PROJECT SUMMARY/ABSTRACT. At least 75% of all drug overdose deaths in the U.S. are related to
opioids. Further, opioid-related mortality rates have increased alarmingly among Black and Latino populations.
People with opioid use disorder have high hospitalization rates, and those who inject opioids have additional
barriers to services and needs that lead them to receive the majority of their care in hospitals. Consequently,
the hospital is a crucial point of contact for people who use opioids (by non-injection or injection), including
Black and Latino individuals. Harm reduction services (HRS) in hospital settings (psychosocial/environmental
and pharmacological HRS) are relatively new, but they have the potential to substantially reduce opioid-related
mortality rates. However, HRS are rarely offered in hospitals, and when offered, may not be offered equitably.
Thus, inadequate and unequal distribution of HRS in hospital settings may be significant impediments to the
health, quality of life, and equitable prevention of overdose among those who use opioids. Addiction medicine
consult programs in hospitals are based on harm reduction theory, but their success in providing HRS
comprehensively and equitably is not well understood. The proposed study builds on such a program located
in public hospitals in New York City, called Consult for Addiction Treatment and Care in Hospitals (CATCH,
R01DA045669). CATCH examines rates of initiation of medication for opioid use disorder and post-discharge
services, but not HRS or inequities by route of use (non-initiation vs. initiation) and race/ethnicity. Yet
information on HRS can be found in the clinical notes section of the electronic health record (EHR), although
efficient methods of using EHR data are needed. The proposed study will use natural language processing
(NLP) and EHR data from the three largest sites enrolled in CATCH (N=982 patients, N˜ 3930 notes, N=9
providers/site.) This proposal has two objectives: Aim 1 will develop and validate a NLP system to identify
patterns of harm-reduction language and other relevant data, using a subset of clinical notes (N=300 notes).
Aim 2 will use the NLP system and entire collection of clinical notes to characterize providers’
recommendations of psychosocial/environmental and pharmacological-based HRS (as well as MOUD). First,
rates of MOUD and HRS recommendations by providers will be described. Then, racial/ethnic differences on
HRS and MOUD recommendations will be modeled using conditional logistic regression, followed by an
examination of route of drug use differences (non-injection, injection). To evaluate the secondary outcomes,
patients’ uptake of psychosocial/environmental and pharmacological HRS and MOUD, the study will describe
the uptake of HRS and MOUD and examine rates of uptake by race/ethnicity and route of drug use through
fixed-effects multinomial logit models. This F31 application aligns with NIDA's strategic plan. If funded,
statistical and research methods training will be a key component of this study, providing vital training in NLP,
harm reduction, and hospital-based interventions for overdose prevention.