Methods: The analytic approach for this NAMRS recurrence analysis comprised several major steps. First, we appended raw single-year NAMRS datafiles to construct multi-year datafiles, and performed basic data processing and cleaning procedures. Second, we applied state-level, client-level, and perpetrator-level inclusion/exclusion criteria, then conducted an initial assessment of data quality. Third, we grouped overlapping investigations into episodes and standardized information from across the investigations that comprised each episode to create episode-level variables. Fourth, we conducted final quality checks using a probabilistic matching technique to support the validity of linking both clients and perpetrators over time. We conducted univariate, bivariate, and multivariate analyses of recurrence. The main multivariate analyses used Bayesian logistic regression to predict 12-month recurrence in a multi-state model and in separate single-state models. The multi-state model used multilevel methods to properly account for clustering in the data (i.e., clients within states).
The initial sample included data on 1,211,360 APS episodes that closed between Fiscal Year (FY) 2016 and FY 2019 from across 19 states. These episodes contained information on 946,477 unique APS clients and 100,119 unique perpetrators.
Results: About one in five of all clients in this analysis experienced at least one episode of recurrence. Nearly all clients (99%) had 5 or fewer episodes over the four-year period. Results from the multivariate models indicate that no client demographics consistently predicted recurrence, except for gender. The presence of any substantiated maltreatment and most maltreatment types increased the chance of recurrence. In particular, self-neglect was associated with the greatest risk of recurrence across maltreatment types. Additionally, clients with longer episodes were less likely to experience recurrence. One of the most powerful and interesting findings from the multivariate models was the association between case closure reason and recurrence. Clients who declined APS were more likely to recur, perhaps due to unaddressed safety issues. Clients whose APS investigation and protective services case completed were also more likely to recur, perhaps due to complex needs of the client requiring longer-term help.
Conclusions/Implications: The findings identify key predictors of recurrence, which may be helpful for APS programs to better recognize and serve clients at high risk of returning to APS after their cases close. The findings also highlight the ambiguity surrounding the concept of recurrence in the APS field and the need for further research to determine the circumstances in which “recurrence”, defined as returning for services, may be considered a negative or a positive outcome for APS systems and clients.