Integrating APS Knowledge into an Algorithm to Identify Cases for MDT Review

Background and Purpose: Case review multidisciplinary teams (MDTs) bring together professionals from health, social service, and justice systems to address complex cases of abuse, neglect, self-neglect, and exploitation of older adults and younger adults with disability or vulnerability. MDTs are a promising approach in addressing complex cases of adult maltreatment, offering support to adult protective services (APS) workers, improving access to interventions such as prosecution and guardianship/conservatorship, and advancing person-centered priorities. However, MDTs nationwide face challenges identifying cases for review. This presentation reports on the development of an algorithm-driven process to systematically identify APS cases that appear appropriate for review by an MDT.

Methods: This project was undertaken in conjunction with San Francisco APS to develop cases for two of their MDTs: one addressing cases involving maltreatment by a perpetrator and one addressing high-risk self-neglect cases. To construct the algorithms, a retrospective analysis of four years of APS case data was augmented by insights received during six listening sessions held with 48 study partners (MDT members and APS staff). APS case data included intake, client, suspected abuser, and case process information. Listening sessions gauged perspectives on cases’ suitability for MDT review, with transcripts coded using constant comparative analysis. Guided by the listening session results, the algorithm was developed using best practices for machine learning, including logistic regression and CART (classification and regression tree), randomly selected training and test data (with a 3:1 split), and bootstrapping the train/test sample, to ensure the stability of the algorithm construction.

Results: Listening session themes related to case selection were Client, Case Details, and Service Utilization Patterns. Client included physical or cognitive impairment, high-risk of abuse or other harms, need for assistance from multiple service systems, and limited motivation to make changes. Case Details referred to aspects of the abuse such as concurrent self-neglect, client protectiveness of the suspected abuser, client denial of the abuse, and limited criminal conduct. Service Utilization Patterns were recurrence within APS, attempting multiple solutions, inability to provide services, and referral by medical professionals, fire departments, or paramedics. The listening sessions guided selection of three target variables to be used for algorithm development: (1) MDT presentation, (2) MDT presentation or potentially negative outcomes (e.g., multiple recurrent cases), and (3) MDT presentation, potentially negative outcomes, or hypothesized characteristics of good cases. Algorithm performance metrics (e.g., area under the curve [AUC]) were used to gauge the ability of the algorithm to correctly predict cases with a given outcome.

Conclusions and Implications: This mixed-method approach to algorithm construction provides an example of community-engaged research being applied to develop knowledge and guide the practice of service delivery for vulnerable older adults and other victims of maltreatment. It will be implemented in practice and tested as a decision support system to augment and improve APS workers’ clinical decisions on when to send cases for review by an MDT. If successful, the implementation of this algorithmic approach will improve APS clients’ access to the enhanced services of an MDT and may result in improved outcomes.