Current alcohol treatment largely relies on self-report data, which can be convenient and low-cost; however, there are limitations [1-3]. Reporting alcohol consumption over long periods may lead to recall bias [4]. Self-report can be made even more challenging when working with an alcohol-dependent population [5]. The current gold standard of alcohol assessments and alcohol research are self-report instruments, such as the Timeline Follow-Back [6] and the Alcohol Use Disorders Identification Test [7]. In recent years, various wearable transdermal alcohol sensor (TAS) devices have been developed. These devices measure alcohol consumption from alcohol vapors excreted through the skin via sweat, known as transdermal alcohol concentration (TAC), and can be worn on the wrist or ankle. Wearing a device all day for long periods allows for regular, repeated measurements and data capture in real time [8,9]. Thus, the TAS device is less likely to miss episodes of alcohol consumption than breath alcohol measurement. It is noninvasive, objective, and low-maintenance and allows behavior to be captured in real-world contexts [10]. These advantages address some of the limitations of other methods currently used, such as breathalyzers and blood and urine tests. Another advantage is that this technology has the potential to communicate with a smartphone [10-12]. Data captured can be uploaded over a mobile network and delivered to the patient, clinician, and researcher in near real time. This means reduced time and resources required by staff in addition to potentially more comprehensive and accurate data collected.

There is a range of TAS brands in various stages of development and validation, some available directly to the public. The 2 devices that have been most widely validated and reported on are SCRAM and WrisTAS [10,13-18]. More recently, Skyn from BACtrack, ION Wearable (rebranded as Proof) from Milo Sensors, and BARE and ORBIS from Smart Start have been developed. BACtrack uses a fuel cell alcohol sensor, whereas ION uses an enzymatic electrochemical biosensor cartridge [10,11]. BACtrack can be used as a stand-alone wearable or integrated into the band of a smartwatch [10,12] and, although it uses the same technology as WrisTAS, BACtrack uses a newer generation with improvements in performance and size [10]. At the beginning of 2021, Smart Start announced the development of ORBIS, a TAS with GPS monitoring technology designed to look like a smartwatch. Instead of fuel cell technology, this device uses a transdermal sensor that allows for continuous monitoring and has cellular communication embedded, removing the need to pair with a smartphone. This device is currently being tested through independent research and clinical and pilot trials [19]. These changes in TAS design and use—for example, linking to a smartphone app to display real-time data—address some of the limitations of the older devices. However, these new generations are in earlier stages of development, and researchers have only just begun exploring their validity, reliability, and usability [10-12].

The accuracy of these devices is evaluated by comparing TAC output with self-reported alcohol consumption via Timeline Follow-Back or drinking diaries, blood tests (blood alcohol concentration [BAC]), or breath alcohol concentration (BrAC). When comparing with self-report, the participant typically wears the device and notes how much alcohol they consume and when. This can allow for drinking in the participant’s own environment and for typically greater, self-dosing drinking to be recorded. However, there can be limits to self-report data, including accuracy of recall, adherence to recording, and social acceptability reporting bias, particularly in clinical populations [1-5]. Sensitivity, also known as true positive rate, measures the number of positive drinking events as indicated by TAC that are true positive drinking events when compared with self-report, BAC, or BrAC. Specificity, also known as true negative rate, measures the number of events where alcohol consumption was not detected and there was no alcohol consumed when compared with self-report, BAC, or BrAC [20,21]. Most studies report correlation or sensitivity and specificity data of the TAS device.

The data collected and calculated from TAC (peaks of use, time to peak, and area under the curve [AUC]) can be compared with the measurements collected via breathalyzer or blood tests (BrAC or BAC). Although TAS devices can automatically take readings at predefined time points, owing to the need for frequent administration of breathalyzer readings or the need for blood tests, studies using these comparisons typically require laboratory settings. This means that there are typically fixed-dose amounts of alcohol given by a research team, and the data are taken during a limited period (a few hours). TAC, BrAC, and BAC data are then statistically analyzed to determine the correlation. It would be optimal for TAS devices to perform with high accuracy in both laboratory and natural, real-world drinking situations.

van Egmond et al [15] conducted a systematic review in 2020 exploring the validity of wearable TAS devices. However, their review only included papers published since 2013 that used TAS devices and validation measures obtained from the devices and provided correlation or detection rate measures. This meant that papers identified by van Egmond that included Milo ION, MOX sensors, and a study on BACtrack were not included in their review [15] because of lack of validation measures. For our review, we decided to include all papers that met our inclusion criteria for investigating the use of wearable TAS devices, with no time constraints. We aimed to explore a broad overview of TAS technology within the field and demonstrate the growth in the range and development of devices.

This systematic review aims to investigate the current knowledge by systematically identifying and evaluating the existing literature on the use of TAS devices in clinical and nonclinical populations, alone or in conjunction with a psychosocial intervention. The primary objective is to assess the level of accuracy of TAS devices. There is a linked review paper investigating the acceptability and feasibility of the devices (Brobbin et al, unpublished data, 2022).

This systematic review was conducted according to the PRISMA-P (Preferred Reporting Items for Systematic Reviews and Meta-Analyses Protocols) guidelines [22]. This protocol has been registered in PROSPERO (CRD42021231027). On review of the results, it was decided that the findings of the systematic review should be reported in 2 papers: one focusing on the accuracy outcomes and a second one focusing on the acceptability and feasibility outcomes.

Studies meeting all criteria were included: full-text original studies published in peer-reviewed journals, written in English, and using a wearable transdermal sensor device reporting accuracy outcomes. For the purpose of this review, a wearable TAS is defined as a wearable device that can measure alcohol consumption from alcohol vapors excreted via the skin. There were no restrictions on publication year or participant clinical and demographic characteristics. Data based on conference abstracts, dissertations, and gray literature were not included.

Bibliographic databases included CINAHL, Embase, Google Scholar, MEDLINE, PsycINFO, PubMed, and Scopus. Searches were carried out between February 1, 2021, and February 8, 2021 (Multimedia Appendix 1 [10,11,13,23-51] and Multimedia Appendix 2). The searches were supplemented by cross-checking the reference lists of key publications, related systematic reviews, and included papers.

All identified titles and abstracts were screened in Covidence (Veritas Innovation Ltd) to identify studies that potentially met the inclusion criteria. From this list, the full text was retrieved and assessed by a reviewer (EB); any doubts were discussed with a second reviewer (SH). Any disagreement was discussed with a third reviewer (PD). A data extraction form was created and pilot-tested with the first 5 included studies and refined as necessary (Multimedia Appendix 3). EB extracted the data independently, and the second reviewer (SH) completed the entry check for accuracy. Any discrepancies were resolved through discussion with the third reviewer (PD).

All outcome measures reported, both objective and self-reported, were extracted. The primary outcome was the accuracy of wearable TAS devices for measuring alcohol consumption compared with other methods (self-report, BAC, and BrAC; definitions of these measures are reported in Multimedia Appendix 4 [14,23,25-28,32,37,52]). This included data on correlations, sensitivity and specificity, percentage or amount of unsuccessful data points collected, and any time lag differences for the sensor to reach peak TAC compared with peak BrAC and delay in peak time from drink consumption.

We used the Mixed Methods Appraisal Tool (MMAT) as it is designed for appraisal in reviews that include a range of designs (qualitative, quantitative, and mixed methods) [53]. EB completed this independently, and any queries were discussed with a second reviewer (SH). Any disagreements were resolved through discussion (PD).

To draw conclusions about the included studies, we developed a synthesis of study characteristics. The data are summarized using a structured narrative description for accuracy measures, and we report the available data reported for correlations, sensitivity and specificity, failure rates, and time lag on SCRAM, WrisTAS, and BACtrack, with a separate section for other TAS model studies (with only 1 study on each of these). Not all studies reported accuracy data on all these measures. A meta-analysis was not possible because of the methodological heterogeneity.

After removing duplicates, a total of 125 papers were screened; 31 (24.8%) were excluded at the title and abstract screening, and 94 (75.2%) full-text papers were assessed. Of those 94 papers, a total of 64 (68%) were then excluded (the reasons are provided in Multimedia Appendix 1). There were 7 additional papers identified through citation searching, of which 2 (29%) were included (Multimedia Appendix 1). The final sample included 32 publications (Figure 1).

Of the 32 studies, 19 (59%) used a SCRAM device [13,23-40], 7 (22%) used a WrisTAS device [34,41-46], 5 (16%) used BACtrack Skyn [10,29,30,40,47], 1 (3%) used an ION Milo sensor [11], 1 (3%) used Quantac Tally [10], 1 (3%) used a wearable Internet of Things (IoT) sensor [48], 1 (3%) used an MOX sensor [49], 1 (3%) used a proton-exchange membrane (PEM) fuel cell sensor [50], and 1 (3%) did not name the TAS used [51]. Some studies (4/32, 13%) used more than one version of the device (eg, SCRAMx and SCRAM-II), and 16% (5/32) used more than one brand of device (Table 1).

A large proportion of studies (29/32, 91%) were conducted with adults in good health. Only 9% (3/32) of the studies included participants who were diagnosed alcohol-dependent clinical populations. Most of this research was conducted in the United States, with 94% (30/32) of the studies located there. The earliest paper included was from 1992, but most studies (21/32, 66%) were published as of 2015 (Table 1).

There were 1228 participants enrolled in total in the included studies, with 1147 included in the procedure or analysis. Therefore, 81 participants who were enrolled were not included in the results (for reasons such as withdrawing or missing data). In total, 1 paper (1/32, 3%) was still in the early stages of data collection for one of their studies and so these participant numbers were unknown [10]. Not all studies included detailed information on the participants’ age, sex, and ethnicity. From the information provided, it could be seen that the participants’ ages ranged from 18 to 58 years, most studies included women and men (27/32, 84%; 2/32, 6% included men only; 1/32, 3% included women only; and 2/32, 6% were unknown) and, for most studies (19/32, 59%), White participants represented a high proportion of the sample (11/32, 34% were unknown; Table 1).

All studies apart from 2 (30/32, 94%) met a minimum of 3 out of 5 criteria [10,23] (Table 1). This is due to Alessi et al [23] not providing details about randomization and participant information, and the study by Wang et al [10] was difficult to score owing to incomplete data collection as their study was still ongoing at the time of publication. With the MMAT, exclusion of low–methodological-quality studies is discouraged [53]. The MMAT scores for each study are provided in Multimedia Appendix 5 [10,11,13,23-51]. All the records selected for data extraction were considered to be at low risk of bias. Owing to the nature of many of the studies included, blinding of participants and personnel was not possible; in some studies (3/32, 9%), there were clear differences in demographics or amount of alcohol provided between groups [30,41,44] or incentives provided within contingency management studies [25] where personnel were required to know participant allocation. Similarly, in many studies (17/32, 53%), there was only 1 group of participants all completing the same task, so randomization or blinding was not required or possible [10,13,24,28,31-34,40,42,43,45,46,48-51], or there was only 1 participant [11]. Some studies (16/32, 50%) did not provide clear information on participant data, randomization, incomplete outcome data, and selective reporting, resulting in potential bias because of limited information in the paper [10,11,24,25,29,30,34,40,41,43-46,48-50].

The aim of this review was to assess the current knowledge on the accuracy of TAS devices. We identified 32 papers, few of which reported the use of this technology in clinical populations. Of the 32 papers, we identified only 3 (9%) studies that used alcohol-dependent participants. Most studies used either SCRAM (19/32, 59%) or WrisTAS (7/32, 22%). However, there were some studies that investigated BACtrack Skyn (5/32, 16%) and others (6/32, 19%). This is a growing field, and the production and investigation of additional TAS devices supports this. This review found that wearable TAS devices could detect alcohol consumption with moderate to strong accuracy over various periods. However, factors such as the amount of alcohol consumed, the environment (laboratory and self-dose real-world setting), age, and where the device is worn must be considered. The findings differed across the TAS brands included, and studies on each brand reported different limitations; for example, time lag compared with BrAC or BAC, data file errors, or device failure.

The SCRAM, WrisTAS, and BACtrack studies showed reasonably high data capture rates but demonstrate that this is <100% [26,27,32,34,41,42,44]. There were also reports of device failure, malfunctions, noise, and tampering, all of which reduced the amount of successful data capture [29,30,43]. It was noted over time that SCRAM lost accuracy, most likely because of water accumulation [34]. The main feature of using these devices compared with other tests, such as breathalyzers or urine tests, is the capacity for continuous measurements, so this is an important benefit of TAS devices. There has been some investigation into factors such as sex and BMI on episode detection, with no conclusive relationship determined [26,28].

SCRAM devices had the lowest amount of recorded failure rates of the included devices [29,30,34,42], but the time lag to peak alcohol concentration was considerably longer than that of the other brands of TAS devices [10,13,29,34,40,45,47]. Although SCRAM, WrisTAS, and BACtrack showed positive correlations with self-report, BAC, and BrAC, the accuracy of determining the amount of alcohol appeared to be stronger in BACtrack than in SCRAM in some studies [29] but not in others [40]. WrisTAS showed the weakest support for accuracy [34,42]. There were fewer studies on BACtrack compared with SCRAM and WrisTAS, and the studies included that used BACtrack mostly used a device prototype. However, the preliminary results appear promising. Similar to WrisTAS, there seems to be a high failure rate and noise in the data files, but there is support for a shorter time lag compared with SCRAM as well as a strong positive correlation with BAC or BrAC, and it maintains the ability to distinguish between high and low doses [29,40,47].

Approximately 3% (1/32) of the studies compared SCRAM and WrisTAS in laboratory and ambulatory settings. When the participants self-dosed, the true positive rates for both devices increased and, the higher the peak BAC level, the higher the rate of true positive TAC [34]. Although SCRAM more accurately estimated self-dosing and WrisTAS more accurately estimated laboratory dosing, neither was accurately estimated in both settings. This difference between devices is something that should be investigated further and could affect preference for brand of device for different uses in different settings [34].

Our findings relating to SCRAM, WrisTAS, and BACtrack were largely in line with the review findings of van Egmond [15]. Data from these TAS devices were found to moderately to highly positively correlate with self-report and BrAC, with WrisTAS and BACtrack devices showing higher malfunction, failure rates, and noise within data files compared with SCRAM. Van Egmond [15] did not include papers exploring other TAS devices.

There was 1 individual paper (1/32, 3%) on each of the other 6 wearable TAS devices [10,11,48-51]. The development of other devices suggests that this is a rapidly evolving technology. The unnamed device in the study by Davidson et al [51] demonstrates the ability of this technology to accurately measure low BAC and the general agreement between TAC, BAC, and BrAC data. Li et al [48] is the only study that compares data taken from the same device but worn on different body parts. The data found suggest that the location of the device on the user does affect the measurements captured, which is important to consider.

This review highlights the small but growing number of studies investigating the accuracy of wearable TAS devices. With only a small number of studies testing more than one brand or even the same setting, intervention, or population, it is difficult to draw direct comparisons. Most studies (29/32, 91%) included healthy adults, and the length of time studied ranged from a day to a few months. If the ultimate objective is for these devices to be used within the clinical population or the criminal justice system for extended periods, this manner of use needs to be explored further. No included study was conducted in a criminal justice setting with offenders. However, there is a growing appeal for this technology to be used in this environment; for example, the use of SCRAM in the UK criminal justice system for alcohol-related offenses [54].

A mix of study designs was included in both the ambulatory and laboratory settings. Although it is useful to conduct studies in controlled laboratory environments to investigate validity, reliability, and accuracy, the target use is not necessarily a laboratory environment. Hence, how the devices perform in vivo with populations other than those studied must be investigated. This would then be able to inform the devices’ long-term effectiveness in future treatment, interventions, research, and policy.

There are currently a small number of studies in this area, but research on the use of this technology is growing and, owing to technological advancements, the accuracy and ability of these devices is improving considerably. What is needed is for research to expand into other populations, such as clinical populations and offenders within the criminal justice system, to examine their accuracy and reliability in the intended target populations and contexts. Although the accuracy outcomes for this technology are promising, there is a limit to this research because of the mostly laboratory and short-duration study design.

The use of wearable TAS devices is becoming more accepted and appealing to society, as evidenced by their increasing implementation in the criminal justice system and increasing research. The implications of this review are that we need to investigate the engagement in real-world settings, where transdermal sensors are intended to be implemented, with the target populations. This can then inform clinical interventions, treatment, research, and use.