Cognitive style and drinking to cope: A prospective cohort study

Background and Aims: Having a negative cognitive style may lead someone to feel hopeless about his or her situation and be more likely to engage in coping-motivated drinking. We, therefore, aimed to investigate the association between cognitive style and drinking to cope. Design: Prospective cohort study. Setting: The former Avon Health Authority in South West England. Participants: A total of 1681 participants of the Avon Longitudinal Study of Parents and Children. Measurements: Participants completed cognitive style questions at age 17 and a subset of drinking to cope questions at age 24. We used linear regression to test the association between cognitive style and drinking to cope, controlling for confounders. Alcohol consumption and dependence scales were included in a secondary analysis. Findings: A 20-point increase (that was the standard deviation of the exposure variable) in cognitive style score at age 17 was associated with an increase of 0.24 in drinking to cope scores at age 24 after adjustment for confounding variables (95% CI) = 0.08 – 0.41, P = 0.003). We found no evidence of an association between cognitive style and alcohol consumption (coefficient = 0.03, 95% CI = − 0.08 – 0.14, P = 0.591) before or after adjustment. There was evidence for an association with alcohol dependence, but this was not present after adjusting for confounders (coefficient = 0.01, 95% CI = − 0.04 – 0.05, P = 0.769). Conclusions: In young adults in England, there appears to be a positive association between negative cognitive style and subsequent drinking to cope.


INTRODUCTION
Problematic alcohol use often starts during adolescence [1,2]. Although long-term heavy drinking can lead to problems such as stroke [3], cancer [4] and heart disease [5], it is also linked to mental health problems, relationship breakdowns, impaired social relationships and employment dismissal [6]. Reducing hazardous alcohol use in early life is important and may prevent later development of alcohol-related problems.
Different motivational factors for drinking can produce different patterns of use and health outcomes, so exploring the motivations to drink would inform interventions to support those most at-risk of alcohol-related problems [7]. Research has found the association between alcohol use and mental health problems in young people is because of problematic use of alcohol as opposed to the quantity consumed [8]. Using alcohol to cope with problems can increase the risk of long term alcohol-related problems compared to other drinking motivations such as to socialise [9,10], and this association is maintained even when controlling for alcohol consumption [11]. Mental health problems such as depression and anxiety may increase the likelihood of drinking to cope, because of people using alcohol to deal with underlying negative emotions and problems. [12,13]. It is possible that coping-motivated drinking provides short-term relief from symptoms of low mood, therefore, negatively reinforcing the idea of drinking to cope. However, although alcohol use may provide relief from depression in the short-term, research has found that those who use substances to cope, even at subclinical levels, are less likely to work on their difficulties [14] meaning their depression may be less likely to improve. Moreover, people who use substances to cope with their difficulties are at higher risk of worsening depression over time [15]. It is likely that this population may be stuck in a 'vicious cycle', where depression is causing higher alcohol use, which in turn is causing higher levels of depression. It is important to investigate risk factors for coping-motivated drinking so that interventions can be targeted to support people before their drinking becomes problematic.
Existing evidence from the depression literature has suggested that negative cognitive style can create an underlying vulnerability to environmental stressors and increases risk of later depression [16].
Cognitive style is based on the hopelessness theory of depression and explores the causal attributions for negative life events. For example, if someone with a negative cognitive style fails a test, they may attribute this to internal factors (i.e. 'I am stupid'), stable factors (i.e. 'I will never pass') and global factors (i.e. 'I fail at everything'). Negative cognitive styles are associated with later depressed mood [17,18] and anxiety [19,20]. It is possible that someone with a negative cognitive style may engage in more negative coping strategies, such as alcohol misuse, because of the mechanism of learned helplessness [21].
Learned helplessness is the idea that someone has no control over negative situations, and is largely linked to negative cognitive style [21]. Making internal, global and stable attributions to events could result in a feeling of helplessness and inability to change, and therefore, may make someone more likely to engage in negative coping behaviours. There is also a link between alcohol use and helplessness [22] and uncontrollable events [23]. Therefore, it is possible that negative cognitive style increases likelihood of drinking through the mechanism of learned helplessness.
To our knowledge, only one study has examined the relationship between cognitive style and drinking to cope [24]. An association was found between negative cognitive style and higher drinking to cope, however the study used a convenience sample of university students (n = 182), and the study was cross-sectional so a temporal relationship could not be assessed. Longitudinal research is needed in a larger, more representative sample.
This study investigated the prospective association between cognitive style at age 17 and drinking to cope at age 24. To our knowledge this is the first cohort study examining whether negative cognitive style is associated with later drinking to cope. We also examined alcohol consumption and dependence, as secondary outcomes.
The total scores could, therefore, range between 0 and 24. Nondrinkers were assigned a score of 0; high scores indicated a higher likelihood of drinking to cope. Mean scores and standard deviations for each item on the drinking to cope scale are shown in Supporting information Table S1.

Alcohol consumption/dependence
The Alcohol Use Disorders Identification Test (AUDIT) [28]

Potential confounders
We adjusted for the following potential confounders [18,30]: sex, parental social class (based on the Registrar's General classification and grouped into manual and non-manual; when the social class of each parent differed the higher level was taken), maternal education (measured by the mother's highest qualification level when the child was born), maternal depression (measured using the Edinburgh Postnatal Depression Scale [31]) and maternal age (measured in years when the child was born). We also adjusted for depression and anxiety at 17 (both measured by the Revised Clinical Interview Schedule [32]), alcohol use at 17 (measured by the AUDIT-10 [28]) and drinking to cope score at 17. Depression, anxiety and alcohol use were added to the model as a separate set of adjustments because we cannot exclude the possibility that they were on the causal pathway from negative cognitive style to drinking to cope.

Statistical analyses
Statistical analysis was conducted on Stata Version 16. The analysis was not pre-registered and therefore, these results should be considered exploratory.

Descriptive statistics
We divided the CSQ-SF scores by the median and reported sample characteristics for all variables according to CSQ-SF scores, using complete data. We repeated these descriptive statistics using all available data for all participants (regardless of whether they had complete data for the exposure, outcomes and confounders) to explore any differences for complete cases compared with all available cases.

Primary outcome
Linear regression models were used for the primary and secondary analysis. Although the drinking to cope score was positively skewed, parametric assumptions were assumed to be met because of the large sample size and the fact that the residuals were normally distributed.
Histograms for the distribution of raw scores and residuals for the drinking to cope scale, AUDIT-consumption and AUDIT-dependence measures are shown in Supporting information Figures S1-S6.
We first conducted a linear regression with the drinking to cope scale as a continuous outcome and CSQ-SF scores as a continuous exposure. We divided the CSQ-SF by 20, its standard deviation, to produce a larger coefficient. The analysis was carried out before and after adjustment for confounders. We calculated the effect sizes for each mode by dividing the mean difference of the outcome by the standard deviation of the outcome. Next, we split the CSQ-SF scores into tertiles, and completed a second analysis with the drinking to cope outcome and the CSQ-tertile variable, to allow for an inspection of non-linearity. We did not report P values for the comparison of the tertiles as P values from subgroups can be unreliable [33]. Finally, we included a quadratic term into the model for each outcome to explore the linearity of the relationship between our exposure and outcome.
Univariable models were run unadjusted, and then were adjusted for: sex, parental social class, maternal education, maternal depression and maternal age. After this we included depression, anxiety, alcohol use and baseline drinking to cope score.
We also re-ran our analysis using the three subscales of the CSQ (internality, globality and stability) to explore any difference.

Secondary outcomes
For the secondary analysis, we repeated the above analyses, using the AUDIT-consumption and AUDIT-dependence scores in two separate models as our outcome measures. Although the AUDIT-consumption measure was normally distributed, the AUDITdependence measure was positively skewed. However, it was decided that linear regressions would be used throughout the secondary analysis because of the large sample size and the normal distribution of the residuals.

Sensitivity analyses
We repeated our main analysis excluding any non-drinkers (people that scored 0 on the AUDIT-consumption scale) to ensure that this sample did not skew any associations found. We split all outcome measures; first, by a median split, and then by the top 20% compared to bottom 80% and we re-ran our analysis using logistic regression, with the same adjustments as in the main analysis.
We also repeated our analysis using the Diagnostic and Statistical Manual of Mental Disorders (DSM-IV) criteria [34] for alcohol dependence as our outcome measure using data from the complete case sample.

Missing data
To address the possibility that missing data biased our results, we re-ran our all our models using a sample based on everyone with complete exposure data and imputed missing data in the primary and secondary outcome and the confounder data, increasing the sample size to 3881. We used multiple imputations by chained equations and imputed 50 data sets [35]. Our imputation models included all variables used in the main analysis plus auxiliary variables. The auxil-

Descriptive statistics
Our final sample included 1681 complete cases (those with data for exposure, outcomes and confounding variables) ( Figure 1). Comparisons between the complete cases and the rest of the ALSPAC sample are shown in Supporting information Table S2.
People with higher CSQ-SF scores had higher depression, anxiety, AUDIT and drinking to cope scores at 17 (

Drinking to cope
Drinking to cope scores ranged from 0 to 19, with a mean of 4.24 (SD = 3.51).  Table 2). Our findings were similar when the CSQ-SF was split into low, medium and high tertiles ( Table 2).
When repeating the analysis using the three subscales of the CSQ, we found that stability had the largest association

AUDIT-consumption and AUDIT-dependence
We found no evidence for an association between CSQ-SF score and AUDIT-consumption score in either the unadjusted model  Table 3). There was also no evidence for an association when the CSQ was split into low, medium and high tertiles (Table 3).
We found some evidence of an association between an increase  (Table 4). Our finding remained similar when examining the CSQ-SF in low, medium and high tertiles (Table 4).
When repeating the analysis using the DSM-dependence scale as our outcome measure, we found no evidence for a relationship between cognitive style and DSM alcohol dependence score (Supporting information Table S7).
We repeated our findings excluding non-drinkers (n = 163) and did not find any differences in our findings. We also repeated the analysis using logistic regression, after creating binary outcomes for the drinking to cope, AUDIT-consumption and AUDIT-dependence.
The results of the analysis showed no differences in findings depending on the statistical method used (Supporting information Tables S8-S13). There was also no evidence for a non-linear relationship between the CSQ-SF and drinking to cope (P = 0.397), AUDIT-consumption (P = 0.666) and AUDIT-dependence (P = 0.240).
Results based on the imputed sample were the same as to those found using the non-imputed data (Supporting information Tables S14-S22).

DISCUSSION
We found that a more negative cognitive style at 17 was associated with higher drinking to cope scores at 24, with a small effect size, and this remained after adjusting for confounders. However, we did not find evidence of an association between cognitive style at 17 and alcohol consumption or dependence at 24.
T A B L E 1 Characteristics of the sample across high and low CSQ-SF scores for complete cases

Strengths and limitations
To our knowledge, this is the first study examining the association between cognitive style and drinking to cope using a prospective cohort study. We adjusted for a wide range of confounders, and the use of multiple outcome measures, including the drinking to cope, AUDIT and DSM-Dependence scale captures the different aspects of alcohol use, allows for a deeper understanding of the nature of the relationship with cognitive style.
A limitation of the study is that the sample may not fully represent the general population, because the Avon area has a high socioeconomic status. Although we controlled for social class throughout, it would be useful to replicate our research in a less affluent area, because alcohol use is related to lower socioeconomic status [38]. The ALSPAC study is also subject to high attrition. Although we ran multiple imputations to replace missing data, which had little influence on the results, the imputed sample would still be less representative than the broader ALSPAC sample. However, within cohort associations should remain valid even when the sample is not truly representative of the population. Residual confounding can never be ruled out in an observational study so we cannot be sure of causality in this investigation [39].
The ALSPAC study measured our exposure and outcome measures at 17 and 24, and we, therefore, did not have data to explore any patterns in cognitive style or drinking to cope between these ages. Nonetheless, age 17 is a time when high alcohol use is common [40], and by age 24 people have most people have more responsibilities and therefore, may be a time when heavy drinking first becomes problematic [41], meaning this is still an appropriate age group to use for this research.
One possibility is for a cyclical relationship between cognitive style and drinking to cope (i.e. drinking to cope could lead to social consequences that make people feel more out of control, leading them to make more negative attributions), and therefore, the link between cognitive style and drinking to cope could be more complex than our findings suggest. Some other psychological processes, such as affect dysregulation, could also have been potential confounders [42,43]. Additionally, we were not able to repeat our analysis using alternative measures of drinking motives aside from drinking to cope scores. Although our drinking to cope measure had good internal consistency, we did not have access to the individual data points for our exposure and secondary outcomes, so could not explore internal consistency for these measures. cope scale asks participants to hypothetically score themselves on a Likert Scale (i.e. almost never/often/sometimes/almost always).
Although both measures have good evidence individually, the difference in wording may lead participants with a negative cognitive style to rate themselves higher on the subjective drinking to cope scale, but not on the objective AUDIT scales, causing differences in our outcome measures.

Mechanisms
The relationship between cognitive style and drinking to cope could be explained by the mechanism of learned helplessness [21].
If someone has a negative cognitive style, this could foster a feeling of helplessness and inability to change, which could explain the decision to use alcohol over alternative coping methods. The fact that the stability subscale had the largest effect on drinking to cope could lead someone to believe that negative events will always happen to them, further exacerbating the idea of learned helplessness. As drinking to cope is also associated with negative mental health outcomes [12,13], these higher rates of negative outcomes could reinforce learned helplessness and negative cognitive style, further increasing coping-motivated drinking.
Our finding that cognitive style did not appear to be related to alcohol consumption could be explained by the fact mental health problems are more strongly associated with problem-use of alcohol, but not necessarily the amount of alcohol consumed [8]. It is important to emphasise that the AUDIT asks about frequency of alcohol use, whereas one can endorse items on the DMQ even if alcohol is used infrequently. Previous research has linked heavy alcohol use with extraversion [45]. However extraverts are less likely to drink for coping motives [46] and are less likely to develop other mental health problems associated with cognitive style [47,48]. Therefore, it is possible that many of the participants drink alcohol for various motives such as social motives, but would not be considered as having an alcohol-related problem or a vulnerability to depression.
Our finding that there did not appear to be an association between cognitive style and alcohol dependence once depression, anxiety and baseline alcohol use were adjusted for was surprising.
One possible explanation is that alcohol dependence occurs later in life so an association may not be demonstrated at the age of 24.
However, another possibility is that the questions asked (i.

Clinical implications
Our finding that negative cognitive style is associated with later drinking to cope has a number of implications. It may be useful to identify people with a negative cognitive style at a young age before they start using alcohol and provide alternatives ways of coping with difficulties and discourage them to engage in copingmotivated drinking. Because cognitive style is associated with other mental health problems, such intervention could lead to broader benefits. There is evidence that cognitive style can be altered using cognitive behavioural therapy (CBT) [49,50], and that CBT for other mental disorders also reduces problem drinking [51,52].
There is, therefore, scope for people seeking help for alcohol problems to target their negative cognitive style using CBT, so that they engage in healthier coping behaviours.

CONCLUSIONS
Our research found evidence for a relationship between cognitive style and drinking to cope. Our findings point toward changes that can be made to support those at risk of problematic drinking now and in the future, helping individuals lead a better quality of life, and relieving some of the financial burden of alcohol problems on the National Health Service (NHS).

ACKNOWLEDGEMENTS
We are extremely grateful to all the families who took part in this study, the midwives for their help in recruiting them and the whole ALSPAC team, which includes interviewers, computer and labora-