PREDICTORS FOR HIDDEN PROBLEM DRINKERS IN GENERAL PRACTICE

Transcription

1 Alcohol & Alcoholism Vol. 3, No. 3, pp , 1996 PREDICTORS FOR HIDDEN PROBLEM DRINKERS IN GENERAL PRACTICE MICHIEL CORNEL*, RONALD A. KNIBBE 1, J. ANDRE KNOTTNERUS, ALEX VOLOVICS 2 and MARIA J. DROP 1 Departments of General Practice, 'Medical Sociology and Statistics, University of Limburg, P.O. Box 616, 6200 MD Maastricht, The Netherlands (Received 20 July 1995; in revised form 19 January 1996; accepted 22 January 1996) Abstract In The Netherlands general practice attenders are not usually questioned about their drinking habits. The objective of this study was to determine to what extent easily available data (e.g. age, gender) can be used to identify categories of patients who are at risk of problem drinking as a preliminary to more intensive screening. Sixteen practices with a total population of 32,000 patients were involved in the study. All problem drinkers known by their GPs and a random sample of one in ten patients not thought to be problem drinkers were admitted to the study at their first surgery visit during a 1year period. A screening questionnaire was used to find hidden problem drinkers amongst the individuals thought to be nonproblem drinkers. The overall response rate was 91% (n = 1405). Problem drinking was detected in 6% (n = 82) of the group regarded by the GPs as nonproblem drinkers (n = 1283). Male gender, smoking, life events and chronic social problems were the strongest nonalcoholrelated predictors of hidden problem drinking. We conclude that a preselection of patients with a greater risk of problem drinking can be made without information related directly to alcohol. Casefinding in this category is much more effective and probably much more acceptable both to the GP and the patients, than the screening of all patients. INTRODUCTION It has been suggested that the general practitioner (GP) is in a good position to identify and treat problem drinkers. The GPs are in close contact with their patients, which enables them to identify problem drinkers at an early stage. Early detection tends to lead to better treatment outcome, perhaps even with minimal intervention (Wallace et al, 1988; Anderson and Scott, 1992). Prevalence estimates from representative Dutch population surveys show that ~ 10% of the population can be identified as problem drinkers (Garretsen and Knibbe, 1985). Morbidity figures from general practice indicate that ~ 12% of the general practice population are diagnosed as problem drinkers (Lamberts et al., 1987; Registration Network Family Practices, 1992). The gap between population survey estimates and GPs' estimates can partly be explained by the perhaps more restrictive definition of problem drinking used by GPs. However, it is quite likely Author to whom correspondence should addressed. that GPs do not recognize a substantial proportion of their patients who are problem drinkers. Doctors tend to stereotype problem drinkers and most of them are not trained to talk to their patients about problem drinking. They seldom give advice on the misuse of alcoholic beverages and its consequences. Patients are generally afraid of being labelled as alcoholics, which has negative connotations. There are, moreover, no adequate diagnostic means for the early detection of problem drinking (Cornel and Van Zutphen, 1989). Laboratory markers are not valid for detection (Wallace and Haines, 1985; Beresford et al., 1990). The use of questionnaires for screening purposes does not easily fit, as a first step procedure, into the working style of the GP. This paper concentrates on characteristics of hidden problem drinkers which GPs can usually observe during normal practice. Studies on characteristics of problem drinkers in general practice populations have so far been largely restricted to those already recognized as problem drinkers (Hore and Wilkins, 1976; Buchan et al., 1981; Rush and Brennan, 1990). Buchan et al. (1981) and Rush and Brennan Medical Council on Alcoholism

2 288 M. CORNEL et al. (1990), however, did use a control group of nonproblem drinkers. It is questionable whether data from known problem drinkers apply to problem drinkers not identified as such by their GP and hence whether these data can help in the detection process. Only one study has used a questionnaire to detect unrecognized problem drinkers and compared them with a control group (Nicol and Ford, 1986). The abovementioned studies mainly concentrated on health profiles. Reasons for encounter and diagnoses in general practice are of little use in the detection of hidden problem drinkers (Cornel et al, 1995). However, more easily available data, like gender, age, life events and social problems, might also be used by the GP to identify the category of patients with a greater risk of problem drinking (e.g. hidden problem drinkers). If such characteristics appear to be related to problem drinking, GPs will be able to identify, in a nonobtrusive way, a category of patients with a clearly increased risk of problem drinking. This cohort of patients can then be asked in more detail about their drinking and drinkingrelated problems, possibly with the help of a questionnaire. Compared with a strategy in which all patients are screened, the potential advantages of this more restrictive approach are obvious. Not only is such a strategy likely to be more effective, it will probably be more acceptable to both the GP and the patients. The central question of this study is to what extent easily available data contribute to the identification of patients with a higher risk of problem drinking. In our analysis, we distinguish four levels of information on the basis of availability to the GP. These levels include gender (level 1), life events (level 2), smoking (level 3) and reported alcohol consumption (level 4). MATERIALS AND METHODS Background This study is part of a larger observational study of problem drinkers in general practice, in which problem drinkers identified as such by the GPs, problem drinkers not identified by the GPs and true nonproblem drinkers were compared. This paper concentrates on the patients who were originally identified as nonproblem drinkers. Patients Over a period of 1 year, 16 general practitioners, with a total patient population of 32,000 people, were involved in the study. The study included patients aged 16 and over who attended during surgery hours at least once within the specified 1year period. The GPs marked all patient cards at their first visit during the 1year observation period. These individuals were then categorized as either known problem drinkers, all of whom were admitted to the main study, or as nonproblem drinkers, one in ten of whom was admitted to the study. Measurements The patients selected for the study received a questionnaire from their doctors containing questions about social background variables, drinking, smoking, drug use, physical and social wellbeing and level of alcohol consumption. The overall response rate was 91%. The questionnaire comprised a screening instrument for the detection of problem drinking based on existing screening instruments, CAGE (Mayfield et al., 1974), SMAST (Selzer et al, 1975) and SAAST (Davis et al, 1987), and some additional screening questions (Cornel et al, 1994). Scaling techniques were used to decide which of the three screening instruments, or what combination of items, could be used for our study. The first step entailed calculating Cronbach's Alphas, a measure of the reliability of a scale, for each of the three screening instruments separately and for the combination of all items on alcoholrelated problems. The second step was a Rasch analysis, a sophisticated scaling technique based on the Rasch model which aims to establish the unidimensionality of a set of questions (Gustafsson, 1977; Wright and Stone, 1979; Molenaar, 1983). The Rasch model is a stochastic model based on the Guttman scale. The latter is based on the following principle: there is a group of persons with a property to be measured, and a group of questions that measures aspects of this property. If the Guttman scale is perfect, n + x questions can be ordered in such a way that all persons with score n score positively on the n easiest questions ('easy' meaning that the aspect occurs more frequently). With a score n + 1, the same n questions are scored plus the next easiest question.

3 PROBLEM DRINKERS IN GENERAL PRACTICE 289 Table 1. The 'Rasch scale': questions in order of increasing 'difficulty' (frequency of positive scores in parentheses) 1. Have you ever felt the need to cut down on your drinking? (260) 2. Do you ever drink to forget your worries? (157) 3. Do close relatives ever worry or complain about your drinking? (118) 4. Have you ever been told by a doctor to stop drinking? (116) 5. Have people annoyed you by criticizing your drinking? (105) 6. Do you ever have a few drinks before you go to a party? (95) 7. Do you ever skip meals when you are drinking? (94) 8. Are you (always) able to stop drinking when you want to? (92) 9. Can you stop drinking without a struggle after one or two drinks? (89) 10. Have you ever had a drink first thing in the morning to steady your nerves or get rid of a hangover? (64) 11. Do you ever drink in the morning? (63) 12. Have you ever been in hospital because of drinking? Was drinking part of the problem that resulted in your hospitalization? (46) 13. Have you ever gone to anyone for help about your drinking? (45) 14. Did you ever hide your drinking? (41) 15. Do you ever drink to optimize your thinking? (35) 16. Do you ever drink in order to work better? (24) 17. Have you ever neglected obligations, your family or your work for two or more days in a row because you were drinking? (24) 18. Have you ever lost a job because of your drinking? (14) This is a very stringent requirement for unidimensionality, which does not allow for biological variation. The Rasch model has the same characteristics but, unlike the Guttman scale, does allow for random variation. It appeared that a combination of 18 questions formed a perfect Rasch scale (Table 1). When applying a cutoff point of 3 (score 3 and higher), estimated prevalence of problem drinking was 7% according to our criterion. CAGE, SMAST and SAAST (cutoff points of 2, 3 and 3 or more positive answers) yielded prevalences of 8, 6 and 6%, respectively. The Cohen's Kappas of CAGE, SMAST and SAAST for our criterion for the study population were 0.76, 0.76 and 0.82 respectively, indicating that the categories of patients identified with these instruments overlap to a large extent. Patients can be categorized into four groups (Table 2). Since the main aim of this study is to find predictors of hidden problem drinkers, this paper concentrates on the 1283 patients who were originally identified as nonproblem drinkers. The response rate in this group was 93%. Variables For purposes of presentation, four levels of information were distinguished on the basis of availability to the GP. The first level contains the basic patient characteristics gender, age, marital status (married or cohabiting versus other), living alone versus living with others, low versus medium and high educational levels, unemployment or disability pension. Age is an interval variable, the others are categorical. The second level included information obtainable from the patient record or information usually known to the GPs, such as life events ('have you recently experienced a serious or sad event that still affects you?'), chronic illness ('do you have a chronic illness or handicap?'), use of sleeping pills and tranquillizers, as well as a scale of chronic social problems and consultation rates for the observation period. Only the last two are interval variables. Chronic social problems were measured using the Longterm Difficulties Questionnaire, a 12item list including various social problems such as those concerning work, school, financial means and personal relationships (Hendriks et al, 1990). Problems were regarded as minor, intermediate or major, scoring 1, 2, or 3 points, respectively. The sumscore of all problems was used as a measure of social problems (Cronbach's Alpha = 0.66). The third level included smoking or nonsmoking, a categorical variable. This was regarded

4 290 M. CORNEL et al. Table 2. Distribution of problem drinkers according to the GPs and screening test GP assessment Positive Screening tests Negative Total Positive Negative Total as a separate level, since collecting information about smoking requires active questioning by the GPs. The fourth level contained information about drinking, e.g. the level of alcohol consumption (measured in standard glasses of 10 g 100% alcohol), drunkenness during the past year (zero versus al) and knowing people with drinking problems. Obtaining this information requires additional effort, and such questions are usually asked only after suspicion of problem drinking has been raised. Analysis These four levels of information were used in four models to predict the presence of problem drinking in the data set. The relation between each of the variables and problem drinking was assessed by simple analysis, resulting in crude odds ratios with 95% confidence intervals for the categorical variables and means for the interval variables. Differences between means were tested using the MannWhitney test. Multiple logistic regression with backward elimination of statistically nonsignificant terms from the four full models was used to select discriminating variables. Insignificant terms according to the Wald test were removed (P > 0.05), unless the variable was significant in the preceding model. Interaction terms of gender and age with all variables were selected by stepwise forward selection. Significance testing between the logistic models was performed by the likelihood ratio test. Analysis was performed using the BMDP program. RESULTS The crude analysis shows (Table 3) that, of the categorical variables, male gender, life events, use of sleeping pills, smoking, drunkenness during the past year and knowing problem drinkers were Table 3. Numbers and percentages of problem drinkers and nonproblem drinkers in relation to categorical variables, unadjusted odds ratios (OR) with 95% confidence intervals () Variables Male gender Marital status, unmarried and not cohabiting Living alone Low educational level Unemployed or disabled Life events + Chronic illness + Sleeping pills + Tranquillizer use + Smoking + Drunkenness + Knowing problem drinkers + n Problem drinkers (II = 82) % Nonproblem drinkers (n = 1201) n % Unadjusted OR ^

5 PROBLEM DRINKERS IN GENERAL PRACTICE 291 Table 4. Mean and standard deviation values of problem drinkers and nonproblem drinkers in relation to interval variables Variables Age Social problems Consultations/year Drinks/day * 8 observations missing. f 71 observations missing. Problem drinkers (n = 82) Nonproblem drinkers (n = 1201) Mean SD Mean SD * f Significance of the intergroup difference P significantly associated with problem drinking (odds ratio a2). Being unemployed or unable to work (disability pension) was also significantly associated with problem drinking but the strength of the association was smaller. Tranquillizer use, 'not being married or not living together' and 'living alone' showed a tendency to an association with problem drinking. Low educational level was not associated with problem drinking. Age and consultation rate (Table 4) did not differentiate between problem drinkers and nonproblem drinkers. Problem drinkers reported significantly more social problems and drank more alcohol than nonproblem drinkers. The results presented above concern patients who were not problem drinkers according to their GPs. Patients known by their doctors as problem drinkers (n = 122) were characterized by being mostly male (79.5%), unmarried (35.2%), living alone (30.6%), reporting important life events (47.5%), smokers (83%) and reporting drunkenness in the past year (62.5%). They had a mean age of 45.7 years, consulted 3.1 times a year and reported 7.2 drinks a day. The prevalence of known problem drinkers is 1%. In the first multiple logistic model (Table 5) only gender and age were found to be significant terms. 'Being unemployed or receiving a disability pension', which was significant in the crude analysis, was no longer significant after the other variables had been taken into consideration. The second model (Table 5), containing data that could be obtained from the patient record, was significantly better (likelihood ratio test, P < 0.01). Life events and social problems were the significant variables that added information. Chronic illness, use of sleeping pills or tranquillizers and consultation did not add any information about the probability of problem drinking. The third model (Table 6), which expanded the second model with the sole addition of smoking, was significantly better (likelihood ratio test, P < 0.01). In the fourth model (Table 6), data about consumption level and drunkenness yielded a relatively high level of information, as is apparent from the difference in log likelihood with model 3 (likelihood ratio test, P < 0.01). Age and life events were no longer significant, nor was knowing problem drinkers. Of the product terms with gender and age which were added to the full model 4, only 'gender * use of sleeping pills', 'gender * consumption level' and 'age * smoking' were found to be significant (likelihood ratio test, P < 0.01). Coefficients of these terms were 3.688, and , respectively. After inclusion of these three terms, the difference in log likelihood for the full model 4 amounted to This is a minor increase compared to the other steps. Figure 1 plots the relation between the percentage of incorrectly predicted nonproblem drinkers (false positives) and the percentage of correctly predicted problem drinkers (true positives) at different cutoff points of the estimated posterior probability for the four models. The diagonal represents the situation in which an instrument has no predictive power. As the predictive power increases, the curve moves to the top left part of the diagram. The area under the curve is a measure of the discriminative power of the model for this patient population (Hanley and McNeill, 1982). The four curves differ significantly. A relatively large improvement is seen when information about

6 to s Table 5. Logistic regression with problem drinking as the dependent variable and basic patient characteristics (model 1), in combination with information obtainable from the patient record (model 2) as independent variables Variables Gender (1 = man, 0 = woman) Age (/10 years) Marital status (not married/not living together = 1, married/living together = 0) Living alone (alone = 1, with others = 0) Education (low = 1, medium/ high = 0) Work (unemployed/disabled = 1, other = 0) Life events (1 = yes, 0 = no) Social problems (score) Chronic illness (1 = yes, 0 = no) Sleeping pills (1 = yes, 0 = no) Tranquillizer use (1 = yes, 0 = no) Consultations/year Constantt log likelihood Full model ± 1.02 (2 SD) Reduced model ± 0.84 (2 SD) * ORs have been computed as the natural antilogarithm of the logistic regression coefficient. t For the constant the logistic regression coefficient is given Full model ± 1.20 (2 SD) Reduced model ± 0.98 (2 SD) z m r^

7 Table 6. Logistic regression with problem drinking as the dependent variable and basic patient characteristics, information obtainable from the patient record and smoking (model 3), in combination with information about drinking (model 4) as independent variables Variables Gender (1 = man, 0 = woman) Age (/10 years) Marital status (not married/not living together = 1, married/living together = 0) Living alone (alone = 1, with others = 0) Education (low = 1, medium/ high = 0) Work (unemployed/disabled = 1, other = 0) Life events (1 = yes, 0 = no) Social problems (score) Chronic illness (1 = yes, 0 = no) Sleeping pills (1 = yes, 0 = no) Tranquillizer use (1 = yes, 0 = no) Consultations/year Smoking (1 = yes, 0 = no) Drinks/day Drunkenness (1 = yes, 0 = no) Knowing problem drinkers (1 = yes, 0 = no) Constantt log likelihood Full model U ± 1.30 (2 SD) Reduced model ± 1.07 (2SD) * ORs have been computed as the natural antilogarithm of the logistic regression coefficient. t For the constant the logistic regression coefficient is presented. OR' Full model ^ ± 1.79 (2 SD) Reduced model ± 1.45 (2 SD) *o 73 fs w i 2 z 73 t/2 Z o m zm» sa 9 o m

8 294 M. CORNEL et al Model 1 Model 3 % false positive Model 2 Model 4 Fig. 1. Prediction of problem drinking in the study population with the help of the 4 logistic models. The percentage of incorrectly predicted nonproblem drinkers (false positives) is plotted against the percentage of correctly predicted problem drinkers (true positives) at different cutoff points of the estimated posterior probability for the four models. As the predictive power increases so the curve moves to the top left of the diagram. Model 1; model 2; model 3 and model 4. drinking is added. DISCUSSION Our findings support the hypothesis that background data and data from the patient record are useful for the GP in detecting problem drinkers, even if they have no direct information regarding their patients' alcohol consumption. In order to find discriminating variables for hidden problem drinkers in this GP population, those problem drinkers already identified as such by the GPs were excluded. The crude analysis shows that some 'nonalcohol variables' were clearly related to problem drinking. Among the dichotomous variables, male gender had the highest odds ratio (32), followed by smoking, life events and the use of sleeping pills. The prevalence of chronic social problems was also significantly greater amongst problem drinkers. Buchan et al. (1981) and Rush and Brennan (1990) found that patients identified by their GPs as problem drinkers had higher consultation rates. We observed no difference in consultation rate between unrecognized problem drinkers and nonproblem drinkers. Buchan's and Rush's findings for known problem drinkers as regards gender, social problems and smoking showed the same trend as ours. Nicol and Ford (1986), who detected unrecognized problem drinkers by means of a questionnaire, found only unemployment to be

9 PROBLEM DRINKERS IN GENERAL PRACTICE 295 significantly associated with alcohol problems. They found no association with presenting complaints. The relation between variables directly related to drinking, such as consumption level, drunkenness and knowing problem drinkers, was clearly demonstrated in our study. The multiple logistic models, taking the relationships between the separate variables into consideration, showed a gain in information from model 1 to 4. Model 1, with only social background variables, was supplemented with data from the patient record in model 2 and with smoking in model 3. Model 4 also contained information about drinking. A relatively important factor was information about smoking. The largest amount of extra information, however, was gained by asking about drinking behaviour. Three interaction terms, 'gender * use of sleeping pills', 'gender * consumption level' and 'age * smoking' were found to be significant, although the information gain was small. For predictive purposes, interaction terms complicate the model unnecessarily. The results of the logistic regression allow the risk of problem drinking in the study population to be estimated for any given combination of parameters. Knowing only gender and age, it can be estimated, with the help of model 1, that a man aged 30 has a 13% probability of being a problem drinker, while a woman of the same age has only a 2% probability. Model 3 can be used to estimate that a 30yearold man who smokes, who has experienced a life event in the past year and who has a score of 5 on the social problem list, has a 42% probability of being a problem drinker. This probability is high enough for serious suspicion of problem drinking, so that the GP should ask specifically about drinking and/or drinkingrelated problems. A woman with the same profile has an 8% probability of being a problem drinker, but if she does not smoke the probability decreases to 2%. Gender is the most important general determinant not apparently related to drinking. It should be borne in mind, however, that the study population included few female problem drinkers (n = 13). Thus, conclusions about this population may not be fully applicable to a female population. Caution seems justified when this model is used to support the detection of female problem drinkers. Although several risk factors result in the same relative increase in the probability of being a problem drinker as for males, the absolute probability of being a problem drinker remains small. Asking about drinking adds a great deal of information. According to the model, the man from model 3, with the additional information that he has been drunk during the past year and has 10 drinks a day, has a 72% probability of being a problem drinker. The prevalence of hidden problem drinkers was 6% in this study population. The predictive power of the abovementioned variables depends in part on the prevalence of problem drinking in the population. A higher prevalence, for instance in a population of hospitalized patients, would result in higher positive predictive values and lower negative predictive values. In a population with a lower prevalence of problem drinking the opposite would occur. Depending on the GPs' working style they can ask every patient directly about drinking problems or drinking behaviour or they can perform casefinding by limiting the questions to a category of patients that are more at risk. If a GP finds it difficult to ask a patient about drinking, they might introduce the problem by informing a patient about the relation between life events, social problems, smoking and problem drinking in order to initiate discussion. Compared with other studies the distinguishing characteristic of our study is that it shows the feasibility of restricting screening to a selection of patients. This outcome does not disqualify the use of validated questionnaires like CAGE (Mayfield et ai, 1974; Wallace and Haines, 1985), MAST (Selzer et ai, 1975), SAAST and AUDIT (Saunders et ai, 1993). However, whereas in these studies all patients were screened, our study shows that restriction of screening to a selection of respondents is feasible. In our opinion this outcome greatly enhances the willingness of the GPs to be more methodical about the detection of problem drinkers among their patients. Acknowledgements We are indebted to the late Wim van Zutphen, who was the initiator and first leader of the project. We would like to thank H. M. J. Jochems, H. M. J. Boesten, M. Dirx, P. Bergmans, F. L. M Soomers, J. M. S. SoomersTurlings, P. H. M. Passage, M. C. M. Thomassen, A. v Deelen, A. J. C. Schlosser, P. G. J. v Aubel, A. G. T. H. v Hoof, H. J. Berendsen, L. J. W. Bongenaars and H. Schiffers, general practitioners, for their invaluable help.

10 296 M. CORNEL et al. The study was supported by a grant from The Netherlands Organization for Scientific Research (NWO). REFERENCES Anderson, P., Scott, E. (1992) The effect of general practitioners' advice to heavy drinking men. British Journal of Addiction 87, Beresford, T. P., Blow, F. C, Hill, E., Singer, K. and Lucey, M. R. (1990) Comparison of CAGE questionnaire and computerassisted laboratory profiles in screening for covert alcoholism. Lancet 336, Buchan, J. C, Bucki, E. G., Deacon, G. L. S., Irvine, R. and Ryan, M. P. (1981) Problem drinkers and their problems. Journal of the Royal College of General Practitioners 31, Cornel, M. and Van Zutphen, W. M. (1989) Recognition of problem drinkers and the role of the general practitioner. Canadian Family Physician 35, Cornel, M., Knibbe, R. A., Van Zutphen, W. M. and Drop, M. J. (1994) Problem drinking in a general practice population: the construction of an interval scale for severity of problem drinking. Journal of Studies on Alcohol 55, Cornel, M., Knibbe, R. A., Drop, M. J., Knottnerus, J. A. and Van Zutphen, W. M. (1995) The medical profile of unidentified problem drinkers in general practice: test of an hypothesis. Alcohol and Alcoholism 30, Davis, L. J., Hurt, R. D., Morse, R. M. and O'Brien, P. C. (1987) Discriminant analysis of the SelfAdministered Alcoholism Screening Test. Alcoholism: Clinical and Experimental Research 11, Garretsen, H. F. L. and Knibbe, R. A. (1985) Two Dutch surveys on problem drinking. Medical Science Law 4, Gustafsson, J. E. (1977) The Rasch model for dichotomous items: theory applications and a computer program. Report no. 63. Institute of Education, University of Goteborg, Fack S431, 20 Molndal, Sweden, Hanley, J. A. and McNeill, B. J. (1982) The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 143, Hendriks, A. A. J., Ormel, J. and Van de Willige, G. (1990) Langdurige moeilijkheden gemeten volgens zelfbeoordelingsvragenlijst en semigestructureerd interview. Een theoretische en empirische vergelijking. Gedrag en Gezondheid 18, Hore, B. D. and Wilkins, R. H. (1976) A generalpractice study of the commonest presenting symptoms of alcoholism. Journal of the Royal College of General Practitioners 26, Lamberts, H., Brouwer, H., Groen, A. S. M. and Huisman, H. (1987) Het transitiemodel in de huisartspraktijk. Supplement May 1st 1987 on ICPC rubrics P15, P16. Huisarts & Wetenschap 30, Mayfield, D, McLeod, G. and Hall, P. (1974) The CAGE questionnaire: validation of a new alcoholism screening instrument. American Journal of Psychiatry 131, Molenaar, I. W. (1983) Some improved diagnostics for failure of the Rasch model. Psychometrica 48, Nicol, E. F. and Ford, M. J. (1986) Use of the Michigan Alcoholism Screening Test in general practice. Journal of the Royal College of General Practitioners 36, 409^*10. Registration Network Family Practices (1992) Health Problems and Diagnoses in Family Practice. University of Limburg, Maastricht, The Netherlands. Rush, B. and Brennan, M. (1990) Is the health profile of problem drinkers different from that of other patients? The Journal of Family Practice 31, Saunders, J. B., Aasland, O. G., Babor, T. F., De La Fuente, J. R. and Grant, M. (1993) Development of the Alcohol Use Identification Test (AUDIT): WHO collaborative project on early detection of persons with harmful alcohol consumption II. Addiction 88, Selzer, M. L., Vinokur, A. and Van Rooijen, L. (1975) A selfadministered Short Michigan Alcoholism Screening Test (SMAST). Journal of Studies on Alcohol 36, Wallace, P. and Haines, A. (1985) Use of a questionnaire in general practice to increase the recognition of patients with excessive alcohol consumption. British Medical Journal 290, Wallace, P., Cutler, S. and Haines, A. (1988) Randomised controlled trial of general practitioner intervention in patients with excessive alcohol consumption. British Medical Journal 297, Wright, B. D. and Stone, M. H. (1979) Best Test Design. Mesa Press, Chicago.