Is frequency of drinking an indicator of problem drinking? A psychometric analysis of a modified version of the alcohol use disorders identification test in Switzerland

Abstract

Background: To study the psychometric properties of a modified Alcohol Use Disorders Identification Test in the multilingual context of Switzerland. Methods: Data were obtained from the most extensive health survey to date in Switzerland, with a response rate of 69%. Seven core items of AUDIT were used. The first two items of AUDIT (frequency of drinking, quantity of drinking) were not used in their original form, but reconstructed from a quantity–frequency instrument (QF) measuring alcohol consumption, and categorized according to the AUDIT. The third AUDIT item (frequency of binge drinking) was similarly not used in its original form, but 8+ drinks instead of 6+ drinks was used. All 10 items of the modified AUDIT were completed by 10 321 subjects — 6677 in the German-speaking region and 3644 in the Latin-language (French, Italian) regions. The dimensionality of the modified AUDIT was modelled by confirmatory factor analysis. The contribution of each item to the total modified AUDIT score was investigated by the prevalence of positively screened subjects when items were deleted, and factor loadings of the unidimensional model. Results: In Switzerland the modified AUDIT fitted neither a unidimensional nor the hypothesized tri-dimensional model. Consumption items, especially frequency of drinking, showed almost no correlation with items measuring alcohol-related problems, but the latent constructs ‘harmful drinking’ and ‘alcohol dependence’ were highly correlated, indicating a shared underlying factor. Frequency of drinking was the item that most influenced whether an individual screened positively or not. Except for the correlation of frequency and binge drinking, results were almost identical for both linguistic regions. Conclusions: The current form of the modified AUDIT may have to be changed for screening in Switzerland, as the study raised questions about the suitability of the frequency-of-drinking item as an indicator of a screening device for alcohol-related problems in this country.

Introduction

The Alcohol Use Disorders Identification Test (AUDIT) has been developed by an expert group for the World Health Organization (WHO) as a screening instrument to detect the presence of early alcohol-related problems before dependence has occurred (Babor and Grant, 1989, Saunders et al., 1993b). Though designed primarily for use by health workers in primary care, it has also proved to be a useful screening instrument in non-clinical populations (Babor et al., 1989). Thus, in addition to its use in a variety of primary care and clinical settings, such as emergency rooms (Cherpitel, 1995), ambulatory care sections (Conigrave et al., 1995), or with people with substance-use disorders (Skipsey et al., 1997), the AUDIT has been used in other populations. Examples are often groups considered to be higher risk for alcohol problems, such as unemployed people (Claussen and Aasland, 1993), rugby players (Quarrie et al., 1996), college students (Fleming et al., 1991), and workers (Seppä et al., 1995).

AUDIT can also be integrated into general-population surveys (Saunders et al., 1993b); surveys in Australia (Fleming, 1996), Canada (Ivis et al., 2000), Finland (Holmila, 1995), and Mexico (see Medina-Mora et al., 1998) constitute examples of this use. Use of the AUDIT in regions with different drinking styles is consistent with one of the stated purposes of WHO in developing AUDIT (Saunders et al., 1993a, Babor et al., 1989): to achieve comparability between different cultures.

Prior to the survey reported here, we know of no use of AUDIT in any general-population study in Switzerland. Switzerland is of particular interest, however, in that it comprises different linguistic regions with different drinking cultures: largely Mediterranean, wine-drinking, in the French- and Italian-speaking parts; and mostly beer-drinking in the German-speaking part (Fahrenkrug and Gmel, 1998). This permits within-country tests of comparability between cultures. Drinking styles can be characterized as largely social-ceremonial and restorative in the Swiss–German areas and largely convivial in the French/Italian-speaking parts (Cahannes and Müller, 1981). To give an example from the most recent Swiss Health Survey, from which the data for this study were obtained, Swiss germanophone drinkers reported on an average 171 drinking occasions per year and French/Italian speaking 254 (F=348.5, P<0.001, η=0.181). Differences related mainly to frequency of wine consumption (German-speaking 111, and French/Italian-speaking 199; F=576.3; P<0.001, η=0.230); the difference in frequency of beer consumption was less (German 66 and French/Italian 73; F=7.0, P<0.01, η=0.026). However, French/Italian-speaking Swiss drank slightly less per occasion (24 g of pure ethanol) than the germanophones (26 g; F=29.6, P<0.001, η=0.053). These differences in drinking cultures may influence patterns of response to AUDIT items.

Various studies have tested the validity and reliability of the core instrument of AUDIT, a 10-item questionnaire (for an overview see Allen et al., 1997). Most evidence concerns concurrent validity, construct validity, discriminant validity, predictive validity and internal consistency. The concurrent validity of AUDIT showed generally high and significant correlations with other screening instruments, such as CAGE (Seppä et al., 1995), and with laboratory tests (Bohn et al., 1995, Conigrave et al., 1995, Allen and Litten, 1998). Construct-validity studies have established that AUDIT scores correlate highly and significantly with risk factors, drinking patterns and behaviour, and drinking consequences (Bohn et al., 1995, Lapham et al., 1998). Studies on discriminant validity, using different reference standards have found that AUDIT performs with high sensitivity and specificity, usually superior to other screening instruments (Bohn et al., 1995, Cherpitel, 1997, Piccinelli et al., 1997, Skipsey et al., 1997, Clements, 1998, Bradley et al., 1998, Steinbauer et al., 1998). It also scored higher on AUC (Area Under the Curve) at least for current alcohol-use disorders (Barry and Fleming, 1993). Moreover, the predictive validity of AUDIT is as good as, or even better than, that of laboratory markers (Conigrave et al., 1995). It has shown good results for internal consistency assessed by Cronbach's α, and a single-factor structure (Fleming et al., 1991, Barry and Fleming, 1993, Schmidt et al., 1995, Skipsey et al., 1997).

Despite the high α values of AUDIT, Allen et al. (1997) suggested investigating its dimensionality by factor analysis and analysis of heterogeneous samples. They suggested two main topics for further research: (1) verifying the dimensionality of AUDIT; and (2) relating differences in dimensionality to features of samples, including cultural characteristics. The present paper considers both topics. It is to be noted that the analysis takes place in Switzerland, which was not one of the original AUDIT sites. It thus permit a unique site for testing convergent validity (Saunders et al., 1993b).

A theoretical issue concerns the scoring of AUDIT. Its 10 items are grouped into three conceptually distinct domains: amount and frequency of drinking (three), alcohol dependence (three), and harm caused by alcohol (four), which may suggest a three-factor structure (Babor et al., 1989). In practice, the AUDIT has been employed mainly as a unidimensional screening instrument, using a summary measure of the 10 items (ranging from 0 to 40) with a cut-off point usually of 8+ or 11+ (Cherpitel, 1995, Skipsey et al., 1997). Thus, for a consistent decision rule it is necessary to assume a single underlying construct of AUDIT and no differential weighting of items — i.e. all items deserve equal emphasis (Wright and Feinstein, 1992).

Hence, as regards the dimensionality of AUDIT two different underlying factor structures have been suggested: either unidimensional or tri-dimensional. When a screening instrument is interpreted by adding up the scores of individual items with the same response scale, this implies: (1) unidimensionality; and (2) attachment of the same weight to each item. Without the assumption of unidimensionality, the instrument score would represent a summation of unlike entities. The second is intuitively clear, because a simple sum can be interpreted to mean that each item has the same weight of one (Lord and Novick, 1968, Wright and Feinstein, 1992). Both implicit assumptions can be tested formally with confirmatory factor analysis: items should load on a single underlying factor and should have equal loadings (Jöreskog, 1971, Jöreskog, 1979). Equality of loadings ensures that for a given cut-off point, such as 8+, it does not matter whether a problem drinker's score has been derived, for example, from high scores on the first three items, high scores on the last three items, or medium scores on all items.

A screening instrument need not always be unidimensional. If, for example, a combination of different symptoms or signs is indicative of a diagnosis, a screening instrument could be multidimensional, and one consideration in constructing AUDIT was to include items representing hypothetically distinct dimensions to extend coverage of relevant domains and its predictive validity (see also Allen et al., 1997). Multidimensionality would require: (1) homogeneity of loadings for each factor; (2) low intercorrelation of factors to assure distinct dimensions; and (3) attachment of weights to each dimension to make possible the combination of screening results from different dimensions (e.g. to determine whether a positive screening of the consumption dimension is sufficient to detect problem drinking).

Although the developers of AUDIT used exploratory factor-analysis of a 150-item interview schedule with samples of patients in six countries (Saunders et al., 1993b), its internal consistency has been the subject of only little subsequent research with factor analysis, especially confirmatory factor analysis. We know of no research with heterogeneous samples of general populations. Exploratory factor analysis has yielded inconsistent results. In a population of female drug-dependent prisoners, El-Bassel et al. (1998) found one factor accounting for 67.3% of the variance. Skipsey et al. (1997) reported a single-factor structure explaining 64% of variance in a sample of drug-dependent patients. Medina-Mora et al. (1998) found a two-factor structure in a sample of male workers. In summary, exploratory factor analysis has suggested no generally simple factor structure for AUDIT, even in homogeneous populations. On the basis of these considerations, the study reported here was designed to test the following hypotheses:

• AUDIT constitutes either a unidimensional or a tri-dimensional screening instrument in the general population of Switzerland.

• The internal factor structure of AUDIT is the same for both Swiss drinking cultures.