Risk Adjusting Residential Treatment Outcomes with Clinical Factors Neil Jordan, PhD Interim Director, Mental Health Services & Policy Program Department of Psychiatry & Behavioral Sciences Northwestern University Feinberg School of Medicine Introduction ► Increasing emphasis on accountability, qualityy of care of mental health services q ► Government funders are using outcomes data to assess provider/program performance & influence payment rates ► Not all providers are alike, nor do their caseloads remain the same over time Differences in outcomes across providers reflect more than just differences in quality Risk Adjustment (RA) ► Method M th d th t accounts that t ffor group diff differences when h comparing non non--equivalent groups on outcomes ► Takes T k into i t accountt ffactors t nott iin th the provider’s id ’ direct control that are related to outcomes ► Risk Ri k variables i bl are those th that th t influence i fl outcomes t but are not part of treatment ► Good G d risk i k adjustment dj t t models d l allow ll for f more equitable and valid comparisons of performance than using unadjusted values Risk Adjustment in Mental Health ► Still emerging Adult psychiatric hospitalization (Banks 1999) Short--term rehospitalization (Hendryx 2001) Short Seclusion & restraint incidents (Schacht 2003) Consumer satisfaction (Greenberg 2004) Youth community mental health services (Ogles 2008) Youth residential treatment (McMillen 2008) Why is RA Uncommon in MH? ► Lack of necessary data Outcomes across multiple providers/programs Preferably not generated and reported by providers Predictor variables measured before the intervention (to serve as risk adjustors) Existing administrative data (because primary d t collection data ll ti for f RA would ld be b very expensive) expensive i ) Lack of comprehensive clinical data Study Objective ► McMillen’s M Mill ’ 2008 study d off RA in i youth h residential treatment recommended including l d measures off clinical l l functioning, f MH diagnoses, and/or MH need as risk adjustors ► Purpose: to determine whether the addition of these types of variables improves a RA model for youth residential treatment (RT) outcomes Study Population ► Youth in state custody receiving RT services for behavioral health p problems in Illinois ► 45 residential treatment and group home providers participating in a pay for performance initiative, excluding Those that serve youth with intellectual disorders Shelter placements Data Source Source, Study Sample ► Administrative data from Illinois Department of Children & Familyy Services ► RT placement during 2007 or 2008 and clinical data prior to RT spell (n=1486) (n=1486) ► Sample characteristics: 61% male Mean age = 14.5 14 5 55% African African--American Outcomes ► Treatment stability (TS): % of days at the facility and not on run, in a psychiatric hospital, or incarcerated ► 180 180--dayy p post-discharge postg placement p stabilityy (PDPS) ( ) (dichotomous) “Favorable” discharge g to less restrictive setting… g ► Foster care, independent living, transitional living ► Less restrictive residential or group home setting ► Placement in chronic mental illness setting …and sustained subsequent placement for 180 days Primary Risk Adjustors – Prior Child Systems Involvement ► Measures Aggressive symptoms and antipsychotic use Juvenile detention or corrections Runaway Psychiatric hospitalization ► Dichotomous Primary Risk Adjustors - CANS ► ► ► ► ► Child and d Adolescent Ad l t Needs N d & Strengths (Lyons, 1999) Clinical outcomes management tooll used d for: f Assessing progress within placement Determining appropriateness for placement change p g Completed by trained rater Continuous; item score ranges from 0 (no need for action) to 3 (need for immediate action) Used most recent CANS prior to beginning of RT placement ► 59 items it across 6 needs d domains: Behavioral/emotional needs (d (depression) i ) Risk behaviors (danger to others) Traumatic stress symptoms (numbing) Trauma experiences (emotional abuse) Life domain functioning (social) Acculturation (cultural stress)) stress Analytic Methods_1 Methods 1 ► Exploratory factor analysis of CANS items ► Factor analysis Principal p factors extraction Varimax rotation ► Yielded 77-factor solution: Externalizing Trauma Sexuality School problems Juvenile justice interactions Acculturation Internalizing Analytic Methods_2 Methods 2 ► Regression R i models d l Ordinary least squares (TS) Logistic L i ti regression i (PDPS) ► Also adjusted for: Demographic D hi characteristics h i i (age, ( gender, d child’s hild’ geographic origin) Placement characteristics (spell length length, severity level and/or specialty population served by unit where child is placed, program’s geographic location) Prior placement in residential treatment Base Model (Prior Child System Involvement Variables Only) Post-Discharge P t Di h Placement Stability Treatment Stability Risk Adjustor Β 95% CI Β 95% CI Aggressive symptoms & antipsychotic use -.024* (-.042, -.005) -.35 (-.79, .09) Detention or corrections -.034*** (-.050, -.017) -.53* (-.95, -.12) Runaway -.040*** (-.059, -.020) -.58* (-1.03, -.13) P Psychiatric hi t i h hospitalization it li ti -.035*** 035*** ( 055 -.016) (-.055, 016) -.50* 50* ( 95 -.05) (-.95, 05) *p< 05 **p<.01, *p<.05, **p< 01 ***p<.001 ***p< 001 Base Model + Clinical Factors Post-Discharge Placement Stability Treatment Stability Risk Adjustor Β 95% CI Β 95% CI Aggress symptoms/antipsychotic use -.022* (-.041, -.003) -.39 (-.85, .07) Detention or corrections -.031** (-.050, -.013) -.33 (-.79, .12) Runaway -.036*** (-.056, -.016) -.48 (-.96, .002) P Psychiatric hi t i h hospitalization it li ti -.030** 030** ( 050 -.010) (-.050, 010) -.56* 56* ( 1 03 -.08) (-1.03, 08) Externalizing -.004 (-.013, .005) -.40** (-.63, -.16) Trauma -.006 ((-.015,, .004)) .15 ((-.08,, .38)) Sexuality -.002 (-.011, .008) -.01 (-.25, .24) School problems -.015** (-.025, -.005) -.24 (-.49, .01) Juvenile justice interactions -.004 (-.016, .007) -.16 (-.45, .12) Acculturation -.001 (-.009, .008) .08 (-.12, .29) Internalizing - 016** -.016 ((-.027, 027 -.006) 006) .02 02 ((-.23, 23 .28) 28) *p<.05, **p<.01, ***p<.001 Preliminary Conclusions ► Prior child system involvement variables explain unique variation in treatment stability and postpost-discharge placement stability ► Clinical needs variables explain additional variation a at o in tthese ese outcomes outco es Treatment stability: internalizing, school issues Post Post--discharge placement stability: externalizing ► Prior child system involvement variables do not appear to be proxies for clinical needs Limitations ► Risk adjustment doesn’t create perfectly equivalent q groups g p or achieve the purism p of random assignment ► Best models explain only 15% 15%--23% of variation in outcomes ► CANS rates child and adolescent needs, not diagnoses Next Steps ► Other items methods for incorporating clinical needs CANS item level: classification and regression trees analysis (to identify significant interactions) interactions) Rasch analysis (to derive a total clinical needs score) ► Explore OLS alternatives for treatment stability model ► Examine p predictive power p of best models Acknowledgements ► Collaborators Jielai Ma, PhD, Northwestern University Richard Ri h d Epstein, E t i PhD, PhD Vanderbilt University Scott Leon, PhD, Loyola University Chicago Andy Zinn, PhD, Chapin Hall at the University of Chicago Alan Morris, Morris PsyD PsyD,, University of Illinois Illinois--Chicago Deann Muehlbauer, MPH, University of Illinois Illinois--Chicago Kathleen Kearney, JD, University of Illinois at Urbana--Champaign Urbana p g ► More collaborators Christopher Larrison, PhD, University of Illinois at Urbana--Champaign Urbana Gary McClelland, PhD, Northwestern University Brice Bloom Bloom--Ellis, Ellis MSW, MSW LCSW, Illinois Department of Children & Family Services ► Funding Support Children’s Bureau, Administration for Children & Families US Dept of Health Families, & Human Services Illinois Department of Children & Familyy Services