Technology-mediated just-in-time adaptive interventions (JITAIs) to reduce harmful substance use: a systematic review

Background and Aims: Lapse risk when trying to stop or reduce harmful substance use is idiosyncratic, dynamic and multi-factorial. Just-in-time adaptive interventions (JITAIs) aim to deliver tailored support at moments of need or opportunity. We aimed to synthe-size evidence on decision points, tailoring variables, intervention options, decision rules, study designs, user engagement and effectiveness of technology-mediated JITAIs for reducing harmful substance use. Methods: Systematic review of empirical studies of any design with a narrative synthesis. We searched Ovid MEDLINE, Embase, PsycINFO, Web of Science, the ACM Digital Library, the IEEE Digital Library, ClinicalTrials.gov, the ISRCTN register and dblp using terms related to substance use/mHealth/JITAIs. Outcomes were user engagement and intervention effectiveness. Study quality was assessed with the mHealth Evidence Reporting and Assessment checklist. Findings: We included 17 reports of 14 unique studies, including two randomized controlled trials. JITAIs targeted alcohol (S = 7, n = 120 520), tobacco (S = 4, n = 187), cannabis (S = 2, n = 97) and a combination of alcohol and illicit substance use (S = 1, n = 63), and primarily relied on active measurement and static (i.e. time-invariant) decision rules to deliver support tailored to micro-scale changes in mood or urges. Two studies used data from prior participants and four drew upon theory to devise decision rules. Engagement with available JITAIs was moderate-to-high and evidence of effectiveness was mixed. Due to substantial heterogeneity in study designs and outcome variables assessed, no meta-analysis was performed. Many studies reported insufficient detail on JITAI infrastructure, content, development costs and data security. Conclusions: Current implementations of just-in-time adaptive interventions (JITAIs) for reducing harmful substance use rely on active measurement and static decision rules to deliver support tailored to micro-scale changes in mood or urges. Studies on JITAI effectiveness are lacking.


INTRODUCTION
With improved mobile hardware, software and computational power, individual-level data on substance use triggers can be collected, processed and actioned in or near real-time.A large body of research using technology-mediated ecological momentary assessments (EMAs) in people's daily lives indicates that lapse risk in people attempting to quit or reduce harmful substance use is idiosyncratic (i.e. it differs between individuals), dynamic (i.e. it fluctuates over time) and multi-factorial (i.e. it is driven by multiple variables, such as urge to smoke, negative affect and contextual cues) [1][2][3][4][5][6][7].For examplehighlighting the dynamic and multi-factorial nature of lapse risk-in smokers attempting to quit, experiencing a strong urge to smoke was, on average, associated with 20% greater odds of lapsing near the quit date, with odds increasing by a further 30% one week after the scheduled quit attempt.Negative affect, however, was significantly associated with the odds of lapsing near the quit date, but this association levelled off shortly thereafter [4].To highlight the idiosyncratic nature of lapse risk, a series of N-of-1 observational studies to examine factors associated with day-to-day alcohol consumption in individuals with a history of alcohol dependence found that different psychological and social factors were important for different individuals [7].Justin-time adaptive interventions (JITAIs) aim to provide tailored support to users at moments of 'need' (e.g.there is a need for support due to low self-regulatory capacity) or 'opportunity' (e.g.there is an opportunity to act positively in line with one's goals) [8,9].Due to the idiosyncratic, dynamic and multi-factorial nature of lapse risk in individuals attempting to quit or reduce harmful substance use, JITAIs are poised as particularly suited to the delivery of lapse prevention support.
There is no consensus definition of what a JITAI is; although they typically harness mobile technology to deliver support, the mode of delivery is not necessarily a defining feature.Hardeman and colleagues propose that JITAIs can be defined in terms of three characterizing features: (i) the intervention corresponds directly to a need for support in real-time (e.g. the user is at risk of smoking lapse due to experiencing high levels of stress) or an opportunity to act positively in line with one's goals, (ii) the content or timing of the support is tailored to that real-time need or opportunity (e.g. the intervention is tailored to the most prominent lapse risk trigger, such as stress) and (iii) the support is automatically triggered by the system (e.g.app, website, health-care professional, peer) and not directly by the users themselves [10].Others have argued that JITAIs can also be usertriggered (e.g.pushing a button within an app or requesting a 'CRAVE' or 'LAPSE' message from an automated text message system) [11].
Nahum-Shani and colleagues propose that JITAIs are defined by their constituent parts, which include (i) decision points (i.e. points in time at which an intervention may be delivered), (ii) tailoring variables (i.e.input used to inform decisions as to when or how to intervene for each individual), (iii) intervention options (i.e. the available change strategies or delivery modes) and (iv) decision rules (i.e.rules that systematically link decision points, tailoring variables and intervention options) [8].Furthermore, some have highlighted that JITAIs are interventions which consider individual change trajectories over time (e.g. from undesired to desired states), taking into account micro-(e.g.weather, stress), meso-(e.g.seasonality, motivational cycles) and/or macro-scale changes (e.g.life transitions such as becoming a parent, retirement) [8,12] (see also http://osf.io/n3scx).
A scoping review of JITAIs within addiction science and related study designs (e.g. the micro-randomized trial) has recently been conducted [13]; however, to date, there has been no systematic and comprehensive review of decision points, tailoring variables, intervention options, decision rules, user engagement and intervention effectiveness of current implementations of technology-mediated JITAIs for reducing harmful substance use.Such a review would be useful for informing the development of new JITAIs and the optimization of existing ones.We therefore aimed to address the following research questions, taking an inclusive approach to the definition of JITAIs: 1. What decision points, tailoring variables, intervention options and decision rules are used in current implementations of technologymediated JITAIs for reducing harmful substance use? 2. Which study designs have been used in the development, optimization and evaluation of JITAIs that aim to reduce harmful substance use?
3. What is the uptake of, engagement with and effectiveness of JITAIs for reducing harmful substance use?

Study design
This review was informed by the Cochrane Handbook of Systematic Reviews of Interventions [14] and adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) checklist [15].A protocol was pre-registered on the Open Science Framework (https://osf.io/e9hcj)and on the international Prospective Register of Systematic Reviews (PROSPERO) (https://www.crd.york.ac.uk/ prospero/display_record.php?ID=CRD42019142019).

Population
We included studies with participants with harmful substance use, including (but not limited to) tobacco, cannabis, alcohol, cocaine or heroin use.As we aimed to provide an overview of the characteristics of JITAIs, interventions targeting participants of any age, in any setting (e.g.primary care, schools) were included.

Intervention
We included JITAIs designed to reduce harmful substance use (e.g.tobacco, cannabis, alcohol, cocaine, heroin), delivered by any type of technological system (e.g.websites, text messages, apps, wearables).Although we were also interested in capturing JITAIs delivered by non-technological systems (e.g. a family member, peer, health-care professional), our search strategy was not specifically designed for this purpose.We took an inclusive approach and considered an intervention to be a JITAI if the primary sensing and delivery mechanism satisfied the following conditions: (i) the intervention corresponds directly to a need for real-time support or an opportunity to act positively in line with one's goals and (ii) the content or timing of the support is tailored to that real-time need or opportunity [10].We included systems where the delivery of support was automated and in response to either EMAs delivered at decision points ('active measurement') or location/sensor data ('passive measurement') [8].We considered an intervention to be a JITAI if the majority of the support it was designed to deliver met our definition.However, a text message intervention in which users could trigger support directly by requesting a 'CRAVE' message but where the majority of the support was not adapted to a real-time need or opportunity did not meet the inclusion criteria for this review.In addition, to distinguish JITAIs from one-off substance use screening and brief advice (sometimes referred to as 'just-in-time interventions', or JITs), we considered an intervention to be a JITAI only if the support was delivered repeatedly over a period of time (i.e. more than once per month).Interventions targeting multiple substances/behaviours were included providing that data could be extracted on the substance use component.

Comparison
Due to the descriptive focus of the review, interventions with any type of (or no) comparator were included.

Outcomes
Included studies had to report at least one of the following empirical qualitative or quantitative outcomes: user engagement (e.g.uptake, use, acceptability, liking) or intervention effectiveness (e.g.reduction in urges, smoking cessation, alcohol reduction).

Study designs
Studies of any design (e.g.qualitative and quantitative studies with data from development work, pilot and feasibility studies, evaluation studies) were included provided that a prototype intervention had been developed.Conceptual or methodological papers with no empirical data (including early user studies without a prototype intervention) were not included.Although we recognize that conceptual papers are helpful for addressing the questions of decision points, tailoring variables, intervention options and decision rules, we were interested in summarizing current implementations of JITAIs in this review.

Electronic searches
Electronic and hand-searches were conducted in January 2020.As technology-mediated JITAIs first started to appear in the literature during the second half of the 2000s [16], articles published in or after 2000 were included.Where possible, the language index was set to restrict the search to articles available in English.We searched Ovid MEDLINE, Embase, PsycINFO, Web of Science, the ACM Digital Library, the IEEE Digital Library, ClinicalTrials.gov, the ISRCTN register and the dblp computer science library (https://dblp.uni-trier.de/).We Terms were searched for in titles and abstracts as free text terms, word stems (e.g.smok$) or as index terms (e.g.Medical Subject Headings, Subject Heading Words, Keyword Heading Words), as appropriate.See Supporting information, File S1 for the full electronic search strategy.

Searching for other sources
We used first-order reference chaining and drew on expertise within the review team to identify additional articles of interest.

Selection of studies
Articles identified via the electronic and hand searches were merged with EndNote and duplicate records were removed.The first and second author independently screened (i) titles, (ii) abstracts and (iii) full texts against the pre-specified inclusion criteria.In line with the PRI-SMA checklist, reasons for exclusion were recorded at the full text stage [15].Discrepancies were resolved through discussion and by consulting the last author if required.

Data extraction and management
A data extraction form was developed by the first and second authors to extract information on (i) study design (e.g.qualitative study, microrandomized trial); (ii) delivery setting (i.e.country, immediate delivery context); (iii) participant characteristics (e.g.age, gender, educational attainment, type of substance, level of dependence, mental and physical health comorbidities); (iv) delivery platform (e.g.smartphone app, health-care professional) and, where appropriate, operating system; (v) whether an existing platform was deployed for intervention delivery; (vi) whether in-house or external developers were used to build the platform; (vii) whether treatment was stand-alone or delivered in adjunct to other support; (viii) payment schedule for participation, if payment was provided (e.g.flat payment, payment per EMA); (ix) intervention options, coded against the Behaviour Change Technique (BCT) taxonomy version 1 [17]; (x) presence of engagement features, as specified in [18]

Quality appraisal
Given the anticipated diversity of study designs, the mHealth Evidence Reporting and Assessment (mERA) checklist [19] was judged as relevant for assessing the quality of studies, including whether or not formative research and user testing has been carried out and reported, and whether barriers to intervention uptake had been considered.
Each checklist item was scored as 'fully reported', 'partially reported' (only some evidence reported) or 'not reported' [19].The quality appraisal was conducted by the first author, with the second author independently rating a random subset (i.e.10%) of included studies.
Discrepancies were resolved through discussion and by consulting with the last author if required.

Data synthesis
Given the diversity of study designs in the included studies, a narrative synthesis was conducted.Results are presented separately for studies with similar study designs that targeted similar behaviours.
In an unplanned analysis, the first and second authors coded factors that hindered or negatively influenced engagement (i.e.'barriers') and factors that promoted or positively influenced engagement (i.e.'facilitators') with JITAIs from qualitative or quantitative data presented in the included papers.We used a combination of deductive and inductive coding, with data coded against Perski and colleagues' conceptual framework of engagement [20], where possible.

Study selection
After removing duplicates, a total of 1047 records were identified through the electronic search.After full text screening, 14 studies (presented across 17 papers) were included in the evidence synthesis (see Fig. 1).

Study and participant characteristics
The majority of studies (10 of 14; 71%) were conducted in the United States, with the remaining studies conducted in the United Kingdom (two of 14; 14%) and Australia (two of 14; 14%) (see Table 1).The identified JITAIs targeted alcohol consumption (seven of 14; 50%), tobacco smoking (four of 14; 29%), cannabis smoking (two of 14; 14%) or a combination of alcohol and illicit drug use (one of 14; 7%).Study designs deployed were single-or two-arm, non-randomized pilot studies (five of 14; 36%), two-or three-arm pilot/feasibility RCTs (five of 14; 36%), two-or three-arm RCTs (two of 14; 14%) and mixed-methods designs in which the analysis of app usage data was combined with qualitative interviews (two of 14; 14%).None of the included studies reported the use of N-of-1 observational or experimental designs, or micro-randomized trial designs.
Studies included a median of 57 participants (range = 15-119 713) who were aged between 15 and 75+ years with a balanced gender distribution (median percentage of female participants = 53%).
JITAIs deployed static (i.e.time-invariant) if-then rules, which were typically based on participants entering a given geographical location (sometimes labelled 'weak spots' by the study authors) or whether a particular psychological or contextual variable, such as negative mood, stress, urges or the presence of others who use drugs/smoke, was reported as present (versus absent) or above a pre-specified threshold.A minority of studies from the same research team (two of 14; 14%) [29][30][31] reported that the if-then rule for triggering support was developed on the basis of lapse risk data from participants in a previous study (i.e. a 'warm start') [2]; the remaining studies did not report deploying data-driven algorithms.None of the included studies considered participant availability or receptivity to just-in-time support in their decision rules.In addition, none of the identified JITAIs adapted the type, frequency or intensity of support according to meso-(e.g.seasonality, motivational waves) or macro-scale changes (e.g.becoming a parent, retirement).
Ease of use, message frequency and perceived usefulness were reported as common facilitators to engagement.Where reported, many participants felt that the real-time prompts were quick and easy to complete [26,28,34], that the prompt/message frequency was just about right [25,29] and that the real-time messages were informative and useful for keeping track of and/or reducing substance use [22,25,27,28,31,33,35].
Reported barriers to engagement included technical issues, message frequency and lack of novelty.For example, participants reported location-triggered alerts going off in the wrong place, not always being triggered when supposed to or taking a long time to recognize that the participant had entered a pre-specified location [22,25].Participants in one study reported issues with an inconsistent EMA prompting schedule [26].Some participants felt that they received too many prompts/messages [25,27,35] or that the message content was too repetitive [26,27].Less frequently-reported barriers to engagement included low perceived personal relevance and Not reported unintended consequences.For example, participants in one study felt that it would have been more useful to receive alerts linked to particular events or when experiencing negative emotions (rather than when dwelling in specific locations) [21].Unintended consequences of JITAIs included reminding participants of smoking (when they had not been thinking about it) [25] and messages being perceived as guiltinducing or condescending [27].

JITAI effectiveness
Twelve studies reported on the JITAI's effectiveness (see Table 3), with the majority using linear and/or generalized mixed-effects models that accounted for the nested data structure.Outcome variables assessed were heterogeneous and the majority of studies did not report being sufficiently powered to detect differences in substance use or abstinence rates between groups.

Quality of included studies
None of the 14 studies reported full details for all 16 quality criteria.

DISCUSSION
This systematic review provides an overview of decision points, tailoring  Second, although important methodological and statistical advances to support JITAI development, testing and optimization have been made -including the multi-phase optimization strategy, microrandomized trials, supervised and unsupervised machine learning [38,39]-few studies identified in our review made use of such innovative approaches.Therefore, researchers, developers and practitioners interested in JITAIs should be supported to adopt relevant new methodological and statistical skills and/or ensure that such expertise is available within multi-disciplinary JITAI project teams.
Third, our review identified two primary ways in which JITAIs determine whether the user is in need of support: active measurement via EMAs or passive measurement, such as via location sensors.
Although engagement with EMAs and intervention messages was moderate-to-high across the included studies (indicating that user engagement is not itself a key barrier), payment was typically provided for completing EMAs or follow-up assessments.We therefore need evidence as to whether participants will also engage with EMAs outside controlled study settings where no payment is provided.A move from active to passive sensing of physiological or ecological indicators of lapse risk (e.g.heart rate variability [40,41], step count, weather) is also an important avenue for future research, with potential for reducing user burden and costs associated with financial incentives for completing EMAs.On the other hand, based on available data, the process of completing active measurements, such as EMAs, can help people reflect on their cravings, mood, etc., which may contribute to an enduring learning experience beyond the use of the JITAI itself [42].There are also important ethical considerations that need to be accounted for when deciding between active versus passive T A B L E 3 Competing demands (−); ease of use (+); perceived usefulness (+) 63% of participants signed up for 6 or more events and the majority completed surveys for all 6 events Random effects mixed models The JITAI group showed a small but non-significant increase between baseline and follow-up in the mean number of standard drinks consumed at the most recent heavy drinking occasion (mean = 12.5 versus mean = 12.7) JITAI = just-in-time adaptive intervention; EMA ecological momentary assessment; IQR = interquartile range.standards (referred to by some scholars as the 'Wild West' of digital health) [43].A related area is the use of sensors or digital devices for passive detection of key outcomes of interest, including smartphoneenabled carbon monoxide monitors to verify tobacco smoking abstinence [44], gesture recognition software on smartwatches to identify cigarette (or cannabis) smoking behaviour [45], transdermal alcohol sensors [46] or alcohol and cannabis sensors in the form of 'tattoos' and rings [47,48].This may simultaneously help reduce user burden and improve confidence in results, yet requires the same careful considerations as discussed above in relation to the passive sensing of physiological or ecological indicators of lapse risk.

Strengths and limitations
The strengths of this systematic review include a comprehensive search of nine medical, psychology, engineering and human-computer interaction databases, substantial team expertise (as indicated by several team members having contributed to papers included in the review), having two reviewers independently screen studies for inclusion, the coding of JITAI content against available taxonomies and conceptual frameworks and a quality appraisal of included studies against the mERA checklist.
However, our review also had several limitations.First, the electronic search and paper screening process was challenging due to the lack of a consensus definition of JITAIs and may mean we did not capture all relevant studies.For example, our reliance upon a specific three-part definition to help determine which studies to include meant that we excluded studies with a 'just-in-time' (JIT) reminder at time-points pre-specified by the user (as opposed to tailored support in response to EMAs delivered at decision points or location/sensor data) [49,50].Second, we decided against including conceptual or methodological papers without any empirical data due to our focus on current implementations of JITAIs.However, such papers may have provided additional insight into JITAI decision points, tailoring variables and decision rules, and the types of study designs that are useful for devising these.Third, although we consider the use of a quality appraisal tool a strength of the review, it was challenging to judge the quality of included studies due to insufficient reporting, particularly with regard to the infrastructure required to run JITAIs (e.g.specific hardware, software, size of message banks) and intervention options.

CONCLUSIONS
JITAIs for reducing harmful substance use tend to rely upon active combined search terms related to substance use (e.g.alcohol, tobacco, cocaine), mHealth (e.g.digital interventions, apps) and JITAI features (e.g.just-in-time interventions, adaptive interventions, personalization, tailoring).Search terms were piloted and refined to achieve balance between sensitivity and specificity.An academic librarian was consulted for the validation of the databases and the final search terms.

(( 1 )( 2 )( 3 )( 5 ) 32 ( 7 ) 27 ( 11 ) 50 ( 13 )
and if so, what type) of support to deliver: active measurement (i.e.ecological momentary assessments; EMAs) or passive measurement (i.e. the phone's location sensors, including the global positioning F I G U R E 1 Preferred Reporting Items for Systematic Reviews And Meta-Analyses (PRISMA) flow-chart of included studies T A B L E 1 Characteristics of included studies Authors (year) Attwood et al.Businelle et al. (2016); Hébert et al. (2018) United States Tobacco Adults recruited from a safety-net hospitalbased smoking cessation clinic Single-arm, non-randomized feasibility study 59 Hébert et al. (2020) United States Tobacco Adults recruited from a publicly available smoking cessation clinic Three-arm, pilot randomized trial 81 (27 in JITAI arm) (4) Dulin et al. (2014); Gonzalez & Dulin (Gustafson et al. (2014) United States Alcohol Adults taking part in a residential alcohol use disorder treatment programme Two-arm RCT 349 (170 in JITAI arm) (6) Hoeppner et al.Ingersoll et al. (2014); Ingersoll et al. (2015) United States Alcohol and illicit drugs Adult patients recruited from two non-urban HIV clinics with histories of substance use Two-arm, pilot RCT 63 (33 in JITAI arm) (8) Naughton et al.O'Donnell et al. (2019) Australia Alcohol Community dwelling young adults Two-arm, pilot RCT with post-intervention telephone interviews 45 (25 in JITAI arm) (10) Shrier et al. (2014) United States Cannabis Adolescent patients from two clinics affiliated with a paediatric hospital Single-arm, non-randomized feasibility study Shrier et al. (2018) United States Cannabis Adolescents and young adult patients from an urban children's hospital Three-arm, pilot RCT 70 (27 in JITAI arm) (12) Suffoletto et al. (2018) United States Alcohol Young adults presenting to an urban emergency department Single-arm, non-randomized feasibility study Weitzel et al. (2007) United States Alcohol College students Two-arm, pilot RCT 40 (20 in JITAI arm) (14) Wright et al. (2018) Australia Alcohol Young adults recruited from the young adults alcohol study, an observational cohort study Three-arm RCT 269 (90 in JITAI arm) JITAI = just-in-time adaptive intervention; RCT = randomized controlled trial.

( 13 ) 2 ) 7 ) 2 )( 3 )( 4 )( 5 )( 7 ) 10 )
Weitzel et al. risky drinking (≥ 5 drinks in a single session in the past 3 months) Not reported JITAI = just-in-time adaptive intervention; RCT = randomized controlled trial.T A B L E 2 Characteristics of JITAIs Authors (year) JITAI hardware delivery platform Name of JITAI Whether JITAI was delivered with existing software Whether software was developed in-house Whether treatment was stand-alone (1) Attwood et al.Businelle et al. (2016); Hébert et al. had the option to access at least four face-to-face counselling sessions and were offered medication (e.g.nicotine patch, varenicline) (3) Hébert et al. were offered nicotine replacement therapy (i.e.nicotine patch, nicotine gum) (4) Dulin et al. (2014); Gonzalez & Dulin (app was offered alongside treatment as usual, but this varied across residential treatment programmes and none offered patients coordinated continuing care following discharge (6) Hoeppner et al.Ingersoll et al. (2014); Ingersoll et al. (2015) Mobile phone (provided to participants) O'Donnell et al. received two 1-hour motivational enhancement therapy sessions with a trained counsellor, delivered face-to-face (11) Shrier et al. (2018) Mobile phone compatible with study software (phones loaned to 55 participants) MOMENT No Not reported No; participants received two 1-hour motivational enhancement therapy sessions with a trained counsellor, delivered face-to-face (12) Suffoletto et al. -in-time adaptive intervention; EMA ecological momentary assessment.T A B L E 2 (Continued) Authors (year) JITAI intervention duration Incentive structure Theory used to inform development of JITAI (1) Attwood et al.Businelle et al. (2016); Hébert et al. cards for completing each in-person assessment visit and based on the percentage of random and daily assessments completed over the study period, with a minimum requirement of 50% assessments completed to receive any payment ($40-120) Not reported Hébert et al. (2020) 5 weeks $30 gift card for attending and completing the pre-quit, quit date, and 4-week post-quit follow-up visit and $50 for completing the 12-week post-quit visit.At the 4-week post-quit visit, participants received additional compensation based on the percentage of random and daily diary EMAs that they completed.Those who completed 50% to 74% of all prompted EMAs over the 5-week EMA period received $50, those who completed 75-89% received $100, and those who completed 90% or more received $150 Not reported Dulin et al. (2014); Gonzalez & Dulin (2015) 6 weeks $60 at each of the baseline and 6-week follow-up assessments.In addition, participants received $5 for each day they completed a daily interview of alcohol consumption and cravings Not reported Gustafson et al. was grounded in self-determination theory, which posits that three types of needs (autonomy, competence, relatedness) contributes to an individual's adaptive functioning.The JITAI feature was designed specifically to promote competence (6) Hoeppner et al. (2019) 6 weeks $35 per online survey (2, 6 and 12 weeks after the quit date), and up to $36 per EMA week (1 week prequit and 1 week upon completion of the 6-week survey), with a maximum total of $287 per participant Not reported Ingersoll et al. (2014); Ingersoll et al. and behaviour skills (IMB) model of adherence and social action theory (8) Naughton et al. (2016) 4 weeks £10 shopping voucher for taking part in the qualitative interview Learning theory and a taxonomy of smoking-related behaviour change techniques (9) O'Donnell et al.Shrier et al. (2014) 4 weeks Compensation for travel and remuneration of up to $280 in gift cards, depending on proportion of study activities completed

14 )( 2 ) 6 ) 14 )( 2 ) 8 ( 6 ) 3 (
weeks (users select time in programme) $40 upon completion of a 3-month follow-up survey Not reported (13) Weitzel et al.Wright et al. (2018) 12 weeks $10 per completed event.If all 6 events were completed, a bonus of $20 was given.Participants received $20 for completing the follow-up survey.Participants who completed all 6 events and the follow-up interview received $100 in cash or voucher Motivational interviewing and brief intervention theory JITAI = just-in-time adaptive intervention; EMA ecological momentary assessment.T A B L E 3 Decision points, decision rules, tailoring variables, user engagement and effectiveness of JITAIs Authors (year) Type(s) of data used to trigger real-time support Decision rule(s) for triggering real-time support Whether decision rules were static or adaptive Tailoring variables used to personalize real-time support (1) Attwood et al. (2017) GPS If entering a pre-specified geographic location ('weak spot'), then trigger a support message Static Not reported Businelle et al. (2016); Hébert et al. (2018) EMA If lapse risk score ≥ 1.0, then trigger a support message.The lapse risk score was calculated as follows: (urge -stress; smoking urge; cigarette availability; motivation to quit.Messages were tailored to the highest rated trigger.Where multiple triggers were equally highly rated, one message was delivered with preference given to negative affect/ stress, smoking urge, cigarette availability and motivation to quit (in the given order) (3) Hébert et al. (2020) EMA Same as Businelle et al. (2016) Static Same as Businelle et al. (2016) (4) Dulin et al. (2014); Gonzalez & Dulin (2015) GPS If crossing a pre-specified geographic boundary, then trigger alert (Hoeppner et al. (2019) EMA If mood = BAD or craving = HIGH, then trigger support message Static Mood; craving (7) Ingersoll et al. (2014); Ingersoll et al. (2015) EMA If mood = 0-2 (bad mood) or skies = rainy/cloudy/snowy, then trigger support message Static Substance use; mood (8) Naughton et al. (2016) Android Location Services, which uses multiple location sensors including the GPS If entering or dwelling (defined by Q Sense as 3 hours or more) in a pre-specified geofence, then trigger support message Static Geofence-triggered support messages were tailored based on average values of features specific to each geofence, collected during smoking reports (location type, strength of urge, mood, perceived stress, presence of other smokers).Other support messages drew from a pre-populated database that matched the user's 11-item demographics (9) O'Donnell et al. (2019) EMA If the user indicates that they are drinking or intend to drink, then trigger support message Static Goals (to reduce alcohol use or harm); affect (positive or negative); social context (alone or with others) (10) Shrier et al. (2014) EMA If the user reports one of their top 3 triggers for use, desire to use, or Static Top 3 triggers for each participant were selected from lists of types of (Continues) T A B L E 3 (Continued) Authors (year) Type(s) of data used to trigger real-time support Decision rule(s) for triggering real-time support Whether decision rules were static or adaptive Tailoring variables used to personalize real-time support recent use, then trigger a support message companions (alone; family; friends, etc.), locations (home, school, work, etc.), activities (work/chores, school/ homework, hanging out/socializing, etc.) and feelings (annoyed, anxious, bored, excited, happy, etc.) (11) Shrier et al. (2018) EMA Same as Shrier et al. (2014) Static Same as Shrier et al. (2014) (12) Suffoletto et al. (2018) EMA If confidence is < 4, then trigger selfefficacy boost message Static Willingness to set reduction goal, confidence, drinking behaviour (13) Weitzel et al.Wright et al. (2018) EMA If-then rule Static Plans to eat, location, time, mood, planned drinking, cumulative drinking, planned spending, cumulative spending, adverse events JITAI = just-in-time adaptive intervention; EMA ecological momentary assessment; IQR = interquartile range.T A B L E 3 (Continued) Authors (year) Barriers (−) and facilitators (+) to users' engagement Summary of users' engagement with JITAI Analytical technique used for estimation of JITAI effectiveness Effectiveness of JITAI (1) Attwood et al. (2017) Low perceived personal relevance (−) 14% of users chose to define a drinking 'weak spot' in week 1.JITAI use rapidly declined over time Not reported Not reported Businelle et al. (2016); Hébert et al. (2018) Message frequency (+) 87% of all prompted EMAs were completed.On average, participants received 102.1 (SD = 23.7)automated intervention messages Generalized linear mixed models A total of 41% (24/59), 17% (10/59), 31% (18/59), 27% (16/59), 22% (13/59) and 20% (12/59) of participants met biochemically confirmed abstinence criteria at the quit date, week 1, week 2, week 3, week 4, and week 12 follow-up visits, respectively.Messages focused on coping with smoking urges corresponded to significantly greater reductions in urges, as compared with messages that were not tailored to smoking urge (β = −0.62,P < 0.001).Stress-focused messages corresponded to significantly greater reductions in self-reported stress, as compared with messages that were not tailored to stress (β = −0.31,P < 0.001).Messages tailored to reduce easy access to cigarettes corresponded to greater reductions in self-reported cigarette availability compared with messages not specifically tailored to reduce easy access to cigarettes (β = −0.21,P < 0.001) (3) Hébert et al. (2020) Perceived usefulness (+) 84% of all prompted EMAs were completed.On average, participants received 145 treatment messages χ 2 tests or analyses of variance A total of 26% (21/81) of participants were biochemically confirmed abstinent at 4 weeks post-quitting (Smart-T2: 6/27, 22%, QuitGuide: 7/27, 26%, usual care: 8/27, 30%), and 17% (14/81) participants were biochemically confirmed abstinent at 12 weeks post-quitting (Smart-T2: 6/27, 22%, QuitGuide: 4/27, 15%, usual care: 4/27, 15%).There were no significant differences in smoking abstinence between treatment groups at any time-point T A B L E 3 (Continued) Authors (year) Barriers (−) and facilitators (+) to users' engagement Summary of users' engagement with JITAI Analytical technique used for estimation of JITAI effectiveness Effectiveness of JITAI (4) Dulin et al. (2014); Gonzalez & Dulin (2015) Technical issues (−); perceived usefulness (+) The JITAI feature was accessed 6.0 times on average (SD = 2.1) Linear mixed models No results reported specifically for the JITAI feature.The LBMI-A app resulted in a significant increase in percent days abstinent, and a significant reduction in percent heavy drinking days and drinks per week between the baseline assessment and the 6-week follow-up (5) Gustafson et al. specifically for the JITAI feature.Patients in the A-CHESS group reported significantly fewer risky drinking days compared with patients in the control group for the intervention and follow-up period and at months 4 and 12, but not month Hoeppner et al. (2019) Not reported Most participants reported their mood (84%) or craving (88%) at least once when prompted Linear mixed modelsAbstinence rates were 45% at the 2-week follow-up (7-day abstinence), 56% at the 6-week follow-up (30-day abstinence), and 47% at the 3-month follow-up (30-day abstinence) (7) Ingersoll et al. (2014); Ingersoll et al. (2015) Lost/stolen hardware (−); lack of time (−); perceived usefulness (+) The response rate for prompts focusing on substance use was 67% Mixed effect model There were no significant differences between groups in days with substance use post-intervention (8) Naughton et al. (2016) Forgetting (−); technical issues (−); perceived accuracy (+); message frequency (+/−); perceived usefulness (+); unintended consequences (−) A total of 202 geofence-triggered messages [aggregated mean delivery rate per day of 3.0 (SD 0.8) per participant] were delivered.A total of 1109 support messages were delivered by the app [mean = 85.O'Donnell et al. (2019) Lack of novelty (−); technical issues (−); ease of use (+) Participants responded to 68% of prompts.On average, participants engaged with the app on 22.1 days (SD = 9.7) out of a possible 28 days Mixed effect models There was no significant main or interaction effect of time or group on the frequency of risky drinking or alcohol-related harm (10) Shrier et al. (2014) Ease of use (+); low perceived burden (+) The response rate for momentary reports was 64% during the baseline week, 50% during the 2 weeks of Wilcoxon's signed-rank test The odds of using cannabis following top-3 trigger exposure were reduced by almost 50% at follow-up versus (Continues) T A B L E 3 (Continued) Authors (year) Barriers (−) and facilitators (+) to users' engagement Summary of users' engagement with JITAI Analytical technique used for estimation of JITAI effectiveness Effectiveness of JITAI the intervention and 64% during the follow-up week baseline (OR = 0.54, 95% CI = 0.31-0.95,P = 0.03).Compared with baseline, average daily frequency of cannabis use tended to be less during the intervention (RR = 0.78, 95% CI = 0.60-1.02,P = 0.07) and at follow-up (RR = 0.73, 95% CI 0.49-1.08,P = 0.11).Percentage of days abstinent over the past 30 days increased slightly, but nonsignificantly, from baseline to followup (37.9 versus 47.3%, P = 0.13) (11) Shrier et al. (2018) Message frequency (−); perceived usefulness (+) A median (IQR) of 35.1% (24.6-60.4%) of the momentary reports and 57.1% (39.3-85.2%) of the diaries were completed Linear and generalized mixed effects models There was a significant arm-by-phase interaction effect, with a greater decline in momentary cannabis desire with MOMENT, compared with MET-only.Cannabis use on momentary reports also decreased over the study, with odds of use in the intervention and follow-up phases significantly lower than in the baseline phase (OR = 0.46, 95% CI = 0.28-0.76and OR = 0.31, 95% CI = 0.19-0.51,respectively).However, the arm-by-phase interaction was not significant (12) Suffoletto et al. (2018) Not reported Response rates to EMAs were, on average, 82.3% for the first 4-week intervention block, 75.3% for the second 4-week block and 72.8% for the third 4-week block Random effects models All groups, except for those enrolled in the study for +12 weeks, significantly reduced their maximum number of drinks consumed on any weekend day.However, those who selected to enrol for 12+ weeks had lower baseline drinking levels (13) Weitzel et al. (2007) Message frequency (−); lack of novelty (−); perceived usefulness (+); unintended consequences (−) 12 participants were sent messages on 12-14 of study days, 3 on 9-11 days, and 5 on 5-8 days.Half of the participants reported reading 98%-100% of the messages Analyses of covariance Participants in the treatment group reported drinking significantly fewer drinks per drinking day compared with participants in the control group during the study period when responding on the hand-held computer, but not on the pen-andpaper follow-up surveys intervention content, development costs and data security.Similar to Hardeman and colleagues' recent review of JITAIs to promote physical activity [10], as research into JITAIs is in its early stages (both in terms of the quality of current implementations and the strength of available evidence), it is premature to comment on the effectiveness of JITAIs for reducing harmful substance use.However, our review highlights important conceptual and empirical gaps for researchers, developers and health-care professionals, as discussed below.Current state of the field and recommendations for future work First, there is no consensus definition of what JITAIs are and how to develop them, with a minority of extant studies relying on theoretical predictions or observational/experimental data from prior participants to devise decision points, tailoring variables, intervention options and/or decision rules.The utility of JITAIs designed to reduce harmful substance use will depend largely upon their ability to account for the observed idiosyncratic, dynamic and multi-factorial nature of lapse risk [1-7]; yet current JITAI implementations do not facilitate real-time optimization for individual users.Therefore, prior to investing in large-scale RCTs, wecontend that further systematic and creative conceptual and computational work-with insights from the former feeding into the latter and vice versa-is required to make progress on JITAI effectiveness.
(Continued) Authors (year) Barriers (−) and facilitators (+) to users' engagement Summary of users' engagement with JITAI Analytical technique used for estimation of JITAI effectiveness Effectiveness of JITAI (14) Wright et al. (2018) measurement: active measurement has the advantage of those being supported by the JITAI being aware of what data are being gathered but comes at the cost of requiring more time and effort, while passive measurement has the advantage of reducing participant burden but risks being more intrusive into a person's life, often without their full awareness or understanding of what information is being gathered about them, for what purpose and how to control or opt out from such tracking.These tensions are not easily solved and likely requires-just like the development of JITAIs themselves-careful consideration of the characteristics of the population being served by the JITAI, their context and other idiosyncrasies.Therefore, the contribution of active versus passive sensing of key variables of interest within JITAIs to their effectiveness (including how to gather high quality data in an ethically responsible manner) needs to be explicitly studied.Although JITAIs developed within academic or clinical settings need to comply with ethical requirements such as clear disclosure of what data are being collected and their intended use, we note that JITAIs are also developed within commercial settings, with different ethical measurement and static decision rules to deliver real-time support tailored to micro-scale changes in mood or urges.Evidence from large-scale studies on JITAI effectiveness is lacking.There is a need for further conceptual work on what JITAIs are and how to develop them, methodological and statistical training for researchers and developers and research examining ethically responsible use of passive sensors for detecting variables of interest.DECLARATION OF INTERESTS O.P., F.N. and J.B. are unpaid members of the scientific committee for the Smoke Free app.J.B. has received unrestricted research funding from pharmaceutical companies (Pfizer and J&J) who manufacture smoking cessation medications to study smoking cessation.M.B. is an inventor of the Insight mHealth Platform, which was used to develop the Smart-T2 app.He receives royalties related to the use of Insight.E.T.H. and E.B.H. have no conflicts of interest to declare.
Quality appraisal with the mHealth Evidence Reporting and Assessment (mERA) checklist T A B L E 4