Leveraging mHealth and Wearable Sensors to Manage Alcohol Use Disorders: A Systematic Literature Review

Background: Alcohol use disorder (AUD) is a condition prevalent in many countries around the world, and the public burden of its treatment is close to $130 billion. mHealth offers several possible interventions to assist in the treatment of AUD. Objectives: To analyze the effectiveness of mHealth and wearable sensors to manage AUD from evidence published over the last 10 years. Methods: Following the Kruse Protocol and PRISMA 2020, four databases were queried (PubMed, CINAHL, Web of Science, and Science Direct) to identify studies with strong methodologies (n = 25). Results: Five interventions were identified, and 20/25 were effective at reducing alcohol consumption. Other interventions reported a decrease in depression and an increase in medication compliance. Primary barriers to the adoption of mHealth interventions are a requirement to train users, some are equally as effective as the traditional means of treatment, cost, and computer literacy. Conclusion: While not all mHealth interventions demonstrated statistically significant reduction in alcohol consumption, most are still clinically effective to treat AUD and provide a patient with their preference of a technologically inclined treatment Most interventions require training of users and some technology literacy, the barriers identified were very few compared with the litany of positive results.


Rationale
Alcohol use disorder (AUD) is characterized by the inability to stop or control alcohol use despite social, occupational, or health consequences [1]. Approximately 85.6% of people aged 18 years and older in the U.S. reported they drank alcohol, 69.5% reported they drank in the last year, and 54.9% reported they drank in the last month. In a survey of primary care providers in the European Union, AUDs were prevalent in 11.8% of the population, which is 1.6 times the population estimate [2]. AUD is specifically attributed to 735,153 deaths in 2019, but indirectly associated with 7,599,264 when alcohol-related deaths are taken into consideration [3]. In the U.S., AUD is associated with $120 billion per year in medical costs in the US, and $7.6 billion in the EU [3,4].
Telemedicine is defined by the World Health Organization (WHO) as healing at a distance through information and communication technologies (ICT) [5]. Telemedicine provides clinical support, overcomes geographical boundaries, involves ICT, and has a goal to improve health outcomes. Telemedicine comes in many forms, but wearable sensors can be connected to apps on mobile devices. When these wearable sensors provide clinical data to providers, this falls under the scope of mHealth.
Treatments for AUD can be both inpatient and outpatient, and they often must be tailored to the individual [1]. Wearable sensors have the ability to observe behavior and physiological constructs and combine them with location tracking. Tracking gait and Healthcare 2022, 10, 1672 2 of 21 sweat can provide feedback on abstinence and intoxication [6,7]. The geographic location can provide pre-programmed text messages to warn against danger areas (proximity to establishments that sell alcohol) [8].
In general, a systematic literature review is conducted to summarize recent science on a particular subject. A continuous growth of research combined with the rapid growth of technology compels scientists to systematically summarize available research and synthesize evidence. These products form the basis for funded research, and they can provide a foundation for modifying evidence based practice. As of the writing of this systematic review, 13 funded grant opportunities exist in the area of alcohol use disorder in the USA alone. Technology often serves as a fulcrum of change, and many mHealth solutions exist to help manage alcohol use disorder. A systematic literature review at the intersection of mHealth and the treatment of alcohol use disorder seemed timely. A systematic review in 2020 analyzed 32 articles over a 5-year period [9]. This study found half of the interventions reported improvements in at least one outcome (reduced cravings, or alcohol use). Only two of the interventions utilized wearable sensors. The remainder were feedback apps for craving management, coping assistance, and tailored feedback [9].
Another systematic review published in 2020 analyzed 22 articles over 10 years [10]. The study team found that most interventions resulted in a positive outcome (reduced depression, increased satisfaction, increased accessibility, increase quality of life, and decreased cost. Interventions included mobile health apps, eHealth (computer programs), telephone intervention, and 2-way video [10].

Objectives
The purpose of this review is to analyze the effectiveness of mHealth and wearable sensors to manage AUD, compared with the outcomes of the same conditions under traditional, face-to-face (in person) treatment, from evidence published in peer-reviewed and indexed journals over the last ten years. Effectiveness will be measured as improvements in AUD cravings, decrease in alcohol consumption, and a positive rating in patient satisfaction.

Eligibility Criteria
Articles for analysis were published in the last 10 years in peer-reviewed academic journals, and published in the English language. They must include participants who are adults (18 years of age or older). Preferred methods were true experiments (RCT, etc.), but quasi-experimental, non-experimental, and qualitative studies were also accepted. Other systematic reviews were not accepted so as not to confound the results. Works that did not mention wearable sensors or mHealth to treat AUD were excluded. Studies with participants under age 18 were excluded. Studies that did not report results were excluded.

Information Sources
Four data sources were queried: PubMed (MEDLINE), Cumulative Index of Nursing and Allied Health Literature (CINAHL), Web of Science, and Science Direct and a focused journal search in the Journal of Addictive Medicine. These databases were chosen because they are well known, exhaustive, and easily accessible by those who want to duplicate the research. MEDLINE was excluded from all searches except PubMed. Searches were conducted on 8 January 2022.

Search Strategy
Our study team used the Medical Subject Heading (MeSH) feature of the National Library of Medicine to create a Boolean search string that combined key index terms: (mhealth OR telemedicine OR "mobile app" OR biosensors) AND ("alcohol use disorder" OR "AUD"). We used the same search string in all databases and the focused journal search. As close as databases would allow, we used the same filter strategies.

Selection Process
We used the Boolean search string in all databases, filtered the results, and screened the abstracts for applicability, in accordance with the Kruse Protocol [11]. The Kruse Protocol defines a systematic methodology to conduct an exhaustive summary of evidence and report in accordance with the PRISMA standard. Studies were removed that did not address the objective statement.

Data Collection Process
We used a standardized Microsoft Excel spreadsheet as a data extraction tool collecting additional data fields at each step. The Kruse Protocol standardized the spreadsheet. We used a series of three consensus meetings to confirm the group of studies for analysis, conduct the thematic analysis, and perform additional analysis [12]. Abstracts were screened and studies were analyzed by at a minimum two reviewers.

Data Items
The Kruse Protocol dictated we collect the following fields of data at each step: DB Source, Date of publication, author names, title, participants, experimental intervention, results, medical outcomes, study design, sample size, bias within study, effect size (Cohen d), sensitivity, specificity, F1, country of origin, statistics uses, patient satisfaction, barriers to adoption, strength of evidence and quality of evidence.

Study Risk of Bias Assessment
Each reviewer noted observed bias and assessed the quality of each study using the Johns Hopkins Nursing Evidence Based Practice tool (JHNEBP) [13]. This was done because bias can limit the external validity of studies [14].

Effect Measures
This study included both qualitative and quantitative studies. Due to the fact that we accepted this range of methodology, we were unable to standardize summary measures, as would be performed in a meta-analysis. Measures of effect are summarized in tables for those studies in which it was reported.

Synthesis Methods
This subheading is for meta-analyses-NOT for systematic reviews. It will be removed by the editor prior to publication.

Reporting Bias Assessment
The overall ratings of strength and quality from the JHNEBP tool provided an assessment of the applicability of the cumulative evidence. Observations of bias were discussed for their implications on their reported results.

Additional Analyses and Certainty Assessment
We performed a narrative or thematic analysis of the observations to convert observations into themes (an observation that occurred multiple times became a theme) [12]. We calculated a frequency of occurrence and report this in an affinity matrix. Reporting the frequency provided confidence in the data analyzed. Figure 1 illustrates the study selection process from the four databases and one targeted journal search. Using established methods, we calculated a kappa statistic (k = 0.96, almost perfect agreement) [15,16]. Figure 1 illustrates the initial search results of 786 and how we filtered and screened these down to the group for analysis (n = 25).  Figure 1 illustrates the study selection process from the four databases and one targeted journal search. Using established methods, we calculated a kappa statistic (k = 0.96, almost perfect agreement) [15,16]. Figure 1 illustrates the initial search results of 786 and how we filtered and screened these down to the group for analysis (n = 25).

Risk of Bias within and across Studies
Because of the high number of RCTs and true experiments in the group of articles analyzed, the JHNEBP quality assessment tool identified 18/25 (72%) as Strength of Evidence I. Only 7/25 (28%) were classified as Strength of Evidence III. Similarly, the strong methodology, large sample sizes, and consistency of results caused the JHNEBP tool to identify 23/25 (92%) as Quality of Evidence A. Only 2/26 (8%) were classified as Quality of Evidence B.
Reviewers also made note of internal and external bias in the studies. All articles were conducted in either one or multiple regions of only one country, which is an indication of selection bias. This threatens the internal validity of the study. Furthermore, 10/25 (40%) observations of sample bias were identified because the sample used a disproportionate percentage of one gender or race. This form of bias threatens the external validity of the results.

Results of Individual Studies
Following the Kruse Protocol, reviewers recorded independent observations during data extraction. These observations were discussed in Consensus Meeting number two. Through the discussion of observations, a thematic analysis was performed to make sense of the data [12]. Reviewers identified themes and performed a second data extraction to ensure no themes were omitted. Table 2 tabulates the themes identified in the literature. Appendices A and B provide an observation-to-theme match. While there is some overlap between Results, Medical Outcomes, and Effectiveness, reviewers felt it was necessary to report them separately in order to highlight both similarities and differences between the studies. Appendix C provides the other observations made by reviewers (sample size, bias, effect size, country of origin, statistics used, and the JHNEBP observations of strength and quality of evidence).

Results of Syntheses
This subheading is for a meta-analysis, not for a systematic review. This section will be removed by the editor before publishing.

Additional Analysis and Certainty of Evidence
Affinity matrices were created to summarize the frequency and probability of occurrence of each theme or observations. Frequency and probability do not imply importance: They only state the probability the theme or observation would be identified in the group for analysis. As part of the thematic analysis, observations that occurred more than once were identified as themes. All others are listed as individual observations.

Patient Satisfaction
Patient satisfaction was very positive for all studies. The reason for this may have been because participants had already presented themselves for treatment for AUD, therefore, they would be positively disposed toward most interventions. The exact modality may not have negatively affected patient satisfaction. This is a significant error of both internal and external validity, and this variable should not be used to form any conclusions about interventions. Table 3 summarizes the results of studies compared with a control group. Table 2 identifies which studies did not have a control group. Five themes and three individual observations were identified by the reviewers for a total of 68 occurrences in the literature. Reduction in consumption was identified in 15/31 (48%) of the occurrences [18,23,26,27,30,[32][33][34][35][36][37][38][39][40][41]. In 10/31 (32%) of the occurrences, the reduction was statistically significant, but in 5/31 (15%) of the occurrences, it was not statistically significant [21,22,29,31,33]. Three of 25 (10%) occurrences mentioned the intervention caused positive retention in treatment programs [17,24,28]. Furthermore, in 3/25 occurrences, the participants increased self-efficacy and scored better on the AUDIT [23,30,32]. In 2/31 (6%) occurrences, the intervention decreased binge drinking [26,27]. The following are individual observations that could not fit into a theme. One intervention used a Bayesian Network Model to predict relapses. This enabled providers to intervene through text, email, or phone. One intervention highlighted a high rate of acceptance among participants, which may have been related to the fact that participants already volunteered for treatment-the modality may not have played a significant part. One intervention noted positive frontal lobe function which could lead to a decrease in addiction behaviors [19,20,25]. Table 3. Results to the studies.

Results Themes and Observations Frequency
Reduction in consumption [ Table 4 summarizes the medical outcomes observed. Ten themes and two individual observations were recorded commensurate with the adoption of (intervention) for a total of 34 occurrences. Many of these themes were like those highlighted in results. Only differences from results will be reported. Three interventions identified an increase motivation to change behavior as a result of the intervention. This occurred in 3/34 (9%) observations [28,34,38]. A high number of observations were unable to be fit into themes. One article mentioned a reduction in craving for alcohol. One mentioned an improved rate of depression indicators. One mentioned an improvement in dependence on alcohol. One highlighted an increase in medication compliance [17][18][19]33]. Effectiveness Themes and Observations Table 5 summarizes the effectiveness themes and observations. Eight themes and six individual observations were recorded by reviewers for a total of 50 occurrences. Many of these themes overlapped with study results and medical outcomes. Only the differences will be reported. In four of the interventions, it was highlighted that these are equally as effective at treating AUD, so the decision to choose one method over the other could fulfil a patient's preference, and this preference may increase the success of the intervention [18,21,22,33]. Two interventions were highlighted as low cost [17,18]. Two interventions resulted in sustained abstinence from drinking [34,35]. One intervention was noted as exceptionally good at providing education about AUD and healthy habits [24]. Barriers to the Adoption of mHealth and Wearable Sensors to Manage AUD Table 6 summarizes the barriers to the adoption of mHealth and wearable sensors to manage AUD. Four themes and two observations were reported for a total of 34 occurrences.

Interactions between Observations
Overall, mHealth apps mostly resulted in a reduction in alcohol consumption or a reduction in cravings [17][18][19][20]24,26,28,30,33,[36][37][38][39]41]. The mHealth + telephone interventions had the same effect [35,40]. The mHealth SMS interventions had mixed results: They both reduced consumption of alcohol, but only one was a statistically significant decrease [22,23]. The telephone interactive voice intervention also showed mixed results: They all decreased alcohol consumption, but not all were statistically significant [21,29,32,34]. Finally, the mHealth with serious games or cognitive training showed the most promise with a younger population. This intervention also showed a decrease in alcohol consumption, and one of them highlighted an increase in frontal lobe function, which is theorized will decrease addiction [25,27,31].
Future research should explore why some of these interventions did not demonstrate a statistically significant reduction in alcohol consumption. There may have been customization of SMS messages or tailoring of the apps to cater to preferences of the patients. This may increase the efficacy of the intervention and decrease prevalence of AUD.

Limitations
Several reviewers were used in this study to control for confirmation bias. Unfortunately, a review is often limited to what can be found. To ensure studies were of high quality, we only accepted studies that had been published, however, this subjects the study to publication bias because we did not consider grey literature. The studies selected all exhibited small instances of selection and sample bias, which affect their internal and external validity, respectively. We only selected published articles from the last 10 years because technology advances so rapidly. Had we looked back 15 years, we may have identified additional themes in the literature.

Conclusions
mHealth and wearable sensors are effective tools to decrease alcohol consumption, increase self-efficacy and self-determination, and provide overall treatment of AUD. The evolution of studies on this topic has slowly grown over time. mHealth technology may require additional training of users at both ends, but its low cost and efficacy outweigh the disadvantages. Although some interventions are not statistically different from traditional care, the use of mHealth and wearable sensors may fulfill the preference of a patient and increase the success of treatment.

Institutional Review Board Statement:
This study used no human subjects, so it is IRB exempt.

Informed Consent Statement: Not applicable.
Data Availability Statement: Data from this study can be obtained by asking the lead author.

Conflicts of Interest:
The authors declare no conflict of interest.   McKay et al. [40] telephone and smartphone-based intervention mHealth + telephone Use of telephone or smartphone was effective in treating AUD Reduction in consumption Improved rates of alcohol dependent persons

Reduction in consumption
O'Grady et al. [41] mHealth app mHealth app Provider-facing technology is effective alcohol intervention services and increase access to care in low-and middle-income countries.
Reduction in consumption Improved rates in alcohol dependent persons

Reduction in consumption
Appendix B Table A2. Observation-to-theme conversion (Effectiveness, and Barriers).

Measures of Effectiveness Effectiveness Themes Barriers to Adoption Barrier Themes
McTavish et al. [17] Low cost intervention, good acceptance, decreased cravings, increased autonomy and self-determination