Abstract
Objective
To evaluate the association between use of newer glucagon-like peptide-1 receptor agonists (GLP-1 RAs; semaglutide, tirzepatide) and alcohol-related hospitalisations among adults with alcohol use disorder (AUD) and type 2 diabetes (T2D) or obesity.
Retrospective cohort study using target trial emulation.
Setting
Electronic health record data from a collective of US healthcare systems.
Participants
Adults with AUD and T2D or obesity who started a newer GLP-1 RA (semaglutide or tirzepatide) or a relevant active comparator between 1 January 2018 and 31 December 2024.
Interventions
Initiation of a newer GLP-1 RA compared with an active comparator across four target trials: (1) anti-diabetic medication (ADM) trial, (2) anti-obesity medication (AOM) trial, (3) medications for alcohol use disorder with T2D (MAUD-T2D) trial, and (4) medications for alcohol use disorder with obesity (MAUD-obesity) trial.
Main outcome measures
Time to first alcohol-related emergency department visits or hospitalisation within 1 year of treatment initiation. Non-alcohol-related hospitalisation was assessed as a negative control outcome. Propensity score based methods (weighting and matching) were used to control confounding and Cox proportional hazards models were used to estimate treatment effects.
Results
A total of 40 703 adults met study criteria, including 18 676 in the ADM trial, 9391 in the AOM trial, 8942 in the MAUD-T2D trial and 11 198 in the MAUD-obesity trial. Initiation of a newer GLP-1 RA was associated with a lower hazard of alcohol-related hospitalisation in the ADM trial (HR 0.74, 95% CI 0.62 to 0.89 vs sulfonylureas; HR 0.78, 95% CI 0.65 to 0.92 vs other ADMs), the AOM trial (HR 0.68, 95% CI 0.54 to 0.85 vs other AOMs), the MAUD-T2D trial (HR 0.37, 95% CI 0.29 to 0.46) and the MAUD-obesity trial (HR 0.35, 95% CI 0.26 to 0.47).
Conclusions
Among adults with AUD and T2D or obesity, initiation of newer GLP-1 RAs was associated with a lower observed risk of alcohol-related hospitalisation.
Keywords: Research Design; Substance misuse; Obesity; Diabetes Mellitus, Type 2
STRENGTHS AND LIMITATIONS OF THIS STUDY.
Conducted four separate target trials in clinically distinct target populations by employing a target trial emulation framework with a new-user active comparator design to reduce bias and confounding.
Use of large, multi-system US electronic health record dataset to enhance generalisability.
Use of propensity score based methods and negative control outcomes analysis.
Potential for residual confounding due to unmeasured socioeconomic and lifestyle factors.
Introduction
Excessive alcohol use is a leading cause of preventable mortality in the US, with more than 178 000 attributable deaths annually and an estimated economic burden exceeding $200 billion.1–4 Alcohol use disorder (AUD) occurs in 10% of US adults,5 yet only 2% of adults with AUD receive medication-assisted treatment.5–7 While efficacious,8 current US Food and Drug Administration (FDA)-approved medications for AUD (MAUD)—acamprosate, disulfiram and naltrexone—have tolerability and adherence challenges that limit their real-world effectiveness.9–11
Glucagon-like peptide-1 receptor agonists (GLP-1 RAs) are used primarily for the treatment of type 2 diabetes mellitus (T2D) and obesity. A growing body of observational evidence suggests that GLP-1 RAs may also reduce alcohol consumption and related adverse outcomes.12 13 However, evidence from randomised clinical trials remains limited and mixed. In a recent randomised trial of semaglutide, reductions were observed in drinks per drinking day and alcohol craving, but there was no significant effect on overall alcohol consumption or number of drinking days.14 These findings may reflect differences in study design and populations, including the inclusion of non-treatment-seeking individuals. Observational studies of varying quality have also reported reductions in AUD-related clinical outcomes, including alcohol-related hospitalisations and recurrent AUD diagnoses, although these studies may lack generalisability.12 13 15 16
Important gaps remain in understanding the potential impact of GLP-1 RAs on AUD. These include a lack of high-quality observational studies using rigorous and consistently defined populations. In addition, given its recent approval, few studies have included tirzepatide, a dual glucose-dependent insulinotropic polypeptide (GIP) and GLP-1 RA. Tirzepatide has demonstrated greater effectiveness on glycaemic and weight-related outcomes compared with semaglutide.17–19 While these metabolic effects are not directly related to alcohol use, they may reflect broader effects on reward pathways and appetite regulation that could also influence alcohol consumption and related outcomes. Accordingly, we employed a target trial emulation framework to estimate the effect of newer GLP-1 RAs (semaglutide and tirzepatide) on alcohol-related hospitalisations for adults with AUD, using four target trials involving clinically distinct populations.
Methods
Data
This study used a subset of Truveta Data.20 Truveta provides access to daily updated and linked electronic health record (EHR) data from a collective of 30 US healthcare systems, representing care delivered to more than 120 million patients across geographically and demographically diverse populations. The dataset reflects provider-based clinical care rather than a closed or fully enumerated population (eg, a health maintenance organisation). Individuals are included based on healthcare utilisation (ie, documented encounters), irrespective of insurance status; thus, patients with any form of coverage, as well as those without insurance (eg, self-pay), are represented. Data are captured across inpatient, outpatient and emergency settings, reflecting care delivered across the continuum, including both primary and secondary care depending on documentation within participating systems.
The Truveta database includes data related to demographics, encounters, diagnoses, vital signs (eg, weight, body mass index, blood pressure), medication requests (prescriptions) and laboratory tests and results (eg, haemoglobin A1c, blood alcohol concentration). In addition to EHR data, medication dispensing data (via e-prescribing) include fills for prescriptions written both within and outside Truveta constituent healthcare systems, resulting in greater observability of patients’ medication history. Medication dispense histories are updated at the time of the encounter and include fill dates, NDC or RxNorm codes, quantity dispensed and days of medication supplied.
Truveta Data are normalised into a common data model through syntactic and semantic normalisation.20 Truveta Data are then de-identified by expert determination under the Health Insurance Portability and Accountability Act Privacy Rule. Once de-identified, data are available for analysis in R or Python using Truveta Studio. Data for this study were accessed on 1 April 2025. As this study used only de-identified data, it did not require institutional review board approval under US regulations. Patients and the public were not involved in this study.
Study design
A target trial emulation framework was used to compare on-treatment alcohol-related hospitalisations for adults with AUD who newly initiated a GLP-1 RA (semaglutide or tirzepatide) or an active comparator medication. Newer GLP-1 RAs were compared with alternatives using four separate target trials (table 1), defined by a disease cohort (T2D vs obesity without T2D) and a treatment-related reason (related to T2D/obesity vs AUD). This yielded four trials: (1) anti-diabetic medication (ADM) trial, (2) anti-obesity medication (AOM) trial, (3) medications for alcohol use disorder with T2D (MAUD-T2D) trial, (4) medications for alcohol use disorder with T2D obesity (MAUD-obesity) trial. The four target trials reflect distinct clinical contexts for initiating GLP-1 RAs, enabling complementary inference about their impact on AUD. ADM (T2D) and AOM (obesity) trials include patients starting metabolic treatment and not actively seeking AUD care. They estimate associations between GLP-1 use and alcohol-related outcomes in routine care, assessing effects independent of AUD treatment-seeking. Because T2D and obesity are the primary FDA-approved indications, these trials mirror real-world prescribing.
Table 1.
Target trials and arms. Target trials are defined by the disease cohort (T2D or obesity without T2D) and (presumed) reason for seeking treatment (non-AUD or AUD)
| Component | ADM trial | AOM trial | MAUD-T2D trial | MAUD-obesity trial |
| Eligibility criteria |
|
|
Same as ADM trial, except:
|
Same as AOM trial, except:
|
| Treatment strategies |
|
|
|
Same as MAUD-T2D trial |
| Index date and assignment procedure | Index date is first trial medication dispensed; patients assigned to baseline-compatible strategy with confounding adjustment to emulate randomisation | Same as ADM trial | Same as ADM trial | Same as ADM trial |
| Follow-up | From index date until earliest of: discontinuation (60-day gap), switching to GLP-1 RA, last encounter before 1 April 2025, or 365 days | Same as ADM trial | Same as ADM trial | Same as ADM trial |
| Outcomes | Same as AOM trial plus exploratory: AST/ALT at 9 months |
|
Same as AOM trial | Same as AOM trial |
| Estimands | On-treatment ATE | On-treatment ATE | On-treatment ATT (among MAUD treated) | On-treatment ATT (among MAUD treated) |
| Data analysis plan | Multinomial propensity model; stabilised IPTW+IPCW; weighted Kaplan–Meier; Cox models with robust SEs | Same as ADM trial | Logistic propensity model; 1:1 nearest-neighbour matching; IPCW; weighted Kaplan–Meier; Cox models with robust SEs | Same as MAUD-T2D trial |
ADM, anti-diabetic medication; ALT, alanine aminotransferase; AOM, anti-obesity medication; AST, aspartate aminotransferase; ATE, average treatment effect; ATT, average treatment effect among the treated; AUD, alcohol use disorder; BMI, body mass index; DPP4, dipeptidyl peptidase four inhibitor; ED, emergency department; GLP-1 RA, glucagon-like peptide-1 receptor agonist; HbA1c, haemoglobin A1c; ICD-10, international classification of diseases, 10th revision; IPCW, inverse probability of censoring weights; IPTW, inverse probability of treatment weights; SGLT2i, sodium-glucose cotransporter-2 inhibitor; T2D, type 2 diabetes.
In contrast, MAUD trials include patients with markers of more severe AUD initiating AUD treatment, approximating a treatment-seeking population. They estimate associations between GLP-1 RAs and alcohol-related outcomes relative to approved AUD medications. Collectively, these designs evaluate GLP-1 RAs across clinically relevant populations with different treatment intent, baseline risk, health status and AUD severity.
Each trial included a clinically distinct population with expected differences in underlying health status, AUD severity and treatment-seeking context. This retrospective observational cohort study follows the STROBE reporting guidelines.21
Study population
Each trial included a new-user cohort of adults with an AUD diagnosis and a cohort-specific condition of interest (T2D or obesity, respectively) between 1 January 2018 and 31 December 2024. AUD was defined by the presence of an AUD-related diagnosis in the previous 2 years. Initiation of a trial-specific medication, defined by pharmacy dispensing, served as the study index event. New use was defined by a previous 2-year negative history of dispensing and administration of trial-specific medications.
T2D was defined by the presence of T2D diagnostic codes within the previous 2 years. For inclusion in the T2D cohorts, a baseline haemoglobin A1c (HbA1c) was also required (up to 1 year before the index event), though no restrictions were made on the HbA1c value. Obesity without T2D was defined by a body mass index (BMI) ≥30 kg/m2 at baseline, using the most recent BMI in the 12 months prior to the index date. Patients with T2D were excluded from the obesity cohort. To improve observability of historical and follow-up information for this study, the population was restricted to patients with at least two outpatient office visits in the previous 2 years. All code lists are provided in online supplemental table S3.
bmjopen-16-7-s001.pdf (1.4MB, pdf)
For the MAUD trials, additional restrictions were applied to emulate patients likely to seek treatment for AUD. History of AUD was defined more narrowly, excluding broader and less severe alcohol-related diagnostic codes (international classification of diseases, 10th revision (ICD-10) F10). In addition, history of AUD was required within the year prior to the index date, rather than within 2 years as used in other trials. As such, the MAUD cohorts represent more restricted subsets of the broader T2D and obesity populations defined for the ADM and AOM trials.
Comorbidities, previous medication use and utilisation were assessed using a 2-year lookback window. Study design timelines and patient flow diagrams are depicted in online supplemental figure S1–S5.
Outcomes
Patients were followed from the index date (treatment initiation) for up to 1 year to identify alcohol-related emergency department (ED) visits or hospitalisations, defined as emergency department or inpatient encounters with either alcohol-related diagnosis (including AUD, alcohol withdrawal and/or other diagnoses related to acute alcohol use; codes in online supplemental table S3) or testing for blood alcohol, ethyl glucuronide or ethyl sulfate levels. Among those with AUD, alcohol-related testing in acute care settings was presumed to be suggestive of the potential involvement of alcohol in their visit. Test values themselves were not considered because values are highly sensitive to time elapsed relative to alcohol consumption.22 Unlike claims data, diagnosis positions are not consistently captured in the EHR, particularly for outpatient encounters, including ED visits. As such, some visits captured as outcomes in this study likely represent visits with alcohol-related reasons rather than visits specifically for alcohol-related reasons.
To test specificity to alcohol-related hospitalisations, non-alcohol-related hospitalisations were also compared as a negative control outcome. Non-alcohol-related hospitalisations were defined as all ED and inpatient encounters other than those meeting the above criteria for an alcohol-related hospitalisation. We would expect no association between GLP-1 RAs and the negative control outcome; any observed association may indicate residual confounding. Additional details on negative control outcome selection are provided in the online supplement (page 4).
In addition to hospitalisation outcomes, changes in hepatic biomarkers (aspartate aminotransferase (AST), alanine aminotransferase (ALT)) were explored at 9 months on treatment for patients with available values in the ADM trial. AST and ALT are indirect markers of excessive alcohol intake and can be elevated for up to 2–3 weeks after consumption22; however, they are notably imperfect markers of alcohol consumption, subject to both low sensitivity and specificity for heavy drinking.22 23 Additional methods and results are provided in the online supplement (pages 5–6 and 15–16).
Patients were censored at medication discontinuation, initiation of a GLP-1 RA (or a newer GLP-1 RA for those starting on an older GLP-1 RA), the last encounter before 1 April 2025 or 1 year from the index event, whichever occurred first. Medication discontinuation was defined as 60 days without medication on hand, based on fill dates and days’ supply per fill.
Statistical analysis
Treatment selection
Population balancing methods were used to address non-random treatment selection. Different balancing approaches were used for the ADM and AOM trials, compared with the MAUD trials, due to differences in the target population and estimand.
For the ADM and AOM trials, the estimand of interest was the on-treatment average treatment effect (ATE), so that estimates would generalise to the full trial population. The propensity to initiate a newer GLP-1 RA, relative to other trial medications, was estimated using multinomial regression, inclusive of a variety of factors plausibly related to treatment selection, including demographics, clinical factors, comorbidities and utilisation (complete list provided in online supplement, page 5). We calculated stabilised inverse probability of treatment weights (IPTW), truncated at the 99th percentile to reduce the impact of extreme weights.24 25
For MAUD trials, the target population included patients actively seeking treatment for AUD. Therefore, the estimand of interest was the on-treatment, average treatment effect among those treated (ATT) with MAUD, with the expectation that effects generalise to the cohort of patients who would otherwise receive treatment with approved MAUD. Propensity scores were estimated as the likelihood of initiating MAUD, relative to newer GLP-1 RAs, using logistic regression. Additional covariates were included to adjust for differences in AUD severity and recency (complete list in online supplement, page 5). Because weighting approaches yielded poor balance between groups (likely due to the requirement that all patients receive some weight), 1:1 nearest neighbour propensity score matching was applied (with a calliper of 0.05), pairing a patient treated with MAUD to a similar patient treated with a newer GLP-1 RA.26 27 Unmatched patients were not included in the analysis.
Informative censoring
Inverse probability of censoring weights (IPCW) were applied to account for informative censoring, where patients remaining on treatment differed from those who were censored.28 The probability of artificial censoring (due to medication discontinuation, switching or loss to follow-up (last encounter)) before 365 days was first estimated using logistic regression. The model considered demographic, clinical and utilisation factors, as well as the exposure group and the index year. Stabilised and truncated IPCW were calculated as the inverse probability of artificial censoring.
For ADM and AOM trials, combined weights were calculated as the product of IPCW and IPTW. For MAUD trials, IPCW was applied to the propensity score matched population.
Outcomes models
The probability of on-treatment ED visit or hospitalisation by 365 days was extracted from weighted Kaplan–Meier curves. Unweighted curves were also plotted for comparison (online supplemental figure S7). Cox proportional hazards models with robust standard errors were used to estimate the on-treatment hazard of alcohol-related hospitalisation between treatment groups. The same approach was used to estimate the hazard of non-alcohol-related hospitalisations between groups. For all Cox models, we assessed the proportional hazards assumption using Schoenfeld residuals and visual inspection of residual plots. These analyses did not indicate meaningful departures from the primary analysis. E-values were calculated to estimate the magnitude of unmeasured confounding required to negate the observed treatment effects.29
As a sensitivity analysis, we repeated the primary analysis in each trial using a more specific outcome definition restricted to alcohol-related ED visits and hospitalisations identified by diagnostic codes only (excluding events defined solely by laboratory testing). This analysis was conducted to evaluate the robustness of findings to potential misclassification of outcomes based on laboratory testing alone.
Results
In total, 40 703 patients met the criteria for at least one trial. This included 18 676 in the ADM trial (newer GLP-1 RA: 4051 (22%); older GLP-1 RA: 2205 (12%); sulfonylurea: 5277 (28%); other ADM: 7143 (38%)), 9391 in the AOM trial (newer GLP-1 RA: 3548 (38%); older GLP-1 RA: 620 (7%); other AOM: 5223 (56%)), 8942 in the MAUD-T2D trial (newer GLP-1 RA: 5599 (63%); MAUD: 3343 (37%)) and 11 198 in the MAUD-obesity trial (newer GLP-1 RA: 2023 (18%); MAUD: 9175 (82%)) (table 2, online supplemental table S1).
Table 2.
Characteristics of patients at baseline before balancing.
| ADM trial (n=18 676) | AOM trial (n=9391) | ||||||
| Characteristic | Newer GLP-1 RA (n=4051) |
Older GLP-1 RA (n=2205) |
Sulfonylurea (n=5277) | Other ADM (n=7143) | Newer GLP-1 RA (n=3548) |
Older GLP-1 RA (n=620) |
Other AOM (n=5223) |
| Age | 57.9 (11.5) | 56.8 (11.6) | 60.3 (12.1) | 61.2 (11.9) | 53.0 (12.7) | 49.5 (12.5) | 47.4 (13.1) |
| Female | 1483 (37%) | 803 (36%) | 1196 (23%) | 1595 (22%) | 1877 (53%) | 387 (62%) | 3476 (67%) |
| Race | |||||||
| Black | 396 (10%) | 306 (14%) | 520 (10%) | 955 (13%) | 268 (8%) | 73 (12%) | 461 (9%) |
| White | 3093 (76%) | 1497 (68%) | 3548 (67%) | 4733 (66%) | 2812 (79%) | 465 (75%) | 3909 (75%) |
| Other or unknown | 562 (14%) | 402 (18%) | 1209 (23%) | 1455 (20%) | 468 (13%) | 82 (13%) | 853 (16%) |
| Ethnicity | |||||||
| Hispanic or Latino | 466 (12%) | 237 (11%) | 651 (12%) | 763 (11%) | 277 (8%) | 57 (9%) | 522 (10%) |
| Not Hispanic or Latino | 3454 (85%) | 1870 (85%) | 4263 (81%) | 5929 (83%) | 3129 (88%) | 539 (87%) | 4528 (87%) |
| Unknown | 131 (3%) | 98 (4%) | 363 (7%) | 451 (6%) | 142 (4%) | 24 (4%) | 173 (3%) |
| Any college on record | 1601 (40%) | 742 (34%) | 1242 (24%) | 1778 (25%) | 1934 (55%) | 333 (54%) | 3015 (58%) |
| Baseline health | |||||||
| Baseline HbA1c | 7.4 (1.8) | 8.2 (2.1) | 8.4 (2.1) | 7.9 (2.0) | – | – | – |
| Baseline BMI | 36.7 (7.4) | 35.6 (8.1) | 31.6 (7.1) | 31.8 (7.2) | 38.3 (6.7) | 39.7 (7.3) | 37.1 (6.3) |
| Unknown | 579 | 373 | 811 | 1170 | – | – | – |
| Baseline ALT | 36.1 (26.1) | 35.5 (24.7) | 36.7 (27.4) | 33.2 (24.1) | 35.4 (25.5) | 33.2 (22.5) | 33.4 (25.3) |
| Unknown | 333 | 216 | 475 | 608 | 710 | 160 | 1206 |
| Baseline AST | 31.0 (24.0) | 30.2 (22.8) | 32.8 (26.1) | 30.8 (23.8) | 29.8 (22.3) | 29.2 (23.3) | 29.9 (24.6) |
| Unknown | 310 | 220 | 485 | 597 | 701 | 157 | 1197 |
| Comorbidities | |||||||
| Elixhauser comorbidity score | 9.5 (10.4) | 10.6 (10.9) | 10.9 (11.3) | 13.3 (11.8) | 5.2 (10.1) | 4.7 (10.8) | 3.4 (9.9) |
| Cirrhosis | 384 (9%) | 240 (11%) | 669 (13%) | 915 (13%) | 194 (5%) | 46 (7%) | 240 (5%) |
| Chronic kidney disease | 1008 (25%) | 719 (33%) | 1634 (31%) | 2827 (40%) | 391 (11%) | 78 (13%) | 517 (10%) |
| Chronic liver disease | 557 (14%) | 325 (15%) | 936 (18%) | 1242 (17%) | 298 (8%) | 69 (11%) | 419 (8%) |
| Cardiovascular disease | 1278 (32%) | 761 (35%) | 1761 (33%) | 2980 (42%) | 375 (11%) | 77 (12%) | 423 (8%) |
| Obesity | 3658 (90%) | 1922 (87%) | 4110 (78%) | 5755 (81%) | 3548 (100%) | 620 (100%) | 5223 (100%) |
| Obstructive sleep apnea | 1827 (45%) | 869 (39%) | 1332 (25%) | 2334 (33%) | 1348 (38%) | 204 (33%) | 1376 (26%) |
| Hypertension | 3505 (87%) | 1914 (87%) | 4604 (87%) | 6413 (90%) | 2383 (67%) | 387 (62%) | 2764 (53%) |
| Hyperlipidemia | 3373 (83%) | 1822 (83%) | 4213 (80%) | 5983 (84%) | 2070 (58%) | 268 (43%) | 2126 (41%) |
| Major depressive disorder | 1342 (33%) | 890 (40%) | 1460 (28%) | 2017 (28%) | 1148 (32%) | 229 (37%) | 2367 (45%) |
| Opioid use disorder | 242 (6%) | 185 (8%) | 332 (6%) | 398 (6%) | 202 (6%) | 54 (9%) | 472 (9%) |
| Other substance use disorder | 345 (9%) | 320 (15%) | 541 (10%) | 826 (12%) | 313 (9%) | 99 (16%) | 861 (16%) |
| Previous or concurrent medications | |||||||
| Metformin | 2699 (67%) | 1530 (69%) | 3884 (74%) | 4741 (66%) | 324 (9%) | 103 (17%) | 232 (4%) |
| MAUD | 350 (9%) | 160 (7%) | 236 (4%) | 355 (5%) | 575 (16%) | 109 (18%) | 953 (18%) |
| Opioid | 1743 (43%) | 1100 (50%) | 2462 (47%) | 3636 (51%) | 1282 (36%) | 271 (44%) | 2342 (45%) |
| Opioid antagonist | 699 (17%) | 355 (16%) | 787 (15%) | 1186 (17%) | 704 (20%) | 144 (23%) | 1141 (22%) |
| SSRI | 1399 (35%) | 888 (40%) | 1539 (29%) | 2092 (29%) | 1535 (43%) | 301 (49%) | 2590 (50%) |
| AUD severity | |||||||
| AUD duration (years) | 3.1 (3.0) | 3.0 (3.1) | 2.7 (2.7) | 3.0 (3.0) | 2.9 (2.8) | 3.1 (3.1) | 2.6 (2.6) |
| Encounter dates with AUD diagnosis noted, previous 2 years (n) | 3 (1-9) | 4 (2-11) | 3 (1-9) | 4 (1-10) | 3 (1-8) | 3 (1-9) | 3 (2-9) |
| Most recent AUD diagnosis (years before index) | −0.4 (0.5) | −0.4 (0.5) | −0.4 (0.5) | −0.4 (0.5) | −0.5 (0.5) | −0.5 (0.5) | −0.4 (0.5) |
| Met narrow AUD definition (other than F10) | 3355 (83%) | 1954 (89%) | 4721 (89%) | 6271 (88%) | 2737 (77%) | 553 (89%) | 4504 (86%) |
| Had any alcohol-related hospitalisation in previous 2 years | 948 (23%) | 691 (31%) | 1707 (32%) | 2584 (36%) | 763 (22%) | 199 (32%) | 1674 (32%) |
| Years in Truveta (since first encounter) | 12.6 (11.9) | 12.0 (11.7) | 11.2 (10.8) | 11.4 (11.0) | 12.8 (11.8) | 14.0 (12.0) | 12.6 (11.7) |
| Outpatient visits in previous 0–12 months (n) | 9 (5-18) | 11 (5–23) | 8 (4-17) | 10 (5–20) | 7 (3-14) | 9 (4-19) | 8 (4-18) |
| Outpatient visits in previous 13–24 months (n) | 7 (3-16) | 9 (3-21) | 6 (2-14) | 7 (2-16) | 5 (2-12) | 6 (2-14) | 6 (2-15) |
Quantitative variables are expressed as mean (SD). Categorical variables are expressed as number (percentage). Utilisation counts are expressed as median (IQR). Durations refers to the time (years) since first evidence. Other race includes Asian, American Indian or Alaska Native, Native Hawaiian or Other Pacific Islander, other race, unknown or declined to answer.
ADM, anti-diabetic medication; ALT, alanine aminotransferase; AOM, anti-obesity medication; AST, aspartate aminotransferase; AUD, alcohol use disorder; BMI, body mass index; GLP-1 RA, glucagon-like peptide-1 receptor agonist; HbA1c, glycated haemoglobin; MAUD, medications for alcohol use disorder (includes acamprosate, disulfiram, and naltrexone); SSRI, selective serotonin reuptake inhibitor.
Patient characteristics differed across trials (table 2, online supplemental Table S1). Patients with T2D were older and less likely to be female (ADM trial: mean age 59.7; 27% female, MAUD-T2D trial: mean age 57.8; 30% female), compared with those in the obesity cohorts (AOM trial: mean age 49.6; 61% female, MAUD-obesity trial: mean age 49.0; 45% female). Patients in the MAUD trials had markers of more recent and more severe AUD, as indicated by the number of visits with AUD diagnoses and the proportion of patients with previous alcohol-related hospitalisations. Across all trials, the distribution of initiation time differed across exposure groups, with users of the newer GLP-1 RAs having started their treatment more recently (online supplemental figure S6).
Among alcohol-related hospitalisations occurring after treatment initiation, most were identified using diagnostic codes (alone or in combination with laboratory testing), with the remainder identified by laboratory testing alone. In the ADM trial, 55.1% were identified using diagnostic codes and 44.9% by laboratory testing alone; corresponding proportions were 56.3% and 43.7% in the AOM trial; 64.0% and 36.0% in the MAUD-T2D trial; and 70.1% and 29.9% in the MAUD-obesity trial, respectively.
ADM trial
Before balancing, patients initiating a newer GLP-1 RA were more likely to be white and female, with markers of better health overall (as indicated by a lower Elixhauser comorbidity score,30 lower prevalence of most comorbidities and a lower baseline HbA1c). IPTW achieved good balance, with all standardised mean differences <0.1 (online supplemental Table S2).
After weighting, alcohol-related hospitalisation within 1 year of treatment occurred in 11.9% (95% CI 9.9% to 13.8%) of patients on a newer GLP-1 RA, 10.3% (95% CI 8.6% to 12.1%) of patients on an older GLP-1 RA, 14.6% (95% CI 13.3% to 16.0%) of patients on a sulfonylurea and 14.0% (95% CI 12.9% to 15.1%) of patients on other ADMs (DPP4i or SGLT2i) (figure 1A). Newer GLP-1 RAs were associated with a lower hazard of alcohol-related hospitalisation compared with sulfonylureas (HR 0.74, 95% CI 0.62 to 0.89; e-value 2.0) and other ADMs (HR 0.78, 95% CI 0.65 to 0.92; e-value 1.9) (figure 2A), but not when compared with older GLP-1 RAs (HR 1.09, 95% CI 0.87 to 1.37).
Figure 1.
Time to alcohol-related hospitalisation. Y-axis represents the probability of being event free (1−survival), x-axis represents time since initiation. Patients were censored at discontinuation, switching to a GLP-1 RA or last encounter. Panels (a) and (b) use IPCW*ICTW weights. Panels (c) and (d) use IPCW weights on a matched population. ADM, anti-diabetic medication; AOM, anti-obesity medication; GLP-1 RA, glucagon like peptide-1 receptor agonist; IPCW, inverse probability of censoring weights; IPTW, inverse probability of treatment weights; MAUD, medications for alcohol use disorder; T2D, type 2 diabetes.
Figure 2.
Hazard of alcohol-related and non-alcohol-related hospitalisations relative to initiation of a newer GLP-1 RA (semaglutide and tirzepatide). Panel A shows alcohol-related hospitalisations; Panel B shows non-alcohol-related hospitalisations. ADM, anti-diabetic medication; AOM, anti-obesity medication; GLP-1 RA, glucagon like peptide-1 receptor agonist; MAUD, medications for alcohol use disorder; T2D, type 2 diabetes.
In a sensitivity analysis limited to outcome events identified by AUD diagnostic codes (excluding laboratory-only events), findings were largely unchanged. After weighting, outcome incidence was 6.5% (95% CI 5.0% to 8.0%) for newer GLP-1 RAs, 6.2% (95% CI 4.7% to 7.6%) for older GLP-1 RAs, 8.8% (95% CI 7.7% to 9.9%) for sulfonylureas and 7.7% (95% CI 6.8% to 8.5%) for other ADMs. HRs remained similar in magnitude, with slightly wider CIs; compared with newer GLP-1 RAs, HRs were 1.04 (95% CI 0.77 to 1.40) for older GLP-1 RAs, 0.72 (95% CI 0.57 to 0.92) for sulfonylureas and 0.81 (95% CI 0.64 to 1.03) for other ADMs.
In exploratory analyses of hepatic biomarkers at 9 months, older GLP-1 RAs and other ADMs were generally associated with smaller reductions in ALT and AST compared with newer GLP-1 therapies (online supplemental figures S8 and S9). Differences were statistically significant for ALT with older GLP-1 RAs (estimate 0.14, 95% CI 0.01 to 0.26) and for both ALT (0.10, 95% CI 0.01 to 0.19) and AST (0.09, 95% CI 0.01 to 0.18) with other ADMs, while differences for AST with older GLP-1 RAs (0.12, 95% CI −0.01 to 0.25) and for sulfonylureas were not statistically significant.
AOM trial
Before balancing, patients initiating a newer GLP-1 RA were older, less likely to be female and had a higher Elixhauser comorbidity score. IPTW achieved good balance, with all standardised mean differences <0.1 (online supplemental Table S2).
After weighting, alcohol-related hospitalisation within 1 year of treatment occurred in 10.1% (95% CI 8.2% to 11.9%) of patients on a newer GLP-1 RA compared with 12.5% (95% CI 8.3% to 16.5%) of patients on an older GLP-1 RA and 12.8% (95% CI 11.1% to 14.5%) of patients on other AOMs (figure 1B). Newer GLP-1 RAs were associated with a lower hazard of alcohol-related hospitalisation compared with other AOMs (HR 0.68, 95% CI 0.54 to 0.85; e-value 2.3), but not when compared with older GLP-1 RAs (HR 1.12, 95% CI 0.76 to 1.66) (figure 2A).
In sensitivity analyses restricting outcome events to those with AUD diagnosis codes, weighted event probabilities were 4.3% (95% CI 3.1% to 5.5%) for newer GLP-1 RAs, 4.8% (95% CI 2.5% to 7.0%) for older GLP-1 RAs and 6.3% (95% CI 5.2% to 7.4%) for other AOMs. Corresponding HRs were 0.68 (95% CI 0.50 to 0.92) for other AOMs and 0.92 (95% CI 0.56 to 1.50) for older GLP-1 RAs, consistent with the primary analysis.
MAUD-T2D trial
Before matching, patients with T2D initiating approved MAUD (n=3343) differed from those initiating a newer GLP-1 RA (n=5599) in several ways. While demographics were largely similar, patients initiating MAUD had a lower BMI and HbA1c and higher levels of hepatic biomarkers (AST and ALT) associated with heavy alcohol use.22 23 They also had several markers of more severe AUD, with more visits in the previous 2 years with AUD, AUD diagnoses more recently and a higher likelihood of alcohol-related hospitalisations in the previous 2 years. MAUD initiators were also more likely to have other substance use disorder and previous or concurrent use of opioid antagonists. Propensity score matching yielded a balanced sample of 3254 patients that approximated the MAUD population, with all standardised mean differences <0.1 (online supplemental Table S2). Unmatched patients included those on both newer GLP-1 RA (n=3972) and MAUD (n=1716).
Among matched patients, 1324 (81%) of those initiating MAUD were censored due to discontinuation before 365 days compared with 786 (48%) patients initiating a newer GLP-1 RA.
After matching and IPCW weighting, alcohol-related hospitalisation occurred in 13.5% (95% CI 11.3% to 15.7%) of patients on a newer GLP-1 RA and 31.4% (95% CI 25.8% to 36.5%) of patients on MAUD (figure 1C), yielding a HR of 0.37 (95% CI 0.29 to 0.46; e-value 4.9) (figure 2A).
When restricting the outcome to visits with an AUD ICD code, weighted event probabilities were 5.9% (95% CI 4.5% to 7.4%) for newer GLP-1 RAs and 22.1% (95% CI 17.1% to 26.8%) for MAUD, yielding a HR of 0.26 (95% CI 0.19 to 0.36).
MAUD-obesity trial
Before matching, patients with obesity initiated on approved MAUD (n=9175) were younger, less likely to be female and more likely to be white compared with patients initiated on a newer GLP-1 RA (n=2023). While those initiated on MAUD had a lower Elixhauser comorbidity index, they had a higher prevalence of substance use disorder and greater previous prescribing of opioid antagonists. Similarly, they had a higher rate of AUD-related hospitalisation in the previous year and higher AST and ALT levels at baseline. Propensity score matching yielded a balanced sample of 3330 patients that approximated the MAUD population, with all standardised mean differences <0 (online supplemental Table S2). Unmatched patients included those on newer GLP-1 RA (n=358) and MAUD (n=7510).
Among patients with obesity, 1364 (82%) who initiated MAUD discontinued treatment before 365 days compared with 870 (52%) of those initiating a newer GLP-1 RA (online supplemental Table S3).
After matching and IPCW, alcohol-related hospitalisations occurred in 8.0% (95% CI 6.0% to 10.0%) of patients on a newer GLP-1 RA and 20.4% (95% CI 16.2% to 24.5%) of patients on MAUD (figure 1D), yielding a HR of 0.35 (95% CI 0.26 to 0.47; e-value 5.2) (figure 2A).
Findings were similar in sensitivity analyses restricting the outcome to events defined by an AUD diagnostic code. Weighted event probabilities were 3.5% (95% CI 2.3% to 4.8%) for newer GLP-1 RAs and 14.5% (95% CI 10.2% to 18.6%) for MAUD, corresponding to a HR of 0.23 (95% CI 0.16 to 0.35).
Negative control outcomes
Non-alcohol-related hospitalisations did not differ significantly between groups in the ADM and AOM trials (figure 2B, figure 3A). In the MAUD trials, however, treatment with a newer GLP-1 RA was associated with a lower risk of non-alcohol-related hospitalisation compared with MAUD in the MAUD-T2D trial (HR 0.70, 95% CI 0.53 to 0.93; e-value 2.3) and MAUD-obesity trial (HR 0.54, 95% CI 0.41 to 0.73; e-value 3.1) (figure 2B).
Figure 3.
Time to non-alcohol-related hospitalisation, balanced populations. Y-axis represents the probability of being event free (1−survival), x-axis represents time since initiation. Patients were censored at discontinuation, switching to a GLP-1 RA or last encounter. Panels (a) and (b) use IPCW*ICTW weights. Panels (c) and (d) use IPCW weights on a matched population. ADM, anti-diabetic medication; AOM, anti-obesity medication; GLP-1 RA, glucagon like peptide-1 receptor agonist; IPCW, inverse probability of censoring weights; IPTW, inverse probability of treatment weights; MAUD, medications for alcohol use disorder.
Discussion
This target trial emulation study found a significantly lower rate of alcohol-related hospitalisations for adults with AUD on semaglutide or tirzepatide compared with non-GLP-1 RA comparators across four clinically distinct cohorts with and without T2D. Effect estimates were broadly consistent in the ADM and AOM trials, and similar patterns were observed in the MAUD trials among patients with and without T2D. In exploratory analyses of hepatic biomarkers in the ADM trial population, changes in AST and ALT were broadly similar across groups. These biomarkers are non-specific indicators of liver injury and may be influenced by factors beyond alcohol use, limiting their sensitivity to detect changes in alcohol consumption. Accordingly, the absence of clinically significant differences does not necessarily contradict the observed reductions in alcohol-related hospitalisations.
This study extends the work of previous studies by including information on tirzepatide, which has been limited in previous studies, and using clearly defined active comparators with strong control for confounding. Despite study design differences, findings for the primary outcome are consistent with previous reports noting reduced alcohol use for patients on a GLP-1 RA. The findings for newer GLP-1 RAs in the ADM and AOM trials (HRs ranging from 0.68 to 0.78 depending on the trial) parallel effects observed in a nationwide Swedish registry study that evaluated the effect of semaglutide on AUD hospitalisations (HR 0.64, 95% CI 0.50 to 0.83)13 and a US study that evaluated recurrent AUD hospitalisations among patients with T2D (HR 0.61, 95% CI 0.50 to 0.75).15 The slight attenuation of treatment effects observed in this study may result from lesser unmeasured confounding in ADM and AOM trials. Taken together, existing evidence from observational studies suggests a potential benefit of GLP-1 receptor agonists on alcohol-related outcomes, while randomised trial evidence remains limited and mixed. This combination of promising but inconclusive findings highlights ongoing clinical equipoise and underscores the need for adequately powered randomised trials.
Findings from the MAUD trials should be interpreted with greater caution. In these analyses, there was evidence of residual confounding, as reflected by the observed association between GLP-1 receptor agonist use and reduced risk of non-alcohol-related hospitalisations, a negative control outcome for which no true association would be expected. In addition, the MAUD trials were characterised by high rates of treatment discontinuation, with more than 80% of patients in comparator groups censored during follow-up, increasing the potential for bias despite the use of inverse probability of censoring weights.
This study has several strengths. First, multiple trials were emulated to reflect clinically distinct target populations, increasing the accuracy and clinical relevance of the findings. Further, each trial included a large and generalisable population. This is important because AUD is a heterogeneous disease31 and patients with AUD who participate in clinical trials often lack generalisability, due in part to comorbidity exclusions and instability.32 Third, inclusion of a negative control outcome further supports specific benefits of GLP-1 RAs on alcohol-related hospitalisations. While GLP-1 RAs were also associated with decreased risk of non-alcohol-related hospitalisations in MAUD trials, the magnitude of this effect was smaller than that for alcohol-related hospitalisations.
This study has several limitations. First, AUD is likely under-captured in clinical practice due in part to stigmatisation.33 34 When documented, it may also be recorded variably within and across healthcare systems, with prior studies noting lower reporting in areas of high social deprivation.35 Second, residual confounding remains an important consideration. Although we used active comparator designs and applied propensity score weighting and matching to achieve good balance on measured covariates, unmeasured differences, particularly related to socioeconomic status, underlying clinical stability or AUD severity, and healthcare engagement, may persist. This is especially relevant given that newer GLP-1 receptor agonists are higher-cost therapies, and patients with access to these medications may differ systematically from comparator groups. While e-values indicated that relatively strong unmeasured confounding would be required to fully explain the observed associations, residual confounding cannot be excluded. In addition, as with all observational studies, the absence of blinding precludes assessment of expectation effects, which may have influenced health-seeking behaviours or outcomes. Third, alcohol-related outcomes were defined using diagnosis codes and laboratory testing for alcohol exposure, which may be subject to misclassification and under-capture. We did not have a severity measure for AUD, including novel biomarkers to ascertain a quantitative measure of alcohol use. Finally, this study was limited to on-treatment analyses which censored patients at discontinuation or initiation of a comparator medication. As such, inclusion of post-baseline treatment information has the potential to introduce bias and limit generalisability. Importantly, on-treatment effects were selected because they tend to better approximate clinical trial results in real world analyses given differences between these populations.36 However, treatment effects observed in clinical trials may not reflect those expected in an average clinical setting, given well-described differences between trial versus non-trial populations.32
Conclusion
In this target trial emulation study, initiation of newer GLP-1 RAs among patients with AUD was associated with a lower observed risk of alcohol-related hospitalisation, with similar associations across populations with T2D and obesity. These findings may suggest a potential role for GLP-1 RAs in the context of AUD. Further research, including randomised trials, is needed to evaluate the effectiveness of GLP-1 RAs for AUD.
Supplementary Material
Footnotes
Contributors: All authors contributed to the study conceptualisation, study design, review and revision of the manuscript, and approved the final manuscript as submitted. PJR conducted analyses and drafted the initial manuscript. EW and EH reviewed the study code. EH reanalysed data during the manuscript revision process. EH is the guarantor of the manuscript.
Funding: This research was funded by Truveta.
Disclaimer: The funder had no role in the design and conduct of the study; collection, management, analysis and interpretation of the data; preparation, review or approval of the manuscript; and decision to submit the manuscript for publication. Study authors conducted this study while employed by Truveta.
Competing interests: PJR is a former employee of Truveta and a current employee of Eli Lilly. This work was conducted and submitted prior to her employment with Eli Lilly. Eli Lilly had no role in this study. JBL declares relevant committee service (unpaid) for the American Heart Association and grant funding from the American Heart Association (NA). HM received research funding from the National Institute on Aging (K01AG070329). APK receives research funding from the American Heart Association (NA). EH and EW are employees of Truveta, which provides access to the data used in this study. DD and NS are former employees of Truveta. All other authors declare no competing interests.
Patient and public involvement: Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Provenance and peer review: Not commissioned; externally peer reviewed.
Supplemental material: This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.
Data availability statement
Data may be obtained from a third party and are not publicly available. Data used in this study are not publicly available and can be obtained from Truveta.
Ethics statements
Patient consent for publication
Not applicable.
References
- 1.Kranzler HR. Overview of Alcohol Use Disorder. AJP 2023;180:565–72. 10.1176/appi.ajp.20230488 [DOI] [PubMed] [Google Scholar]
- 2.Alpert HR, Slater ME, Yoon Y-H, et al. Alcohol Consumption and 15 Causes of Fatal Injuries: A Systematic Review and Meta-Analysis. Am J Prev Med 2022;63:286–300. 10.1016/j.amepre.2022.03.025 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Sacks JJ, Gonzales KR, Bouchery EE, et al. 2010 National and State Costs of Excessive Alcohol Consumption. Am J Prev Med 2015;49:e73–9. 10.1016/j.amepre.2015.05.031 [DOI] [PubMed] [Google Scholar]
- 4.Esser MB, Sherk A, Liu Y, et al. Deaths from Excessive Alcohol Use - United States, 2016-2021. MMWR Morb Mortal Wkly Rep 2024;73:154–61. 10.15585/mmwr.mm7308a1 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Results from the 2023 national survey on drug use and health: detailed tables. Center for Behavioral Health Statistics and Quality; 2024. Available: https://www.samhsa.gov/data/report/2023-nsduh-detailed-tables [Google Scholar]
- 6.Han B, Jones CM, Einstein EB, et al. Use of Medications for Alcohol Use Disorder in the US: Results From the 2019 National Survey on Drug Use and Health. JAMA Psychiatry 2021;78:922–4. 10.1001/jamapsychiatry.2021.1271 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Mark TL, Kranzler HR, Song X, et al. Physicians’ opinions about medications to treat alcoholism. Addiction 2003;98:617–26. 10.1046/j.1360-0443.2003.00377.x [DOI] [PubMed] [Google Scholar]
- 8.Maisel NC, Blodgett JC, Wilbourne PL, et al. Meta-analysis of naltrexone and acamprosate for treating alcohol use disorders: when are these medications most helpful? Addiction 2013;108:275–93. 10.1111/j.1360-0443.2012.04054.x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Fuller RK, Branchey L, Brightwell DR, et al. Disulfiram treatment of alcoholism. A Veterans Administration cooperative study. JAMA 1986;256:1449–55. [PubMed] [Google Scholar]
- 10.Garbutt JC, Kranzler HR, O’Malley SS, et al. Efficacy and tolerability of long-acting injectable naltrexone for alcohol dependence: a randomized controlled trial. JAMA 2005;293:1617–25. 10.1001/jama.293.13.1617 [DOI] [PubMed] [Google Scholar]
- 11.Walker JR, Korte JE, McRae-Clark AL, et al. Adherence Across FDA-Approved Medications for Alcohol Use Disorder in a Veterans Administration Population. J Stud Alcohol Drugs 2019;80:572–7. 10.15288/jsad.2019.80.572 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Subhani M, Dhanda A, King JA, et al. Association between glucagon-like peptide-1 receptor agonists use and change in alcohol consumption: a systematic review. eClinicalMedicine 2024;78:102920. 10.1016/j.eclinm.2024.102920 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Lähteenvuo M, Tiihonen J, Solismaa A, et al. Repurposing Semaglutide and Liraglutide for Alcohol Use Disorder. JAMA Psychiatry 2025;82:94–8. 10.1001/jamapsychiatry.2024.3599 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Hendershot CS, Bremmer MP, Paladino MB, et al. Once-Weekly Semaglutide in Adults With Alcohol Use Disorder: A Randomized Clinical Trial. JAMA Psychiatry 2025;82:395–405. 10.1001/jamapsychiatry.2024.4789 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Wang W, Volkow ND, Berger NA, et al. Associations of semaglutide with incidence and recurrence of alcohol use disorder in real-world population. Nat Commun 2024;15:4548. 10.1038/s41467-024-48780-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Quddos F, Hubshman Z, Tegge A, et al. Semaglutide and Tirzepatide reduce alcohol consumption in individuals with obesity. Sci Rep 2023;13:20998. 10.1038/s41598-023-48267-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Frías JP, Davies MJ, Rosenstock J, et al. Tirzepatide versus Semaglutide Once Weekly in Patients with Type 2 Diabetes. N Engl J Med 2021;385:503–15. 10.1056/NEJMoa2107519 [DOI] [PubMed] [Google Scholar]
- 18.Aronne LJ, Horn DB, le Roux CW, et al. Tirzepatide as Compared with Semaglutide for the Treatment of Obesity. N Engl J Med 2025;393:26–36. 10.1056/NEJMoa2416394 [DOI] [PubMed] [Google Scholar]
- 19.Rodriguez PJ, Goodwin Cartwright BM, Gratzl S, et al. Semaglutide vs Tirzepatide for Weight Loss in Adults With Overweight or Obesity. JAMA Intern Med 2024;184:1056–64. 10.1001/jamainternmed.2024.2525 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Truveta . Our approach to data quality. Truveta; 2025. Available: https://www.truveta.com/resources/whitepaper/truvetas-approach-to-data-quality/ [Google Scholar]
- 21.Langan SM, Schmidt SA, Wing K, et al. The reporting of studies conducted using observational routinely collected health data statement for pharmacoepidemiology (RECORD-PE). 2018. Available: https://www.bmj.com/content/363/bmj.k3532.full [DOI] [PMC free article] [PubMed]
- 22.Harris JC, Leggio L, Farokhnia M. Blood Biomarkers of Alcohol Use: A Scoping Review. Curr Addict Rep 2021;8:500–8. 10.1007/s40429-021-00402-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Alatalo P, Koivisto H, Puukka K, et al. Biomarkers of liver status in heavy drinkers, moderate drinkers and abstainers. Alcohol Alcohol 2009;44:199–203. 10.1093/alcalc/agn099 [DOI] [PubMed] [Google Scholar]
- 24.Robins JM, Hernán MA, Brumback B. Marginal structural models and causal inference in epidemiology. Epidemiology 2000;11:550–60. 10.1097/00001648-200009000-00011 [DOI] [PubMed] [Google Scholar]
- 25.Hernán MA, Robins JM. Estimating causal effects from epidemiological data. J Epidemiol Community Health 2006;60:578–86. 10.1136/jech.2004.029496 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Rosenbaum PR, Rubin DB. The central role of the propensity score in observational studies for causal effects. Biometrika 1983;70:41–55. 10.1093/biomet/70.1.41 [DOI] [Google Scholar]
- 27.Rosenbaum PR, Rubin DB. Constructing a Control Group Using Multivariate Matched Sampling Methods That Incorporate the Propensity Score. Am Stat 1985;39:33–8. 10.1080/00031305.1985.10479383 [DOI] [Google Scholar]
- 28.Joffe MM. Administrative and artificial censoring in censored regression models. Stat Med 2001;20:2287–304. 10.1002/sim.850 [DOI] [PubMed] [Google Scholar]
- 29.VanderWeele TJ, Ding P. Sensitivity Analysis in Observational Research: Introducing the E-Value. Ann Intern Med 2017;167:268–74. 10.7326/M16-2607 [DOI] [PubMed] [Google Scholar]
- 30.Elixhauser A, Steiner C, Harris DR, et al. Comorbidity measures for use with administrative data. Med Care 1998;36:8–27. 10.1097/00005650-199801000-00004 [DOI] [PubMed] [Google Scholar]
- 31.Carroll KM. The profound heterogeneity of substance use disorders: Implications for treatment development. Curr Dir Psychol Sci 2021;30:358–64. 10.1177/09637214211026984 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Blanco C, Olfson M, Okuda M, et al. Generalizability of clinical trials for alcohol dependence to community samples. Drug Alcohol Depend 2008;98:123–8. 10.1016/j.drugalcdep.2008.05.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Grüner Nielsen D, Andersen K, Søgaard Nielsen A, et al. Consistency between self-reported alcohol consumption and biological markers among patients with alcohol use disorder – A systematic review. Neuroscience & Biobehavioral Reviews 2021;124:370–85. 10.1016/j.neubiorev.2021.02.006 [DOI] [PubMed] [Google Scholar]
- 34.Hoonpongsimanont W, Ghanem G, Chen Y, et al. Underreporting of Alcohol Use in Trauma Patients: A Retrospective Analysis. Subst Abus 2021;42:192–6. 10.1080/08897077.2019.1671936 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Waddell EN, Leibowitz GS, Bonnell LN, et al. Practice-Level Documentation of Alcohol-Related Problems in Primary Care. JAMA Netw Open 2023;6:e2338224. 10.1001/jamanetworkopen.2023.38224 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Wang SV, Sreedhara SK, Schneeweiss S, et al. Reproducibility of real-world evidence studies using clinical practice data to inform regulatory and coverage decisions. Nat Commun 2022;13:5126. 10.1038/s41467-022-32310-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
bmjopen-16-7-s001.pdf (1.4MB, pdf)
Data Availability Statement
Data may be obtained from a third party and are not publicly available. Data used in this study are not publicly available and can be obtained from Truveta.