This is an open-access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on http://www.jmir.org/, as well as this copyright and license information must be included.
In recent years, interventions that deliver online personalized feedback on alcohol use have been developed and appear to be a feasible way to curb heavy drinking. Randomized controlled trials (RCTs) among the general adult population, however, are scarce. The present study offers an RCT of Drinktest.nl, an online personalized feedback intervention in the Netherlands.
The aim of this study was to assess the effectiveness of computer-based personalized feedback on heavy alcohol use in male adults.
Randomization stratified by age and educational level was used to assign participants to either the intervention consisting of online personalized feedback or an information-only control condition. Participants were told as a cover story that they would evaluate newly developed health education materials. Participants were males (n = 450), aged 18 to 65 years, presenting with either heavy alcohol use (> 20 units of alcohol weekly) and/or binge drinking (> 5 units of alcohol at a single occasion at least 1 day per week) in the past 6 months. They were selected with a screener from a sampling frame of 25,000 households. The primary outcome measure was the percentage of the participants that had successfully reduced their drinking levels to below the Dutch guideline threshold for at-risk drinking.
Intention-to-treat analysis showed that in the experimental condition, 42% (97/230) of the participants were successful in reducing their drinking levels to below the threshold at the 1-month follow-up as compared with 31% (67/220) in the control group (odds ratio [OR] = 1.7, number needed to treat [NNT] = 8.6), which was statistically significant (χ2
1 = 6.67,
Personalized online feedback on alcohol consumption appears to be an effective and easy way to change unhealthy drinking patterns in adult men, at least in the short-term.
International Standard Randomized Controlled Trial Number: NTR836; http://www.trialregister.nl/trialreg/admin/rctview.asp?TC=836 (Archived by WebCite at http://www.webcitation.org/5ytnEz2vp)
The present study aims to examine the effect of Drinktest (www.drinktest.nl), an online personalized feedback intervention targeted at heavy drinking adults in the Netherlands. It is important to inform heavy drinkers about the possible consequences of their drinking behavior. After all, heavy alcohol consumption is highly prevalent worldwide [
Unfortunately, it is not easy to reach heavy drinkers with face-to-face interventions. There often is a shortage of health care professionals who can deliver the interventions, even in resource rich countries. Moreover, heavy drinkers may be reluctant to discuss their drinking behavior [
Offering interventions online may help to solve this problem. Online interventions designed to decrease alcohol consumption have proven to be feasible instruments to reach heavy drinkers and are generally well received [
Most of the evidence regarding the effectiveness of online alcohol interventions is collected in studies aimed at college students. In a recent meta-analysis, Carey and colleagues [
However, for population segments other than students, the results found in literature are not yet conclusive. A recent meta-analysis focusing on online alcohol and tobacco interventions in the general population suggested positive outcomes, that is, an overall effect size (Cohen’s d) of 0.22 (95% confidence interval [CI] 0.14 - 0.29) for the alcohol interventions [
A number of studies did report positive effects of online self-help alcohol modules in the general adult population. A recent meta-analysis [
The present study examines the effectiveness of Drinktest, a single 10-minute online session in which tailored feedback is delivered, with no therapist involved. (See
Besides Drinktest, only one other single-session Internet-based personalized feedback intervention (Check Your Drinking or CYD) was examined in an RCT directed at the general adult population [
Drinktest was developed by the Netherlands Institute for Health Promotion and Disease Prevention (NIGZ). Drinktest offers brief personalized feedback regarding in an individual’s personal alcohol consumption patterns. The intervention consists of various components: overview of mean weekly alcohol intake, associated health risks, self-help guidelines to reduce alcohol intake, normative feedback to compare one’s own alcohol consumption to the level of one’s own cohort. A first version of Drinktest was found to effectively reduce alcohol intake in women but not in men [
A screening questionnaire was administered to all men aged 18 to 65 (n = 9000) in two nationally representative panels consisting of 25,000 households that can receive online questionnaires. Our questionnaire contained the Quantity-Frequency Variability index of alcohol intake (QFV) [
In total, 817 men fulfilled the inclusion criteria and were willing to consider participation in the study. Additional participants were recruited through advertisements in national newspapers, to which 70 eligible men responded. All 887 men were contacted by telephone and asked to participate. After indicating their understanding that the research included a visit to the university, a total of 450 out of the 887 (50.7%) men contacted agreed to participate and gave informed consent. After 1 month, 413 participants were successfully followed-up. Of the 37 out of the 450 (8.2%) lost to follow-up, 2 had moved away and 35 did not respond. After 6 months, 403 participants were followed up successfully. Of the 47 out of the original 450 (10.4%) lost to follow-up, 4 had moved away, 41 did not respond, and 2 had died.
Flow chart
At screening, participants were told a cover story to reduce the risk of response bias stemming from social desirability. Participants were told that they would judge newly developed educational materials addressing one of three possible life style topics: alcohol, smoking, or exercise. They were then told that they were randomly assigned to the alcohol group and that they were invited to evaluate the materials irrespective of their actual alcohol intake. They also received information on the procedure of the study, which consisted of one visit to the university and three written questionnaires at 0, 1, and 6 months to be filled in at home. Those responders who were eligible and willing to participate were contacted by telephone to explain to them again what participation would entail and to schedule the appointment at the behavioral laboratory. It was not revealed to the participants that their inclusion in the study was based on the degree of their alcohol intake. They were then randomized to either the computer-based personalized feedback (experimental condition) or the control group.
Prior to the appointment, participants received an informed consent form and a baseline questionnaire at their home address and were asked to bring both to their appointment. Written and signed consent was thus ensured. The baseline questionnaire included items measuring alcohol consumption and demographic characteristics.
On arrival, participants received a short standardized instruction on how to use the computer and Internet site (experimental condition) or how to read the leaflet (control condition). Participants were seated individually in a soundproof room for 20 minutes in order to ensure exposure to the educational materials and to reduce the effect of possible extra-experimental factors. Materials were offered to participants identical to real-life setting in order to maximize external validity of the study. Participants in the experimental condition completed the test online and were given the opportunity to make a printed copy of their personalized feedback. Participants in the control condition read the leaflet on paper in full color print and could take a copy home afterwards.
Next, participants received a short evaluation questionnaire with dummy questions to maintain the cover story and to determine participants’ evaluation of the educational materials. They were then given the first part of their payment (€25).
At 1 month and at 6 months after studying the educational materials, participants received postal questionnaires. The first follow-up included questions about drinking behavior as well as determinants of alcohol consumption. The second follow-up also contained measures of alcohol consumption to assess effect maintenance. All items regarding alcohol consumption included in the measurements at different time points were identical so that comparisons could be made over time. After completing and returning the last follow-up, participants received the second payment (€25).
In order to minimize dropout rates, participants were sent a first reminder to return the follow-up questionnaire after 2 weeks, were sent a second reminder with a new questionnaire after another week, and were contacted by telephone for a third reminder if needed. After the last follow-up period had ended, participants received a standard letter explaining the true objectives of the study. The study was approved by the medical ethics committee of Erasmus Medical Centre in Rotterdam (reference number MEC-2006-343).
Participants in the experimental condition received brief personalized feedback on alcohol use through the website www.drinktest.nl. The test is designed for adults who consume alcohol regularly or excessively and invites them to explore the possible negative consequences of their drinking behavior. The aim is 2-fold: prevention of heavy drinking and reduction of alcohol intake in heavy drinkers. In the first part of the test, respondents are asked to report their weekly alcohol consumption and number of binge drinking occasions and to indicate whether they think they consume too much alcohol and whether they intend to reduce their alcohol intake in the future. Based on this information, respondents receive the first part of the advice, which covers possible consequences of their drinking behavior. The first part of the feedback also includes a normative component in which participants can compare their alcohol consumption to that of others in the same age and gender bracket. Previous literature has revealed that including normative feedback in brief interventions aimed at reducing alcohol intake has favorable effects because people generally overestimate alcohol intake of others and underestimate their own alcohol intake [
In the second part, the participants are asked questions concerning their drinking moments, drinking pattern, self-efficacy, attitude, and intention (behavioral stage according to the transtheoretical model [
An early version of the intervention was designed in 2002, and its effectiveness was evaluated in a randomized controlled trial from 2005 to 2006 [
Participants in the control condition were given a standard brochure (“Facts About Alcohol” [
Randomization was conducted using a computer random number generator in the Statistical Package for the Social Sciences (SPSS), version 15.0. (SPSS Inc, Chicago, IL, USA). Randomization was stratified by age and educational level to ensure a good balance of these prognostically relevant characteristics of the participants across the experimental conditions. The condition to which participants were assigned was revealed to research assistants once recruitment was complete. All participants were blinded to assignment by providing them with a cover story (see “Procedure” section above).
The primary outcome measure was heavy drinking, defined as alcohol consumption exceeding the guidelines for low-risk drinking: an average of more than 20 alcohol units per week (excessive drinking) and/or more than 5 units on a single occasion on at least 1 day per week (binge drinking). Alcohol units per day per week were assessed with the Dutch version of the QFV [
The trial was powered to detect changes in alcohol consumption comparable to those found in a previous trial [
To check whether randomization had resulted in two comparable groups, logistic regression analysis was used with condition as the dependent variable and a set of possible confounders (among them age and level of education) as predictors. Following the CONSORT (consolidated standards of reporting trials) statement all our analyses were conducted in agreement with the intention-to-treat (ITT) principle while imputation was used to deal with loss to follow-up. Imputation of missing values was done using the expectation-maximization algorithm of Little and Rubin [
Furthermore, we repeated all the analyses described above for completers only and when imputation was carried out using the last observation carried forward method. Finally, in order to assess who benefited most from the intervention at the 6-month follow-up, interaction terms were computed by calculating the products of the intervention dummy (intervention versus control) with four dichotomous variables: (1) age (18-44 vs > 44), (2) education (high vs low, ie, academic or college degree versus lower levels), (3) weekly alcohol units at baseline (< 28 versus ≥ 28), and (4) binge drinking at baseline (at least once per week versus less frequently). We then entered these interaction terms together with the corresponding main effects into the logistic regression model.
All tests were conducted at alpha = .05 (two-sided) except for the check on randomization, which was done at alpha = .10 to ensure that also relatively small baseline differences between groups in terms of age and level of education would be detected. Data were analyzed using SPSS version 15.0.
Participant characteristics and primary outcome measures at baseline
Experimental Condition |
Control Condition |
Test Result |
|
||
|
|
.87 | |||
Mean (SD) | 40.6 (15.2) | 40.3 (15.1) | |||
|
χ²2 = 1.2 | .54 | |||
Low, n (%) | 36 (15.7) | 43 (19.5) | |||
Medium, n (%) | 70 (30.4) | 62 (28.2) | |||
High, n (%) | 124 (53.9) | 115 (52.3) | |||
|
χ²1 = 0.6 | .44 | |||
Living with partner, n (%)b | 113 (49.6) | 101 (45.9) | |||
|
χ²1 = 0.02 | .89 | |||
Paid employment, n (%)b | 128 (56.1) | 125 (56.8) | |||
|
|
.60 | |||
Mean (SD)c | 30.9 (19.2) | 31.7 (14.3) | |||
|
|
.49 | |||
Mean (SD)d | 2.1 (1.3) | 2.2 (1.3) |
a Low = elementary or high school, medium = occupational certificate, high = university or college degree
b Includes 2 missing values
c A standard unit of alcohol contains 10 grams of ethanol
d Frequency of binge drinking defined as frequency of consuming more than 5 units of alcohol on at least one single occasion per week
Loss to follow-up after 1 month was 8.2% (37/450) and was evenly distributed across both conditions (n = 18 in the experimental condition and n = 19 in the control condition) (χ²1 = 0.98,
Comparison of the participants’ characteristics at baseline between those successfully followed up and those lost during follow-up
Followed up |
Lost to Follow-up |
Test Result |
|
||
|
|
.003 | |||
Mean (SD) | 41.2 (15.0) | 34.2 (14.8) | |||
|
χ²2 = 0.87 | .65 | |||
Low, n (%) | 70 (17.4) | 9 (19.1) | |||
Medium, n (%) | 116 (28.8) | 16 (34.0) | |||
High, n (%) | 217 (53.8) | 22 (46.8) | |||
|
χ²1 = 5.29 | .02 | |||
Living with partner, n (%)b | 199 (49.6) | 15 (31.9) | |||
|
χ²1 = 0.63 | .43 | |||
Paid employment, n (%)b | 229 (57.1) | 24 (51.1) | |||
|
|
.09 | |||
Mean (SD)c | 30.8 (17.0) | 35.2 (17.9) | |||
|
|
.29 | |||
Mean (SD)d | 2.1 (1.3) | 2.3 (1.2) |
a Low = elementary or high school, medium = occupational certificate, high = university or college degree
b Includes 2 missing values
c A standard unit of alcohol contains 10 grams of ethanol
d Frequency of binge drinking defined as frequency of consuming more than 5 units of alcohol on at least one single occasion per week
Change in success rates of adherence to the low-risk drinking guideline at the 1-month follow-up
Experimental Condition | Control Condition | ||||||||||
n | % Success | n | % Success | OR | 95% CI | NNTa | χ² |
|
|||
|
|||||||||||
Total sample |
230 | 42.2 | 220 | 30.5 | 1.7 | 1.13-2.46 | 8.6 | χ²1 = 6.7 | .01 |
||
Completers-only | 207 | 41.5 | 195 | 29.2 | 1.7 | 1.14-2.60 | 8.1 | χ²1 = 6.6 | .01 | ||
Total sample |
230 | 37.4 | 220 | 25.9 | 1.7 | 1.14-2.56 | 8.7 | χ²1 = 6.8 | .01 | ||
|
|||||||||||
Total sample |
230 | 45.7 | 220 | 37.3 | 1.4 | 0.97-2.06 | 11.9 | χ²1 = 3.3 | .07 |
||
Completers-only | 195 | 44.1 | 188 | 36.7 | 1.3 | 0.90-2.05 | 13.5 | χ²1 = 2.2 | .14 | ||
Total sample |
230 | 37.4 | 220 | 31.4 | 1.3 | 0.88-1.93 | 16.7 | χ²1 = 1.8 | .18 |
a NNT = numbers needed to treat
b EM imputation = imputation based on the expectation-maximization algorithm
c LOCF imputation = imputation based on last observation carried forward
At 6 months after studying the educational materials, even more participants in both conditions decreased their alcohol consumption to below the limits of heavy drinking, that is, 45.7% (105/230) in the experimental condition and 37.3% (82/220) in the control condition, but the difference between the conditions was no longer significant (χ²1 =3.3,
Analyses of the predictor-by-treatment interaction effects showed that favorable treatment response at the 6-month follow-up was not modified by any of the patient characteristics as measured at baseline, that is, age (Wald test = 3.03, df = 1,
The results of this study show that computer-based personalized feedback is successful in decreasing the percentage of male heavy drinkers in the short run. After 1 month, participants who received the intervention were more successful than controls in bringing down their alcohol consumption, even to within the guideline norms for low-risk drinking (42% versus 31%). However, after 6 months, the success rates were 46% versus 37% for the intervention and the control condition, respectively, and did not reach statistical significance, either under an intention-to-treat analysis or the completers-only analysis.
Our findings lend partial support to the idea that computer-based personalized feedback has a more favorable effect on the reduction of heavy drinking than a standard brochure on alcohol consumption, at least in the short-term. The initial effect of the intervention is further confirmed by the number needed to be treated (NNT), estimated at 8.6, which is comparable to, for example, the NNT found in Riper and colleagues (where NNT was equal to 8.5) in a more intensive intervention directed at the same population (ie, a general population >18 years of age) [
These results appear to confirm previous findings on the effectiveness of online alcohol interventions [
The findings should be seen in the light of the limitations and strengths of our study. To begin, a substantial percentage of participants in both conditions decreased their alcohol consumption to below the Dutch guidelines for low-risk drinking. An explanation for this may be that compared with those not willing to participate in the study (dropout before randomization was 50%), participants may have been more motivated to change their alcohol intake. The fact that subjects thought they would evaluate educational materials irrespective of their actual alcohol intake, however, turns this into an unlikely explanation of the favorable effect. Also, the repeated alcohol questions participants were asked to fill out at different time points may have had an intervention-effect. Neither the possible high motivation of our subjects nor the intervention effect of the alcohol measures, however, explains the differences in alcohol consumption at follow-up between the intervention and control condition. Moreover, the effect size found in our study is similar to those reported in face-to-face inventions aimed at adults in the general population [
The overall loss to follow-up in our study was limited to 10%. This is a low percentage compared with the average loss to follow-up of 35% as found in meta-analyses on online alcohol interventions [
A limitation of this study is that we relied on self-reported measures. However, Del Boca and Darkes [
Due to the nature of the educational materials in our study, blinding of participants was not possible. This may have led respondents to underreport their alcohol consumption at follow-up. We tried to minimize the possibility of social desirability bias in the primary outcome by using a cover story. Respondents were not informed of the true objectives of the study until after the last follow-up questionnaires were received.
An additional drawback of the present study may be that the sample was limited to adult men, and the results may, therefore, not simply be generalized to women. However, restricting the present study to men was a deliberate choice, since in our previous study with an earlier version of the intervention we demonstrated that it was beneficial to women but not to men [
Finally, it may be seen as a limitation of our study that we invited participants to visit our laboratory, which may threaten the external validity. This may, for example, have induced participants to spend all the available time actually reading the tailored advice, whereas if they were to conduct the intervention in their home environment, participants might not complete the entire intervention due to possible distractions or time constraints. Similarly, in a private setting, participants may not be able to ask for help, whereas in the laboratory situation a research assistant was present if they needed any support on the use of the computer and the Drinktest website. However, at this stage it was also a conscious choice to conduct a randomized controlled trial under laboratory conditions in order to assess the efficacy of the intervention after which an effectiveness study with a pragmatic randomized trial could be conducted. Furthermore, a recent meta-analysis [
Personalized online feedback on alcohol appears to be an effective and fairly easy way to change unhealthy drinking patterns in adult men, at least in the short-term. Drinktest.nl yearly draws about 90,000 male visitors. Of these, 70% (63,000) report to be heavy drinkers, and 40% (25,200) of these heavy drinkers complete the test, implying that they actually receive the complete tailored feedback. Based on the NNT at the 6-month follow-up, assuming that the revised Drinktest will attract the same number and type of visitors each year and assuming that conducting Drinktest in a private setting would generate the same effects, this would imply that more than 2000 men per year (n = 2117) will successfully reduce their alcohol intake during at least 6 months as a consequence of spending 10 minutes of their time on Drinktest.nl. Offering personalized feedback on alcohol through highly accessible Internet sites may thus contribute to generating health gains at the population level in an efficient and economically affordable way. In fact, Smit and colleagues calculated that introducing evidence-based eHealth interventions such as Drinktest into the Dutch health care system would substantially improve the cost-effectiveness of the system for alcohol use disorders overall [
The study was funded by the Netherlands Health Research Council (ZonMw), Grant # 50-50110-98-235. We thank Lotte Ploegmakers, Anouk Vogelzang, Soenita Ganpat, Monique de Hoog, Esther Beekman, and Estelle Wienk for their help with the data collection and all participants for their contribution to this study. Also thanks to Rob Bovens, Odile Smeets, Anke Oenema, and Reinout Wiers for their helpful comments to previous versions of this manuscript as well as to Ingmar Franken for letting us use “his” behavioral lab at the Erasmus University Rotterdam.
None declared
Screenshot of Drinktest.nl
confidence interval
consolidated standards of reporting trials
Check Your Drinking
intention-to-treat
Netherlands Institute for Health Promotion and Disease Prevention
number needed to treat
odds ratio
Quantity-Frequency Variability (index of alcohol intake)
risk difference
randomized controlled trial