Alcohol consumption

Excessive alcohol consumption is a major risk factor for disease, disability and mortality and has been identified as a causal agent in more than 200 disease and injury conditions.1 Higher alcohol consumption has been found to be associated with lowered life expectancy,2 and according to the WHO,3 harmful alcohol consumption is related to approximately 3 million annual deaths globally. Among the population aged 15–49 years, alcohol has been identified as the leading risk factor for death and disability-adjusted life-years.4 Alcohol is by far the most used and misused psychoactive substance in the workforce,5 and 1–3 out of 10 employees can be characterised as risky drinkers in need for interventions,6–9 that is, having a consumption pattern that increases the risk for social, legal, medical, occupational, domestic and economic problems.10 Even though adverse consequences of alcohol tend to accumulate in concordance with increased consumption,2 4 it is far from straightforward to establish an appropriate threshold distinguishing between no/low-risk and risky drinking. Whether a particular drinking pattern or consumption level can be conceived of as risky, depends on several factors, such as: (1) effects of alcohol consumption interact with other individual characteristics, such as general health, sociodemographic, physiological and other lifestyle factors11 and (2) any level of drinking may be risky given certain circumstances, such as when being pregnant, operating heavy machinery and taking medications known to interact with alcohol.12 International drinking guidelines, often expressed in terms of a number of alcohol units during a specific time frame, vary considerably across countries, and moreover, even standard drink sizes vary internationally.12 In both research and clinical practice, thresholds for risky drinking are often applied based on scores on composite instruments, assuming a more complex relationship between alcohol and health, such as a score of 8 or higher on the Alcohol Use Disorders Identification Test (AUDIT).10 13

Alcohol can affect mood as well as cognitive and psychomotor performance. Psychopharmacological and experimental workplace simulation studies have explored effects of alcohol intoxication on performance, generally suggesting little consistent impairment at low to moderate intoxication levels (blood alcohol content (BAC) 0.01%–0.08%), while at higher BAC levels (≥0.09%) impairment seems to increase quite linearly with task complexity.14–17 For comparison, one standard UK drink approximates a BAC of 0.02% for a male (age: 40 years; body weight: 80 kg) or 0.04% for a female (age: 40 years; body weight: 60 kg).18 For both, a BAC of ≥0.09% would be surpassed after three drinks. In a 6-hour time window, a BAC of ≥0.09% would be present after nine (male) or six (female) drinks. Hangover episodes, defined as an adverse mental and physical state experienced after heavy drinking when the BAC level returns to zero (p.85)5 include symptoms that may be related to performance decrements, such as headache, nausea, drowsiness and sensitivity to light/sound.15 19 20

Alcohol consumption may influence activity performance in a variety of domains, including the occupational sphere. Regarding employees’ alcohol consumption, one may distinguish between workforce overall alcohol consumption (consumption regardless of context) and work-related alcohol consumption (consumption prior to or during the workday, as well as in contexts directly related to the work environment or the employment relationship).5 21–23 According to Frone’s integrative conceptual model of employee substance use and productivity, not showing up at work (absenteeism) and arriving late at work (tardiness) are primarily believed to be affected by off-the-job drinking, while leaving work early and reduced work performance are thought mainly to be due to on-the-job drinking, that is, drinking within 2 hours before work, during breaks or while performing the job.5 24 However, the model does allow for possible cross-over effects between contexts. Off-the-job drinking ‘may indirectly affect performance outcomes to the extent that it causes off-the-job substance impairment, which when carried into the workplace becomes workplace impairment’ (p. 134).5 An association between employees’ alcohol consumption and absenteeism is quite well established in the literature,25 while alcohol-related presenteeism stands out as a far more under-researched topic.

Presenteeism

Presenteeism has been defined in a variety of ways and the concept somewhat suffers from a ‘definitional creep’ (p. 521).26 Two distinct traditions in presenteeism research have been identified.26 27 The first tradition has primarily emphasised the exploration of presenteeism determinants and studied presenteeism as a chosen behaviour or personal choice. In this perspective, presenteeism is defined as the act of ‘showing up for work even when one is ill’ (p. 519)26, or ‘the phenomenon of people who, despite complaints and ill health that should prompt rest and absence from work, are still turning up at their jobs’ (p. 503).28 Hence, presenteeism may be conceived as an alternative to absenteeism and, as such, even as a health-promoting measure within a return-to-work framework.29 The second tradition has been more oriented towards consequences of this behaviour, in particular related to productivity loss. Researchers in this tradition have defined presenteeism as ‘decreased on-the-job performance due to the presence of health problems’ (p. 548)30, ‘the health-related productivity loss while at paid work’ (p. 351)31, or ‘the measurable extent to which health symptoms, conditions and diseases adversely affect the work productivity of individuals who choose to remain at work’(p. 2).32 Evidently, the first tradition treats presenteeism as a behaviour, regardless of its consequences, while the second tradition claims that adverse performance outcomes are inherent in the conceptualisation of presenteeism.

It is plausible to conceive that a variety of health conditions do not result in productivity impairment, and from an organisational perspective, it may be argued that situations in which employees attend work while sick become of interest primarily when performance decrements are involved. In this systematic review, we consider presenteeism as reduced on-the-job performance due to health problems.30 As such, presenteeism constitutes a link between on-the-job productivity and employee health,30 addressing the grey area between optimal work performance and the absence of productivity (ie, absenteeism).26 Within this frame, alcohol-related presenteeism can be conceptualised as the presence of a positive association between alcohol consumption and impaired work performance (or conversely as a negative association between alcohol consumption and work performance). Alcohol-related presenteeism is thus operationalised as the product of a relationship between two variables (exposure: alcohol consumption; outcome: work performance) rather than a single variable (attending work while sick), rendering it possible to retain the notion of work performance as inherent in the phenomenon of presenteeism without conflating cause and effect.

Performance outcomes at work comprise several phenomena related to productivity. The concept of presenteeism is most directly associated with task performance. However, performance may as well be related to contextual performance (such as working extra hours and helping coworkers), counterproductive behaviour (such as workplace aggression and property damage) and issues related to job safety, such as injuries resulting from accidents (p. 132).5 A recent Norwegian study revealed that employees’ alcohol consumption was a major concern relating to safety issues,33 and several studies support an association between alcohol and occupational injuries.34–36 However, in the context of the present review, we focused on work performance related to task performance, which can be conceived of as most directly related to on-the-job productivity.

Absenteeism and presenteeism have been found to be moderately correlated and related by baseline presenteeism being a risk factor for future absenteeism.37 Several authors have argued that presenteeism may carry more substantial societal costs than absenteeism. Hemp stated that ‘the illnesses people take with them to work (…) usually account for a greater loss in productivity because they are so prevalent, so often go untreated, and typically occur during peak working years. Those indirect costs have long been largely invisible to employers’ (p. 2).38

Known predictors of presenteeism include diseases and disorders (eg, musculoskeletal problems, depression and anxiety), certain individual characteristics (eg, gender, age, job satisfaction, stress and family status) and factors related to the organisational environment (eg, employment security, work schedules, workload, managerial support, corporate culture and leadership style).27 Knowledge of mechanisms underlying presenteeism is, however, still quite limited. In particular, the impact of individual health risks or combinations of risks should be researched more extensively.30

Rationale and aim

Some studies have explored alcohol-related presenteeism, either directly or indirectly. There is, however, a lack of synthesised knowledge, rendering it difficult to assess the evidence of a possible association between employee alcohol consumption and work performance. In their review of relationships between psychological, physical and behavioural health and work performance, Ford et al found alcohol consumption to be weakly associated with work performance problems.39 However, this conclusion was based solely on 12 studies identified in two scientific databases in 2011. It seems imperative to generate new accumulated knowledge in order to aid in deciding whether and how workplace interventions and Workplace Health Promotion Programs (WHPP) should include an emphasis on alcohol consumption.

The aim of this review was to explore whether evidence in the research literature supports the notion of alcohol-related presenteeism, that is, whether evidence supports an association between employee alcohol consumption (overall, as well as work related) and impaired work performance.

Protocol and registration

This review is registered in the International prospective register of systematic reviews and is part of the Norwegian national Workplace Interventions preventing Risky Use of alcohol and Sick leave (WIRUS) project. Original research from the WIRUS project is published elsewhere.9 23 40

Eligibility criteria

Studies exploring alcohol-related presenteeism, that is, the relationship between alcohol consumption (exposure) and work performance (outcome) among employees (population), were included in this review. Included studies had to satisfy the following criteria: (1) type of study (observational study, eg, case–control, prospective cohort or cross-sectional study); (2) type of participants (the study reported results from a sample of employees, defined as all salaried persons between 16 and 70 years of age, both workers and managers, regardless of employment sector or branch); (3) type of measures/tests (the study reported one or more statistical test(s) of a relationship between a measure of alcohol consumption and a measure of work performance); (4) type of publication and language (the study was reported as a full-text empirical research article published in English or a Scandinavian language in a peer-reviewed scientific journal); and (5) time (the study was published year 1990 or later).

Studies were excluded if they (1) reported results from samples in which employees were mixed with other groups (eg, full-time students and unemployed), unless results were reported independently for each group and/or (2) reported tests where alcohol and/or work performance were analysed in combination with other factors (eg, if on-the-job performance was analysed in combination with absenteeism within a wider productivity variable). Time restrictions were set a priori due to drinking behaviour, in particular, resulting from complex and interacting antecedents that are susceptible to changes over time.24 41 42 Hence, very old studies may suffer from low external validity.

Literature search

A primary database search strategy (based on a MEDLINE structure; see online supplementary file 1) was developed and applied in seven scientific databases (MEDLINE, Web of Science, PsycINFO, CINAHL, AMED, Embase and Swemed+). Where necessary, the search strategy was adapted to each database. The primary (MEDLINE) strategy comprised a total of 31 steps, of which 20 were abstract-level text searches, 7 were based on Medical Subject Headings (MeSH) terms (Medical Subject Headings, topics or similar terms), and the remaining were combinations of results applying Boolean operators (OR; AND). First, studies relating to the population (employees) were searched for (employee*; employed; worker*; workforce; work [MeSH]; employment [MeSH]), followed by studies relating to the exposure (alcohol consumption) (alcohol*; drink*; drunk*; hangover; “hang over”; alcohol drinking [MeSH]; binge drinking [MeSH]; drinking behaviour [MeSH]) and the outcome (work performance) (presenteeism; “job productiv*”; “work productiv*”; “job capacity”; “work capacity”; “job ability”; “work ability”; “job impair*”; “work impair*”; “job performance”; “work performance”; presenteeism [MeSH]; work performance [MeSH]). Finally, search blocks for population, exposure and outcome were combined. Database search results were transferred to EndNote.

No restrictions were imposed at the search stage. The primary search strategy was pilot tested by three reviewers prior to conducting the main searches. Databases were initially searched in September 2017. An updated search was conducted in October 2018. Additionally, reference lists in included studies were hand searched for potential relevant studies.

Study and data selection

After searching the seven databases, hand searching in reference lists in included studies and removing duplicates, identified studies were screened for relevance on a title/abstract level. Study selection was based on the results of combining the three main search blocks in the database search strategy (population, exposure and outcome). For quality assurance of the search strategy and eligibility criteria, the first 20 studies were independently screened by three reviewers. The remaining studies were independently screened by two reviewers. Initial disagreements on eligibility were resolved through discussion. The reviewers reached consensus. Hence, it was not necessary to consult with a third reviewer. Potentially relevant studies were independently assessed in full-text format for eligibility by two reviewers. Initial disagreements were resolved through discussion, without the need for consulting a third reviewer.

Data extraction

Data from the included studies were extracted independently by two reviewers. Disagreements were resolved through discussion, without the need to consult a third reviewer. We were unable to locate standardised extraction forms appropriate for this review. Therefore, we developed and applied two extraction forms.

First, on a study characteristics extraction form, the following pieces of information were extracted from each included article: title, author(s), year of publication, characteristics of study sample, study setting, number of participants included in the study (study sample size), gender and age distribution, study design, data collection method(s), information on the measures of exposure and outcome and the number of tested associations relevant to the review research question. Second, on an association characteristics extraction form, the following pieces of information were extracted about each relevant association: type of statistical test, number of participants included in association (association sample size), effect size, p value and/or CI and information on the measures of exposure and outcome. Extracted data were entered in spreadsheets for further analysis.

Quality assessment

Searches indicated that studies fulfilling the inclusion criteria were characterised by different designs and by containing several statistical associations between alcohol consumption and presenteeism. Included studies were characterised by exploring broader aims related to health and productivity, while this review emphasises the relationship between alcohol and work performance in particular. Hence, it was deemed inappropriate to conduct overall quality assessment of each study. Instead, relevant tested associations in the included studies were assessed on two key domains: (1) sample size (low quality=<500; moderate quality=500–999; high quality=≥1000) and (2) risk of confounding (level of adjustment, the extent to which associations between exposure and outcome were controlled for possible confounding variables: low quality=unadjusted or unclear; moderate quality=adjusted for individual or work-related/environmental factor(s); high quality=adjusted for individual and work-related/environmental factors). The sample size thresholds were based on the assumption that alcohol-related presenteeism is a relatively low-prevalent phenomenon in the workforce. The study of rare events requires greater statistical strength than the study of frequent events.43 Samples consisting of less than 500 observations were defined as small. Sample size categorisations were similar to thresholds applied in a recent association-based review of alcohol-related absenteeism.25 Each association was ascribed an overall quality judgement (low, moderate or high) based on the assessment of the two key domains, according to the ‘worst score counts’ algorithm recommended by the COSMIN (COnsensus-based Standards for the selection of health Measurement INstruments) guidelines.44 Hence, an association’s overall score was equal to its lowest domain assessment. High-quality associations were thus characterised by being based on at least 1000 observations and being adjusted for individual (eg, gender, age, personality, disease conditions and drug use) as well as work-related/environmental factors (eg, work position, work schedule and job characteristics).

The quality assessment procedure was pilot tested on a random sample of 10 associations. Quality assessments were performed independently by two reviewers. Consensus was reached, and initial disagreements were resolved through discussion, without the need for consulting a third reviewer.

Analysis

Measures of exposure (alcohol consumption) as well as measures of outcome (work performance) displayed considerable heterogeneity between the included studies. As a result of the heterogeneous nature of the included data, meta-analyses were deemed inappropriate. Included data (associations) were instead analysed with frequency tables and cross-tabulations. First, associations were sorted into a frequency table by quality level and overall association characteristics. Next, four contingency tables were constructed in order to explore properties of the identified associations more thoroughly: (1) direction and significance, (2) quality and direction, (3) publication year and quality and (4) significance and quality. The four 2×2 tables were analysed by means of ORs (with 95% CIs) and χ² tests of independence (with phi coefficients). Finally, measurements of alcohol consumption and work performance applied in the included studies were categorised into subgroups.

Patient and public involvement

No patients or public were involved in this review study.

Overview of the evidence

Searches in the seven databases resulted in 540 articles (MEDLINE: n=135; Web of Science: n=128; PsycINFO: n=63; CINAHL: n=22; AMED: n=3; Embase: n=189; Swemed+: n=0). Hand searching in reference lists resulted in an additional nine articles. After duplicate removal (n=282), a total of 267 unique articles remained. Application of the eligibility criteria resulted in exclusion of 158 studies, leaving 109 potentially relevant articles.

Eighty-three studies were excluded after being subjected to full-text assessment. The vast majority of these were excluded as a result of not reporting a statistical test of an association between alcohol consumption and work performance (n=52) or because of publication type (n=24). Articles not reporting tests of associations were typically characterised by: (1) not studying variables that conceptually could be defined as alcohol consumption and/or work performance and (2) analysing alcohol consumption and/or work performance in combination with other factors, rendering it impossible to isolate the association of interest. Alcohol being analysed in combination with smoking/other lifestyle factors and work performance being analysed in combination with absenteeism constitute typical examples. Articles excluded on the basis of publication type were typically conference papers. The study selection process resulted in 26 studies satisfying all inclusion criteria and is presented in figure 1.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

PRISMA flow chart of the study selection process. PRISMA, Preferred Reporting Items for Systematic Reviews and Meta-Analyses.

The 26 included studies were based on data from 92 730 employees from a total of 15 countries (Australia, China, Czech republic, Denmark, Finland, Greece, Ireland, Japan, the Netherlands, Norway, Portugal, Slovenia, Sweden, Switzerland and the USA). Employees in the USA constituted the samples in half of the studies (13 of 26). The vast majority of studies (21 of 26) were based on cross-sectional research designs. A total of 132 associations between alcohol consumption and work performance were tested in the 26 included studies. Characteristics of the included studies are presented in table 1. Characteristics of the included associations are presented in online supplementary file 2.

Quality of the included data

Ninety-three of the 132 associations (71%) were based on samples smaller than 1000 employees. Approximately half of the associations were unadjusted (n=63; 48%), while 29 associations (22%) were adjusted for individual factors as well as for work-related/environmental factors. By applying the ‘worst score counts’ algorithm, 80 associations (61%) were judged as being of low quality, 38 associations (29%) were of moderate quality, while 14 associations (11%) were characterised by high quality. Results from quality assessment of the included associations are presented in online supplementary file 3.

Direction, significance, quality and time

One hundred and two of the 132 tested associations (77%) indicated a positive relationship between alcohol consumption and work performance, that is, implying that higher levels of consumption were associated with higher levels of performance impairment. Approximately half of these (n=56, 55%) were statistically significant. The majority of positive associations was judged to be of low quality (n=70, 69%), followed by moderate (n=23, 22%) and high quality (n=9, 9%). For instance, in a sample of employees in the USA, Kirkham et al45 found that risky drinking, as measured with the CAGE questionnaire,46 was associated with impaired work performance, measured with the Work Limitations Questionnaire,47 both overall (ID36, β=0.20, p<0.001) as well as among those aged <45 years (ID37, β=0.22, p<0.001) and ≥45 years (ID38, β=0.20, p<0.001). Among Finnish employees, Pensola et al48 found that high hangover frequency (at least six hangovers during the past 12 months), compared with low frequency (no alcohol or less than six hangovers during the past 12 months), was associated with moderate or poor self-reported work ability (ID41, PRR (prevalence rate ratio)=1.15, 95% CI: 1.0 to 1.3). In a study of Norwegian employees, Aas et al40 found that higher binge drinking frequency (measured with a single item from the AUDIT)10 13 was positively related to the experienced degree of impaired work performance (measured with a single item from the Work Productivity and Activity Impairment questionnaire)49 during the past 7 days (ID127, β=0.06, p<0.01).

Twenty-five of the 132 tested associations (19%) indicated a negative relationship, that is, implying that higher levels of alcohol consumption were associated with lower performance impairment (higher work performance). Only two of these associations were statistically significant, and both of these were of low quality. These two associations (ID66, r=0.10, p<0.01, and ID68, r=0.09, p<0.01, in Friedman et al50) tested the relationship between duration of alcohol use and overall work performance and found that longer duration, as opposed to shorter duration, was associated with higher work performance.

Five associations (4%) were not possible to classify as either positive or negative. They were characterised by: (1) finding no differences in work performance between compared alcohol consumption groups (ID102, M_diff=0.0, p=0.68, in Moore et al51; ID130, OR=1.00, p=ns, in van den Berg et al52); (2) by finding significant differences between multiple consumption groups but without a consistent positive/negative pattern (ID28, unclear effect size, p<0.001), and ID29, unclear effect size, p=0.03, in Kim et al53); or (3) by finding a J-shaped pattern where abstainers scored comparable with moderate-level drinkers on impaired performance (ie, higher than low-level drinkers) but still lower than heavy drinkers (ID98, unclear effect size, p<0.05, in Moore et al51). The identified associations, sorted by quality level and overall association characteristics, are presented in table 2.

Positive associations were considerably more likely than negative associations to be statistically significant (OR=14.00, 95% CI 3.1 to 65.5; χ² (1, n=127)=17.80, p=0.000, phi=0.37). However, negative associations were less likely than positive associations to be of low quality (OR=0.22, 95% CI 0.1 to 0.6; χ² (1, n=127)=11.37, p=0.001, phi=−0.30). Furthermore, recent studies (≥year 2000) were more likely than older studies (<year 2000) to be of moderate or high quality (OR=2.95, 95% CI 1.30 to 6.79; χ² (1, n=132)=6.96, p=0.008, phi=0.23). There was no significant relationship between whether associations were significant and whether they were of moderate/high or low quality. The four 2×2 contingency tables are presented in table 3.

Measurements of alcohol consumption and work performance

Categorisation of the applied measurements of alcohol consumption in the 26 included studies revealed eight subgroups: (1) consumption status (eg, current alcohol drinker (yes/no), applied in Yu et al54); (2) drinking frequency (eg, number of times drunk during past 3 months, applied in Ames et al21; typical frequency of alcohol consumption during past year, applied in Aas et al40); (3) drinking intensity (eg, average number of alcohol drinks during the past week, applied in Adler et al55); (4) drinking volume (eg, monthly frequency × typical quantity during past 30 days, applied in Blum et al56); (5) binge drinking (eg, binge drinking (six or more drinks on a single occasion) frequency during past year, applied in Aas et al40); (6) hangover (eg, frequency of hangover episodes at work during past year, applied in Ames et al21); and (7) composite instruments comprising several aspects of consumption, such as frequency, intensity and alcohol problems (eg, the AUDIT,10 13 applied in Richmond et al57); and (8) alcohol-related diagnosis (eg, DSM-IV diagnosis of alcohol abuse, applied in Lim et al58).

The 26 included studies contained a total of six work performance measurement categories: (1) overall work performance/impairment (eg, supervisor ratings of overall work performance, applied in Lowmaster and Morey59; self-reported current work performance compared with lifetime best, applied in Furu et al60; Work Limitations Questionnaire sum score,47 applied in Kirkham et al45); (2) domain-specific work performance/impairment (eg, Work Limitations Questionnaire subscale Time management,47 applied in Adler et al55); (3) impaired performance quantity (eg, number of days working below a normal level of performance during past 12 months, applied in Fisher et al61; estimated per cent impaired performance during past week, applied in Boles et al62); (4) impaired performance frequency (eg, frequency of impaired performance episodes during past 12 months, applied in Schou et al63); (5) prognosis of work performance (eg, self-assessed probability of good work performance within frame of 6 months, applied in Karlsson et al64); and (6) work performance status (eg, impaired work performance during past 4 weeks (yes/no), applied in Yu et al54). The identified associations, sorted according to measurements of alcohol consumption and work performance, are presented in table 4.

In the 132 included associations, the most frequently applied alcohol measurement was drinking intensity (n=28, 21%) and composite instruments (n=27, 20%). Overall work performance/impairment (n=67, 51%) and quantity of impaired performance (n=35, 27%) were the most frequently utilised work performance measures. When exploring the group of associations characterised by being significant positive and of moderate or high quality (n=18), the vast majority of these (n=15) applied either hangover (n=9) or composite instruments (n=6) as alcohol consumption measures.

Implications

Overall, this review provides support for the notion of alcohol-related presenteeism, that is, that employee alcohol consumption may be associated with performance decrements at work. Research has, although often demonstrating somewhat mixed results, shown that employees’ alcohol consumption is related to occupational outcomes, including absenteeism and occupational injuries.25 34–36 The results of this review on alcohol-related presenteeism imply that impaired work performance may be an additional detrimental occupational outcome related to alcohol consumption. As such, this review provides further support for targeting alcohol consumption within workplace interventions aimed at improving employee health and productivity, rather than implying that interventions should specifically target presenteeism behaviour. Further research is necessary for determining whether and how presenteeism should be targeted directly in interventions.

It is not possible to draw firm conclusions regarding the relationship between alcohol consumption and work performance. The majority of identified evidence was of low quality as a result of low power (small sample sizes) and/or risk of confounding. Moreover, the majority of identified studies were cross-sectional, and thereby unable to draw causal inferences about the relationship between exposure and outcome. Above all, this review implies the need for further research. First, future research would benefit from studying alcohol-related presenteeism by means of more robust study designs that better enable exploration of causal mechanisms and development over time. A more thorough exploration of alcohol as a risk factor for impaired work performance could be done by means of retrospective case–control studies, where historical data sources containing information on alcohol consumption (such as medical records) are used in order to compare work impaired (cases) with non-impaired employees (controls). How the relationship between alcohol and work performance develops over time can be explored with prospective cohort studies, where researchers can follow and compare risky and non-risky drinkers with repeated measurements of work performance.

Second, both alcohol consumption and work performance are conceptualised and measured very differently across current studies. Such heterogeneity makes it difficult to explore findings in the literature by means of meta-analyses. Progress in the field seems to hinge on researchers’ ability to reach more agreement on how to conceptualise these variables and measure them using instruments with satisfactory psychometric properties. This seems particularly true for the concept of presenteeism. According to an expert panel from the American College of Occupational and Environmental Medicine (p. 351),31 productivity instruments should be supported by scientific evidence, be applicable to the specific work setting, support decision making and be practical. Ospina et al73 concluded that the following three instruments were most strongly supported by evidence: The Stanford Presenteeism Scale (six-item version),74 the Endicott Work Productivity Scale75 and the Health and Work Questionnaire.76 Regardless of design, future research would benefit from measurement triangulation. For instance, alcohol consumption could be measured with a validated self-report composite measure (eg, the AUDIT measuring both consumption and alcohol-related harm, or the abbreviated AUDIT-C measuring only consumption),10 13 items separating off-the-job and on-the-job drinking and hangovers, and an alcohol biomarker test (such as the carbohydrate-deficient transferrin test). Work performance could be measured with a validated self-report composite instrument (eg, the Stanford Presenteeism Scale),74 as well as with supervisors’ ratings of employee work performance and, where possible, register data on task performance. Measurement triangulation may provide more valid measures as well as enabling exploration of a potential correspondence between consumption contexts, impairment contexts and performance outcomes.

Third, future research would benefit from taking possible mediators and moderators of the relationship between alcohol and work performance into account, such as sociodemographic, general health, work related and other lifestyle factors.

Methodological considerations

This review has some limitations. First and foremost, due to the heterogeneous nature of the identified data, we were unable to perform meta-analyses on the included data.

Second, it may be considered a limitation that this review used associations and not studies as the unit of interest. Associations were deemed the appropriate unit of interest in this review for two reasons: (1) included studies were characterised by exploring broader aims related to health and productivity, while this review specifically aimed at exploring the relationship between alcohol consumption and work performance and (2) in several studies, multiple associations between alcohol consumption and work performance were tested (often with different measures and subgroups within each study).

Third, this review did not use a previously validated critical appraisal tool (CAT) for assessment of included primary research. One reason for this is that studies based on different study designs were included in the review. At present, there exists no generic gold standard CAT for application across study designs.77 78 A second reason is that the current review emphasised associations rather than studies as the unit of interest. Hence, it was deemed more appropriate to develop a parsimonious and conservative quality assessment system in which each association was evaluated based on power (sample size) and risk of confounding (level of adjustment). Deliberately, we chose a conservative approach to quality assessment by ascribing each association an overall score in accordance with the ‘worst score counts’ algorithm. Such an approach is in line with the COSMIN guidelines.44

Fourth, articles published before 1990 were not eligible for inclusion in this review. This exclusion criterion was set a priori as a result of old studies having limited external validity due to changes in drinking behaviour over time. Time restrictions were imposed at the study selection stage, not in the literature search phase of the review. This decision was made in order to be able to assess the magnitude of potentially relevant research published prior to 1990. Seventeen articles from the 1980s were excluded in the title/abstract screening. However, these articles did not satisfy all the other inclusion criteria and were, thus, not exclusively excluded based on year of publication. Hence, we do not find it very likely that relevant studies published before 1990 have been missed.

Fifth, we chose to use the concept of presenteeism in line with researchers who define it in terms of decreased on-the-job productivity due to health problems.30 Such an understanding does ascribe valence to the phenomenon, that is, a behaviour contributing to lost productivity that may carry negative influence on the overall work environment.79 We are, however, aware of differing opinions among scholars regarding conceptualisations of presenteeism. Different definitions have different strengths and weaknesses. According to Johns,26 a proper definition should: (1) neither ascribe motives nor consequences to presenteeism and (2) avoid conflating cause and effect by perceiving productivity loss itself as presenteeism. To some extent, we do agree with such objections against a productivity-based definition. A more open understanding, such as simply ‘showing up for work even when one is ill’ (p. 519),26 does not ascribe a certain valence to the phenomenon, nor does it presuppose or exclude any particular consequence. We believe, however, that in a socioeconomic and organisational perspective, situations in which employees attend work while ill become of interest primarily when performance decrements are in fact involved. In order to avoid conflating cause and effect, we operationalised alcohol-related presenteeism as the product of a relationship between two measurable variables, that is, alcohol consumption (predictor/exposure) and work performance (outcome).