Brief alcohol intervention is an umbrella term for a variety of interventions that are characterized by limited intensity or duration, and the common goal of motivating recipients to reduce their alcohol consumption. Examples encompass counseling, advice, and motivational feedback, which can be delivered face-to-face or remotely through digital technologies. There is clear evidence that brief interventions in general can reduce alcohol use in adults who exceed low-risk drinking limits [1-5]. This also applies to interventions that deliver automatically generated, personalized motivational feedback [6-10]. The use of digital technologies such as expert system software, that is, computer programs that automatically evaluate individual information and compile tailored feedback based on a set of predefined rules [11], enables the provision of highly individualized and motivation-enhancing feedback in a reliable and resource-efficient manner [12]. The feedback itself may be delivered web-based [13], through apps [14], short messages [15], letters [9], or a combination thereof [16], whereas the common core is a digitalized expert system.

Crucial questions remain concerning the applicability of brief alcohol interventions such as personalized motivational feedback to the broader population. Low-risk drinking limits [ 17] imply a threshold below which the risk of disease or premature death is marginal and considered acceptable. Therefore, low-risk drinkers were virtually not represented in the brief intervention trials to date.

However, studies emerging in recent years challenge the common practice of excluding them from interventions to reduce alcohol use. These studies revealed that methodological problems (eg, misclassification and uncontrolled confounding) that bias drinking risk estimates downward were common in past research, suggesting that the risk associated with low-risk alcohol intake may be greater than previously estimated [18-21]. Even minimal drinking amounts have been linked to an increased risk of adverse brain and cardiovascular outcomes, as well as prevalent cancers [22-25]. The large proportion of low-risk drinkers in the population might imply high numbers of cases of disease and death below previously recommended thresholds for at-risk drinking [26], which are neither marginal nor acceptable. Therefore, the greatest impact on health can be made by addressing alcohol use in the population as a whole. Accordingly, the latest (inter)national drinking guidelines do not differentiate between low-risk and at-risk alcohol consumption anymore [27-29].

There is a paucity of evidence regarding the long-term effects of brief alcohol interventions. Only very few studies included in the most recently published systematic reviews and meta-analyses investigated treatment effects after more than 1 year [1,3-8]. The impact of low-threshold brief interventions may not only unfold in the weeks directly following the intervention. It may take time before the applied behavior change techniques affect people’s motivation and ultimately, their behavior [7,30,31]. The most influential theories of health behavior change posit that intention is the best predictor of actual change [32-35]. However, the promotion of motivation to change does not guarantee a healthier lifestyle, as is well-documented in the intention-behavior gap [36]. Individuals may reside in a motivational state, contemplating the idea of drinking less alcohol, for a prolonged period of time before an actual change in alcohol consumption manifests itself, or possibly never does. Translated to the prevention context, the effects of brief alcohol interventions may require time to become measurable. To provide an example, reflecting on the pros and cons of drinking, re-evaluating this decisional balance from time to time, eventually setting the goal of drinking less, as well as reaching this goal might be a protracted process [37,38].

The burden of alcohol-related harm is unequally distributed in the general population, with certain groups, such as those with low levels of education, experiencing disproportionately high levels of alcohol-related morbidity and mortality [39-42]. Although the mechanisms underlying this so-called alcohol harm paradox are not well understood yet [43], reducing alcohol-attributable social inequalities is a fundamental public health goal. Alcohol prevention efforts should therefore be systematically evaluated for their equity impact [44], ie, their ability to reduce (positive impact), maintain (neutral impact), or increase (negative impact) health-related social inequalities. On the one hand, lower-educated groups are less likely to accept an offered lifestyle intervention [45,46]. On the other hand, there is considerable evidence that brief interventions are at least as beneficial for alcohol users with low as they are for those with higher socioeconomic status [6,47,48]. Regarding the brief alcohol intervention investigated in this study, evidence after 12 months suggested that alcohol users with lower school education benefited, while those with higher school education did not [49]. An important question is whether this positive equity impact could be sustained over a longer period of time.

In this study, a general population sample of adults who reported any alcohol use in the past year received either computer-generated individualized feedback or assessment only. No clear evidence of efficacy was observed after 1 year [50]. However, the intervention was found to reduce drinking among low-risk drinkers 6 months after enrollment [50]. Given these promising findings after 6 months against the lack of evidence regarding the differential efficacy of brief alcohol interventions in subgroups with different drinking patterns, in particular, those below low-risk drinking limits, we aimed to further explore how intervention effects might differ between low-risk and at-risk drinkers.

The first aim of this study was to compare the change in self-reported alcohol use between the intervention and control groups from baseline to 3 and 4 years after study start. Based on increasing effects of a similar intervention after 2 years in German hospital inpatients [9,51], we expected lower alcohol use in participants randomized to the intervention compared to the control group. The second aim was to explore to what extent the intervention effects, after 3 and 4 years, were moderated by alcohol use severity and school education.

The data of this randomized controlled trial using a general population sample of alcohol users showed no evidence of effects of individualized feedback based on a computer expert system after 4 years. Unexpectedly, we observed drinking reductions in the control group and no change in the intervention group 4 years after study start. Long-term intervention effects were not moderated by alcohol use severity or school education.

It is puzzling and somewhat paradoxical that individualized feedback might have produced potentially negative long-term effects, considering that (1) there was no difference between study groups after 1 [50] and 3 years; (2) the intervention was designed according to the principles of motivational interviewing [56], using benevolent and appreciative language to address the participants’ motivation and beliefs about their alcohol use; (3) preventive intervention are usually assumed to have few risks; and (4) evidence on the existence of adverse and unintended effects of brief alcohol interventions is scarce [64].

Among the studies included in the meta-analyses and reviews published in the last years [1,3,4,6,65-68], only 1 trial was identified that reported potentially adverse effects of a brief intervention targeting alcohol use [69]. After 3 and 6 months, the proportion of high-risk drinkers in a sample of patients with tuberculosis in South Africa decreased more strongly in the control group (treatment as usual) than in the brief counseling intervention group [69]. However, comparability appears to be very limited due to differences in study characteristics. The almost complete lack of studies reporting unintended brief intervention effects may be due to publication bias or to the fact that brief interventions usually operate at such a low motivational threshold that adverse effects seem unlikely. This makes it all the more important to disentangle the long-term effects mentioned earlier.

There are at least 3 possible explanations for our findings. First, the feedback provided in the intervention letters may not have had the desired effect, or even an opposite effect for some participants. Two-thirds of the sample were low-risk drinkers, a group that has been largely absent from brief intervention trials to date. The feedback may have been insufficiently tailored to the needs of this group. The motivational processes involved in reducing alcohol consumption may differ for different subgroups of drinkers. For instance, personalized normative feedback assumes that people underestimate their drinking in relation to their peers [67]. This may not be the case for all members of our population-based sample, which included a multitude of drinking patterns. Rather than enhancing the motivation to drink less, the feedback may have provided some participants with a form of justification for their drinking. If that is the case, future interventions may better omit risk levels such as low-risk or at-risk drinking that are no longer supported by evidence [18-25]. Second, participation in the follow-up assessment after 4 years was the lowest in the trial (59%) and selective. Although our analysis strategy can deal with data missing at random [70], the possibility of systematic bias due to data not missing at random can never be completely ruled out [71]. Incorporating nonignorable missingness into latent growth models is based on untestable assumptions and can result in a high degree of uncertainty [72] as was the case in our sensitivity analyses (Multimedia Appendix 2). Third, the follow-up after 4 years was conducted between April 2022 and June 2022, a period during which the German general population had experienced 2 years of restrictions and constantly shifting life circumstances due to the COVID-19 pandemic. The pandemic affected people’s alcohol consumption [73,74], their health literacy [75], as well as their trust in science [76]. During the follow-up interviews, some participants spontaneously indicated that they had retained the feedback letters and re-read them from time to time. Although we were unable to investigate this more rigorously, participants in the intervention group might have re-evaluated their feedback during the COVID-19 pandemic. Albeit highly speculative, it is possible that this could have facilitated undesired effects through an unintended false sense of reassurance or influenced the accuracy of self-reported drinking in the intervention group but not in the control group.

Temporary positive intervention effects among low-risk drinkers [50] and participants with a low or medium educational background [49] were not sustained over the longer term. In a high consumption country such as Germany [77], where the social pressure to drink is high, individualized feedback might need to be provided more frequently and consistently to achieve prolonged drinking reductions.

The public health potential of computer expert systems as brief intervention tools lies in their capacity to reliably provide highly individualized feedback to large groups of alcohol consumers, with relatively low effort once the expert system has been developed. This is particularly relevant given that brief alcohol interventions should not be limited to individuals who exceed former low-risk drinking limits, considering the health risks associated with drinking patterns below these limits [18-24]. Consequently, the target group of behavior-directed alcohol prevention encompasses a substantially larger share of the population than before. Motivational feedback based on digitalized expert systems is suitable for this purpose due to its efficiency and translational potential. In an increasingly digitalized world, the possibilities for providing visually appealing and easily accessible expert system feedback are manifold and growing constantly. In contrast, more terrestrial ways of feedback delivery, such as written letters sent via postal mail as used in the present study, may become more salient, leaving a lasting impression in a world characterized by mostly digital ways of communicating. Future studies could explore the role of delivery mode on how motivational feedback is processed and evaluated by recipients.

This study represents one of the first attempts to incorporate new evidence regarding the risk from low drinking amounts [18-21] into a universal preventive intervention targeting all people who drink alcohol, regardless of how much. High participation and retention rates were achieved with a large population-based sample, which was comparable to the general population regarding the prevalence of problematic alcohol use [78]. The intention-to-treat analysis, including several sensitivity analyses, ensures the reliability of the conclusions drawn from the study data. However, the following limitations need to be acknowledged. First, the participation was selective, potentially introducing bias through nonignorable missingness. Second, alcohol consumption was self-reported using a quantity-frequency approach, which is known to be subject to underreporting by respondents [79]. Third, the assessment of the primary outcome may not have been sensitive and fine-grained enough to detect changes among participants with infrequent drinking patterns. Fourth, the PRINT trial was originally designed to investigate intervention effects after one year and the required sample size was calculated based on group differences within this period of time. Following the acquisition of renewal funding, we were able to explore long-term intervention and moderation effects. Since the long-term follow-ups after 3 and 4 years were not initially scheduled in the PRINT trial, the results should be considered as exploratory in nature, with potentially limited statistical power.

This study contributes to the literature on brief interventions by providing insight into the long-term effects of computer-generated individualized feedback letters. While previous studies have largely focused on individuals who consumed alcohol above a certain risk threshold, this study expands the scope to include all drinkers.