The Alcohol Use Disorders Identification Test (AUDIT) was published in 1989 by a study group of the World Health Organization (WHO) (Saunders et al., 1993a; b) and recommended for screening and brief intervention in medical care (Babor et al., 2001; Dasarathy et al., 2019; Gonzalez et al., 2020). Since then, it has been used worldwide including Africa (Atkins et al., 2021), China (Li et al., 2011), and Russia (Neufeld et al., 2021). The number of publications increased steeply (Babor & Robaina, 2016), and several reviews about the AUDIT are available (Allen et al., 1997; Berner et al., 2007; Higgins-Biddle & Babor, 2018; Lange et al., 2019; Reinert & Allen, 2002, 2007). Evidence proved the AUDIT to represent a psychometrically valid and reliable assessment tool across different medical settings and in a general adult population sample (Moehring et al., 2018).

Purpose of the AUDIT was to screen for alcohol use disorders in an easy and cost-saving manner which may be practiced in different settings all over the world. The ten AUDIT questions cover four zones (Babor & Robaina, 2016) and three domains (Babor et al., 2001). The zones are defined by sections of the AUDIT sum score. They indicate alcohol dependence, harmful and hazardous in addition to low risk alcohol consumption. Among the domains, the first represents the drinking behavior including frequency and quantity of alcohol drinking and frequency of having had six or more drinks on one occasion. Item 1 additionally indicates whether the respondent consumed any alcohol in the last 12 months. The second domain represents symptoms of alcohol dependence (frequency of not having been able to stop drinking, frequency of not having been able to comply with expectations of others or to fulfill role obligations because of having drunk alcohol, frequency of having needed a first alcoholic drink in the morning). The third domain includes consequences of alcohol consumption (frequency of having felt guilty because of drinking, frequency of memory loss according to the night before due to alcohol drinking, frequency of having caused injury under the influence of alcohol, frequency of others having been concerned about alcohol drinking of the person). Each of the domains 1 and 2 is ascertained by three, domain 3 by four questions (Babor et al., 2001).

Given the briefness and practicability of the AUDIT, the question arises whether the AUDIT may predict time to death. According to a large body of evidence, alcohol use disorders are related to early death (John et al., 2021; Roerecke & Rehm, 2013). However, less is known about whether screening of alcohol use disorders by the AUDIT may predict mortality risk. One meta-analysis included seven studies on the prediction of mortality by the AUDIT that had been published in the years 1989 to 2016 (Kuitunen-Paul & Roerecke, 2018). The follow-up periods were limited to less than 5 years in the majority of the seven studies. Not all of these used a random adult population sample. Among the study participants who had consumed alcohol in the last 12 months prior to the data gathering, those with an AUDIT score 8 to 40 (at-risk drinkers) had a relative risk 1.24 (95% confidence interval 1.12 − 1.36) to die within the follow-up period compared to the study participants who had an AUDIT score 1 to 7 (moderate drinkers) (Kuitunen-Paul & Roerecke, 2018).

To date, there is limited evidence of predicting mortality by the AUDIT which is due to a lack of random general population samples and of follow-up time larger than 5 years. Little is known about the prediction of mortality by the zones and domains of the AUDIT. The zones may serve as a simple measure of the severity of alcohol use disorders. The domains may add to the discussion whether the alcohol drinking behavior domain might make the alcohol dependence and alcohol consequences domains unnecessary and whether high alcohol consumption might suffice to determine alcohol use disorders (Rehm et al., 2013). The aim of this study was to analyze the AUDIT among a random adult general population sample with respect to total and disorder-specific mortality after 20 years. In detail, three questions were to be answered: first, do the AUDIT zones predict mortality; second, whether the zones may predict specific mortality with cardiovascular, cancer, or other disorders; and third, do all domains add to the prediction of mortality?

Methods

Sample

A random adult population sample aged 18 to 64 years had been drawn in northern Germany (Meyer et al., 2000a) including rural and urban areas. Names and addresses of residents were provided by the residents’ registration offices. Every resident in Germany has to be registered by law in the data files of the registration offices. All persons in the sample lived in private household. Among the 5829 individuals eligible for the baseline study, 4093 (70.2%) interviews had been completed July 1996 to March 1997, and 4075 were analyzed. The interview had been conducted in the households of the respondents or any other preferred location. The interviewers had been trained for the purpose of the interview. The date of the interview had been proposed by letter and was customized according to the preferences of the respondent. A mortality follow-up was conducted from April 2017 until April 2018 (John et al., 2020). Among the 4075 study participants with complete baseline data, for 47 vital status, i.e., whether the person was dead or still alive, could not be proven. For 4028 (98.8% of 4075) study participants, vital status was ascertained. Among them, 447 had abstained from alcohol in the past 12 months prior to the baseline assessment according to the first question of the AUDIT. A data analysis of the abstainers may be found elsewhere (John et al., 2021). The remaining 3581 persons had consumed alcohol in the past 12 months prior to the contact at baseline. This is our final sample for the reported data analysis.

Assessments

Baseline

The AUDIT was presented as a questionnaire to the study participants. They filled it in at the end of the interview and while the interviewer was present (Rumpf et al., 2002). The possible AUDIT sum scores range between 1 and 40. The score was divided into four zones as recommended (Babor & Robaina, 2016; Babor et al., 2001). Accordingly, zone 1 encompasses the values 1 to 7 and indicates low risk drinking, zone 2 (values 8 − 15) hazardous, and zone 3 (values 16 − 19) harmful alcohol use. Zone 4 (values 20 − 40) may indicate alcohol dependence. The zones were defined to indicate risk levels and needs for interventions such as simple advice when a respondent had an AUDIT sum score in zone 2 or advice plus brief counseling and monitoring in case of an AUDIT sum score in zone 3. The possible value range of domain 1 (AUDIT questions 1 − 3) was 1 to 12, of domain 2 (AUDIT questions 4 − 6) it was 0 to 12, and of domain 3 (AUDIT questions 7 − 10) it was 0 to 16.

Mortality Follow-Up

Mortality was ascertained as total and as diagnosis-specific mortality. For total mortality, we used vital statistics data. In Germany, a death case including the name, the last address, and the date of death has to be registered in the residents’ registration files at the last place of residence by law. Vital statistics data were retrieved from these files. For diagnosis-specific mortality, the death certificate information was used. Based on the information of the residents’ registration office about the date of death, we retrieved the death certificates from the local health authorities at the place of residence of the study participant. In Germany, in each case of death, a physician has to fill in a standardized death certificate. It includes the date of death and four sections of diagnostic information based on the judgement of the certifying physician.

Diagnosis-specific mortality was assessed using the four sections of diagnostic information of the death certificates: disorder that elicited death, disorders that contributed to death, basic health disorders, and further health disorders. Since the information in the four sections could not be assumed to be clearly distinct in their contribution to death, we considered all disorders irrespective of their diagnostic segment. There were 15 diagnostic items at maximum to be filled in by the certifying physician and 11 diagnostic items of the autopsy. We analyzed diagnostic information when endorsed in any of the 26 diagnostic items. Among all diagnoses, we grouped under cardiovascular disorder: all heart disease, all blood vessel disorders, and all cerebrovascular disorders. Among the remaining death cases, we grouped all with any cancer diagnosis to cancer. After that, the remaining death cases were allocated to a group of other disorders: any gastrointestinal, respiratory, brain, acute disorders (injury, suicide, accidents, other death cases due to violence), or no diagnosis. The disorders were documented according to the WHO International Classification of the Diseases, version 10 (World Health Organization, 2019).

Data Analysis

First, we analyzed the AUDIT sum score and the zones as predictors of total mortality and used the possible value ranges. In addition, we divided zone 1 into two ranges of values in order to define low risk drinking. Second, we tested the AUDIT zones as predictors of diagnosis-specific death. Among the 3581 persons of the final sample, 454 death cases occurred. These were used for the analysis of total mortality. For 25 death cases, no death certificate was available, and 1 did not contain information. We used the remaining 428 death certificates for the data analysis of diagnosis-specific mortality. Third, we analyzed the AUDIT domains as predictors of total mortality. We used the domains with their entire range of values. In addition, we collapsed the values of each of the domains into four ranks with similar proportions of death cases. We used Cox Proportional Hazards analysis and give the hazard ratio (HR) with 95% confidence interval (CI). The Cox proportional hazards assumption was tested using Schoenfeld residuals (Bellera et al., 2010; Flynn, 2012). For diagnosis-specific mortality, we used competing-risks regression analysis to take into account that the risks for one diagnosis group may be impeded by the risks of other diagnosis groups (Pintilie, 2007). We specify the subhazard ratio (SHR) and CI. All hazard and subhazard ratios were adjusted for age and sex. All data analysis was performed using STATA 17.0 (StataCorp LP, 2021). The data analysis was not pre-registered.

Results

The final sample of 3581 persons included 1758 females (Table 1). The mean age was 41.29 years (95% confidence interval 40.87 − 41.71). The age and sex distribution of the sample corresponded to the age and sex distribution of the general population at age 18 to 64. The differences between the sample and the general population were 0.2% according to sex and 1.8% at maximum according to age groups (18 − 29, 30 − 44, 45 − 64 years) (Meyer et al., 2000b). The empirical value range of the AUDIT was 1 to 37. Among males, 10.64%, and among females 2.84% had a value in AUDIT zones 2 to 4 (Table 1).

Table 1 AUDIT zones and sex at baseline
Full size table

The AUDIT sum score and zones turned out to predict total mortality (Table 2). Each of zones 2 to 4 was related to a 70% higher HR than the respective lower zone after adjustment for age and sex. Study participants in zone 4 had an HR 3.96 (1.86 − 8.41) compared to the study participants in zone 1 after adjustment for age and sex. Among the deceased study participants who had any cardiovascular disorder documented in the death certificate, the data revealed increased HRs for zones 2 to 4. Each of these was related to time to death by an 84% higher risk than in the zone below. Study participants in zone 4 had an HR 6.19 (2.87 − 13.35) compared to the study participants in zone 1 in the group with cardiovascular death after adjustment for age and sex. Those without cardiovascular disorder but with cancer did not show a relation to time to death. Among the study participants with cardiovascular disorder and among participants with disorders other than cardiovascular or cancer, those in zone 2 had a higher SHR than the persons in zone 1.

Table 2 Alcohol Use Disorders Identification Test at baseline and deceased study participants 20 years later
Full size table

Having divided zone 1 into AUDIT sum score 1 to 4 and sum score 5 to 7 made a difference according to time to death. Among alcohol consumers with a sum score 5 to 7, 14.91% had been deceased, whereas these were 11.28% among those with a sum score 1 to 4. The HR for total mortality among persons with a sum score 5 to 7 was 1.22 (0.95 − 1.55) compared to those with sum score 1 to 4, adjusted for age and sex. Among study participants with any cardiovascular disorder involved in death, those with a score 5 to 7 had a higher HR (1.40; 1.03 − 1.90) than those with a score 1 to 4.

Each of the three AUDIT domains predicted total mortality (Table 3). This was the case if the three domains were tested separately or if two domains were tested within one model. For all three domains in one model (Table 3, model 5), data revealed that domains 1 and 2 remained as predictors, whereas domain 3, consequences of alcohol consumption, did not. All combinations of the four domain ranks except two included one or more study participants (Table 4).

Table 3 Alcohol Use Disorders Identification Test at baseline and total mortality
Full size table
Table 4 Alcohol Use Disorders Identification Test domains, number of persons
Full size table

Discussion

This study has four main findings. First, the AUDIT zones predicted mortality in a dose–response manner. Second, the data revealed a particularly strong relation between AUDIT scores and cardiovascular mortality. Third, even within low risk drinking as defined by authors of the AUDIT, we found increased cardiovascular mortality. Fourth, the domains predicted mortality.

The AUDIT predicted total mortality in a dose manner with a 70% higher HR per zone. This finding clearly speaks in favor of the AUDIT in the prediction of time to death. It confirms previous evidence as reported by a meta-analysis (Kuitunen-Paul & Roerecke, 2018). By using a random adult population sample and a 20-year mortality follow-up, our findings add significantly to the previous evidence which was based on follow-up time periods limited to less than 5 years in five of the seven studies that had been reported (Kuitunen-Paul & Roerecke, 2018). The authors of the meta-analysis found a relative risk 1.24 (95% confidence interval 1.12 − 1.36) for those with AUDIT scores larger than 7 to die within the follow-up period of time compared to those who had a score 7 or less (Kuitunen-Paul & Roerecke, 2018). Our findings speak in favor of higher hazards of early death. One reason may be the longer time span of 20 years in our study compared to less than 5 years in most of the studies in the meta-analysis (Kuitunen-Paul & Roerecke, 2018). Our findings clearly speak in favor of the predictive power of the AUDIT given a longer lifetime period. Our data also revealed the strong power of alcohol use disorders to limit life expectancy. The consideration of all zones makes the dose–response relation with time to death likely.

According to diagnosis-specific mortality, we found a particularly strong relation of the AUDIT with cardiovascular mortality. This result confirms findings of a meta-analysis concerning death cases among patients with alcohol use disorders after treatment. It shows that the most common health disorders involved in death were cardiovascular disorders (Abdul-Rahman et al., 2018). This can be suggested to be reasonable. According to evidence, alcohol drinking is a potential causal factor in different cardiovascular disorders including arrhythmia, atrial fibrillation, alcoholic cardiomyopathy, atherosclerosis, and hypertension (Statescu et al., 2021). According to hypertension, our findings of increased risk among those with a rather low AUDIT score of 5 to 7 having an increased risk of early death corresponds to results from a meta-analysis evidencing that, among men, drinking 24 gr alcohol per day was related to an increased risk of incident hypertension (Roerecke et al., 2018).

In contrast to cardiovascular disorders, our data did not reveal a relation with mortality for cancers. This contradicts evidence from a 30-year mortality follow-up in which alcohol consumption turned out to predict time to cancer death (Jankhotkaew et al., 2020). The longer time span of 30 years compared to ours of 20 years might have made the difference. The follow-up time of 20 years might not be sufficient in a sample of 18 to 64-year-old residents. In our study, only 93 persons with cancer were found in the data analysis compared to 279 with cardiovascular disorder. Other than in the case of cardiovascular disorders, several cancers occur in male or female reproductive organs only. Females had a low proportion of values in AUDIT zones 2 to 4. Among the 122 females, 32 had cancer of their reproductive organs. This might have been a reason for the AUDIT zones not being related to cancer death. It would be a useful approach to analyze hazard ratios of death for persons with versus without alcohol use disorders among these subgroups of males and females with cancer. However, particularly among females, death cases seemed to be insufficiently prevalent to realize such an analysis. Larger samples might provide proof of a relation with time to death. Other disorders, including gastrointestinal, respiratory, and acute ones, were related to death in the person group of a rather low AUDIT score (8 − 15).

Among study participants with low risk drinking as defined by the AUDIT, those with a sum score 5 to 7 were detected by our data to die earlier than those with a sum score 1 to 4. This supports results according to which an increased risk for cardiovascular disorders such as high blood pressure exists among males with moderate alcohol consumption (Roerecke et al., 2018). Our finding confirms evidence that low drinking amounts may predict health disorders or death (Global Burden of Disease 2016 Alcohol Collaborators, 2018). The result is also in line with research of recent years that disclosed shortcomings of studies which had found lower risks of death among low to moderate alcohol consumers than among alcohol abstainers (John et al., 2021; Naimi et al., 2017; Stockwell et al., 2016; Visontay et al., 2022). Our data speak in favor of decreasing the AUDIT score for low risk drinking to 1 to 4. This corresponds to validated cut-off scores which have been found for the AUDIT (Dybek et al., 2006; Moehring et al., 2019; Rumpf et al., 2002).

According to the AUDIT domains, this study revealed that alcohol consumption may not supersede the alcohol dependence and the alcohol consequences domain. Even when all three domains are included in one model, alcohol dependence remained significant in addition to alcohol consumption. When consumption and dependence or consumption and consequences of alcohol drinking were tested in one model, both domains turned out to be related to time to death. Alcohol consequences and more so alcohol dependence seem to provide additional information in the prediction of early death. This finding adds to the discussion about the need of alcohol use disorders in the prediction of mortality (Rehm et al., 2013). Also, all combinations of the domain ranks except two were fulfilled by one or more persons. This speaks in favor of utilizing all three domains of the AUDIT.

Strengths of this study include the time between baseline and follow-up assessment being longer than 5 years. The sample was a random general adult population sample. The proportion of study participants among those who had been eligible was 70.2%. Limitations include that all baseline data have been collected by self-statements only whereby reporting bias could not be ruled out. In particular, underreporting of alcohol consumption must be considered. However, if we assume underreporting in the AUDIT, it seems all the more convincing that the relationships exist. The AUDIT had been presented by an interviewer as part of an interview. Even a random general population study as ours has the limitation that residents who suffer from alcohol use disorders may be less open to take part in an interview about health risks or complaints. Given these limitations, the results seem to be remarkably convincing in our view. Furthermore, a random adult population sample might produce less reporting bias than clinical samples because social desirability might be less strong. We cannot say if and how this may have biased the results. Moreover, this study aimed to test the predictive power of the AUDIT, not the best prediction of time to death. Therefore, smoking and other behavior-related risks and socioeconomic status data have not been considered. The AUDIT has been developed as a screening instrument for worldwide use. It has to be considered that our data are from a region in one European nation only.

Conclusions

Findings of this general adult population study with a mortality follow-up of 20 years suggest that the AUDIT predicts total and cardiovascular mortality in a dose–response manner. Among the low risk drinking group of alcohol consumers, increased mortality risk was found. The three domains of the AUDIT, alcohol consumption, alcohol dependence, and consequences of alcohol consumption, predicted mortality. Findings of this study confirm the practicability and usefulness of screening alcohol use disorders in a cost-saving manner. Feedback about risks of early death to alcohol consumers is corroborated by the results. They provide support to the screening of and early intervention in alcohol use disorders.

.