Ecodevelopmental and Systemic Modeling and Implementing High Fidelity Interventions in Real World Settings Guillermo Prado1,2, Hilda Pantin1, Seth Scwartz1, Jose Szapocznik1 & Daniel J. Feaster 1,2 1 Center for Family Studies, University of Miami 2 Stempel School of Public Health, Florida International University 1 The Ecodevelopmental Model and Methodological Questions 2 The Center for Family Studies is interested in the role of ecodevelopmental context in the prevention and treatment of adolescent behavior problems, drug abuse, and HIV/AIDS. 3 Ecodevelopmental Theory Incorporates three primary, integrated components: (a) Social Ecological Theory (b) Developmental Theory (c) Emphasis on Social Interactions 4 Social Ecological Theory Bronfenbrenner (1979, 1986) Highlights the Multiple Influences on Adolescent Development, including: (a) Macrosystems (e.g., cultural & societal values) (b) Exosystems (e.g., parents’ exosystemic stressors and social support for parents) (c) Mesosystems (e.g., parental monitoring of peers and collaboration with youth’s school) (d) Microsystems (e.g. family functioning) 5 Developmental Component Emphasizes the changing nature of youth, contexts, & their interdependence over time e.g., family functioning is influenced not only by parents’ current social support & work stress, but also by previous levels of social support & work stress e.g., current family functioning in turn influences both present and future levels of adolescent behavior problems 6 Social-Interactional Component Risk and protection are expressed in the patterns of direct transactions between individuals within and across the different contextual levels e.g., when parents engage in supportive interactions with individuals outside the family, they are more likely to parent their children in supportive rather than harsh ways. 7 An Ecodevelopmental Perspective on Prevention 8 Context of Adolescent Behavior Problems, Drug Use, and Risky Sexual Behavior Social-Cultural Context Parental Resources/Stressors Parents’ Social Support Parents’ Work Stress Cultural Mismatch Family Microsystem Parent-adolescent communication about sex Parent-Adolescent Communication Parental Involvement Family-School Relations Parental involvement in school Monitoring homework School School Bonding Academic Achievement Immigration Policy Positive Parenting Marital Conflict Family support Language Problems Family-Peer Relations Parental monitoring of peers Supervision of situations of sexual possibility Peers Substance use w/ friends Sexually active friends Poverty 9 Ecodevelopmental Model of Problem Behaviors Social Support for Parents Adolescent Acculturation Family Exosystemic Stressors Parent Acculturation Parental Monitoring of Peers Parental-Adolescent Communication about Sex Peer Sexual Behavior Family Relations Early Adolescent Sexual Initiation Social Cognitive Mediators re. Sex Early Adolescent Substance Use Peer Substance Use Early Adolescent Problem Behaviors Social Cognitive Mediators re.Drug Use 10 Implications of Ecodevelopmental Theory for Methodological Development Ecodevelopmental research involves multiple levels of nesting Repeated Observations Individuals Families Ecosystemic levels across developmental stages Statistically, data are non-independent— Substantively people and their social contexts are interdependent 11 Implications of Ecodevelopmental Theory for Methodological Development Ecodevelopmental research involves longitudinal processes Ecodevelopmental processes are interrelated and influence each other over time How do we model these interrelationships over time? Ecodevelopmental processes develop over time, and a snapshot of such a process (e.g., family functioning at baseline) is not accurate How do we account for this in our model? 12 Implications of Ecodevelopmental Theory for Methodological Development Ecodevelopmental research involves influences from multiple systems Ecodevelopmental models involve two and three level interactions How do we model these moderation effects when the variables are observed? How do we model these moderation effects when the variables are latent? 13 Multiple levels of nesting in effectiveness trials Time is nested within adolescents who are nested within families who are nested with therapist who are nested both within treatment & site Treatment crosses site 14 Implementing a High Fidelity Systemic Treatment in Drug Abuse Treatment Centers 15 Plan of Section Clinical Trials Network (CTN) BSFT BSFT Design Issues—Level of Control BSFT Experience in High Fidelity Implementation 16 Clinical Trials Network NIDA funded network to test the effectiveness of efficacious treatments in real world settings As of 9/05: Nodes = 17 States = 34+Puerto Rico CTPs = 152 Protocols = 27 (11 closed to enrollment, 5 in development) Currently 88 CTPs involved in 11 open studies Mission: To implement science-based efficacious treatments in community settings AND to show that these implementations are an improvement over current practice 17 Brief Strategic Family Therapy—BSFT, CTN0014 Brief Strategic Family Therapy is a systemic, process-focused family therapy 4 months with weekly sessions Up to 8 booster sessions over 8 months Focus on changing repetitive (maladaptive) interactions within the family Note that the focus is on underlying processes, not crises May use a crisis as a content focus, but therapy addresses underlying “everyday” processes BSFT has been the focus of over 30 years of research at the Center for Family Studies 18 BSFT-Therapy Components Three major techniques Joining (Engaging Participants into Treatment) • Balanced across all family members • Must join with most powerful Diagnosis (Family Relationships and Roles) Restructuring (Implementing the Treatment Plan) • Work in Present—Enactments • Reframing • Shifting boundaries 19 BSFT CTN0014 8 Sites 60 participants per site on average 480 adolescent participants Drug Use assessed monthly for 12 months Full assessments at Baseline, 4 months, 8 months and 12 months Delinquent Behaviors, Conduct Problems, Sexual risk behaviors, Adolescent Pro-social Activities, & Family Functioning 20 BSFT—Randomization Participants will be randomized to BSFT or the clinic’s standard outpatient treatment Note, randomization is at the individual level, not at the clinic level 21 Design Considerations for Effectiveness Trials Level of Control Homogeneity of Study Population (Szapocznik) Standardization & Monitoring of Treatment (Szapocznik) Standardization & Monitoring of Control (Feaster) Handling of Site Variance (Feaster) Efficacy study—Fixed Effect Effectiveness—Random Effect 22 BSFT: Heterogeneous Treatment Population Inclusion Criteria: 12-17 years of age Any illicit drug use in last 30 days Lives or is expected to with “family” Reside in the same geographic area as CTP Signed consent & assent Exclusion Criteria: Not living with family (halfway house, institution, etc.) Suicidal or homicidal risk must be stabilized, first Current or pending severe criminal charges if likely to lead to incarceration If already receiving drug treatment services 23 Choice for BSFT: Full Control of Experimental Condition Extensive Training and Supervision 5 months of training Weekly supervision Training and Supervision are considered integral to the BSFT model 24 Multiple Levels of Clinical Supervision & Adherence Monitoring Clinical Supervision Weekly conference calls with a BSFT supervisor. Supervision includes: National Clinical Supervisor weekly face-to-face sessions with each clinical supervisor Regularly sits in on selected supervision calls with sites Adherence Ratings of Videotapes videotape review case discussion and planning. Randomly selected sessions trained independent raters in Miami Failure to adhere to the BSFT model Definition-<70% adherence for 3 consecutive sessions Consequences & Corrective Action no new cases until 80% in two consecutive sessions Increased supervision & retraining until meets criteria If criteria not met before conclusion of current cases withdrawn from study (at discretion of clinical supervisor) 25 Therapist Consent and Selection Process Identification of volunteers Consent Demographics Views of Adolescent Drug Abuse – Q-sort Selection Interviews (Site PI, National Study Director/ Coordinator, and BSFT Head Training Supervisor) Family Session Randomization Academic training Years of clinical experience 26 Training Phase for BSFT Therapists Five-month clinical training program Workshops Four 3-day workshops Week 1 (Workshop 1) *Miami Week 3 (Workshop 2) Week 5 or 6 (Workshop 3) Week 13 (Workshop 4) Supervision Weekly group supervision Each therapist will have ½ hour for videotape review and ½ for case discussion Pilot Cases: 2-4 cases for each therapist Certification 27 Implementation Phase Active Cases Caseload builds over time Study caseload will range from 2-8 cases (minimum = 0; maximum = 10) Supervision Weekly group supervision Review of videotapes Review of clinical forms Treatment planning Each therapist (2 active) will have 30-45 minutes for videotape review and 30-45 minutes for case discussion 28 Adherence Ratings By Site Node/Site All Sites 15/301 7/802 7/302 7/604 13/701 9/200 3.5 (3.2, 3.9) 3.5 (3.2, 3.7) - 3.4 (3.0, 3.7) 3.5 (3.3, 4.1) 3.8 (3.7, 4.0) 3.6 (3.1, 4.0) 108/130 (83.1%) 38/46 (82.6%) - 21/26 (80.8%) 30/34 (88.2%) 4/4 (100.0%) 15/20 (75.0%) 3.8 (3.5, 4.0) 41/46 (89.1) 4.1 (4.0, 4.2) 6/6 (100.0%) 4.0 (3.8, 4.0) 16/16 (100.%) 4.0 (3.5, 4.3) 28/31 (90.3%) 3.8 (3.8, 3.8) 1/1 (100.0%) 3.7 (3.6, 4.0) 10/12 (83.3%) 4.0 (3.0, 4.0) 4.0 (3.0, 4.0) 108/112 (96.4%) 44/46 (95.7%) 4.0 (4.0, 4.0) 6/6 (100.0%) 4.0 (4.0, 4.0) 4.0 (3.0, 4.0) 16/16 (100.0%) 29/31 (93.6%) 4.0 (4.0, 4.0) 1/1 (100.0%) 4.0 (3.5, 4.0) 12/12 (100.0%) Domain Average for Blind Ratings Median (25th, 75th) % Good Adherence2 Domain Average for Clinical Ratings Median (25th, 75th) % Good Adherence 3.9 (3.6, 4.0) 102/112 (91.1%) Overall Supervisor’s Ratings Median (25th, 75th) % Good Adherence Good Adherence is 3 29 Adherence Ratings Show adequate adherence Some variability across sites Supervisors’ ratings uniformly higher than than independent raters’ 30 31 Some Statistical Approaches and Areas for Further Research in Systemic Modeling and Design of Effectiveness Trials 32 Plan of Section: Specification of Site Effects & Bounds of Inference Power analysis and Trial Planning Time Structure of Models & Reciprocal Effects Need for Simulation Research Questions about Mplus & Simulation 33 Design Considerations for Effectiveness Trials Level of Control Homogeneity of Study Population (Szapocznik) Standardization & Monitoring of Treatment (Szapocznik) Standardization (Feaster) & Monitoring of Control Handling of Site Variance (Feaster) Efficacy study—Fixed Effect Effectiveness—Random Effect 34 Considerations for Choice of Comparison Condition Standardized Control Group Smaller sample size (sites & participants) High internal validity Lacks ecological validity for CTPs Treatment As Usual Larger sample size (sites & participants) High external validity Highly variable across sites Minimum site size— within site comparisons 35 Choice for BSFT: Treatment As Usual Compare BSFT to the population of treatments in the community Currently drug abuse treatment has considerable variability in treatment approach and implementation 36 Handling of Site Variance Fixed Effect (control for site & remove variance from error): Cannot generalize statistically Random Effect: Can generalize effect beyond the clinics included Both site & site X treatment are random Implications on power of study 37 Choice for BSFT: TAU Differences in TAU at each site implies larger variance of the random Site X Treatment effect Sites need not be randomly assigned, but need to describe the generality of clinics (if not randomly selected) 38 Fidelity of Implementation Variability in fidelity across sites will increase the site and site by treatment effects 39 Specifics of BSFT Statistical Plan Hypothesis 1: BSFT will be significantly more effective than TAU in reducing adolescent drug abuse, defined as the percentage of days with positive drugs use. 40 Example of Expected Trajectories Drug Use Over Follow-up 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 BSFT-Outpatient 0.1 TAU-Outpatient 0 T4 T8 T12 41 Secondary Hypotheses BSFT will be significantly more effective than TAU in : – Reducing: Delinquent behaviors & conduct problems Sexually risky behaviors – Increasing: Prosocial activities (school, employment) Family functioning (parenting, parentadolescent communication) 42 Analysis Strategy Randomization at the Individual Client Level Multi-level Growth Curve with 3 levels Level 1 is within person—Time Level 2 is between persons (and incorporates treatment assignment) Level 3 is between site Controlling for baseline drug use (ANCOVAtype specification) Note: not accounting for all levels of nesting in primary hypotheses (therapist effects examined in post-hoc analyses) 43 Proposed Model Level 1—Time: yijt ij 0 ij1 aijt ijt Time, a, is centered at Month 4 Level 2—Between Adolescents: ij 0 0 j 0 1 j 0 BSFT rij 0 ij1 0 j1 1 j1BSFT rij1 Level 3—Between Sites: 0 j 0 000 u0 j 0 1 j 0 100 u1 j 0 0 j1 001 u0 j1 1 j1 101 u1 j1 000 100 001 101 Grand Mean Month 4 Drug Use in TAU Increment to Month 4 Drug Use for BSFT Grand Mean Slope in Drug Use in TAU Increment to Slope in Drug Use for BSFT 44 Multisite Power Analyses (Raudenbush & Liu, 2001) Procedure assumes single post test (not growth curve) We have 12 monthly drug use assessments However, drug use is not normally distributed 1.0 = 0.050 2 = 0.05 2 J= 8,n= 56, = 0.10 2 J= 8,n= 56, = 0.15 0.9 J= 8,n= 56, 0.8 0.7 P o w e r 0.6 0.5 0.4 0.3 0.2 0.1 0.20 0.40 0.60 Effect Size 0.80 1.00 45 Methodological Issues Most research on design implications of site effects has been done by statisticians with drug trial experience Aim is to prevent site effects or justifying ignoring them if they exist Therefore, little prior evidence published on site variability & particularly on the site by treatment interaction 46 Methodological Issues Power Analysis Lack of prior info makes difficult Lack of software (Raudenbush & Liu, 2001) Simulation in M-Plus Have been doing simulations In SAS, have difficulty with 8 sites identifying the site & site X treatment variance in growth curve framework Specifically, it may not be possible to identify the covariance between the site and site X treatment random components Simulations do not exactly match Raudenbush & Liu 47 Resource Allocation Issue Random site X treatment effect with variability in Treatment as Usual requires a large sample Necessity of effectiveness research if to generalize to new clinics Some disagree (due in part to costs) Simulation to show Type 1 error Look at potential mistakes of policy based on inappropriate overly precise estimates 48 Reciprocal Effects Person-specific cross-lagged panel model Could be formulated in a latent difference score framework Model is important for systemic phenomenon like family interactions Illustrated within a person using coping and distress data 49 “SETA-Fam” Hypotheses b SET a Family Functioning Target Woman’s Outcomes d e c Dotted Line Refers to Hypotheses in SETA Family Members’ Outcomes 50 Within A Person Distress 1 a Distress 2 c e e f Coping 2 e f d d b Distress 4 c f d a Distress 3 c z Coping 1 a b Coping 3 b Coping 4 Test a through d as random effects 51 Within a Family Distressj 1 a Distressj 2 c e d f e Distressk 2 Distressj 4 c f e f d d b a Distressj 3 c z Distressk1 a b Distressk 3 b Distressk 4 Test a through d as random effects 52 Distress M=.61* AR V=.02* Cross Lagged w/AR Random Distress T1 Distress T2 Distress T3 .19* .19* .19* .03 .03 .03 Coping T1 Coping T2 Coping T3 Adj. BIC=2206.9 Coping M=.55* AR V=.03 Distress T4 Coping T4 53 How does Model Fit Compare to Growth Curve Specification? Distress Baseline .77* Coping Baseline -.46* Distress Linear Distress Quadratic V=0 .9999* Coping Linear Adj. BIC=2219.5, *Difficulty with Psi Coping V=0 Quadratic 54 Selected Issues to Address Questions to be addressed: Parameter interpretation Comparative Model Fit/Discrimination Initial Conditions Different time frames both within and between individuals (data is collected on only one time frame, though with variability in exact times of collection) Simulations Question, is it possible to vary parameters in a simulation within Mplus and save the parameter settings to the generated data file? [AR@.3 to .8 by .1]. AR @ 0 to 5 by .5. 55 56 57