Identifying and Comparing Profiles of Treatment Need for Youth in School-Based and Community-Based Programs at a Large Mental Health Agency

Background and Purpose:

In the era of evidence-based practice, organizations are responsible to use data to improve service delivery and client outcomes. Mental health agencies need efficient strategies to organize, analyze, and disseminate large amounts of data (Bates et al., 2014).  Recursive partitioning methods (e.g. Strobl et al., 2009), including random forests and regression-based decision trees, have been identified as innovative approaches to handling large datasets but are rarely used in health services organizations (Cordell et al., 2016). The current study explores patterns of treatment need for youth receiving assessments in a large youth mental health agency and compares profiles of need for youth in school-based and community-based programs.

 

Methods:

Youth entering services (mean age: 13.5 years) included 624 youth receiving school-based services (32% female; 42% African-American / Black, 28% European-American/White, 30% Other) and 673 clients receiving community-based services (39% female; 39% Latino/Hispanic, 30% European-American/White, 31% Other) between 7/1/2015-06/30/2017. Program staff completed a Child and Adolescent Needs and Strengths (CANS; Lyons, 2009) assessment within 60 days of enrollment. CANS items, scored on a 0-3 scale, were dummy-coded to indicate “actionable” scores which should be included in a treatment plan (Lyons, 2009). A Total Actionable Item (TAI) score, indicating overall need, was calculated by summing all actionable items on clients’ initial assessment. Random forests identified the top individual CANS items, and regression-based decision trees yielded combinations of CANS items, associated with higher overall need.

Results:

School-based youth had an average total need (TAI) of 15 points (SD: 7.08; Range: 1-43), while community-based youth had an average total need of 21 points (SD: 8.23; Range: 0-46). For school-based youth, the top actionable items significantly associated with higher need (all p < .001) ranged from Decision Making (36% of youth actionable; average total need: 19 points) to Interpersonal Skills (49% of youth actionable; average total need: 17 points). For community-based youth, these items ranged from Decision Making (55% of youth actionable; average total need: 25 points) to Trauma-Induced Hyper Arousal (32% of total youth actionable; average total need: 26 points).  

Combinations of items that were significantly associated with higher need (all p  < .001) differed between the school-based and community-based samples. School-based youth who were actionable on Recreational Needs, Coping Skills, and Living Situation had an average total need of 29 points, while youth who were actionable on two out of three of these items had a lower average total need (20-23 points), and youth who were not actionable on any of these items had the lowest average total need (11 points). Community-based youth who were actionable on Decision Making, Family Functioning, and Living Situation had an average total need of 29 points, and youth who were not actionable on any of these items had the lowest average total need (13 points).

Conclusions and Implications:

Recursive partitioning models are promising for organizations seeking to analyze large administrative datasets. Implications of these analyses for clinical decision making and organizational resource allocation in mental health service agencies for youth and families will be discussed.