Alcohol use and spousal mental distress in a population sample: the nord-trøndelag health study

Sample

During the years 1995 to 1997, the population aged 20 years or older in Nord-Trøndelag County, Norway, was invited to participate in a health screening survey, the Nord-Trøndelag Health Study (HUNT 2). The health study consisted of a health examination, including measuring blood pressure and cholesterol and a number of other analyses of the blood. Two questionnaires included items on health and illness, health behaviour, life-style, demography and other possible risk factors for health problems. Only the questionnaire data are used in the present study. The first questionnaire, Q1, was sent by mail together with the invitation letter and returned at the site of the health examination. The second questionnaire, Q2, was distributed during the health examination, and the participants were asked to complete it at home and return it by prepaid mail. The sample in the present study consists of participants aged 20-70, as participants over 70 years of age received a special questionnaire version, lacking one of the key items. Of the 77,659 invited individuals aged 20-70, 54,466 (70.1%) returned Q1 and 46,241 (59.5%) returned both Q1 and Q2. A detailed description of the full sample is available elsewhere [13]. The personal identification number assigned to all Norwegian citizens was used by the governmental statistics agency Statistics Norway to identify married or cohabiting couples. Only couples with complete data on all variables of interest after imputations were retained in the analyses, leaving 11,584 couples to be analyzed. These couples will be referred to as the nuclear sample from hereon. Mean age was 48.5 years (SD=11.2) in men and 45.7 years (SD=11.0) in women.

Ethics

The HUNT-Study has been approved by the Regional Committee for Medical and Health Research Ethics for Central Norway and has been performed in accordance with the ethical standards laid down in the 1964 Helsinki declaration. All respondents gave their written consent for their data to be used for research purposes.

Measures

Treatment of missing values

SPSS Missing Value Analysis (MVA), Expectation Maximization (EM) was used to impute values separately for the various set of items (mental distress, alcohol consumption, CAGE). Imputations were made for respondents with valid data for a minimum of 50% of the items of each set of items. The proportion of respondents with non-complete data was reduced from 17.8% to 1.0% for mental distress. Most of these had only one or two missing items. Prior to imputations of the alcohol consumption items, the respondents having reported to be abstainers and who had not completed the alcohol consumption items were scored zero on all the items. In cases where some of the consumption items were blank and other consumption items had valid responses higher than zero, we assumed that blank responses signified no consumption for those beverages and replaced these missing values by zero. For instance, if a person had reported five units of beer and seven units of wine and left liquor open, this was assumed to mean “no liquor”. This recoding (abstainers and blanks in combination with responses >0) reduced incomplete alcohol consumption data from 51.5% to 6.5%. Subsequently, the alcohol consumption items were imputed for respondents with valid data for a minimum of 50% of the items, which reduced missing values from 6.5% to 4.6%. Imputations of CAGE only reduced missing values from 8.5% to 7.7%. Due to the small number of records gained by imputing CAGE, we chose to replace missing values for the individuals who had scored less than the average on the alcohol consumption index with the response indicating absence of alcohol-related problems on the CAGE items. This reduced incomplete/missing CAGE data from 7.7% to 0.8%.

Statistical analyses

Multivariate hierarchical regression analyses (SPSS, Regression, Linear) were run stratified by gender, investigating the predictive value of male and female index persons’ alcohol consumption and alcohol-problems on their spouses’ mental distress. To investigate deviations from linearity for the effect of the trichotomized alcohol consumption variable on spousal mental distress, curve estimation regression analyses were performed. The R² of the functions allowing curve linearity was not higher than the R² of the linear regression analysis, supporting the usage of a trichotomized alcohol consumption variable. Spearman rank-order correlations were run for the index persons’ alcohol predictor and spousal mental distress, giving crude estimates of the associations. In the hierarchical regression analyses, spousal age and spouses’ and index persons’ educations were entered in the first model (model 1). In model 2, alcohol consumption was added to the model. To observe the predictive effect of the CAGE items without adjusting for alcohol consumption, the CAGE items were entered together with the demographic variables from the first model in model 3. In model 4, alcohol consumption and the CAGE items were entered together with the demographic variables to observe the effects of each alcohol predictor when adjusted for the other alcohol predictors.

The effect of alcohol consumption/problems in the index person on mental health in the spouse may be somewhat confounded by the index person’s mental distress, because the index person’s mental distress may have an impact both on their own alcohol consumption/problems and on the spouse’s mental health. Likewise the spouse’s own alcohol consumption or alcohol-related problems may affect both the index person’s alcohol consumption/problems and the spouse’s mental health, and accordingly act as a confounder. Just as likely, however, these variables (the index person’s mental distress and the spouse’s alcohol consumption and alcohol-related problems) may act as mediators. That is, the index person’s alcohol consumption/problems cause distress in the index person and increased alcohol consumption/problems in the spouse, and these again cause distress in the spouse. Research has shown that alcohol abuse most likely causes within-person mental distress, rather than the other way around [17], suggesting that the spouse’s mental distress may act more as a mediator than as a confounder. Research regarding the causality of spousal concordance for alcohol consumption implies that both assortative mating and spousal convergence over time are involved [18], suggesting that alcohol use and alcohol-related problems may both confound and mediate the effect of index persons’ alcohol consumption/problems on spouses’ mental distress. The possible causal pathways between index persons’ and spouses’ alcohol variables and mental distress are presented in Figure 1. In conclusion, whereas model 4 may be somewhat under-adjusted, model 5, including index person’s mental distress and spousal alcohol variables, may be over-adjusted because it probably adjusts for mediator effects.

Spousal and own alcohol consumption and alcohol problems may act as effect modifiers for each other. Thus, interaction terms between index persons’ alcohol variables (alcohol consumption and each of the CAGE items), and between index person and spouse alcohol variables were specified. Demographic characteristics of the spouse or the couple and index person mental distress may also moderate the effects of the alcohol variables. We tested for interaction effects between index person alcohol variables and the following variables: spousal age, average couple education, and index person mental distress. All interaction effects were adjusted for all variables included in model 4.

Descriptive statistics

The distribution of the outcome measure, mental distress, is presented in Figure 2, collapsed across genders. It deviates somewhat from a normal distribution, with skewness 1.31 and kurtosis 2.29.

Results of the hierarchical regression analyses

Male index persons – female spouses

The results from model 1, where the demographic variables were entered together, showed that spousal age and spousal and index person education were significantly related to spousal mental distress. In model 2, index person alcohol consumption was entered into the model but did not significantly predict spousal mental distress. In model 3, index person alcohol consumption was removed from the model, and replaced by index persons’ CAGE responses. Only one CAGE item, “Felt bad or guilty” significantly predicted spousal mental distress. In model 4, both index person alcohol consumption and CAGE were entered simultaneously into the model. When adjusting for demography and CAGE, alcohol consumption significantly predicted a decrease in spousal mental distress. “Felt bad or guilty” was still the only CAGE item significantly predicting an increase in spousal mental distress. In model 5 spouse alcohol variables and index person mental distress were added to the model, to explore potential mediation or confounding. The negative effect of index persons’ alcohol consumption increased in model 5, indicating that some of the effect of alcohol consumption was suppressed in models 1-4. The effect of the CAGE item “Felt bad or guilty” seemed to be completely mediated or confounded by the new variables entered in model 5 and lost its significant effect. “Cut down” was the only significant CAGE item in model 5, predicting a decrease in spousal mental distress. The remaining CAGE estimates decreased in magnitude in model 5. Spouses’ own alcohol consumption was unrelated to their mental distress, when adjusting for all the variables in block 5. All four spousal CAGE items were highly predictive of spouses’ own mental distress.

Female index persons – male spouses

The relationship between female index persons’ alcohol consumption and alcohol-related problems, measured by CAGE, and their male spouses’ mental distress was investigated with the same hierarchical procedure as reported above. Model 1 showed that spousal age and spousal and index person education were significantly related to spousal mental distress. In model 2, alcohol consumption was entered into the model, but did not significantly predict spousal mental distress. In model 3, alcohol consumption was replaced by the index persons’ CAGE items, of which the items “Criticized” and “Felt bad or guilty” significantly predicted an increase in spousal mental distress. In model 4, index person alcohol consumption and CAGE were entered simultaneously. When adjusting for demography and index person CAGE, alcohol consumption significantly predicted a decrease in spousal mental distress. The CAGE items “Criticized” and “Felt bad or guilty” still significantly predicted an increase in spousal mental distress in model 4. In model 5, the effect of index person alcohol consumption increased, and still significantly predicted decreasing spousal mental distress. Two index person CAGE items also significantly impacted the outcome. “Cut down” was negatively associated with and “Criticized” positively associated with spousal mental distress. All CAGE estimates, except “Cut down” decreased in magnitude in model 5. Spouses’ own alcohol consumption was unrelated to their mental distress, whereas all spousal CAGE items were positively related to their own mental distress.

Interaction effects

Interaction terms between index persons’ alcohol variables, between index persons’ and spouses’ alcohol variables, between index persons’ alcohol variables and mental distress, and between index persons alcohol variables and the demographic variables were specified. In total, 49 interaction analyses were run. The significance level was corrected according to the Bonferroni formula, rendering the alpha level at 0.001. This resulted in two significant interaction effects – between 1) female index persons’ and male spouses’ alcohol consumption (p=.001), and 2) “Cut down” in male index persons and “Criticized” in female spouses (p<.001). Two effects were approaching significance, after Bonferroni correction - 3) “Criticized” in male index persons and “Eye-opener” in female spouses (p=.008); and 4) “Felt bad or guilty” and mental distress in female index persons (p=.009).

Simple effects analyses in stratified samples showed 1) that the negative (protective) effect of alcohol consumption among female index persons was stronger for spouses consuming high amounts (B=-.298, p=.001) and very high amounts (B=-.200, p=.012) than for spouses with a low/moderate consumption (B=-.024, p=.495. 2) The negative effect of “Cut down” in male index persons was stronger for female spouses who themselves had been “Criticized” (B=-.679, p=.015) than for spouses who had not been “Criticized” (B=.076, p=.224). 3) The positive (risk) effect of “Criticized” in male index persons on spousal mental distress tended to be stronger for female spouses who had not had an “Eye-opener” to steady nerves (B=.078, p=.029), than for female spouses who had had an “Eye-opener” (B=-.732 p=.421). 4) The positive effect of mental distress in female index persons tended to be stronger when the index person themselves had not “Felt bad or guilty” (B=.466, p=.000) than when the index person had “Felt bad or guilty” (B=.211, p=.026).

Methodological considerations

The existing literature on the relationship between one spouses’ alcohol abuse and the other spouses’ mental distress has for the most part only investigated the relationship between male alcohol abusers and female spouses. Also, several of the previous studies have been based on samples of limited generalizability. The present study is based on data from both male and female spouses, from a large sample, representative of the Norwegian adult population, implying a high generalizability of our results.

However, there are methodological limitations to our study. The response rate for the individuals having returned both Q1 and Q2 (59.5%) may have caused a selection bias. However, a recent attrition study of the HUNT 2 sample showed that high alcohol consumption in a previous HUNT study only predicted non-participation in HUNT 2 moderately well (OR= 1.27 for the top 3% consumption) [22]. Also, even highly selective non-participation only seems to moderately influence associations between variables [23], giving reason to believe that our estimates have not been severely affected by non-participation.

The lack of diagnostic measures rendered us incapable of positively identifying alcohol abusers among the index persons. As shown by the mean values of the high and very high consumption groups, the consumption reported may not be indicative of alcohol use disorders. However, alcohol consumption is usually underreported in population studies [24], making it feasible to assume that the actual consumption is considerably higher than reported. The degree to which underreporting leads to misclassifications depends upon whether or not the underreporting is systematic - that is, whether the amount of under-reporting correlates highly with the real consumption, such that most people report, say, half their real consumption - or non-systematic. The distribution of the consumption in our sample corresponds well with distributions normally observed in alcohol consumption research, in which approximately 2/3 of the population reports drinking less than the average consumption and 15% reports drinking more than twice that of the average [25]. This suggests that the underreporting is systematic, and that the high and very high consumption groups in fact primarily include high and very high consumers – who for the large part will be individuals with alcohol use disorders. By choosing strict criteria for what is considered high and very high consumption, there is a much higher probability of misclassifying a real case as a non-case than of misclassifying a non-case as a case. To avoid substantial attenuation of the results, keeping the case groups relatively free from false positives is much more important than is avoiding some pollution of the large non-case group by (a relative low fraction of) false negatives. Therefore, the choice of strict criteria defining the top five percentile as high and very high consumers with potential alcohol problems is methodologically sound.

Alcohol consumption was measured as current consumption, whereas alcohol-related problems were measured as lifetime problems. The outcome, mental distress, was also measured as current distress. Current disorders are more predictive of co-occurring problems in the spouse [1], which may have deflated the observed association between alcohol-related problems and spousal mental distress. Furthermore, individuals abstaining from alcohol were asked to skip the CAGE, which may have caused previous alcohol abusers to skip this measure given that they were abstaining at the time of the survey. It was not possible to separate steady heavy drinkers from binge drinkers in this data material. Previous research has found that spouses of steady heavy drinkers experience less mental distress than spouses of binge drinkers – although the total alcohol consumption of steady heavy drinkers in general is higher than that of binge drinkers [6]. Being able to distinguish effects of steady heavy drinking from effects of binge drinking on spousal distress would have added information to our results.

The mental distress index was composed of two separate mental health measures, the HADS and the CMD, which has not previously been used in a combined version, and of which only the first has previously been validated. However, we still judged the face validity of the combined index to be better than that of the HADS alone, because the HADS depression items seem not to include negative emotionality. The correlation between the HADS scores and the scores from the combined index was 0.96 for men and 0.97 for women, so clearly the content of the HADS was not radically changed after the inclusion of the CMD items. Also, the internal consistency of the measure increased after including the CMD. Accordingly, even if our measure was not strictly validated, we must assume that the validity was as least as good as for the HADS instrument.

Our design is not suited to decide about the causal direction between alcohol abuse and spousal mental distress. Although not evidence-based, we would judge a causal direction from alcohol abuse in one of the spouses to mental distress in the other as more plausible than the reversed causal pathway, but there are probably also people who drink to drown problems related to their spouse’s’ poor mental health.

Finally, and related to the problem of unknown causality, our results could be confounded by effects of assortative mating, primarily for alcohol consumption, for which there is a strong spouse correlation. The association between alcohol abuse in spouse A and mental health in spouse B could partly reflect assortative mating for alcohol abuse together with a relationship between own alcohol abuse and own mental health. However, recent results based on HUNT data suggest that most of the observed spouse similarity for alcohol consumption reflects convergence during the spouses’ life together rather than assortative mating. Only a minor part of the spouse similarity seems to have been present at the time the spouses started to see each other [26]. Also, the association between own alcohol abuse and mental health is moderate (Table 1), implying that an important confounding by assortative mating is unlikely.