A systematic and comprehensive literature search was performed in collaboration with research librarians at the Karolinska Institute University Library. The search strategy covered inclusion and exclusion criteria were formulated according to both PICO (Population, Intervention, Comparison, Outcome) and PEO (Population, Exposure, Outcome) frameworks, see the Supplementary Material for details. Except for research focusing on humans, no restrictions regarding population or comparison were used. Intervention or Exposure was interventions aimed at preventing or reducing use, excessive use and addiction (abuse and/or dependence) to alcohol, narcotics, doping or tobacco, that are delivered digitally via computer, tablet, smart phone or equivalent. Outcomes included were outcomes that can be related to abstinence from use or reduced use of alcohol, narcotics, doping or tobacco. Outcomes related to experiences of or preferences regarding digital interventions were also included.

In the first stage, a test search was run to identify the relevant MeSH and free-text search terms, the results of which were compared to a set of known studies. Next, the preliminary search strategy was revised to increase sensitivity. The search was performed in three databases on 2021-11-19: Medline, Web of Science Core Collection, and PsychInfo. The database-specific search strategies were created by combining terms related to clinical presentation (e.g. “tobacco OR nicotine OR smoking”) with terms related to technology (e.g. “SMS OR social media OR web OR internet”) and study aim (e.g. “prevent OR treatment”). The initial search identified n=11,062 hits published before 2015 and n=17,053 hits published in 2015 or later. After using a validated method for deduplication (21), the search hits were reduced to n= 6,691 publications before 2015, and n=9,981 publications in/2015 or later. The search was re-run on 2023-01-23, at which time an additional= 3,936 publications were added. The final number of screened studies published 2015 or later, thus totaled n=13,917. See the Supplementary Material for exact search terms used with each database and number of resulting hits.

See Figure 1 for flow-chart. The final, deduplicated search results were imported into the Rayyan online software (www.rayyan.ai), which was used for subsequent processing. In the first stage, all n=13,917 records were assessed manually for initial topic eligibility by screening the title and (if needed) the abstract. A random sample of 600 titles were assessed by at least two of the authors who were blinded to each other. The assessment differed between authors regarding 50 titles and the differences were resolved in consensus discussions. This resulted in n=8,881 records being excluded for not being about interventions, the intervention not being digital or not targeting alcohol, narcotics, doping or tobacco. Examples of excluded publications covered research on electronic cigarettes (without any digital intervention), digital data collection (outside the context of a digital intervention), and on pharmaceutical systems. Inclusion and exclusion criteria used to assess eligibility can be found in the Supplementary Material .

In the second stage, the abstracts of the remaining n=4,690 records were screened in detail, resulting in a final sample subjected to coding. To ensure a directed analysis covering features of key clinical and public health importance, a deductive approach was used in which abstracts were coded with respect to five characteristics: type of study, type of technology, type of substance, theoretical grounding, and target group/arena. Within those pre-defined main categories, codes were generated inductively, to ensure that the codes represented the presumed diversity in the research field. If the information in the abstract was insufficient to code type of study, technology or substance, the full text was retrieved (n=169). All randomized trials were additionally coded for type of outcome measure and level of prevention. Level of prevention used pre-determined categories; 1. universal for interventions that prevent use or potentially harmful use of substances in populations, 2. selective for interventions that prevent harmful use in risk groups 3. indicated for interventions that prevent harm from use of substances among individuals at risk, 4. treatment for interventions that reduce harmful use or treat individuals who have developed substance use disorders (22). From identified meta-analyses and systematic reviews information about number of included studies and effects of digital interventions were extracted.

All authors participated in the coding. Codes were derived directly from the texts and every new code was communicated to the whole group via the Rayyan tool. Similar and conflicting codes were discussed in a dedicated chat and in regular meetings, thereby facilitating investigator triangulation (23). A random sample of 615 abstracts were coded by at least two of the authors separately and a preliminary coding scheme was created. Conflicts in the coding of 92 of these abstracts were discussed and resolved with all coders. In addition to the manual coding of each manuscript all identified codes, including variations, were additionally used as search terms in Rayyan to secure that all abstracts relevant for each code were identified. The most common codes in each category and some additional examples identified codes in each category are presented under results. A full list of codes and categories can be found in the Supplementary Material . The effects of digital interventions in each category as reported in included systematic reviews are summarized to highlight the results found in the most recent and most comprehensive reviews.

In addition to presenting descriptive statistics pertaining to the research questions, a bibliometric analysis was conducted, comparing the number of included publications included per year with the total number of publications per year in PubMed; these comparison numbers were retrieved using the easyPubMed R package (24).

The most common type of design was randomized studies (n=739). This category included randomized controlled trials, randomized clinical trials and randomized experiments. The randomized category also included pilot trials that were randomized (n=163), as well as a smaller number of factorial (n=33) and non-inferiority trials (n=6). The second most common category was development and testing of digital interventions (n=708), which include acceptability and/or feasibility (n=391), engagement (n=95) usability (n=44), and participatory-design (n=26) studies. The category quasi-experimental and observational studies (n= 344) included observational (n=94), quasi-experimental (n=74), pre-post (n=65), non-randomized pilot (n=146) and case-control (n=4) studies. Qualitative methodology was described in n=330 of the publications, including focus groups (n= 49), interviews (n=48), content analysis (n=45) and mixed methods (n=79). Secondary analyses of previously published trials was the topic of n=230 publications. Other types of designs included economical evaluations (n= 43), implementation studies (n=27) and case-studies (n=54). Of the included publications, n=374 were other reviews, of which n=172 were systematic reviews and n=68 meta-analyses. The search also identified several protocols for trials (n=305) and conference presentations, from which no findings had been published (n=140). See Table 3 for the major study types per substance category.

The terms used to describe the technology component of the digital interventions included “digital” (n=393), “electronic” (including “e-health”, n=378) or “technological” (including “tech”, n=62). The most common technology category was interventions delivered over the internet (n=1068), typically referred to via terms such as internet, online, or web (i.e., world wide web) interventions. Another common category was mobile interventions (n=795), which included smartphone apps typically native to iOS or Android. Other interventions were described as primarily text- based (n=423), usually delivered through SMS, email or other message services. Telehealth (n=370) includes concepts such as telemedicine and telepsychiatry, which usually involves healthcare contact via video or telephone, sometimes also combined with internet interventions, apps or text messages. In connection with the COVID-19 pandemic, telehealth and mobile interventions have become more common. Video conferencing was explicitly mentioned in n=65 of the studies. Computers or tablets (n=306) with installed applications or other offline-accessibility, were found to be have been used in healthcare or school settings.

Many studies combine several different technologies in the same intervention. Some included studies used so called blended interventions (n=181) that combine digital interventions with face-to-face contact. Others used digital interventions only as a smaller part of a larger intervention package (n=86). Comparisons between digital and face-to-face interventions were made in n=63 of the studies. Additional technologies covered included videos (n=64), virtual reality (VR, n=25) or gamification (the inclusion of game-like elements in non-game settings, n=57). A subset of studies featured wearable sensors (n=85). New ways of working via the internet also make it possible to be present where the intended target group already is; among the coded studies there were examples of using social media (n=180) and discussion forums (n=61). See Table 4 for information on the different main categories of digital technologies that were found to have been used. The number of studies using different technologies per year is shown in Figure 3 , while overlaps between different technologies is visualized in Figure 4 .

The main substance categories identified were tobacco (n=1234), alcohol (n= 1156), and other drugs (n= 678). The category tobacco covered primarily publications about smoking (n=1054) or unspecified tobacco use (n=138), but also included vaping (e.g., electronic cigarettes, n=16), waterpipe (n=3) and smokeless tobacco (n=8). The drug category included all addictive substances that were not nicotine nor alcohol. The most frequent substances in the drug category were cannabis (n=133) and opioids (n=268). Other drugs covered were doping in sports (n=6), prescription medication (n=27), and illicit drugs (n=82), including stimulants (n=39). The category health-behavior (n=269) was created to collect studies that were not solely about substance use, but in which substance use was one focus or outcome, including.e.g. prescription drug use among users of a Behavioral Skills-Based Virtual Reality Program for Chronic Low Back Pain (25) or a review of electronic mental health interventions for Indigenous youth (26). In recent years, an increasing number of publications regarding opioids was observed. See Figure 5 for the development of number of studies per substance each year and Figure 6 for overlaps between substances.

Theor ethical grounding of the studied intervention could be coded in n=2,016 (66%) of the included abstracts. The most common theoretical grounding was Cognitive Behavioral Therapy (CBT, n=271) which included e.g., community reinforcement approach (CRA, n=18), acceptance and commitment therapy (ACT, n=35). Another common category was brief interventions (n=279) which also included combinations with screening (SBI, n=99) and referral to treatment (SBIRT, n=43). The feedback category (n=177) included interventions using personal normative feedback (n=76) or feedback on blood alcohol concentration (n=41). Monitoring (n=109) included both self-monitoring (n=24) and ecological momentary assessment/interventions (EMA/EMI, n=38). The term Social interventions (n=160) was used to collect interventions with a prominent social component, like forums (n=61), social network (n=13) or peer support (n=70). Some interventions were based on different forms of cognitive training interventions (n=92), such as cognitive bias modification (n=56) or cue exposure (n=15). The Adherence to treatment category (n=208) included both referral (n=53) and adherence to different parts of treatment, e.g. medication (n=136). Contingency management which was the focus of n=45 studies, is used to both increase adherence to treatment and decrease substance use. Other mentioned theories used in the studies included the trans-theoretical model (also referred to as Stages of change, n=26), theory of planned behavior (n=23) and social cognitive theory (n=14). A few publications explicitly mentioned the use of so-called behavior change techniques, BCTs (n=28). Table 5 shows the main categories of methods used in digital interventions.

The included articles describe digital interventions that target a wide range of populations; from the entire population of a country or speakers of a certain language to special risk groups or specific settings. The biggest special population targeted was health care (n=471) including conditions like cancer (n=43), cardiovascular (n=48), diabetes (n=14) or HIV (n=100) and settings like primary care and emergency care. A related category was mental health (n=202) that included e.g., depression (n=42), trauma (n=30) and psychosis (n=10). Young adults were categorized together with students (n=420) e.g., at college (n=126) or university (n=40). The category family included many different sub-categories; adolescents (n=223), parents (n=50), pregnant or post-partum (n=87) and concerned significant others of people with SUDs (n=48). Some digital interventions have been adapted to minorities (n=174) including different ethnic/cultural minorities, sexual and gender minorities (n=53) as well as people with disabilities (n=10). Digital interventions have also been tested among disadvantaged people (n=142) that also included individuals that are homeless (n=18), living in rural areas (n=39) or low- and middle-income countries (LMIC, n=22). SUD treatment and recovery were the focus of n=271 publications. This category also included methadone/buprenorphine substitution (n=57) and overdose prevention (n=28). Most interventions were aimed directly at individuals with hazardous or harmful substance use, but a smaller number of studies describe interventions that target whole groups or special settings. These include interventions in schools (n=36) or policy work among restaurants and bars (n=13). See Table 6 for main categories of targeted populations.

Digital interventions for alcohol, tobacco and drugs have been studied in randomized trials at all different prevention levels: from primary, universal prevention (n=82) to tertiary interventions for people with substance use syndromes (including harmful use and dependence, n=104). The most common levels of prevention were indicated prevention for at-risk consumers (n=379) and selective prevention (n=178) for at-risk groups (e.g., students or people with various medical conditions that can be linked to the substances’ effects on physical and mental health).

Outcome measures in the randomized trials included in the overview were primarily various measures of substance use (n=367) or abstinence (n=153) that were measured by self-report or via biochemical measures such as exhalation or saliva, (n=48). Other measures used in studies where substance use was not relevant or possible to measure were e.g., intentions to use (n=44), attitudes (n=30) or knowledge (n=14) around using substances. In studies of the use of digital interventions, adherence were often used as outcome (n=62). In n=28 studies, substance use was only a secondary outcome measure. In some studies, the outcome measure was vaguely or unclearly described in the abstracts (n=37).

This scoping review has some obvious limitations. Congruent with the aim of a scoping review, the very broad scope and associated search strategy entailed that it was not feasible to code the entirety of the surveyed literature in detail. Many publications included only elementary information relevant to digital interventions in the abstract (e.g., specific technology used, theoretical grounding, outcome measures), lowering the level of detailed that could be compiled. For example, tobacco cessation interventions were mostly just described as smoking cessation without any additional information on the methodological or theoretical grounding for the intervention. A review also shows that only half of digital interventions for alcohol mention theory and even fewer used theory to select or develop the intervention (94). The broad scope also meant that quality assessment of the publications included was not possible. Even if many of the systematic reviews and meta-analysis included made their own quality assessments, the statements in the narrative overview of effects of digital interventions should be interpreted with some caution. Non-consistent use of terminology in most of the identified categories made detailed coding even more difficult. Some of the digital interventions included in the current review only used digital components for a small part of a larger intervention. This will likely become more common as digital interventions become increasingly integrated into regular health care settings. It is important to be able to distinguish these from interventions where the digital component is in focus or the main part of the intervention. One suggestion for a common glossary would be to start with a clearer threshold or definition on what constitutes a digital intervention.

This scoping review has several implications for future research. The continued growth of the field revealed by our analysis highlights the need to make reference to recent publications. There is a need for updated reviews and meta-analysis both in general, and in different areas of digital interventions targeting different kinds of substance use among different populations. Many of the randomized studies identified in this review have not yet been included in meta-analysis. Moreover, many of the most recent general meta-analysis have mostly included studies published before 2015 (27–29). Since most of the randomized studies had use and/or abstinence as outcomes, there are good conditions for carrying out future meta-analyses. An obvious implication of the non-consistent use of terminology is that meta-analyses and systematic reviews on more targeted research questions need to continue having broad search term strategies covering a range of terms, to ensure total coverage. For example, now that HTML5 web applications have the same functionality as many smartphone applications and also can in parallel be accessed through traditional web browsers (95) – it would also be valuable to derive definition frameworks that accommodate these technological developments.

The evidence on some common interventions is still inconclusive (30, 32, 38) or even lacking on many newer types of interventions (42, 44), intervention components (46, 48), as well as target groups (44, 60), revealing a need for more randomized controlled trials of digital interventions. Hopefully the broad scope of the current review also can assist researchers in identifying areas with little or no previous research. In addition to the included articles, our review also identified work in the digital and substance use field that was not about interventions. One common such area was monitoring of social media and other web content or analyzed automatically with digital tools to detect trends in the use of, and attitudes towards substance use or compliance with regulations, e.g. online advertising and sales (96). Monitoring of online content, use of Big Data and AI raises many new ethical questions regarding digital interventions targeting substance use that have not yet been studied much (97, 98).