Initiation of Medications for Alcohol Use Disorder among Hospitalized Veterans

Data Sources and Study Population

We analyzed electronic health records, bar-code medication administration data, and pharmacy fill records from the VA Corporate Data Warehouse for all VA hospitalizations for AUD between 2022 and 2023.

We identified AUD hospitalizations with a primary discharge diagnosis International Classification of Diseases-10 codes of F10 and all sub-codes. Veterans with an AUD hospitalization who were transferred between VA hospitals were identified by overlapping dates and merged into single episodes. We excluded hospitalizations resulting in death and discharge to hospice. We also excluded hospitalizations discharged to skilled nursing facilities and non-VA hospitals, where post-hospital medication initiation could not be reliably captured.

As our focus was initiation of MAUD, we excluded hospitalizations of Veterans who were previously prescribed MAUD based upon outpatient pharmacy fills or medication administrations (for injectable naltrexone). For this exclusion, we used a 180-day fixed pre-hospitalization look-back period, conservatively based on prior VA algorithms for identifying cross-sectional medication use. (16,17)

Outcomes: Initiation of MAUD

The primary outcome was initiation of MAUD, which included medication received during hospitalization (inpatient) or a prescription fill or injectable administration within 7 days of hospital discharge. Consistent with prior studies of discharge prescribing, we included MAUD filled or administered within 7 days of discharge as it was likely prescribed by the inpatient team and allows a grace period for patients who do not immediately fill medications immediately after discharge. (18,19) We used the 2021 VA/Department of Defense joint clinical practice guideline on management of substance use disorders to define recommended (i.e., oral naltrexone, injectable naltrexone, or oral topiramate [off-label]) and suggested (i.e., oral disulfiram or acamprosate) pharmacologic therapies for patients with AUD. (4) We did not include gabapentin, which is a third-line treatment, given its frequent use for other indications and weaker evidence base. We identified hospital MAUD exposure based on bar code medication administration data that captures real-time administration of medications to inpatients. Patients who received both oral and injectable naltrexone at discharge were grouped as receiving injectable naltrexone. As secondary outcomes, we examined two additional definitions of MAUD initiation: inpatient receipt of MAUD, and receipt of MAUD within 30 days after hospital discharge.

Patient Characteristics

Given existing disparities in evidence-based prescribing and AUD outcomes, (6,20–22) we examined initiation of MAUD for subgroups defined by age, sex, race, ethnicity, area deprivation index (ADI), and rurality. We categorized age in years as: 18-39, 40-64, and 65 or older. We used VA data on self-identified race and ethnicity to define mutually exclusive categories of non-Hispanic American Indian/Alaska Native, Asian, Black, Multi-racial, Native Hawaiian or other Pacific Islander, or White; Hispanic, any race; or missing race. The ADI is a validated composite measure of neighborhood disadvantage developed from 17 individual census block group-level measures from the American Community Survey. (23) Higher ADI scores represent more disadvantaged neighborhoods, and ADI was examined in quartiles. Rurality was defined by 2010 rural–urban commuting area codes.

We examined alcohol-related, clinical, and hospitalization covariates hypothesized to influence MAUD prescribing decisions. (12–15, 24) Alcohol-related covariates were obtained from the electronic health record in the year before admission including: prior AUD diagnosis, prior opioid use disorder diagnosis, pre-admission Alcohol Use Disorder Identification Test-Consumption (AUDIT-C) score as a marker of AUD severity (25), and remote prior MAUD use (i.e., any fill for MAUD in 181 to 365 days before admission but no refills). Clinical covariates included individual comorbidity indicators identified from the AHRQ Elixhauser comorbidity index (26), quintiles of a claim-based frailty index developed for VA patients (27), the VA Care Assessment Need (CAN) score that predicts 90-day risk of hospitalization or death (28), and indicators for types of healthcare utilization in the prior year (i.e., primary care visits, emergency department visits, and hospitalization admission). Hospitalization covariates included the following: primary discharge diagnosis (alcohol dependence, abuse, or use), length of stay, intensive care unit admission, discharging specialty (medicine, psychiatry, and surgery), receipt of a specialty addiction consult (categorized as a clinician [physician or advanced practice clinician] or non-clinician [social worker, nurse, peer support] consult), discharge disposition (regular, against medical advice, and to a VA domiciliary), and discharge month and year.

Hospital Characteristics

We examined several hospital-level characteristics hypothesized to influence MAUD prescribing: hospital volume (quartiles), hospital complexity which VA defines as high, medium, or low based on combination of patient risk, complexity of clinical programs, and presence of education and research focus, (29) and whether the hospital was in the top quartile of proportion of a) rural patients or b) racial and ethnic minority (non-White) patients. We also examined hospital-level substance-use services including volume of substance use disorder hospitalizations (quartiles), availability of inpatient substance abuse treatment programs, and availability of residential treatment programs (domiciliary).

Statistical Analyses

As each AUD hospitalization is an opportunity to initiate treatment, we considered separate hospitalizations as the primary unit of analysis, with Veterans potentially contributing multiple observations. We calculated cohort descriptive characteristics and calculated proportions initiating MAUD during hospitalization, within 7 days of discharge, and within 30 days of discharge. We next calculated unadjusted hospital rates of MAUD initiation. We then estimated two sets of regression models to address three key questions: 1) Within hospitals, which demographics and clinical characteristics are associated with MAUD initiation, 2) To what extent does variation in MAUD initiation between hospitals reflect hospital-level or hospitalization-level variation, 3) which hospital-level characteristics are associated with MAUD initiation.

To identify characteristics associated with initiation of MAUD within hospitals, we constructed a multivariate logistic regression model which included fixed effects for the demographic, clinical, and hospital characteristics previously described as well as for each unique hospital. Clustered robust standard errors were used to allow for intragroup correlation of hospitalizations within patients (e.g. when patients contributed multiple observations). Post-estimation marginal predicted probabilities and marginal differences are presented. Second, to quantify the relative contribution of hospitalization-level factors versus hospital-level factors on MAUD initiation, we performed a variance decomposition and calculated McFadden’s pseudo-R² for three models: a model using hospital fixed effects only, a model including hospitalization-level covariates only, and the fully adjusted model. The relative proportion of explained variation attributable to patient factors was calculated as the pseudo-R² of the patient model divided by the pseudo-R² of the full model. Third, to identify hospital-level characteristics associated with MAUD initiation, we constructed mixed effect logistic regression which included three nested levels: unique hospitalizations, patient-level (e.g. multiple separate hospitalizations for the same patient), and hospitals (e.g. hospitalizations within the same hospital), hospitalization characteristics, and hospital-level characteristics. We constructed sequential mixed effect models, first including random effects only, second adding hospitalization-level fixed effects, and third adding hospital-level characteristics as fixed effects. We then plotted hospital-level predicted probabilities for MAUD initiation for each mixed effect models and estimated residual variation across hospitals.

Analyses were conducted using Stata v18, with statistical significance assessed using 95% CIs with an α of 0.05.

Role of the Funding Source

The funder had no role in the design or conduct of the study; collection, management, analysis, or interpretation of the data; preparation, review, or approval of the manuscript; or the decision to submit the manuscript for publication.

Initiation of MAUD

MAUD was initiated inpatient or within 7 days of discharge in 8,932 hospitalizations (30.8%) (Figure 1A). For the majority, MAUD was initiated inpatient (69.6%; 6221/8932), but for nearly one-third (2711/8932), MAUD was first received within 7 days of discharge. Of those who received MAUD inpatient, 97.7% subsequently received outpatient MAUD within 30 days of discharge. Over half of patients initiating MAUD inpatient or within 7 days of discharge received oral naltrexone (57.9%), but 13.9% received injectable naltrexone. Less commonly prescribed drugs included acamprosate (16.5%), topiramate (3.8%) and disulfiram (3.6%) (Figure 1B). A few patients (4.2%) were discharged on multiple MAUD. From 7 and 30 days following discharge, an additional 1,595 MAUD initiations occurred (7.9% of 20,109 untreated at discharge).

Demographic and Clinical Factors Associated with MAUD Initiation

MAUD initiation as an inpatient or within 7 days of discharge was less common in older Veterans, more common among female Veterans, and less common in American Indian and Alaskan Native Veterans compared with White Veterans (Table 2). There were no significant differences in MAUD initiation by other racial and ethnic groups, rurality, or quartile of ADI.

Regarding clinical characteristics, MAUD initiation was more common for patients with a prior AUD diagnosis and with remote prior use of MAUD (Table 3). Patients reporting low or no alcohol use on pre-admission AUDIT-C were less likely to initiate MAUD, but among those with higher AUDIT-C scores there were no differences in MAUD initiation. MAUD initiation was less common among patients with greater medical complexity, including those with opioid use disorder, frailty, and receiving ICU care.

Patients on psychiatry services were more likely to initiate MAUD than those on medicine services, and patients receiving an addiction specialty consult were more likely to initiate MAUD. Patients discharged against medical advice were less likely to initiate MAUD. MAUD initiation rates increased moderately in 2023 compared with 2022.

Hospital Variation in MAUD Initiation

Initiation of MAUD varied widely across hospitals, with the median (IQR) unadjusted hospital-level initiation rate of 29.9% (IQR 22.6% to 36.3%); the minimum initiation rate was 5.2% and the maximum was 66.7% (Figure 2). Of the total variation explained by the full model, patient case-mix accounted for 67% and unmeasured, hospital-specific factors accounted for the remaining 33% of the explained variation in MAUD initiation.

Hospital Factors Associated with MAUD Initiation

In mixed effect models that included hospital characteristics, similar associations between patient and hospitalization characteristics and MAUD initiation were observed (Supplemental Table 2). However, characteristics of hospitals, including size, complexity, presence of an inpatient substance abuse program, and presence of a domiciliary program, were not associated with initiation of MAUD. In the 3 sequential mixed effect models including adding hospitalization characteristics to the model with hospital random effects reduced the residual variation from 0.57 (95% CI, 0.40 to 0.74) to 0.44 (95% CI, 0.31 to 0.58) (Supplemental Table 3, Supplemental Figure 2). However, the fully adjusted model adding hospital characteristics did not appreciably further reduce residual variation (0.40 (95% CI 0.27 to 0.52) further indicating the presence of unmeasured, hospital-level factors contributing to the observed hospital variation.