Diagnosis is a foundation of clinical decision making and treatment (
1). Diagnoses, as clinical labels, can produce lasting stigma and, when inappropriate, can produce lasting damage to the individuals who receive a diagnosis (
2). Diagnoses that are stigmatized, such as alcohol use disorder (AUD), can be particularly damaging. Furthermore, misdiagnosis can result in ineffective treatment, inaccurate prognostic assessments, poor outcomes, and distrust of the health care system (
3,
4). Factors that influence the diagnosis of behavior-based conditions include medical conditions, varying symptom presentation, the clinician’s level of education and experience, the patient’s willingness to disclose symptoms, cultural factors, and the application of standardized criteria or assessments (
5). A patient’s race or ethnicity and ethnic and racial differences between patients and providers can also influence diagnostic decisions through explicit or implicit bias (
3,
6,
7), that is, clinicians’ conscious or unconscious prejudices or stereotypes (
8,
9).
Studies in the Veterans Health Administration of the Department of Veterans Affairs (VA) have shown that the rate of clinically recognized AUD is higher among Black and Hispanic veterans than among White veterans (
10). Black veterans are also more likely than White veterans to be identified as needing an intervention (
11) and to receive psychosocial interventions (
11,
12) but less likely to receive pharmacotherapy for AUD (
13). One potential explanation for the observed group disparities is that these groups differ in alcohol consumption patterns (
14); alternatively, AUD may be viewed as a biological illness in White patients but as a behavioral disorder or lifestyle choice among Black and Hispanic patients (
15). While these studies suggest the presence of racial and ethnic bias in AUD diagnosis and treatment, collateral information upon which to assess bias in the diagnosis of AUD, such as recent or lifetime measures of alcohol consumption, was unavailable.
Here we examined the contribution of self-reported alcohol consumption to the likelihood of receiving an AUD diagnosis among Black, Hispanic, and White veterans. To facilitate the identification of individuals with unhealthy alcohol use, beginning in 2007, the VA has routinely screened primary care patients using the consumption subscale of the Alcohol Use Disorders Identification Test (AUDIT-C) (
16,
17), comprising the first three items of the 10-item AUDIT (
18). We examined the association of AUDIT-C scores with AUD diagnostic codes across the three racial and ethnic groups in a national cohort of more than 700,000 veterans. Importantly, we are examining race and ethnicity in the context of differential racialization; that is, because no biological basis for these groups exists, race and ethnicity are best understood as social constructs and proxies for the experience of racism and discrimination. Specifically, we evaluated the relationship between race and ethnicity and AUD diagnosis adjusting for self-reported alcohol consumption; evaluated whether this relationship, if it exists, varies by consumption levels; and evaluated sociodemographic and clinical correlates of an AUD diagnosis. Analyses were stratified by sex to account for biological differences (
19) related to that variable and to examine the intersectionality of race and ethnicity and sex. Based on previous studies that suggest the presence of racial and ethnic bias in diagnosis and treatment within the VA (
10–
13,
20) and that bias may vary by alcohol consumption level (
21), we hypothesized that Black and Hispanic veterans would have a higher frequency of AUD diagnosis than White veterans after adjusting for alcohol consumption and that the frequency of AUD diagnosis among racial and ethnic groups would vary by alcohol consumption level.
Discussion
In this sample of more than 700,000 veterans, we identified a differential frequency of AUD diagnosis by race and ethnicity. The greatest discrepancy was among Black men, who, at all but the lowest and highest levels of alcohol consumption, had 23%–109% greater odds of an AUD diagnosis than White men. Hispanic men had 20%–32% greater odds of an AUD diagnosis than White men. The prevalence of disorders associated with persistent heavy drinking (e.g., alcoholic cirrhosis and hepatitis), whose diagnoses generally rely on objective measures (e.g., laboratory values and ultrasound findings), was similar across the three groups, which suggests that the greater likelihood of an AUD diagnosis among Black and Hispanic veterans was likely not due to different levels of alcohol consumption. The association between race and ethnicity and AUD diagnosis remained after adjustment for alcohol consumption level, alcohol-related disorders, drug use disorders, and other potential confounders.
The sample for which we had EHR data on self-reported alcohol consumption and AUD diagnosis was large enough to account for multiple potential contributing factors. The frequency of AUD here (21% overall) was lower than that in the general population estimate (29%) from the National Epidemiologic Survey on Alcohol and Related Conditions–III (NESARC-III) (
31), and the frequency was lower among both men (22% here vs. 36% in NESARC-III) and women (14% here vs. 23% in NESARC-III). Notably, the VA EHR data are cumulative over approximately 20 years of available data, compared with lifetime estimates in NESARC-III. In a previous VA study (
10), the frequency of AUD was 10% among Black veterans, 7% among Hispanic veterans, and 6% among White veterans, compared with 31%, 24%, and 18%, respectively, in the present study. Despite the use of diagnostic data from a single year in that study, rather than the cumulative estimate from the VA EHR obtained in the present study, both studies found the same order of AUD frequencies by race and ethnicity, which was opposite that found in NESARC-III, where Black individuals had the lowest lifetime AUD prevalence (22%), followed by Hispanic (23%) and White (33%) individuals.
Notably, NESARC-III used a structured diagnostic interview, which is likely to be more accurate (i.e., less biased) than a clinical interview, as is used in the VA, which could also contribute to the higher prevalence of AUD in the general population than in the VA population. Our findings in the VA population highlight differential clinical assessment of AUD by race and ethnicity, and this difference could be due to overdiagnosis of Black veterans, underdiagnosis of White veterans, or, more likely, a combination of the two. Both kinds of misdiagnosis can have harmful effects, because overdiagnosis can be stigmatizing and underdiagnosis can delay treatment. Consistent with the observation that there are disparities in the diagnosis of AUD associated with race and ethnicity, the strength of correlations between AUDIT-C score and AUD diagnosis increased monotonically in both sexes. Specifically, White veterans showed the lowest correlation between AUDIT-C score and AUD diagnosis, Hispanic veterans showed an intermediate correlation, and Black veterans showed the highest correlation. The findings suggest that White veterans are underdiagnosed with AUD. Despite a higher rate of referral and treatment for AUD among Black veterans than among White or Hispanic veterans (
11), the available data do not allow us to determine the net impact of the diagnostic differences on patient outcomes. Any potential benefit of greater treatment rates should not overshadow the central issue that racialized inequity in assessment, particularly of Black patients, appears to exist. Studies are needed to examine the mechanism by which veterans receive an AUD diagnosis and multilevel factors such as bias and systemic racism that likely affect the observed inequity.
The greatest disparity in AUD diagnosis after adjustment for potential confounders occurred at maximum AUDIT-C scores of 3 and 4, near the cutoff for a positive AUDIT-C screen (≥3 for women; ≥4 for men). These findings suggest that, at scores near the threshold, providers are more likely to assign a diagnosis of AUD to Black or Hispanic than White veterans (
6,
32,
33). In a series of experiments that evaluated implicit stereotyping, physicians were more likely to associate stigmatizing medical conditions (e.g., drug use and HIV) with Black than White patients (
6,
32), suggesting that diagnostic disparities may reflect implicit bias. Studies of diagnostic disparities suggest that they could result from the differential presentation of psychiatric symptoms across racial and ethnic groups (
10,
21). Although this perspective could reflect the impact of culture on psychiatric symptom presentation (
34), it also indirectly acknowledges that diagnostic science and practice reference the White experience.
Research has also shown that psychological distress and social disadvantage (including factors such as poverty, racial and ethnic stigma, unfair treatment, and cumulative disadvantage) can contribute to persistent racial and ethnic disparities among individuals with alcohol dependence despite lower levels of heavy alcohol consumption (
21,
35,
36). Although social disadvantage likely mediates or moderates the associations identified in the present study, as observed in other studies using national samples (
21,
35,
36), we did not have access to such measures. The interrelationships of race and ethnicity and social disadvantage and their effects on alcohol-related problems are complex and merit in-depth exploration in the veteran population.
We found that the presence of a diagnosis of a drug use disorder and an AUD diagnosis were highly correlated. These disorders commonly co-occur, both in the VA population (
37) and the general population (
31,
38). In NESARC-III, the prevalence of a concurrent AUD and drug use disorder (which may include cannabis and tobacco use disorders) did not differ substantially by race and ethnicity (Black, 9%; Hispanic, 7%; White, 8%) (
34). However, in the present study sample, Black men were over three times (23%) and Hispanic men were 1.5 times as likely (12%) as White men to have at least one comorbid drug use disorder (8%). This may be because once a patient receives an AUD diagnosis, providers are more likely to query the patient about use of other substances or vice versa. The findings may also reflect implicit bias toward Black and Hispanic veterans, which prompts additional screening for use of other substances in these populations (
6,
39,
40).
More research is needed to understand the source of these differences. To aid in the valid diagnosis and treatment of the use of multiple substances, standardized screening and assessment methods are recommended. This, by itself, however, may not be adequate, as there are multiple examples of racial bias in medicine that occur even when objective tests are used. A commonly cited example is the estimated glomerular filtration rate, where different formulas have been used for Black and White patients. This has led to less access to kidney transplant for Black than White individuals, despite comparable levels of severity of renal disease. Other examples of biased algorithms include those used to predict the risk posed by a trial of labor in women who have previously delivered a baby via cesarean section, the risk of developing breast cancer, the risk of developing a kidney stone, and the use of spirometry to measure lung function, among others (
41).
In our study, women were less likely than men to receive an AUD diagnosis, consistent with population estimates (
42) and findings among veterans (
43). Although women consume less alcohol than men, this difference has been decreasing in recent years (
44). In studies of unhealthy alcohol use, women experience greater alcohol use–related stigma than men (
45), which could impact how providers respond to (
46) and document (
20) alcohol use among women. More research is needed to understand sex and gender differences, and their intersection with race and ethnicity, in substance use reporting and documentation in the medical record.
Similar to the findings among men, at nearly every level of alcohol consumption, Black women were more likely than Hispanic and White women to receive an AUD diagnosis, despite having a similar distribution of alcohol consumption and prevalence of alcohol-related disorders among the groups. There were few differences in the relationship between AUD frequency and alcohol consumption between Hispanic and White women. Where such differences were present, White women had a greater frequency of AUD diagnosis than Hispanic women, consistent with estimates from the 2019 National Survey on Drug Use and Health, where alcohol use prevalence was higher among non-Hispanic White women than Hispanic women (
42).
Our study has several limitations. Despite obvious differences in the frequency of AUD diagnosis by racial and ethnic group, the basis for the discrepancies cannot be ascertained using EHR data, and we did not have information on how diagnoses were made. Second, self-reported measures of alcohol consumption may be subject to recall bias. In two U.S. national surveys (total N >494,000) that used AUDIT-C data from participants who reported past-year drinking (
47) approximately 20% of male and female veterans reported drinking levels that were inconsistent with screening results based on standard cutoff scores. Because the available data did not permit a conclusion to be drawn as to the source of the discrepancies, objective measures of alcohol use (e.g., biomarkers) are needed to validate self-reports. Third, we did not have data on socioeconomic or social disadvantage factors, which may have mediated or moderated the associations identified in this study. Fourth, findings from a sample of U.S. veterans enrolled in a genetic cohort study may not generalize to other populations, including the general veteran population. Lastly, the substantially smaller number of female relative to male veterans provided less statistical power to detect differences among women.
Our study also has notable strengths. The availability of annual assessments of alcohol consumption and informative EHRs enabled us to analyze relationships between measures of alcohol consumption and AUD diagnosis, with analyses that included multiple clinical factors that could influence these associations. Whereas previous studies could analyze data spanning only several years (
10–
13,
20), we analyzed data from individuals’ entire VA EHRs. Second, the large and diverse sample, particularly of men, provided enough statistical power to examine factors that affect the likelihood of an AUD diagnosis.