Behavioral Health Needs of Older Adults Living in Poverty: Machine Learning-Based Predictive Models

Background and Purpose: Growing evidence revealed behavioral health (BH) service needs for older adults (WHO, 2020; Webb, 2020; Novotney, 2019). However, the BH service needs and use of older adults living under 200% of poverty (42%; Cubanski et al., 2021) remained largely understudied. To develop contextually sensitive and effective services for older adults in poverty, this study aimed to identify the characteristics and patterns for older adults’ BH service needs, comparing to those of middle-aged adults.

Methods: This study used one Midwestern state’s 2019 publicly funded BH administrative data. A sample of adults aged 50 and older were identified (total n=23,370; middle-aged adults (aged 50 to 64): n= 18,529, older adults (aged 65 and older): n=4,841). For each individual, the Adult Needs and Strengths Assessment (ANSA; Lyons, 2009) had been completed to facilitate care planning and to monitor outcomes. Machine learning decision tree analysis with a chi-square automatic interaction detection (CHAID) model was used to detect the relationship between BH and age group. Variables included (a) life functioning (e.g., medical/physical, independent living, sleep, social residential stability, and legal involvement); (b) BH needs (e.g., psychosis, depression, anxiety, adjustment to trauma, and substance use); and (c) risk behaviors (e.g., suicide risk, self-harm, and exploitation). Variables were coded as binary (0=no action required; 1= intervention required). Demographic information (i.e., age, gender, and race/ethnicity) was included.

Results: Substance use (SU) and medical/physical (MP) problems were the most important predictors that distinguished age groups, followed by needing assistance to live independently, adjustment to trauma, residential stability, criminal justice involvement, gender, and BH issues, (e.g., impulse control, anxiety, depression, and anger control). For older African Americans, the SU rate (16.04%) was significantly higher than for older non-African Americans (7.93%). Additionally, for non-African Americans with SU, anxiety, trauma adjustment, and impulse control issues, significantly distinguished age groups. Finally, among individuals without SU, gender was an essential factor that differentiated functional needs by age group. Many older women had medical/physical conditions that required assistance (27.98%). In the cross tabulation, middle-aged men experienced high rates of depression (82.88%) and anger control issues (88.15%). The overall model accuracy was acceptable (AUC = .73).

Conclusion and Implications: Different characteristics and patterns of BH and functional needs emerged between middle-aged and adults over 64. Not surprisingly, medical/physical and independent living challenges issues were more common among older adults in multiple contexts.Notably, older African Americas experienced higher rates of SU problems. SU was often associated with trauma adjustment, trauma adjustment, anxiety, impulse control issues, and legal involvement. Middle-aged White adults with SU showed higher rates of anxiety and impulse control than their counterparts, while criminal justice involvement was higher than their counterpart among middle-aged White adults with SU. Along with the importance of BH services for older people, this study demonstrates the need for age and racially/culturally sensitive and BH treatment services and supports.