Page 112 - AIH-2-1

Artificial Intelligence in Health ML approach for client and program effects in OUD retention

1. Introduction In this study, we applied heterogeneous treatment effect
(HTE) estimation methods to understand which workforce
The opioid epidemic continues to adversely impact the diversity characteristics facilitate positive retention effects.
public health system of the United States. The Centers HTE estimation is a machine learning method which was
for Disease Control and Prevention estimates that there originally designed to study variations in the effects of
were over 81,000 opioid-related overdose deaths in 2023. clinical interventions and has been generalized to other
1
Increased opioid use disorder (OUD) treatment retention applications such as public policy and marketing. 22-25
can improve treatment outcomes, including reduced rates Heterogeneous treatment effect (HTE) estimation
of mortality and of relapse. Concurrently, retention rates methods, including causal forests, have been effectively
2-5
in OUD treatment are highly variable between programs applied in fields such as personalized medicine, public
and demographic groups, with 6-month retention rates policy, and marketing. 26,27 In personalized medicine, HTE
commonly dropping below 50% for some groups. Several helps tailor treatments to individual patients, improving
4,6
studies measuring retention in OUD programs have found outcomes by accounting for diverse responses. In public
lower retention rates among minoritized individuals policy, it identifies how different populations are impacted
who identify as Black/African American and as Latino/ by interventions, guiding more equitable policymaking. In
Hispanic (Black and Latinos, hereafter). Other studies marketing, HTE enables businesses to optimize strategies
7-9
have identified subgroup differences between Black, Latino, by understanding how various customer segments respond
and White clients, including variations in predictors of to different campaigns. The strength of HTE methods lies
retention and the treatment outcomes associated with in their ability to handle complex interactions and high-
retention. 10,11 It is therefore important to consider unique dimensional data, offering deeper insights than traditional
differences, particularly of minoritized patients like Black regression models.
clients, when exploring strategies to boost retention rates
in OUD programs. In this work, we adopted a state-of-the-art HTE
estimation method called “causal forest,” to examine the
Past research on the effect of culturally responsive
practices on the retention of Black OUD clients has identified heterogeneous impact of workforce diversity on OUD
treatment retention.
Causal forest is a machine learning
28,29
promising culturally responsive organizational factors, method that extends the random forest framework to
including offering bilingual language services; developing estimate the varying effects of a treatment across different
specific policies and procedures designed to serve minority
clients; and having managers who believe in the importance of subgroups within a population. This method involves
cultural sensitivity. 12-16 Workforce diversity, defined as having constructing an ensemble of decision trees, where each
a higher percentages of Black staff members, is thought to tree is specifically designed to identify splits in the data
that reveal differences in treatment effects between
improve Black OUD clients’ treatment outcomes by fostering subpopulations. To ensure accurate and unbiased estimates,
a culturally responsive treatment environment. 13,17-20 causal forest uses a technique known as “honesty,” where
However, previous studies on the impacts of workforce the data used to determine the optimal splits in the trees
diversity on OUD client retention have looked for simple
associations and have included only a few basic modifying is separate from the data used to estimate the treatment
variables, leading to variable retention outcomes. 16,21 effects. This approach allows for a detailed exploration of
how the impact of a treatment may differ across various
The heterogeneous nature of these results indicates segments of the data.
that workforce diversity may have differential impacts
There are several advantages of this method over
on retention rates in OUD programs with different traditional regression models. First, due to potential high
organizational characteristics. We build on prior studies collinearity and a high false discovery rate, only a limited
that have suggested that workforce diversity in the
absence of other factors, such as high levels of training number of interactions can be included in traditional
and education among staff members, may be insufficient regression models. Second, causal forest provides variance
to improve treatment outcomes. 17,18 Unpacking the for individually-estimated treatment effects, that is, one
heterogeneity in associations between workforce diversity can calculate the asymptotic p-values for the statistical
and treatment retention can help healthcare policymakers, significance of treatment effects for each observation.
leaders of OUD treatment programs, and researchers to By examining HTE, we can untangle the various factors
understand which programs would benefit most from the that may influence how workforce diversity impacts OUD
expansion of workforce diversity, and importantly, the client retention. The benefit of this study to the field of
additional conditions necessary to optimize the benefits of healthcare, and disparities within this field in particular,
workforce diversity. includes informing healthcare policies, and practices, on

Volume 2 Issue 1 (2025) 106 doi: 10.36922/aih.3750