The study was done within the longitudinal project titled, "Women and alcohol in Göteborg" (WAG) 19. The WAG-data is collected as a two phase technique epidemiologic study with an initial screening phase followed by individual face to face interviews (Table 1). The procedures are in concordance with recommendations for studies of low-prevalence diseases 20. The project was carried out in one urban (93 157 inhabitants) and one suburban (105 683 inhabitants) district, in Göteborg, which is the second largest town in Sweden, with a total of 444 553 inhabitants 1995 21. Compared to Göteborg at large the population in the uptake area was younger, healthier with higher incomes, but in most socio-demographic variables it was similar to national rates of Sweden 19. The study sample also includes a clinical sample consisting of consecutive patients visiting all relevant public medical service units in the uptake area; including three mother care units, seven general practice centers, the three emergency wards for internal medicine, surgery and general psychiatry as well as two psychiatric outpatients units.

As a first step in the screening phase, a 13-item questionnaire developed for screening of alcohol problems (Screening, Women and Alcohol in Göteborg, SWAG) was sent, in 1986, to all women born in 1925, 1935, 1945, 1955, and 1965 who were identified in the official population register for District West (population 99 328) of Göteborg on December 31, 1985. The 13 items questionnaire was constructed as a Likert scale with a maximum score of 13 points (indicating alcohol related problems). When validated for screening of female AUD, sensitivity was > 74 %, specificity > 96 % and PPV 40–50 %, which was regarded as satisfactory. A more detailed description of the psychometric properties of SWAG can be found elsewhere 19. Additional birth cohorts (women born in 1970/1975 and 1980) were included in 1995 and 2000, employing the same procedures as in 1986. The screening procedure was the same for the clinical sample, with the exception that the screening questionnaires were distributed during visits to the medical service units. For the analyses, women > 70 years of age were excluded from the clinical sample, since no corresponding age group existed in the population based samples.

After screening, a stratified random sample was selected for interviewing. The stratification groups were made up of all respondents with a score of ≥ 4 points on SWAG, and respondents with a score of 1–3 points, 0 points, and drop outs in the screening phase. The three latter groups were selected with varying percentages in order to form groups of similar sizes. Interviews were done face to face in 1990, 1995, and 2000 respectively. The semi-structured interviews comprised approximately 1000 items and took from 1 1/2 up to several hours to perform. The interview focused on socio-demographic characteristics, alcohol consumption, health issues and included questions about experiences of treatment for AUD. Treatment was defined as seeking treatment specifically for alcohol problems, and refers to any kind of treatment available. The interviewers were clinicians trained in making diagnoses according to the DSM-III-R, but only Axis I, IV and V diagnoses were made. The interviewers were trained until interrater agreement was reached at a diagnostic level, but not necessarily with regard to severity/remission/subtype specifiers.

The present study was based on cross-sectional data from 1990 (clinical and population samples), 1995 and 2000 (population samples). Although the clinical sample was recruited from services used by women representing the population at large, it differed from the population sample in some respects. The women were older, had more children, and slightly fewer were married/cohabiting. More women in the clinical population belonged to the lower social class, and fewer to the upper class. Furthermore, the symptom severity among women with AUD, calculated as the numbers of alcohol related symptoms multiplied with the duration in years, was higher in the clinical sample. There was no difference in educational level, measured as < 10 years of education (Table 2).

We concluded that the differences between the samples would not in any important way interfere with the analysis of predictors for the outcome variables used in this study. Thus we pooled the samples in order to reach enough power in the statistical analysis. The numbers of women diagnosed with AUD needed for analyzing associations of personality traits and treatment-seeking were estimated to approximately 200.

Informed consent was obtained from the interviewees and the procedure was approved by the Research Committee for Ethics at the Faculty of Medicine at Göteborg University (Ö 591–99).

Lifetime psychiatric diagnosis were made with the Diagnostic and Statistical Manual of Mental Disorders, 3^rd d edition (DSM-III – R) 22. After information obtained with the Composite International Diagnostic Instrument-Substance Abuse Module (CIDI-SAM) and/or from patient records, 2 diagnoses were added and 32 were excluded 19. The diagnosis "lifetime AUD" refers to AUD occurring at any time in life (including also adolescence). In some computations, this group is divided into "current AUD" (= AUD during the last 12 months prior to the interview) and "resolved AUD" (= AUD present only before the last 12 months prior to interview). Other drug abuse is not included in the computations, mainly because the illicit drug use is very low in this study and other Swedish studies compared to the use of alcohol.

Diagnoses of anxiety and depressive disorders were likewise made by the interviewers according to the DSM-III-R. We constructed the dummy variables depression and/or anxiety in order to include all diagnoses we considered relevant to each condition and also to increase cell sizes 23. Lifetime anxiety and lifetime depression refers to these diagnoses regardless of when in life they have occurred (including also childhood).

The Karolinska Scales of Personality (KSP) is a personality test originally constructed to measure aspects of personality closely linked to the information processing and arousal systems in the individual in order to find biological correlates of personality traits relevant in psychiatric research 24. Thus this personality test is not intended to give a full description of personality. KSP consists of 135 statements grouped into 15 different scales. Each statement has four response alternatives: "disagree completely", "partly disagree", "partly agree" and "agree completely", with a score of 1, 2, 3 and 4 points, respectively. Some items are reversed so that "disagree completely" gives 4 points instead of 1. The points for each scale are then added up and the raw scores are transformed into T-scores. High scores indicate pathology on all scales except for the Socialization scale 24. The scales included are described in Table 3. The stability of personality traits in the general population sample was tested over 5 years. Correlations between initial assessment and follow-up were high for most KSP scales (r = 0.48 – 0.77), with most values > 0.7, indicating individual stability. The strength of the correlations increased significantly with age 25.

The most common reasons given for not participating in the study were lack of time and low alcohol consumption (in the latter case with the assumption that their responses thus would be of little interest to the researchers). Another reason for attrition was that some of the selected women could not be reached by letter or telephone. A random quarter of the attrition group from the screening phase was invited to the interview phase. They did not differ significantly from those who participated in both the screening phase and the interview phase regarding alcohol consumption and prevalence of AUD, age and education 26. Furthermore, we have fairly good knowledge about internal attrition. Most of the women in this group agreed to take part in a shorter telephone interview, and to these KSP was not administered.

In an earlier study within the WAG project medical files were searched, and as only four new cases were identified in the attrition group, this did not affect prevalence of AUD 19. Thus it seems unlikely that there would be a significant overrepresentation of women with alcohol problems in the attrition group. However, if this had actually been the case, it would have implied that the differences we found were underestimated, not overestimated. Hence, the attrition does not appear to constitute a serious problem in interpreting the findings.

The reference group (1,122) included women who participated with a complete interview, and excluded women with AUD.

We report data as T-scores (transformed scores with a mean of 50 and one standard deviation for the reference population). The KSP is sensitive to age and educational level. We therefore used a linear regression model to estimate the effect of these factors, and added a quadratic model when this improved the fit of the model significantly. The raw scores were then adjusted for age and education and thereafter transformed into T-scores.

Significance tests for comparing two groups regarding personality scores were made with two-sample t-test. P-values < 0.05 are considered statistically significant, and 95% confidence intervals (CI) are given. When comparing two groups regarding occurrence of characteristics, e.g. diagnosis of anxiety, odds ratio with 95% confidence interval were calculated from two-by-two contingency tables.

Multi-variable analyses of factors related to treatment-seeking were performed with logistic regression models within each of the 15 scales using factors such as age, education, social class, and additional diagnoses of anxiety and/or depression as covariates.

Personality traits did not predict treatment-seeking for alcohol problems in a stepwise logistic regression model, after adjustment for age, education, social class, psychiatric co-morbidity, and severity. Women with lifetime AUD (n = 217) differed from women without AUD (n = 1,122) on twelve of fifteen scales of KSP. P-values estimated by two-sample t-test ranged from 0.001 to 0.0016. There were no significant differences in three scales: inhibition of aggression, detachment and social desirability (Table 4).

Among women with lifetime AUD (n = 217), the treated women (n = 42) had significantly higher scores than untreated women (n = 175) on three scales of the KSP; somatic anxiety, muscular tension, and guilt. The observed differences in the other scales were non-significant (Table 5).

More than half of the women with lifetime AUD had resolved their problems (n = 123), meaning that AUD was not identified during 12 months prior to the interview. Of those 19 % (n = 23) had received treatment. The treated women scored significantly higher than the untreated on somatic and psychic anxiety, irritability, tension, and guilt feelings (Table 6). The observed group differences (non-significant) in the other scales were in the expected directions. Compared with the treated women, the untreated women (n = 100) scored more similarly to the reference group (n = 1,122). However, they had significantly higher scores than the reference group on somatic and psychic anxiety, irritability, monotony avoidance, impulsivity and psychastenia scales, and lower scores on the socialization scale. There were no statistically significant differences between treated and untreated women with current AUD.

In women with lifetime AUD, 36% (n = 79) also had lifetime anxiety and 21% (n = 45) had lifetime depression. The corresponding figures in the sample without AUD, were 18% (lifetime anxiety) and 11 % (lifetime depression). Among women with resolved AUD, there were no differences in occurrence of psychiatric disorders (lifetime) between the treated (n = 23) and the untreated women (n = 100).

Among women with unresolved AUD we estimated an OR: 3.1 (95% CI: 1.1 – 8.7) for treated women with current AUD (n = 19), regarding the risk of having lifetime anxiety, compared to the untreated group (n = 75). There were no differences between the groups regarding lifetime depression, OR: 0.98 (95% CI: 0.3 – 3.4).

To our knowledge, no study has previously examined treatment-seeking for AUD in relation to personality traits. One of the advantages of this study was that the data stem from a population based sample, which gave us the opportunity to study women that will never be found in a clinical sample, as well as women who had entered treatment. A two-staged sampling was used since the prevalence of AUD among women is low and the prevalence of treatment-seeking is even lower.

The differences between the treated and untreated women with AUD, adjusted for age and education, are statistically significant for three of the 15 personality scales, which proves that the obtained sample size of about 200 cases, gives sufficient power to detect important associations between treatment-seeking and personality factors. The scale with the largest difference, guilt feelings, showed a mean difference between the two groups of 5 units unadjusted (p < 0.01), but 2.5 adjusted (p = 0.32). Similar relations were found for the other two scales with significant difference unadjusted. We think that the obvious interpretation of this result is that personality factors per se only play a small role in predicting treatment-seeking after controlling for the most important external factors. However, any difference will be statistically significant if the sample size is large enough, so for didactic reasons we performed a simulation study, using a bootstrap technique, to examine at what sample size we could expect to reach 5 % significance with probability 80 % in a multivariate binary logistic model for the three personality factors that were closest to be significant in the models with age, education, severity, psychiatric diagnosis and socio-economic status as covariates. Required sample sizes were approximately 450, 600 and 650, a result which we think agrees with our conclusion that low power is not a problem in this analysis.

In a 5 -year follow-up study personality traits in general were found to be stable among adult women 32. Still, we cannot disregard the possibility that psychiatric disorders can influence personality traits and vice versa.

Pooling a population sample and a clinical sample was needed in order to obtain a sufficient sample size, but this procedure also created limitations of the study. Although the initial comparisons of the samples showed that sufficient similarities for the present analyses existed, the samples differed somewhat. For example did the women in the clinical sample score higher on anxiety, lower on socialization and had higher symptom severity. This may have influenced the results to some extent, but the clinical sample (n = 171) was small compared with the general population samples (n = 1,168), and it is mainly composed of women having attended GP's and maternity care units. These women do not differ greatly from the general population, as most women attend these units in Sweden. The conclusion is that this sample is fairly representative of at least Caucasian women in the Northern Europe.

The way in which personality traits interact with treatment requirements in relation to other factors needs to be studied further. The question remains whether women who never seek treatment should be identified and offered help. We do not think that the high percentage of all the many women who resolved their AUD without help should be interpreted as they did not need or would have benefited from help. These individuals experience difficult problems regarding alcohol (as the diagnoses indicates), and it is always important to try to reduce harm 33. Thus, it is possible that the comparatively low proportion of women who resolve their AUD with the aid of treatment does not reflect the "full" treatment need or even demand for treatment in all women with AUD.