Cost-effectiveness and cost-utility analysis of a web-based computer-tailored intervention for prevention of binge drinking among Spanish adolescents

study sought to assess the cost-effectiveness and of a web-based computer-tailored The sample was drawn from a study evaluating the Alerta Alcohol programme. The population consisted of adolescents aged 15–19. Decision tree analysis was used to estimate costs and health outcomes, as measured by number of BD occasions and quality-adjusted life years (QALYs). Incremental Cost-Effectiveness and Cost-Utility Ratios were also calculated from National Health Service (NHS) and societal perspective and for a time horizon of four months. Multivariate deterministic sensitivity analysis of best/worst scenarios by subgroups was used to account for uncertainty.


Methods
Design, population and sample The sample was part of a two-arm cluster randomized controlled trial that evaluated an intervention aimed at reducing binge drinking among adolescents (aged [15][16][17][18][19] in Andalusian secondary schools through a web-based computer-tailored programme known as Alerta Alcohol (more information about the design of the study may be found in Lima-Serrano et al. (2018) [17]). A total of 1,247 adolescents from 15 public high schools were assessed at baseline (January-February, 2017) and 612 adolescents at 4-month follow up (May-June, 2017). However, because the collection of cost data began later in time, we only had complete data on effectiveness and costs for 367 adolescents that were included in this economic evaluation analysis.

Intervention and comparator
Alerta Alcohol programme provided feedback through preventive messages and personalized information about the bene ts of not consuming alcohol, with the aim of reducing positive attitudes about and excessive consumption of alcohol, while also assessing social in uences and self-e cacy. The tailored messages were based on the I-change model, which integrates elements of various models of social cognition and self-regulation, and assumes that behaviour is the result of the individual's intentions, action plans and abilities [18][19][20]. This intervention comprised six sessions. In initial session or rst session, participants completed a baseline questionnaire which elicited information on demographics, alcohol use behaviours, mediator variables such as motivational determinants (attitude, social in uences, self-e cacy) and cost measures (healthcare and non-healthcare costs). The EQ-5D-5L questionnaire was used to measure health-related quality of life (HRQOL). This initial session was followed by presentation of three different scenarios (sessions 2-3) existing a 1-2-week period between sessions, a fourth session in which adolescents could accept the challenge of not consuming excessive alcohol at an upcoming event, a fth session to evaluate the response to the challenge and a sixth session, scheduled four months after the rst session, to evaluate the intervention. The follow-up questionnaire included the same items that the baseline questionnaire except for the demographic variables. The control group received only the baseline questionnaire and a follow-up questionnaire (sessions 1 and 6) not receiving any active intervention in between.

Model
A cost-effectiveness (CEA) and a cost-utility analysis (CUA) were carried out in order to evaluate the e ciency of the Alerta Alcohol programme compared with not engaging in any active intervention. A decision tree was developed consisted of three situations: the adolescent reduces, maintains or increases the number of BD occasions in the last 30 days (Fig. 1). Each arm of the decision tree resulted in health outcomes and costs. The health outcomes were, for the CEA, the mean of the difference between the number of BD occasions in the last 30 days in the post-intervention and pre-intervention periods, and for the CUA, the mean of the difference in QALYs obtained through the EQ utility index.
A time horizon of four months was evaluated, so no discount rate was applied because of the short time horizon. Subgroup analysis was carried out by age, gender and availability of pocket money. All estimates were calculated from the Spanish National Health Service (NHS) perspective as well as from the societal perspective. The two sets of estimates are presented separately and differentiated. Table 1 shows details of the data required to populate the model for costs and health outcomes. The analysis was conducted using Stata version 14.0 (StataCorp, College Station, TX, USA) and Microsoft Excel version 16.16.5.

Variables
Information on demographics (gender, age, economic situation at home, weekly pocket money, parentsé ducational level), alcohol use behaviours, other substances use, and mediator variables such as motivational determinants (attitude, social in uences, self-e cacy) were collected through the clinical trial using a questionnaire with the purpose of comparable working the two groups (intervention and control group).
Within demographics variables, the economic situation at home was obtained using the question "Of the following situations, which one would you identify with the most?". The response options were converted into a dummy variable, which indicated a value of 1 "good economic situation at home" and a value of 0 "other economic situation". This question was developed ad hoc and used in another study carried out by Lima-Serrano et al. [21].
The weekly pocket money availability was asked by means of the question "How many Euros do you have per week to spend on yourself?" with ve response options: 0 €, 1-10€, 11-20€, 21-30€, more than 30 €. Notwithstanding these amounts were recoded into three categories: 0 €, 1-20€ and more than 20 €; due to proportions of each response category. Similar recode was used in the study carried out by Díaz-Geada et al. [22]. Later, for analyses, this variable was converted to a numerical variable using the mean of each response option.
The parents´ educational level was calculated according to the number of schooling years after answering to a question with the following response categories: "No study", "Primary studies", "Baccalaurate/professional training", "University".
In relation to alcohol use, data about the number of BD occasions in the last 30 days, the alcohol use in the last week, family (father, mother and siblings) alcohol consumption frequency, family binge drinking frequency, peer alcohol consumption frequency and peer binge drinking frequency were asked. These questions as well as other variables related to risk perception and mediator variables such as motivational determinants (attitude, social in uences, self-e cacy) are part of ve scales to measure determinants of BD in Spanish adolescents validated by Lima-Serrano et al. [23].
As regards to other substances use although its frequency of consumption was measured through a selfreported question based on the ESTUDES' questionnaire [5], only costs of cigarettes and shishas were able to obtained from literature review.

Costs
The following direct healthcare and non-healthcare costs related to the Alerta Alcohol programme and BD behaviour were identi ed and measured: (1) intervention costs, (2) direct healthcare costs (i.e., costs for services within the healthcare sector), (3) direct non-healthcare costs (i.e., costs for services outside the healthcare sector), and (4) direct costs to the subject (e.g., costs associated with use of tobacco or other substances) (see Additional le 2 - Table A1 for more details).
To aid the comparative quantitative analysis, the mean unit and annual costs were converted to €2017 using country-speci c or country-group-speci c in ation on average consumer prices.

Outcome measures
For the CEA, the outcome unit measured was measured in terms of reducing the number of BD occasions in the last 30 days by gender, age and available pocket money. Data was provided by the clinical trial. It was analysed using a two-part model ( Table 2).   Note: Average values and standard deviations shown in brackets. *** , ** and * represent the signi cance level at 1%, 5% and 10%, respectively; 1,000 replications were used for bootstrapping and standard errors were clustered at classroom level. For the CUA, the outcome measure was quality-adjusted life years (QALYs), calculated by means of the EQ utility index on the basis of the answers given by the adolescents to the EQ-5D-5L questionnaire. We considered the EQ-5D-5L (adult version) as the appropriate questionnaire version, being this indicated from the age of 15 [24]. In addition, there is no a value set for EQ-5D-Y and then, it interferes with the calculation of QALYs [20]. This index was calculated by using the Spanish value set [25][26][27][28].

Presentation of results
The main model outputs used in this analysis were the incremental cost-effectiveness ratio (ICER) and the incremental cost-utility ratio (ICUR). We used the cost-utility threshold for Spain of €21,000 to €24,000 per QALY [29].

Subgroup and sensitivity analysis
Cost-effectiveness and cost-utility analyses were carried out for three subgroups (de ned by gender (female/male), age (< 17 years old/≥17 years old) and weekly pocket money (€0/€1 20/>€20)) due to differences found in the literature in these subgroups [7,[30][31][32]. Uncertainty was studied through multivariate deterministic sensitivity analysis of best/worst scenarios by the same subgroups mentioned above.

Sample Characteristics
In relation to socioeconomic characteristics, age at beginning of intervention by group (intervention and control group) was statistically signi cant, as well as current job situation of the adolescent´s father. In baseline period, there were statistically signi cant differences in relation to number of glasses of alcohol consumed in outdoor public places and siblings´ BD frequency. In post intervention period, there were statistically signi cant differences in relation to siblings´ alcohol use frequency, adolescent´s father BD frequency and adolescent´s shishas or hookahs use (see Table 3). Note: We show the average values and standard deviations in brackets. *** , ** and * represents statistically signi cant differences at 1%, 5% and 10% between values of variables in intervention group and control group in pre and post-intervention period (2nd and 3rd columns, 4th and 5th columns, respectively). to tra c accidents. Additionally, the number of BD occasions decreased and HRQoL increased in both groups, but the effect was greater in the intervention group than in the control group.
The programme showed a statistically signi cant reduction in number of BD ocassions in the older group (≥ 17 years). Females and those who had available pocket money of between €1 and €20 showed greater adherence to the intervention and a reduction in number of BD occasions.

Incremental cost-effectiveness and cost-utility ratios (ICERs and ICURs)
ICERs differed from both perspectives. Cost of reducing BD occasions by one per month was €16.63 from the NHS perspective. Notwithstanding, the intervention was dominant from the societal perspective resulting in savings of €7,986.37 by one BD occasion averted per month.
With regard to QALYs gained, the intervention was more expensive but also more effective, resulting in an incremental cost of €71.05 per QALY gained from NHS perspective in comparison with the control condition. From societal perspective, this intervention was dominant resulting in savings of €34,126.64 per QALY gained (see Table 4).

Subgroup and sensitivity analyses
An analysis by gender was carried out from both perspectives, but the results did not change at decision level from societal perspective when the effect measure used was QALY nding that the intervention was In relation to pocket money, the intervention proved more cost-effective for those who had no pocket money using both outcome measures, number of BD occasions averted and QALYs gained. However, some differences were found in those who had a pocket money available between €1 and €20, and those who had more than €20.
The best scenario showed that the intervention could be cost-effective in reducing the number of BD occasions from NHS perspective and dominant from societal perspective. As regards QALYs gained, the best scenario showed its cost-effectiveness from NHS perspective and its dominance from societal perspective (see Table 3 and Additional le 3 - Figure A2).

Discussion
To our knowledge, this study is one of the rst studies to examine the cost-effectiveness and cost-utility of a web-based intervention carried out among adolescents with the aim of preventing binge drinking in Spain. The intervention showed cost-effectiveness and cost-utility from both NHS and societal perspective, based on the BD occasions outcome measure and QALYs in comparison with no active intervention.
The inferiority of the intervention for certain subgroups could be explained in part by the nding that baseline consumption for these subgroups was relatively low compared with that of their counterparts.
For instance, the number of BD occasions in the baseline period for the female subgroup and for the older subgroup (17 and older) was markedly higher than for the male subgroup and for the younger subgroup (under 17 years). The major incremental effect obtained by subgroups in terms of number of BD occasions averted per month was for adolescents who had not weekly pocket money available reaching an incremental effect of 0.465. This fact was also found in the study of Crocamo et al. [30] in which a high pocket money availability was a risk factor for young people in relation to the number of BD episodes. On the other hand, the major incremental effect obtained in relation to QALYs gained was for those had a weekly pocket money of more than €20. In relation this nding, Nur et al. [31] found that young people who received an adequate amount of pocket money (understanding "adequate" similar to high) had a higher score in quality of life, speci cally in mental health. Hence, this could explain a better general HRQoL. In addition, the small incremental effect in QALYs could be explained due to heterogeneity found in another studies among binge drinker young people [33][34][35].
The ndings of this study are similar to those of the Alcohol Alert study carried out by Drost et al. (2016), in which, from both perspectives, and particularly, the intervention was more cost-effective in reducing the number of BD occasions per month for older adolescents (aged 17-19 years) than for those who were younger than 17 years old [7].
Given the scarcity of literature related to economic evaluations of behaviour change interventions, one of the main strengths of this study is the cost-effectiveness and cost-utility assessment including both the NHS and the societal perspective [36]. It is known that the societal perspective is dominant over other perspectives [32,[37][38]. Nevertheless, both the choice of this perspective and the way in which it was implemented in the design of the study can be considered important strengths. Another strength of our study is the separate reporting of data for subgroups, since the impact of behaviour change interventions might vary according to contextual factors, as noted by Das et al. (2016) [39].
Notwithstanding these strengths, it is necessary to contextualize the results of this study by taking into account the limitations encountered, the rst being the low response rate in the follow-up period and, in relation to the cost questionnaire, the failure to include questions related to costs from the beginning of the study. The main cause of missing data for the follow-up questionnaire was early completion of classes by vocational training students (whose classes ended before those of the other participants).
Moreover, the date for administering the post-intervention questionnaire fell close to the nal examination period in the schools involved, which made it di cult to ensure that all participants completed the questionnaire. It is known that high attrition rates are common in eHealth interventions [40,41].
A second limitation was that the data collected in the study came from self-reported questionnaires completed by the adolescents and may therefore have been affected by subjectivity. However, previous studies have found that self-reporting of risk behaviours among adolescents and young adults shows good reliability and validity [42][43][44].
Finally, another important limitation was that short-term behaviour change assessments capture little bene t, so we would need to monitor the effect of this intervention in the long term. However, according to a review of economic evaluations of behaviour change interventions [15], only six studies had longer follow-up periods (up to 5 years).

Conclusions
To conclude, computer-tailored feedback is a cost-effective way to prevent binge drinking in terms of reducing the number of BD occasions and of increasing QALYs among adolescents and especially for speci c subgroups of this population. However, in order to capture major changes both in reduction of number of BD occasions and in quality of life and savings in healthcare costs due to a behavioural change intervention, long-term follow-up of the intervention would probably be required. Countries should consider this type of programme to inform the design of public health policies targeting alcohol use among adolescents. The study received approval from the Bioethics Committee of Andalusia from the Ministry of Health and Families of Andalusia (Consejería de Salud y Familias de la Junta de Andalucía), contract PI-0031-2014.Written informed consent was obtained from parents and students prior to participation in the study. The questionnaires were self-completed by the adolescents and con dentiality was ensured.

Consent for publication
Not applicable.

Availability of data and materials
The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Competing interests
The authors declare that they have no competing interests.

Funding
This study was prepared within the framework of the research project "Alerta Alcohol: design, validation and evaluation of the programme of selective prevention of alcohol abuse in adolescents: web-based computer-tailored intervention" (PI-0031-2014), and nanced by the Andalusian Public Foundation