The descriptive analysis of the variables was conducted using the central trend and spread measures, if they followed normal distribution and, for asymmetric distributions, medians and interquartile ranges (IQR) were used. In the bivariate analysis, the group averages were compared with the Student t-test or the Mann-Whitney U test, if the distributions were not normal. In the bivariate analysis of categorical variables, the chi-squared test was used. Uncompleted questionnaires (1.2%) were not analyzed, and atypical timestamp questionnaires (3.4%) were considered faked users and not taken into account for the analysis. Variables with a significant association (P<.05) were considered for entry into the logistic regression model. Multiple logistic regression models were performed to determine crude odds ratios, 95% CI, and their corresponding P values for each drug. Model fit was assessed using the Hosmer-Lemeshow goodness-of-fit statistic where the model was determined a good fit if the P value was not significant.

Stata v14.0 software was used for the log file statistical analysis with unique extended commands to identify unique cases (M Hills), geocode_ip (S Correia) and dc or Data Check (JM Domènech) to verify the integrity of the database with unique identifiers. IP, type of device, type of Web explorer, time spent, and demographic data referred were combined to detect unique users.

From February 24, 2015 to August 24, 2015, the website received a total of 15,867 visits. Figure 1 shows that total unique site visitor as determined by Analytics were 3885, and the test view rate was 2428 (62.50%). A total of 1657 (10.44%) were users who finally completed screening. Participation rate was 1675 of 2242 (74.71%) with a final completion rate of 68.24% (1657/2428).

A total of 76.9% of the access sessions during the period studied took place in the first month after the website was presented, with 3 activity peaks (February 25, launching day, March 4, and March 10, 2015) first one likely related to press release.

A total of 83.1% of the users were from Spain, 2.3% from Mexico, and 2.2% from Argentina. In Spain, the majority of 35.7% of users were from Madrid, followed by Barcelona (8.2%), Bilbao (3.7%), and Valencia (3.5%, see Figure 2).

Users were mainly from 3 internet sources: 40.2% direct traffic (manual access to the website), 33.2% were redirected from another website related to press release, and 20.3% came from a search engine. A total of 6.4% came from social networks. A multilevel regression model was carried out to analyze the possible influence of location clustering in developing any drug risk showing no significant differences (intraclass correlation coefficient, ICC=.01, 95% CI 0.001-0.11).

The average age of the subjects was 37.4 years (standard deviation, SD 12.9) and 59.9% of cases stated to be male. Users could test themselves for drug-related risks and no risks could be showed. Of all the users screened, 21.1% did not have any moderate to high substance-related risks (as opposed to 78.9% with a moderate or high risk in relation to at least 1 substance). The average time spent taking the test was 3 min (IQR: 2-4), with a statistically significant higher time spent for those who screened positive on the test for at least 1 drug. The number of drugs reported was higher for the group who screened positive compared with the group who showed no moderate to high drug-related risks (P<.001); see Table 1.

The most highly consumed drug by men as well as women was tobacco, which all the self-surveyed subjects stated to smoke at some point, followed by alcohol (87.9%) and cannabis (54.1%). We found moderate or serious substance-related risk for tobacco (50.7%), alcohol (26.4%), cannabis (21.8%), and sedatives or hypnotics (11.6%) as the 4 most frequent.

In the subgroup analysis, significant differences were observed based on gender: men had moderate to high risks for alcohol disorders in 17.6% of cases compared with women in 8.8% (P=.001) of cases; in the consumption of cannabis, men had 15.9% cases compared with women in 5.9% (P<.001) cases, and in the consumption of cocaine, men had 3.5% of cases compared with women in 1.5% (P<.001) cases (see Table 2). Male respondents had 1.55 times risk of having alcohol problems (odds ratio, OR 1.55, 95% CI 1.18-2.04; P=.002) and had double the risk for cannabis problems (OR 2.07, 95% CI 1.46-2.92; P<.001); see Table 3.

In the subgroup analysis of age categories, users who screened positive for moderate to high substance-related risks were distributed into groups up to 44 years of age. The highest number of users with drug-related risks was in the 35- to 44-year-old category: tobacco (12.3% moderate, 1.6% serious), alcohol (5.3% moderate; 1.9% serious), cocaine (1.6% moderate), and sedatives/hypnotics (2.6% moderate). The cannabis-related risks were concentrated in the 2 youngest age groups (18-24 and 25-34 years) with 5.7% of all subjects consuming the drug. Growing age increased 3 times risk of developing alcohol problems for people between 45 and 65 years (OR=3.01, 95% CI 1.89-4.79; P<.001). In contrast, increasing age protects against developing cannabis-related problems (OR=0.30, 95% CI 0.17-0.56; P<.001); see Table 3.

With regard to polyconsumption or the consumption of different drugs throughout the same period of time, a moderate or serious risk was observed in 31.6% of the subjects. Number of substances and sex distribution are shown in Table 4. Male respondents, compared with female respondents, were more likely to develop polyconsumption (OR=1.92, 95% CI 1.5-2.47; P<.001) . Those user screening by mobile devices had higher risk of developing problems even when adjusting for potential confounder such as age (OR=1.41, 95% CI 1.1-1.8, P=.006) . Age showed no significant differences regarding poly-consumption.

Table 5 shows the combinations of the most common moderate or serious drugs-related risks for 2 drugs and the percentage of polyconsumers of each pattern. The most frequent combinations found were tobacco and cannabis (31.3%), tobacco and alcohol (16.2%), and tobacco and sedatives (8.4%). There was also a group of subjects with a serious risk for each such as tobacco and cannabis serious (10.4%) and tobacco and alcohol serious (7.2%).

In the first 6 months after the Web-based self-screening for drug consumption website was launched, a total of 15,867 users were recorded, of which 1657 (10.4%) completed the screening. The average age obtained was 37.4 years and 78.9% showed moderate or serious drug-related risks. The most highly consumed drugs stated were tobacco (50.7%), alcohol (26.4%), cannabis (21.8%), and sedatives (11.6%) with men taking more alcohol, cannabis, or cocaine; and young people (aged 18-35 years) using more cannabis. Polyconsumption was observed in 31.6% of the cases.

In terms of the number of users, our website showed data similar to others in the existing literature [17]. Nevertheless, it is true that other single-drug platforms, such as those for only alcohol screening, have achieved a greater scope with larger samples after targeted marketing campaigns also on the internet, as well as by mailings and massive letter box drops of pamphlets [23,24].

Our website user map (see Figure 2) shows that the geographic distribution of users with drug problems are similar to that found in the literature [3], the most common cities still being Madrid and coastal cities (Barcelona, Valencia, and Bilbao). This could be explained either because big cities have higher drug use rates [3] or just because they are major cities and we are getting more respondents from these cities. However, after adjusting in a multilevel regression model, cluster had no influence on developing drug problems.

Users surfing the Web using Google services are giving their demographic characteristics and interest information for analysis according to the Google privacy terms. A total of 52.1% of users could be analyzed, and differences were observed in relation to age and gender. In the data provided by the users, there were 59.9% men, as opposed to 41.2% in the Google sample, and 7.1% of users stated in the test that they were younger (18-24) than when compared with the data obtained with Google. This could be due to the 50% of incomplete information from the Google remaining sample, or if representation is assumed, due to the fact that some people did the test for others.

Drug-use patterns partially differ with respect to consumption prevalence in Spain according to the EDADES [3], in which hypnotics-sedatives were higher than cannabis in the 15- to 64-year-old population and, in our sample, cannabis was still the third highest consumed drug. The characteristics of both samples could explain this difference, given that the upper age ranges are more highly represented in the EDADES study, as opposed to our sample and, in addition, because these are drugs that are initially consumed at middle age. Furthermore, both tests differ since the EDADES questionnaire addresses mostly consumption rates but does not address other type of drug-related risks for nondependent people.

In terms of polyconsumption, there are also differences: the definition used in the EDADES survey refers to the combined consumption of substances in the same period [3]. The ASSIST test questions say, “at some time” and “in the last three months,” therefore, to a certain extent, it already partly includes that information. Moreover, the ASSIST offers an estimation of the clinical and social risks related to substance use. Therefore, as far as terminology is concerned, our polyconsumption is the combination of moderate and serious risk estimates relating to any of 2 or more substances detected. This is clinically relevant as the users are at a greater risk and may need more intense treatment.

This is the first drug-related risks screening website related to all common drugs, using a scale that has been internationally validated by the WHO (ASSIST), adapted to Spanish for Web-based screening.

The results also show a target group for potential intervention relating to the tobacco-cannabis polyconsumption pattern, due to the high percentage detected in certain age groups (18-34 years).

One of the major advantages of this type of tool would be the possibility of reaching populations that would otherwise delay resorting to the health system and also offering primary care doctors and other health care professionals a valid tool for handling the risks related to the consumption of substances before it becomes an established medical condition.

The potential impact of applying overall prevention measurements (screening as secondary prevention), instead of those addressed to specific populations is a public health issue that is currently being debated. The screening in itself would be pointless if counteraction was not offered. There is already evidence of decreasing mortality rates in favor of global intervention after screening general population samples by models designed for cardiovascular disease, as opposed to screening specific population alone [39]. Drug-related interventions such as brief interventions using ASSIST Web-based tools have already proven effectiveness in the English samples compared with in-person brief interventions [15].

One of the main limitations would be the ability to identify sole users and therefore to eliminate duplicating or overestimating the results. Several users could have access from the same device and even from the same network to complete the test, without being able to make a distinction between them. Therefore, the server data were exploited, limited by the anonymized IP address, compared with the data from the device used for access and the time (day and time) that the test was taken and, together with the aforementioned demographic data, an algorithm was established to identify single and valid users. When the users could not be differentiated, an attempt was made to eliminate “testing users” from the analysis to identify erroneous situations such as the screening of many drugs in a short period of time, several attempts by the same user, etc. For users making several attempts, the last answer was accepted as valid.

Another limitation that can be assumed is related to the scope of publication. Users accessed the website mainly because they had heard about it or were redirected by a different website; 1 of every 5 did so after browsing with Google. Given that the environment was created under a research framework publicized by a public body (National Anti-Drug Abuse Plan), access and users came from locations where there was publicity resulting from the presentation. Future studies would benefit by including a broader population to improve external validity.

Another difficulty encountered was the restriction of the screening age to 18 years and older. At the time we developed the website, validation for the adolescent population had not been published. In 2015, Gryczynksi et al [40] validated this tool for adolescents in primary health care, meaning that it could therefore be extrapolated to our general population. Its adaptation to our environment would be simple, as it requires minimal programming changes to allow access to younger users (12-17 years) according to the aforementioned validation.