The Moderating Effect of Tobacco Smoking on the Efficacy of a Computer-Based Brief Alcohol Intervention: Results from a Randomized Controlled Trial Open Access

Evidence showed that computer-based brief alcohol intervention (BAI) has an effect in reducing alcohol consumption among persons who consume alcohol above certain health-risk levels [1‒3] (e.g., >7/14 drinks of 14 g alcohol per week or >3/4 drinks on any day for females and males, respectively [4]). In addition, a Cochrane meta-analysis found small effects of BAIs on heavy episodic drinking [5].

Increasing evidence of the potentially harmful effects of alcohol consumption was followed by changes in alcohol consumption guidelines, which now state that only abstinence from alcohol is safe to prevent disease and death [6]. In addition, studies have shown that low-to-moderate alcohol consumption does not reduce mortality compared to lifetime abstainers [7, 8]. BAIs could be a tool helping persons to reduce their alcohol consumption along these guidelines.

The combination of alcohol and tobacco consumption is highly prevalent in many nations [9], and both behaviors may act synergistically as a risk for disease and death [10, 11]. Tobacco smoking and alcohol consumption can intensify the craving for each other [12, 13]. A potential cross reinforcement to consume the other substance could be explained with common neurobiochemical pathways releasing dopamine in the mesolimbic system [14, 15]. Several studies showed an association between tobacco smoking and heavy episodic drinking in adolescents [16] and adults of different ages [17‒19]. There are multiple reasons why persons practice heavy episodic drinking. Fun, sensation seeking, escaping problems or peer pressure seem to be the main motivations [20‒22]. The main motivations to smoke tobacco seem to be enjoyment or stress relief [23, 24]. Individuals who engage in heavy episodic drinking were found to be more likely to smoke socially compared to those who do not engage in heavy episodic drinking [25]. The combination of the two substances appears to have positive cross-reinforcing effects as they may combine their potential to facilitate social interaction, and negative cross-reinforcing effects as both substances seem to enhance stress-reducing or calming effects [26, 27]. Heavy episodic drinking and tobacco smoking in combination are likely to result from the same motivation as either behavior alone. This motivation might be supported by cross reinforcement mediated by a joint activation of the mesolimbic system and, therefore, could explain why persons practicing heavy episodic drinking seem to be susceptible to the consumption of tobacco.

The existing evidence on the moderating effect of tobacco smoking on BAI efficacy is limited and inconsistent. Persons in alcohol dependence treatment who smoked, compared to persons who did not, had more dependence symptoms, greater treatment dropout [28] and a lower likelihood of maintaining alcohol abstinence over 12 months after treatment [29]. In contrast, a study with Swiss adolescents found a text message-based intervention to be more effective in reducing binge drinking in persons who smoked tobacco compared to persons who did not [30]. Other findings revealed that tobacco smoking status or smoking intensity did not moderate the effects of BAIs among general hospital patients with at-risk drinking [31].

It remains unclear if tobacco smoking status moderates the efficacy of BAI that addresses persons irrespective of quantity, frequency, and consequences of their alcohol use. This study aimed to test whether BAI reduced heavy drinking days (HDDs) less among persons who smoked tobacco than among persons who did not in a sample of adults with alcohol consumption in the past 12 months. Hence, the moderating effect of tobacco smoking status on the outcome of BAI in a general population sample was to be investigated.

This study reports data from a secondary analysis of the randomized controlled trial “Testing a proactive expert system intervention to prevent and to quit at-risk alcohol use” (PRINT). The primary aim of the PRINT study was to test the 12-month efficacy of a computer-based BAI on self-reported alcohol use. A general population sample of adults at the age of 18–64 years who had consumed alcohol in the past year was randomized to an intervention or a control group (allocation ratio 1:1). The intervention group received assessment plus computer-generated feedback letters at baseline, month 3, and month 6. The control group received assessment only. Both groups were followed-up at month 12. The PRINT study was approved by the Ethics Committee of the University Medicine Greifswald (BB 147/15) and was prospectively registered in the German Clinical Trials Register (DRKS00014274). The protocol was published on 9 July 2018 [32]. Primary and secondary outcome data have been published elsewhere [33]. The intervention has been found to result in drinking reductions 6 months after study start, especially among persons with low-to-moderate drinking amounts. However, this effect was not present at month 12. The current analysis is of exploratory nature. This article follows the CONSORT guidelines for reporting randomized controlled trials [34]. A detailed checklist is available as online supplementary material (for all online suppl. material, see https://doi.org/10.1159/000545866).

Overall, 6,645 residents who showed up from April to June 2018 in the waiting area of the municipal registry office in Greifswald, Mecklenburg-West Pomerania, Germany, were invited by study assistants in personal contact to take part in a survey containing the eligibility screening for the PRINT study. Among those who fulfilled the inclusion criteria (age 18–64, sufficient language and reading skills, no notable cognitive impairment or a physical condition that prevented trial participation and no employment at the conducting research institute), 2,947 persons (74% of those eligible for the screening) completed the screening on alcohol consumption in the past 12 months on tablet computers. After the clients had finished the self-administered questionnaire, study assistants informed those eligible for the PRINT trial about its aim and procedures. Clients without a permanent address or telephone number were excluded. Among those who had consumed alcohol at least once in the past 12 months prior to the screening, 1,646 persons (67% of the persons eligible for the trial) participated and were randomized to intervention (n = 815) or control group (n = 831), respectively. More details can be found elsewhere [33].

The study participants were randomized to the intervention and the control group using simple randomization with a 1:1 allocation ratio and a computer-generated list of random numbers. The intervention group received assessment plus computer-generated individualized feedback letters at baseline, month 3, and month 6. The control group received assessment but no feedback at any time (“assessment only”).

The study assistants remained blinded to the allocation. Allocation was not disclosed to participants until they received feedback or not.

The intervention utilized expert system technology to automatically generate individualized feedback. It was an advanced version of an intervention that had proven effective for general hospital patients with at-risk alcohol consumption [2]. The intervention included three contacts. At baseline, participants answered questions about their alcohol consumption and motivational constructs based on the transtheoretical model (TTM, i.e., motivational stage of change, self-efficacy, decisional balance, and processes of change) on a tablet computer. The expert system software analyzed the data in comparison to data from a German adult general population sample, selected appropriated feedback modules, and generated a normative feedback letter. The letter included graphical and textual feedback on alcohol consumption in comparison to persons of the same sex and age and on TTM constructs in comparison to persons in the same motivational stage of change. As described in more detail elsewhere [32, 33], the content and number of feedback elements were tailored to a person’s alcohol-related risk level and motivational stage of change. If appropriate, the letters referred to indicated pages in an accompanying stage-tailored paper manual for further information and advice about low-risk drinking. Study assistants then sent out the letters to the participants by mail. At 3 months, study assistants conducted computer-assisted telephone interviews (CATIs) that asked the same questions as at baseline. The software generated an ipsative feedback letter that included information on own changes in drinking and TTM constructs from baseline to 3 months. The letter was sent by mail. At 6 months, CATIs were again conducted that ask the same questions as at baseline. The software generated a second ipsative feedback letter, reflecting changes from 3 to 6 months, and this was also sent to the participants by mail.

Since the intervention provided feedback based on a participant’s assessment data, completing the assessment was a prerequisite for receiving the intervention. At 12 months, participants were asked whether they had received any feedback on their alcohol consumption within the last 12 months. Those who remembered receiving feedback were asked how many letters they had received. Of the 645 intervention group participants who were followed-up at month 12, 519 (81%) remembered receiving any alcohol feedback. Of the 572 participants who completed all three intervention assessments, 483 (84%) remembered receiving three feedback letters.

The assessment at 3 months was completed by 1,402 (85%) participants, the assessment at 6 months was completed by 1,332 (81%) participants, and the 12-month follow-up assessment was completed by 1,314 (80%) participants. A more detailed flow chart of the PRINT study following the CONSORT criteria can be found elsewhere [33] (Fig. 1).

Data were assessed using either self-administered questionnaires provided on tablet computers at baseline or standardized computer-assisted telephone interviews at months 3, 6, and 12.

The number of HDDs in the last week prior to each assessment was determined using the timeline follow-back method [35]. Participants were asked to indicate the number of alcoholic drinks they had on each of the 7 days prior to the day of assessment. A drink was defined as 14 g pure alcohol (assumed to be equivalent to 0.25–0.3 L beer, 0.1–0.15 L wine/sparkling wine or 4 cL spirits). In accordance with National Institute on Alcohol Abuse and Alcoholism recommendations days with ≥4 drinks on a day for women and ≥5 drinks for men were defined as HDDs [4].

Tobacco smoking status was assessed by the question “Do you smoke currently?” and four possible responses: “No, I have never smoked” (persons who never smoked), “No, I am not a smoker now” (persons who formerly smoked), “Yes, I smoke daily” (persons who currently smoked daily), and “Yes, I smoke sometimes” (persons who currently smoked less than daily).

Covariates were assessed at baseline and included sex, age, school education, and employment status. School education was separated in <10, 10–11 or ≥12 school years. Employment status encompassed full-time employment, part-time employment, being a student, unemployment, and other (being retired, a homemaker or similar).

Using Mplus 8.7, latent growth curve models for count outcomes were calculated to test whether tobacco smoking moderates the efficacy of BAI on reducing HDDs [36]. All available data were used to calculate the models (intention to treat principle) based on a missing at random assumption using full-information maximum likelihood estimator. To decide on the form of the growth curves, rescaled likelihood ratio tests were performed [37] which suggested that a linear latent growth curve model fitted the data best (log-likelihood ratio = 1.45, difference in free parameters = 1, p = 0.22). As shown in Figure 2, the latent growth curve model included two latent growth variables representing the initial level of alcohol use (intercept) and the linear rate of change (linear slope). Repeated measures of HDDs were regressed on growth factors representing the growth trajectory using a Poisson model. The intervention effect on HDDs was calculated by regressing the linear slope on the study group. To evaluate the potential moderator effect of tobacco smoking status (persons who never smoked, persons who formerly smoked, persons who currenty smoked less than daily and persons who currently smoked daily) the linear slope was regressed on study group, tobacco smoking status, and an interaction term between study group and every tobacco smoking status, respectively. A sensitivity analysis was performed by adding covariates to the model.

Among all study participants, 55.9% (n = 920) were females and 44.1% (n = 726) were males (Table 1). The mean age of participants was 31.0 years (SD = 10.8). The initial assessment revealed that 1,085 persons (65.6%) drank at a low-risk level, while 553 (33.9%) drank at risk. Furthermore, 8 persons (0.5%) exhibited signs of probable alcohol use disorder. At baseline, the average number of HDDs in the past week was 0.4 (SD = 1.1), at 3 months 0.4 (SD = 1.0), at 6 months 0.3 (SD = 0.9), and at 12 months 0.4 (SD = 1.0). Among the participants, 52.6% (n = 866) have never smoked, 15.1% (n = 248) have formerly smoked, 10.3% (n = 169) have currently smoked less than daily, and 22.0% (n = 363) have currently smoked daily. The mean number of HDDs in the past week in relation to tobacco smoking status is shown in Table 2. Further information about the study sample can be found elsewhere [33].

The moderating effect of the different tobacco smoking statuses on the efficacy of BAI on HDDs was not statistically significant. Figure 3 shows the 12-month change in HDDs by tobacco smoking status and study group as indicated by incidence rate ratios (within tobacco smoking status group differences). Table 3 shows the 12-month intervention effect on HDDs by tobacco smoking status (between tobacco smoking status group differences) for the moderator analysis and the sensitivity analysis. In the moderator analysis persons who never smoked decreased HDDs significantly, both in the intervention group (9% decrease, p = 0.012) and in the control group (10% decrease, p = 0.007). The difference between these two groups was not significant (p = 0.847). No significant decrease was found among all other groups of persons who currently or formerly smoked tobacco.

Persons who formerly smoked decreased HDDs by 11% (p = 0.111) in the intervention group and by 1% (p = 0.845) in the control group, but the difference was not statistically significant (p = 0.243). Persons who currently smoked less than daily increased HDDs by 11% (p = 0.059) in the intervention group, and by 13% (p = 0.007) in the control group. The difference was not significant (p = 0.782). Persons who currently smoked daily increased HDDs by 6% (p = 0.224) in the intervention group, those in the control group reduced HDDs by 3% (p = 0.383). The difference was not significant (p = 0.125). Adding covariates did not substantially change the estimate of the outcome. The results of the sensitivity analysis can be found in Table 3.

This randomized controlled BAI trial in an adult general population sample of 1,646 persons with alcohol use in the past 12 months revealed four main findings. First, there was no moderating effect of the smoking status on the efficacy of BAI. Second, persons who never smoked reduced HDDs significantly with and without intervention. Third, persons who formerly smoked presented a non-statistically significant tendency to reduce HDDs with intervention but not without intervention. Fourth, persons who currently smoked less than daily tended to increase HDDs with and without intervention. However, this again was statistically not significant.

Smoking status had no moderating effect on the outcome of our computer-based BAI. Our findings are in line with a study that found no moderating effect of the tobacco smoking status on a computer-based BAI in at-risk drinking patients in a general hospital [31]. Only limited evidence has been found on the moderating effect of tobacco smoking on BAI efficacy. Some studies suggested a negative influence of the tobacco smoking status on the alcohol consumption outcome in alcohol use dependence treatment [28, 29]. A study with a general population sample was limited to adolescents and found a greater reduction of HDDs in the group of persons who smoked tobacco compared to persons who did not [30]. To our best knowledge there are no other studies that investigated the effect of tobacco smoking on the efficacy of computer-based BAIs in a sample of adults drinking alcohol at any quantity or frequency.

In our study, persons who never smoked reduced their HDDs. The intervention had no effect on this reduction. Possible explanations are that these individuals may be more aware of health-risk behaviors leading to a healthier lifestyle and higher motivation to change their behavior [38]. Evidence showed a healthier diet for persons who have not smoked compared to persons who have smoked [39] and lower odds ratios for persons who have not smoked in having multiple health-risk factors (such as only short time of physical activity, low fruits and vegetable intake, and alcohol consumption) compared to persons who have smoked [40]. Persons who never smoked among alcohol consumers might have less a load of health-risk behaviors. Therefore, they are probably able to reduce their HDDs without intervention. A reconsideration of the own alcohol consumption habits as an assessment effect could be possible as well [41].

Although not statistically significant, persons in the intervention group who formerly smoked tended to reduce their HDDs, while those in the control group did not. This trend may be explained by greater sensitivity to BAI as they have already succeeded in changing their smoking habits. Personalized feedback might by trend be effective among them because they are probably more aware of consequences due to health-risk behaviors. Studies showed a higher awareness for negative health consequences [42] and a higher tobacco smoking-related self-efficacy [43] for persons who formerly smoked compared to persons who currently smoked. Using the same strategies to quit multiple health-risk behaviors [44, 45] could help persons who formerly smoked to benefit more from BAI.

Persons who currently smoked tended to increase their HDDs, although changes were not statistically significant. Studies found that persons who currently smoked tended to underestimate their personal health risk due to tobacco smoking (optimism bias) [46], even though they report more respiratory symptoms than persons who formerly smoked [42]. Individuals who combine drinking and smoking are likely to require more intensive interventions. For future BAIs it may be useful to investigate the smoking status and the motivation to change this habit. Therefore, persons who smoke and drink could benefit from a more intensive intervention, i.e., a multi-behavioral brief intervention. An intervention that aims to reduce tobacco smoking and alcohol consumption at the same time could reduce the risk of relapse when one substance increases the craving for the other. Multi-behavioral interventions showed the ability to simultaneously reduce binge drinking and tobacco smoking [47]. Furthermore, studies suggested that the combination of smoking cessation treatment and alcohol treatment in alcohol use dependent persons who smoke seemed not to weaken the outcome of the alcohol treatment [48, 49]. Addressing multiple interactions between alcohol consumption and tobacco smoking for disease and death might help raise awareness for the combination of both health-risk behaviors. However, the evidence seems equivocal on whether a multi-behavioral intervention targeting alcohol and tobacco use is more effective than a tobacco-only intervention in maintaining tobacco abstinence [50, 51].

Delivering brief interventions via app on smartphones could be beneficial for a population outside of healthcare settings. Ease of integration into daily routines, immediate feedback on drinking habits, and a proactive approach with reminders to use the app would facilitate the dissemination. In addition, apps could extend their feedback to other health-risk behaviors that interact with alcohol use such as tobacco smoking. A multi-behavioral intervention could be automatically offered to persons who combine drinking and smoking, if they indicate this in the app. Furthermore, it should be offered to persons seeking to reduce multiple health-risk behaviors at once. For persons that formerly smoked the app could explore the key factors of the successfully used strategy to quit smoking. It could use the same elements and support this with individualized motivational messages. Currently, the evidence for app-based interventions seems inconclusive [52‒54]. One study limited to college students found short-term effects of an app on reducing drinking quantity compared to controls [55]. Many studies proactively contacted the participants for follow-up [54, 55]. Achieving high retention rates without proactive contact seems to be an unsolved problem as maintaining a high retention rate in app-based studies seems to be complex [56]. In the future, with the use of artificial intelligence via large language models, even a message-based motivational interview seems to be conceivable [57].

This study has three strengths. First, after established efficacy of BAI among persons who drink at-risk in healthcare settings [5], the PRINT study is one of the first that tested a BAI addressing individuals with alcohol use irrespective of quantity, frequency, and consequences of their consumption in a setting beyond healthcare that provided access to the general population. Second, with the use of latent growth curve analyses, it was possible to handle missing data adequately and, therefore, to include all baseline participants in the analysis. Third, external validity was ensured by the large number of participants and the high participation and follow-up rate in a general population sample. Three study limitations should be noted. First, self-reported data were used. Social desirability bias might be possible and may have resulted in under-reporting of alcohol consumption and tobacco smoking. However, the time line follow-back method is a brief and validated tool to measure alcohol consumption [58]. Second, the PRINT study measured heavy drinking, and the results might not be comparable with binge drinking as the study did not measure the time it took persons to drink their amount of alcohol per drinking occasion in addition to the number of standardized drinks per day. Third, the intervention effects of the primary outcome were small. We wanted to investigate effects of BAIs on heavy episodic drinking in persons drinking on any risk level to collect more evidence before we draw a conclusion about its effectiveness in a general population. Future alcohol intervention studies with stronger effects should test the moderating effect of the tobacco smoking status to find out if persons consuming both substances need a different intervention.

In this study, no statistically significant moderator effect of tobacco smoking status was found on HDDs after brief intervention to reduce alcohol consumption. Our data suggest that the intervention is most likely to be effective among persons who formerly smoked tobacco. For persons who currently smoked tobacco the intervention seems not to be effective.

We thank the participants for repeated participation, the study assistants for data collection and Christian Goeze for software programming.

This clinical trial was registered before patient enrollment (date of registration: 12 March 2018), trial registration No. DRKS00014274 in the German Clinical Trials Register, and approved by the Ethics Committee of the University Medicine Greifswald (BB 147/15). Written informed consent was obtained from individuals to participate in the study.

The authors have no conflicts of interest to declare.

The study was funded by the German Research Foundation (BA 5858/2–1, BA 5858/2–3). The funder had no role in the design, data collection, data analysis and reporting of this study.

Cedric Gerbracht: statistical analysis and article drafting. Sophie Baumann: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, software, supervision, validation, and visualization. Andreas Staudt: data curation, formal analysis, investigation, methodology, project administration, software, supervision, and validation. Jennis Freyer-Adam: conceptualization, methodology, and software. Gallus Bischof: software. Christian Meyer: investigation. Ulrich John: conceptualization and methodology.

The data that support the findings of this study are not publicly available due to terms of written informed consent to which participants agreed but are available from the principal investigator (S.B.) upon reasonable request.