The study design and flow of participants is shown in Figure 1. Ethical approval was obtained from the University of New South Wales Human Research Ethics Committee and consent was obtained electronically from all participants. The study was conducted entirely online with all contact occurring via automated emails. Following online screening, eligible participants were asked to provide informed consent in order to take part and were randomized to one of two conditions delivered over four weekly sessions: (1) the DEAL Project or (2) HealthWatch. Access to the website for each of the programs was for a period of 10 weeks from the point of randomization. Randomization was automated within the online program; therefore, the trial researcher was blind to randomization. This process occurred immediately after the eligibility screener, consent form, and baseline assessment were completed. Block randomization was conducted with a 1:1 ratio; however, due to a programming error (which included test users within the blocks), a group imbalance occurred resulting in 60 participants being randomized to the DEAL Project and 44 randomized to HealthWatch. Participants then completed a baseline assessment on entry to the study and follow-up assessments at posttreatment (5 weeks), and at 3 and 6 months postbaseline.

Recruitment took place between July 2013 and February 2014 using extensive media coverage across Australia, including tertiary institution flyers and street press, radio and newspaper stories, treatment services websites, and paid Facebook and Google advertisements. Individuals were eligible for the study if they (1) were aged 18 to 25 years, (2) reported current moderate depression symptomology (score of ≥7 on the Depression Anxiety Stress Scale [DASS-21-Depression]) [35], (3) were drinking at hazardous levels as measured by the Alcohol Use Disorders Identification Test (AUDIT; score of ≥8) [36], (4) had the ability to access the Internet (either in the private residence of the participant or willingness to use the public library/other suitable venue with Internet access), and (5) were Australian residents. Exclusion criteria included (1) a Psychosis Screener score ≥3 [37], (2) an inability to speak English, (3) serious risk of suicide in the past 2 weeks (serious thoughts of suicide and desire to act), and (4) daily use of cannabis/weekly use of amphetamines.

A structured online assessment was conducted at each of the four assessment time points. Primary outcomes were (1) depressed mood as measured by the Patient Health Questionnaire-9 (PHQ-9) [40] and (2) quantity and frequency of alcohol use as measured by the TOT-AL [41]. The PHQ-9 is a reliable and valid brief measure of depression severity over the past 2 weeks [42]. The self-report measure utilizes a 4-point Likert scale ranging from 0 (not at all) to 3 (nearly every day). Total scores range from 0 to 27. Scores ≤4 are considered to represent minimal depression, scores 5-9 are considered mild, 10-14 are considered moderate, 16-19 are considered moderately severe, and scores ≥20 are considered indicative of severe depression. The TOT-AL has been found to be a reliable and valid online measure of alcohol consumption [41]. The tool uses a dropdown menu of type, brand, and size of beverages consumed each day for the past week. The TOT-AL calculates the cumulative unit content of the drinks consumed over the previous 7 days (1 unit=approximately 8 g ethanol) to generate an overall consumption score (drinks per week) and frequency score (drinking days per week).

Automatic email prompts to complete online follow-up assessments were sent to participants at baseline, posttreatment (5 weeks), and 3 and 6 months postbaseline. Reminder emails were sent if the participant did not complete an assessment within 6 days (three emails per assessment in total). On completion of assessments, participants received an AU $10 iTunes voucher as reimbursement for time.

Data were analyzed using PASW Statistics 18 for Windows (release 18.0.0) [45]. Baseline differences between groups were examined using chi-square (using Yates continuity correction in 2×2 tables to avoid overestimation) and independent-samples t tests.

Analysis of outcomes was undertaken based on intention to treat using a series of generalized estimating equations (GEE) [46]. Baseline scores were modeled and controlled for in the GEE analysis. Linear and negative binomial (with log link) GEE were used to examine outcomes with normal and count distributions, respectively. An exchangeable correlation structure was used for all models. An interaction term for the intervention variable and time (group × time) was included in each model to test for differential change over time. When this interaction was nonsignificant (P≥.05), it was removed from the model and the analysis was rerun. Alpha was set at .05 and a least significant difference method was used for pairwise comparisons. Results are reported as the unstandardized mean difference (beta) with Wald-type 95% confidence interval (95% CI) and chi-square for linear models and the risk ratio (RR) with 95% CI for negative binomial models. Furthermore, standardized effect sizes (Cohen’s d) were calculated for the primary outcomes [47].

The sample consisted of 104 participants (female: 59.6%, 62/104) with a mean age of 21.74 (SD 2.22) years. The mean PHQ-9 score was 16.32 (SD 5.00) and the mean AUDIT score was 17.39 (SD 6.42). Drinking quantities and frequencies were positively skewed with a median of 15.20 standard alcoholic drinks per week (range 0-97) consumed over a median of 2.00 drinking days per week (range 0-7). There were no statistically significant differences between the groups on baseline characteristics (Table 1).

This RCT evaluated the feasibility and preliminary efficacy of the DEAL Project, a Web-based program that aims to reduce depression and alcohol use in 18- to 25-year-olds. The program demonstrated statistically significant greater reductions in depression and alcohol use compared to a control group at posttreatment. Furthermore, the positive outcomes observed among those randomized to the DEAL Project were maintained at 3- and 6-month follow-ups. However, between-group differences at these later time points disappeared because of statistically nonsignificant shifts in both control and treatment groups. There is evidence to suggest that in brief intervention trials, assessment alone may result in improved outcomes either as a consequence of assessment on subsequent self-report (known as the Hawthorne effect) [49] or as a catalyst to mobilize individuals into actual behavioral change [25]. As such, there is the potential that those in the HealthWatch condition may have derived benefit from not only the assessments, but also the thought involved in completing the surveys. Alternately, because this was a sample that sought out this treatment, the control condition may have been intensive enough for some change to be observed when combined with participant motivation for change. Participants may have also accessed other treatments. However, this was not borne out in our data on service utilization over the follow-up period. Treatment deterioration effects may also have led to this lack of between-group differences at 6 months. Finally, natural recovery cannot be disregarded as an alternative explanation for disappearance of differences between the two groups. Nevertheless, it would appear that the DEAL Project was associated with more rapid improvement in depression and alcohol outcomes compared with control.

Overall, mean PHQ-9 depression scores in the treatment group dropped from the “moderately severe” range to just outside the range for “mild depression” at 6 months. This was a clinically significant change [44]. At baseline, the treatment group were drinking, on average, 3 days per week and consuming more than 25 drinks per week. At 3-month follow-up, drinking occasions had halved and participants were drinking just over nine drinks per week. Although this reduced figure is still considered above the recommended range for short-term harm [50], it has potentially large public health implications (especially considering the automated nature of the program), including a reduction in risk of harm to the individual and the societal costs associated with heavy alcohol use, including violence, hospital and emergency department visits, road safety and drunk driving, and lost productivity [51]. The posttreatment effect sizes observed in this study were considerably stronger than previous research; however, at 3- and 6-month follow-ups, effect sizes looked similar to the small and moderate effects found in previous Internet-based trials for single disorder interventions [14,16,19,52,53]. Similarly, the overall effects of the DEAL Project program at 3- and 6-month follow-ups were similar to those of the SHADE program trial [24]. The DEAL Project is briefer, unguided, and delivered completely online, which is likely to increase cost-effectiveness and accessibility for youth.

Despite these various strengths, this study is not without its limitations. Although there was significant interest at the recruitment level (with more than 900 individuals beginning—and 839 completing—the screener over a 6-month period), the recruitment rate was low. However, this was unsurprising considering this was an opportunistic sample (not treatment-seeking). Consequently, on average, individuals randomized to the DEAL Project fully completed less than half of the four sessions offered to them. Of those who completed one session, approximately one-third went on to fully complete the program. This issue with adherence is unsurprising given the unguided nature of the program [54,55]. Despite considerable efforts in the program development stage to optimize adherence [56], this raises some questions around program acceptability and feasibility. Interestingly, the mean number of DEAL Project modules completed was equivalent to the number of modules completed by the younger participants in the previous SHADE trial [57], suggesting that adherence may be a generalized issue for this demographic. The DEAL Project’s brief structure allows for exposure to more key strategies before dropout. Although only one-third of the treatment group completed at least half of the program, module completion refers to full completion and does not account for participants sampling from different modules in a nonlinear way. Furthermore, the reasons for selective attrition are difficult to interpret because they may reflect the contradictory possibilities of dropout due to dissatisfaction versus dropout due to a sense that the individual feels their needs have been met [58]. Further research is needed to examine methods to improve engagement and retention within online programs. Nevertheless, despite the lack of adherence the positive results are encouraging for the utility of brief interventions. Similarly, although not dissimilar to previous studies without therapist/administrative guidance of participants [59,60], the follow-up rates were low. This may limit the generalizability and conclusions that may be drawn from this study. Telephone or face-to-face contact during online trials has been shown to increase adherence; however, this reduces real-world applicability [61]. Nevertheless, the GEE analysis used is robust to this level of loss to follow-up [62] and missingness was found to be at random.

Due to difficulties diagnosing disorders online, the sample was nondiagnostic. Nevertheless, the mean baseline PHQ-9 score was in the moderately severe range and the mean baseline AUDIT score indicated high-risk/harmful levels of use. According to the PHQ-9, more than half the sample had a likely depression diagnosis, whereas two-thirds endorsed at least one alcohol abuse diagnostic criterion and three-quarters endorsed more than two dependence diagnostic criteria. Thus, this was not a clinically insignificant sample. Furthermore, subthreshold conditions have been associated with comparable negative outcomes to full-disorder syndromes, especially in younger populations [63,64], and provide an opportunity for early intervention.

Additionally, this Australian sample may not generalize internationally. The programming error in randomization is a possible study limitation. Nevertheless, few differences were found between the groups at baseline. Finally, as with any study of this kind, there is the potential for self-report bias; evidence suggests self-report provides useful and accurate estimates when conditions are designed to maximize response accuracy [65]. Studies have shown that self-reports of alcohol use correlate with behavioral observations [66]. Furthermore, the anonymity provided by online assessment is likely to be more accurate than other forms of self-report [67].

Overall, the DEAL Project was associated with significant improvements in both depression symptoms and alcohol use among young people with these co-occurring conditions relative to control at posttreatment. However, although within-group improvements were maintained over the 6-month follow-up period, the significant between-group differences were no longer present at long-term follow-up. This study adds useful evidence to both the eHealth and comorbidity treatment fields. Further studies are required to better understand these long-term outcomes and address the program adherence and trial attrition issues that were present in this study.