANS algorithm separately recommended level of care using clinical ratings obtained during CAYIT staffings: Group CANS Algorithm CAYIT Residential Less Restrictive Residential Residential Concordant Discordant Concordant (n=449) Mean / % 14.8 Age Gender Male 62.8% African American Caucasian Hispanic & Other Completed placement Length of placement (Day) Length of wait till placement (Day) 61.5% 32.0% 6.5% 54.6% 363.1 79.6 Ethnicity Predicted Clinical Improvement Greater Less Discordant (n=95) Mean / % p 14.6 n.s. n.s. 60.0% n.s. 58.5% 36.2% 5.3% 56.8% n.s. 354.7 n.s. 84.1 n.s. Introduction → Hypothesis → Methods → Results → Discussion CANS Algorithm Criteria for Residential Treatment: Requires both Risk Behaviors & Emotional/Behavioral Issues Criteria 1. At least 2 or more '3' among the following needs 2. At least 3 or more '2' among the following needs CANS Item 14 46 47 48 49 50 51 52 53 54 55 56 57 58 Adjustment to Trauma Psychosis Attention/Impulse Depression Anxiety Oppositional Conduct Substance Abuse Attachment Eating Disturbance Affect Dysregulation Behavioral Regression Somatization Anger Control 14 46 47 48 49 50 51 52 53 54 55 56 57 58 Adjustment to Trauma Psychosis Attention/Impulse Depression Anxiety Oppositional Conduct Substance Abuse Attachment Eating Disturbance Affect Dysregulation Behavioral Regression Somatization Anger Control 3. A rating of '2' or '3' on Developmental 32 Developmental 4. At least 1 or more '3' among the following risk behaviors 5. At least 3 or more '2' among the following risk behaviors 59 60 61 62 63 65 67 59 60 61 62 63 64 65 66 67 68 69 Suicide Risk Self Mutilation Other Self Harm Danger to Others Sexual Agression Delinquency Fire Setting Suicide Risk Self Mutilation Other Self Harm Danger to Others Sexual Agression Runaway Delinquency Judgment Fire Setting Social Behavior Sexually Reactive Behavior CANS Domain Traumatic Stress Symptoms Behavioral/Emotional Needs Decision Rules Rule #1: Traumatic Stress Symptoms (Criteria 1 or 2 or 3) AND (Criteria 4 or 5) Rule #2: Behavioral/Emotional Needs Consider a specialty Residential Treatment if Rule #1 is met and a ‘2’ or ‘3’ is rated on any of the following items CANS Item Life Domain Functioning Risk Behaviors 32 36 37 63 65 CANS Domain Developmental Life Domain Functioning Medical Physical Sexual Agression Risk Behaviors Delinquency Risk Behaviors Introduction → Hypothesis → Methods → Results → Discussion Clinical Outcome ∆ CANS standardized scores in 5 domains: Mixed Design ANOVA – Group x (Time x Domain) T0: CANS at CAYIT staffing T1: Interim residential CANS T2: 3-6 months residential CANS Placement stability: Survival Analysis Hospitalization rates: Zero-inflated Poisson Regression Between date of CAYIT and placement entry Between placement entry and 6 months after Introduction → Hypothesis → Methods → Results → Discussion CANS Strengths 0.2 0 -0.2 -0.4 -0.6 -0.8 -1 -1.2 0.4 0.4 0.2 0 -0.2 -0.4 -0.6 -0.8 -1 -1.2 T0 T1 Time T2 T0 Risk Behaviors 0.2 0 -0.2 -0.4 -0.6 -0.8 -1 -1.2 -1.4 -1.4 -1.4 T1 Time T2 T0 T1 Time 0.4 0.2 0 -0.2 -0.4 -0.6 -0.8 -1 -1.2 0.4 - 0.2 0 -0.2 Concordant Group Discordant Group -0.4 Note. Higher score denotes greater severity. -0.6 -0.8 -1 -1.2 -1.4 -1.4 T0 T1 Time T2 Behavioral/Emotional Needs Standardized CANS Scores Standardized CANS Scores Life Domain Functioning Standardized CANS Score 0.4 Standardized CANS Scores Standardized CANS Scores Trauma Symptoms T2 T0 T1 Time T2 Introduction → Hypothesis → Methods → Results → Discussion CANS Discordant Standardized CANS Scores 0.4 0.2 Trauma Symptoms 0 Strengths -0.2 Behavioral/ Emotional Needs -0.4 -0.6 -0.8 Life Domain Functioning -1 -1.2 Risk Behaviors -1.4 T0 T1 Time T2 Standardized CANS Scores Concordant 0.4 0.2 0 -0.2 Trauma Symptoms -0.4 Strengths Behavioral/ Emotional Needs -0.6 -0.8 -1.2 -1.4 Note. Higher score denotes greater severity. Life Domain Functioning -1 Risk Behaviors T0 T1 Time T2 Note. Higher score denotes greater severity. Group x Time: p <.05; Group x (Time x Domain): p=.07 T0→T1 or T2: Improvement of Concordant group significantly greater than Discordant group in Behavioral/Emotional Needs, Risk Behaviors, and Life Domain Functioning. By T2: No significant differences between Concordant and Discordant groups across 5 domains. Introduction → Hypothesis → Methods → Results → Discussion Placement Stability Residential Treatment Group Completed Yes No Total Concordant 245 190 449 Discordant 54 40 94 Discordant Group more likely to disrupt (HR=1.13, n.s.) Controlling for other predictors No significant Group differences Prior runaway predictive of disruption (HR=1.63, p<.001) Male predictive of stability (HR=0.59, p<.05) Introduction → Hypothesis → Methods → Results → Discussion Hospitalization Rates Incidence Rate (IR) –/-- Concordant Group –/-- Discordant Group 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 High Risk Hospitalization Low/Medium Risk Hospitalization Between Date of CAYIT and Placement Entry Between Placement Entry and 6 Mo. After Hospitalization rates improved in both groups over time, though greater improvement in Discordant Group During treatment, controlling for other predictors Discordant Group resistant to high risk hospitalization (IRR=.32, p<.01) Male resistant to high risk hospitalization (IRR=.34, p<.001) Introduction → Hypothesis → Methods → Results → Discussion Summary Clinical outcomes supporting the hypothesis and validity of the algorithm were supplemented by less robust findings regarding placement stability and hospitalization rates Caveat:↑Room for clinical improvement in Concordant group suggests regression to the mean, but Baseline differences required to differentiate Group membership Regression to the mean could not make groups equal at T2, or account for all improvement Introduction → Hypothesis → Methods → Results → Discussion Implications Following the CANS algorithm (Concordant decisions) predicted greater clinical improvement in behavior and emotional symptoms CANS algorithm improves residential placement decisions Over-treating wards (Discordant decisions) by ignoring CANS algorithm decisions led to fewer and inconsistent clinical gains Less restrictive, communitybased treatment for overtreated wards Placement decisions based on standardized clinical criteria are encouraged. CANS algorithm needs to be refined and expanded Introduction → Hypothesis → Methods → Results → Discussion Implications Less robust outcome in placement stability and hospitalization rates likely due to lack of comparisons of Concordant/Discordant decisions in lower levels of care and inter-site variation in crisis management, respectively. Future research should Account for placement of origin and wait time for placement Examine predictors of Concordant vs. Discordant decisions Evaluate the long-term impact of decisions beyond immediate placement Expand outcome variables by considering adverse events (e.g., runaway) Introduction → Hypothesis → Methods → Results → Discussion