Authors: Yihua Yue (1Center for Value-Based Care Research, Primary Care Institute, Cleveland Clinic, Mail Code G10, 9500 Euclid Ave, Cleveland, OH 44195, USA;), Michael B. Rothberg (1Center for Value-Based Care Research, Primary Care Institute, Cleveland Clinic, Mail Code G10, 9500 Euclid Ave, Cleveland, OH 44195, USA;), Sudie E. Back (2Department of Psychiatry & Behavioral Sciences, Medical University of South Carolina, Charleston, SC, USA;; 3Ralph H. Johnson VA Healthcare System, Charleston, SC, USA), Olajide Adekunle (1Center for Value-Based Care Research, Primary Care Institute, Cleveland Clinic, Mail Code G10, 9500 Euclid Ave, Cleveland, OH 44195, USA;), Ha T Tran (1Center for Value-Based Care Research, Primary Care Institute, Cleveland Clinic, Mail Code G10, 9500 Euclid Ave, Cleveland, OH 44195, USA;), Phuc Le (1Center for Value-Based Care Research, Primary Care Institute, Cleveland Clinic, Mail Code G10, 9500 Euclid Ave, Cleveland, OH 44195, USA;)
Categories: Article, alcohol use disorder, health disparities, underdiagnosis
Source: Journal of general internal medicine
Authors: Yihua Yue, Michael B. Rothberg, Sudie E. Back, Olajide Adekunle, Ha T Tran, Phuc Le
Alcohol use disorder (AUD) is highly prevalent in the U.S., affecting approximately 10.9% of adults, yet remains underdiagnosed and undertreated.
To estimate the rate of AUD underdiagnosis in clinical practice, identify individual characteristics associated with underdiagnosis, and assess whether receiving a diagnosis increases the likelihood of treatment.
Retrospective cohort study.
Adults aged ≥ 18 years who screened positive for unhealthy alcohol use using AUD Identification Test-Consumption (AUDIT-C) score (≥ 4 for men, ≥ 3 for women). This study used data from the All of Us Research Program, a national, population-based cohort to reflect the diverse U.S. population.
AUD diagnoses were identified using ICD-9/10 codes. Treatment receipt was defined as either medication (disulfiram, acamprosate, or naltrexone prescriptions) or psychotherapy (identified via CPT-4 codes).
We identified 114,511 participants with unhealthy alcohol use (mean age = 50.4 years; 39.3% male). The overall AUD diagnosis rate was 10.1%, with rates of 6.8%, 21.5%, and 41.6% among participants with mild, moderate, and severe AUD risk, respectively. Factors associated with increased odds of receiving an AUD diagnosis were older age, male sex, non-Hispanic White, lower educational attainment and income, unemployment, unmarried status, public insurance coverage, and co-occurring substance use, mental health disorders, and alcohol-related medical conditions. The lifetime treatment rate was 2.55% for medication and 7.08% for psychotherapy. An AUD diagnosis was associated with increased odds of receiving medication (aOR = 10.68; 95% CI: 9.68–11.79) and psychotherapy (aOR = 1.57; 95% CI: 1.46–1.69).
In this cohort of U.S. adults with unhealthy alcohol use, AUD was underdiagnosed across all risk level with females, racial/ethnic minorities, residence in economically deprived areas, and private insurance holders more likely to be undiagnosed. Given the strong association between diagnosis and treatment receipt, these diagnostic disparities are likely to contribute to inequities in care.
Alcohol use disorder (AUD) imposes a significant disease burden in the US, affecting approximately 10.9% (28.1 million) of adults in 2022.^1,2^ AUD is linked to numerous health complications, including injuries and chronic diseases that impact different organ and body systems, and often co-occur with psychiatric disorders, such as major depressive disorder and anxiety disorder.^3,4^ As a result, AUD is associated with approximately 178,000 deaths and costs $259 billion annually.^2,5^
Despite its high prevalence and substantial social and health impact, AUD remains underdiagnosed and undertreated in clinical practice.^6^ The rate of AUD diagnoses was estimated at only 1.0% of all patients with electronic health records (EHRs), far below the expected national prevalence.^7^ Underdiagnosis is particularly common among adults aged 18–34 years.^8^ In addition, Black and Hispanic individuals have higher diagnosis rates than non-Hispanic White individuals in Veterans Affairs (VA) systems but lower rates in non-VA settings.^7,9^ Among those diagnosed, treatment receipt remains low, with only 30% of individuals with a lifetime AUD diagnosis getting treated. Younger adults, racial/ ethnic minorities, and lower-income individuals were less likely to receive treatment than their counterparts.^10^ Multiple barriers, such as stigma, limited knowledge, low perceived need, and financial constraints, can hinder treatment uptake.^11^
Previous research has highlighted disparities in AUD diagnosis, particularly across race, ethnicity, and sex. However, these studies have been limited to specific populations, such as patients in the VA or regional healthcare systems, which may not reflect patterns in the broader US adult population.^7,12^ In addition, a clinical diagnosis can help facilitate timely, evidence-based treatment,^13^ yet no study had quantified the extent to which an AUD diagnosis influences treatment receipt. In this study, we aimed to estimate the rate of AUD underdiagnosis in clinical practice, examine a broader set of factors associated with AUD diagnosis and explore the association between having a documented diagnosis and treatment receipt among US adults with unhealthy alcohol use.
This retrospective analysis utilized de-identified data from All of Us (AoU) v.8 database (N = 633,540 participants as of 9/30/2023). The AoU program is a large-scale, longitudinal research program initiated by the National Institutes of Health (NIH) with the goal of enrolling at least one million participants from diverse backgrounds across the United States.^14^ AoU collected a wide range of data including EHRs, surveys, physical measurements, biospecimens, and digital health data from wearables.^14^ The study protocols were approved by AoU Institutional Review Board (IRB), and all participants provided informed consent.^15^
We included participants aged ≥ 18 years who had EHR data, self-reported sex at birth as male or female, had an area deprivation index (ADI) score—a measure of neighborhood-level socioeconomic disadvantage, and met criteria for unhealthy alcohol use in the past year, defined by an elevated AUD Identification Test-Consumption (AUDIT-C)^16^ score of ≥ 3 for women and ≥ 4 for men.^17^ This criterion has been shown to yield maximized sensitivity and specificity in a large U.S. primary care sample.^17^ The analytic sample selection process is shown in Fig. 1. Participants were further categorized into mild AUD risk (men: AUDIT-C 4–6; 3–6), moderate AUD risk (7–8), and severe AUD risk (9–12).^7^
Primary outcomes included receipt of an AUD diagnosis, medication, and psychotherapy. AUD diagnoses were identified based on International Classification of Disease (ICD) codes for alcohol abuse and dependence (Supplement Table 1). Medication was identified through any prescriptions of FDA-approved medications (disulfiram, acamprosate, and naltrexone). Psychotherapy was defined using Current Procedural Terminology (CPT)−4 and Healthcare Common Procedure Coding System (HCPCS) codes for individual, family, or group-based psychotherapy (Supplement Table 2). We selected codes that represented sessions with “the intentional use of verbal techniques to assess or modify the patient’s emotional well-being” and ≥ 30 min long.^18,19^
Demographic and socioeconomic factors included age, sex at birth, race/ethnicity, education level, employment status, income, and marital status. We included mutually exclusive categories for private, Medicare, Medicaid, Medicare and Medicaid (dual eligibility), VA or military, other, and no coverage. ADI was categorized into quintiles, with the 1 st quintile representing the least and 5th quintile the most economically disadvantageous area.^20^ We included a skip category for missing data in all demographic and socioeconomic variables.
Based on ICD codes (Supplement Table 1), we identified mental health disorders (mood disorders, anxiety disorders, post-traumatic stress disorder, personality disorders, psychosis, attention-deficit/hyperactivity disorder, and eating disorders), substance use disorders (cannabis, hallucinogen, inhalant, opioid, sedative, cocaine, nicotine, and other stimulant use disorders), and alcohol-related medical conditions (alcoholic liver disease, pancreatitis, alcohol-related cardiomyopathy, polyneuropathy, gastritis, myopathy, degeneration of the nervous system, esophageal varices, gastroesophageal hemorrhage, liver cirrhosis, and portal hypertension).^2^ Because preexisting health conditions might affect diagnosis and treatment, we included only those documented prior to the date of the corresponding outcome. This allowed us to assess their role as potential precursors or triggers for clinical diagnosis or treatment, rather than as consequences of AUD.
Descriptive statistics, means and standard deviations (SDs) or counts and percentages, were used to describe participant characteristics overall and stratified by history of AUD diagnoses and treatment receipt. Chi-square test and t-test were used to assess differences between participants with vs. without diagnoses and those with vs. without treatment.
To identify factors associated with AUD diagnoses, we used multivariable logistic regression model incorporating patient characteristics and health conditions. Adjusted odds ratios (aOR) with 95% confidence intervals (CI) were estimated. Since alcohol consumption may influence diagnosis, we additionally used three multivariable models to examine the associations within each level of AUD risk (mild, moderate, and severe).
To examine whether a diagnosis was associated with treatment receipt, we ran two separate multivariable logistic models while also adjusting for patient characteristics and health conditions. For this analysis, we only included AUD diagnoses that occurred no later than treatment initiation and further stratified by AUD risk level. Finally, in sensitivity analysis, we used a more stringent definition of AUD risk, defined as an AUDIT-C score of ≥ 5 for both sexes following the VA and Department of Defense’s guideline.^21^
Among 266,211 participants with non-missing data, 114,511 (57.5%) met inclusion criteria for elevated AUD risk. Mean age was 50.4 years (SD = 17.3), 60.7% were female, and 63.4% were non-Hispanic White (Table 1). Stratified by AUDIT-C score, 85.0%, 9.7%, and 5.3% of participants were in the mild, moderate and severe group, respectively. The overall diagnosis rate was 10.1%, with 6.8% in the mild, 21.5% in moderate, and 41.6% in severe groups (Supplement Fig. 1). Across all AUD risk levels, males had higher diagnosis rate than females.
Compared to participants without a diagnosis, those with an AUD diagnosis had higher AUDIT-C scores (6.48 vs. 4.61) and were older (52.0 vs. 50.2 years). Males and non-Hispanic Black individuals had significantly higher diagnosis rates than females and non-Black groups, respectively. Diagnosis rate was also higher in participants from 3rd to 5th ADI quintiles, those with Medicaid, dual eligibility, and VA insurance, and those with a substance use disorder, mental health disorder, and alcohol-related medical conditions.
In the multivariable model, females were less likely to be diagnosed with AUD than males (aOR = 0.38; 95%CI: 0.36–0.39), so were Hispanic individuals (aOR = 0.88; 95%CI: 0.82–0.94) and other racial/ethnic group (aOR = 0.81; 95%CI: 0.72–0.91) compared to non-Hispanic White participants (Table 2). In contrast, individuals with lower education levels, lower incomes, and those who were unemployed or not working had increased odds of receiving diagnosis.
The relationship between ADI and diagnosis was non-linear. Compared to the least deprived areas (1st quintile), the 2nd and 3rd quintiles had increased odds of diagnosis, while 4th and 5th quintiles had decreased odds. Diagnosis also differed by health insurance coverage. All public insurance types were associated with increased odds compared to private insurance, with VA having highest odds (aOR = 2.57; 95%CI: 2.34–2.83). Finally, participants with substance use disorders (aOR = 2.55; 95%CI: 2.43–2.68) and mental health disorders (aOR = 1.91; 95%CI: 1.83–2.00) were more likely to be diagnosed. Among alcohol-related medical conditions, pancreatitis was associated with the highest odds of diagnosis (aOR = 14.85; 95% CI: 9.78–22.54).
In stratified analyses, most associations remained consistent with the main analysis, with a few exceptions. Non-Hispanic Black individuals with severe AUD risk had a decreased odds of diagnosis compared to non-Hispanic White counterparts (aOR = 0.59; 95%CI: 0.51–0.69). As severity of AUD risk increased, the difference across sexes and the effect of pancreatitis decreased while the difference in 4th and 5th vs. 1 st ADI quintiles widened.
Overall, 2.55% of participants with elevated AUD risk received medication, and 7.08% received psychotherapy. Among those with a documented diagnosis, 16.9% received medication and 22.9% psychotherapy, compared to 0.9% and 5.3%, respectively, of those without a diagnosis. Medication was used by 1.6% of participants with mild, 5.1% moderate, and 13.0% severe AUD risk, while psychotherapy use rates were 6.7%, 8.1%, and 11.7%, respectively.
Participants who received medication or psychotherapy had higher AUDIT-C scores and were more likely to have a diagnosis (Supplement Table 3) than those who were untreated. They were also more likely to be male, unemployed or not working, have lower income, and reside in areas in 1st-3rd ADI quintiles. In addition, treated participants were more likely to have Medicaid, dual eligibility, or VA insurance, substance use or mental health disorders, and alcohol-related medical conditions. Non-Hispanic White participants had the highest psychotherapy treatment rate, while Non-Hispanic Black participants had the highest medication treatment rate.
After adjusting for covariates, having a documented diagnosis was associated with increased odds of receiving medication (aOR = 10.68; 95% CI: 9.68–11.79) (Table 3) and psychotherapy (aOR = 1.57; 95% CI: 1.46–1.69) (Table 4). There was no sex difference regarding receipt of medication, but females were less likely to receive psychotherapy than males (aOR = 0.85; 95%CI: 0.81–0.90). Compared to non-Hispanic White participants, non-Hispanic Black (aOR = 0.83; 95%CI: 0.73–0.93) and Hispanic (aOR = 0.62; 95%CI: 0.54–0.71) participants were less likely to receive medication whereas Hispanic participants were more likely to receive psychotherapy (aOR = 1.09; 95%CI: 1.01–1.18). People with lower incomes also had increased odds of receiving psychotherapy but not medication.
Participants with Medicaid, dual eligibility, or VA insurance, those who were not working or unemployed, and those who were single/divorced/widowed/separated had increased odds of being treated. In contrast, individuals from more economically disadvantaged areas had decreased odds of receiving treatment. Mental health disorder was positively associated with both treatment types, while substance use disorder was negatively associated with receipt of psychotherapy (aOR = 0.73; 95%CI: 0.68–0.78). Pancreatitis was associated with receipt of medication (aOR = 1.77; 95%CI: 1.15–2.72) but not psychotherapy.
In stratified analyses, the association between diagnosis and treatment receipt weakened as AUD risk level increased. For medication, the associations with sex, income, dual eligibility and pancreatitis remained significant only in the mild AUD risk group. Hispanic participants had decreased odds of receiving medication across all AUD risk levels, with the effect size increasing as severity worsened. For psychotherapy, the significant effect of sex, race/ethnicity, and education was observed only in the mild AUD risk group, while the association with VA insurance, ADI quintiles and substance use disorder was enhanced as risk level increased.
Our findings remained consistent after restricting the study sample to individuals with AUDIT-C scores of ≥ 5 (Supplement Table 4).
In this study, we found that 10.1% of AoU participants with elevated AUD risk had an AUD diagnosis recorded in their EHR. Although the diagnosis rate in participants with severe risk was six times higher (41.6% vs. 6.8%) than that in those with mild risk, more than half of those with severe risk remained undiagnosed. Several factors may contribute to this phenomenon. Providers may not routinely screen for or systematically document AUD, particularly in patients without overt clinical symptoms. A study among US primary care providers found that validated alcohol screening questionnaires were used in only 2.6% of primary care visits, and alcohol counseling was documented in 0.8% of visits.^22^ Competing clinical priorities in primary care may result in limited time for thorough substance use assessments when providers prioritize acute conditions over behavioral health.^23^ Stigma surrounding AUD, both from patients and providers, can also lead to missed diagnoses.^24^ Patients may underreport alcohol use, and clinicians may hesitate to formally diagnose AUD due to concerns about labeling or patient resistance.
We also found that variables positively associated with AUD diagnoses included older age, lower socioeconomic status, unmarried status—a proxy for less social support, co-occurring substance use or mental health disorders, and alcohol-related medical conditions. In contrast, female sex, racial/ethnic minority, residency in economically deprived areas, and private insurance were associated with lower odds of receiving diagnosis. Our finding on race/ethnicity align with Ellis et al.,^7^ who reported lower provider-documented AUD among Black and Hispanic than White participants, whereas VA studies showed higher diagnosis rates among racial and ethnic minorities despite similar alcohol consumption level.^9,12^ This difference may be explained by the unique characteristics of VA population, including a higher proportion of men, greater alcohol consumption, and more coexisting substance and mental health disorders.^7^
Importantly, we found that an AUD diagnosis was strongly associated with treatment receipt, with diagnosed individuals having 11 times higher odds of receiving medication and 1.6 times higher odds of receiving psychotherapy. This finding emphasizes the important role of clinical recognition in facilitating access to evidence-based treatments for AUD, particularly medication, which can help individuals reduce unhealthy alcohol consumption and improve health outcomes. Besides diagnosis, lower income increased the odds of receiving psychotherapy while people with 4th or 5th ADI quintiles consistently had decreased odds of receiving either treatment.
The relationship between race/ethnicity and treatment receipt was interesting. Despite a higher crude rate of medication receipt among non-Hispanic Black, adjusted model revealed lower odds of medication receipt than non-Hispanic White, suggesting the confounding effects from other characteristics. In addition, the adjusted association showed Hispanic individuals were less likely to receive medication but more likely to receive psychotherapy than non-Hispanic White. While access to treatment in racial/ethnic minorities could be limited due to provider biases and cultural or linguistic barriers,^25,26^ Latinos were more likely to seek mental health treatment services when they perceived a need for treatment.^27^ In addition, we found that those with Medicaid or dual eligibility were more likely to be treated than those with private insurance.^28^ In our study, participants with VA insurance also had increased odds of treatment and the effect was enhanced when the severity of AUD risk increased. This finding is consistent with the VA’s well-established infrastructure, which includes routine use of AUDIT-C screening tool in primary care, integrated behavioral health services, dedicated addiction programs, and comprehensive coverage for AUD treatments.^29,30^
Our study revealed notable sex differences among individuals with mild risk of AUD. Women had lower diagnosis rate and were less likely to receive psychotherapy than men but more likely to use medication. This may reflect lower perceived risk of alcohol use, as women are twice as likely to believe they can manage issues independently.^30,31^ Gender biases in clinical settings, where AUD is often viewed as more prevalent in men, may also contribute to reduced screening among women.^31^ In addition, women may underreport alcohol use due to social stigma or internalized shame, further reducing the likelihood of receiving a formal diagnosis.^32^ Furthermore, women often experience more severe health consequences from alcohol use at lower consumption levels than men, potentially increasing their motivation to use medication to mitigate harm.^33–35^
Our study has important implications. Pre-visit and standardized screening may be integrated into primary care and emergency settings. Expanding provider education on the importance of early recognition and reducing stigma-related barriers have been shown to improve diagnosis rates.^36^ Leveraging EHR alerts or clinical decision support tools to prompt AUD screening and documentation may enhance recognition and facilitate timely intervention. Moreover, policies that incentivize comprehensive substance use assessments, improve insurance coverages for addiction services,^37^ and expand access to culturally and linguistically competent care^38^ could help bridge the gap in AUD care.
Our study has several limitations. First, it was subject to selection bias because we excluded individuals who had never sought healthcare and had no EHR data. Second, because this is an EHR-based analysis, we could not determine whether medication or psychotherapy was offered but declined by patients. Similarly, we were unable to capture patient engagement in non-medical support, such as Alcoholics Anonymous, which may have treatment-like effects but are not documented in EHR. Third, psychotherapy was identified using CPT and HCPCS codes that are not AUD-specific and may reflect counseling for other behavioral or mental health concerns. However, for individuals with unhealthy alcohol use or AUD, it is likely that alcohol-related concerns were also addressed during these sessions. Fourth, we categorized AUD risk using AUDIT-C scores assessed at AoU program enrollment, which may not reflect the severity at the time of AUD diagnosis or treatment. Fifth, the EHR data lack information on AUD screening practices in different health systems, preventing us from assessing whether screening was associated with improved diagnosis or treatment. Finally, while the AoU database intentionally oversamples underrepresented populations, which was valuable for addiction research, the sample was not nationally representative of the US population, limiting the generalizability of our findings.
In our study, having an AUD diagnosis in the EHR was positively associated with treatment, with diagnosed individuals having almost 11 times higher odds of receiving medication and 1.6 times higher odds of receiving psychotherapy. However, underdiagnosis was common, with almost 90% of all individuals with elevated AUD risk and 60% of those with severe risk being undiagnosed. Factors associated with underdiagnosis included female sex, racial and ethnic minorities, and individuals living in economically deprived areas.
Supplementary Information The online version contains supplementary material available at https://doi.org/10.1007/s11606-025-10089-5.