To reflect the interdisciplinary nature of digital health technologies, we deployed searches in a mixture of medical (PubMed, PsycINFO, Medline, CINAHL, Cochrane), engineering (IEEE Xplore, ACM Digital Library), and general-purpose databases (Web of Science, Scopus). Base terms “ADHD”, “Technology”, and “Adult” were used to initially scan titles and abstracts to identify related terminology. Where possible, we incorporated MeSH, APA, or key word descriptors to index relevant literature categories. Search terms deployed in PubMed are shown in Table 1, with all search queries available to view in the Supplementary Material.

Broad inclusion criteria were set, such that any primary peer-reviewed paper, which evaluated the deployment of a DHT or prototype DHT, to target any aspect of ADHD, in any setting, with adults, was included. By keeping the criteria broad, we sought to include as many examples of DHT development and deployment as possible. We set no limit on the publication date and conducted our last round of searches to identify examples up to and including December 2025. We excluded abstracts, student theses, datasets, theoretical contributions, and technologies deployed solely to participants between 0 and 17 Years Old. Texts were limited to those available in the English language.

In the context of this review, we adopt the definition of a DHT provided by NICE (40). The definition includes technologies, apps, platforms, or software that is intended to benefit an individual's health, or the wider health and social care system. Importantly, it also excludes technologies designed solely for healthcare professional training, or technologies that are only used to facilitate data collection in research studies (44). As such, technologies which were deployed with the intent of researching ADHD neurophysiology only, without an explicit health benefit, are not included in the review.

All searches were exported and collated into EndNote 21 (45). Full results were uploaded to the online review management software rayyan.ai (46) and screened for duplication. Two of the authors (FS, SW) then independently screened the titles and abstracts against the inclusion and exclusion criteria to determine eligibility. Where decisions could not be made from the title and abstract alone, the full text was assessed. Disagreements between the two reviewers were settled through regular discussions following the initial round of screening. Inter-rater reliability was 97.25%. Papers which were screened in were assessed by the primary author, with any exclusions made at this stage noted with reasoning. The citation lists of full-text included papers were examined to identify any further eligible examples.

Key study information including title, abstract, year of publication, primary/secondary outcome(s), study size, and age of participants was extracted from each paper and recorded in a spreadsheet using Microsoft Excel (47).

To chart the collected papers, we adopt the framework proposed by NICE for DHTs (40), wherein technologies are classified by their intended purpose and stratified into tiers based on their overall risk. Key descriptions are summarized in Table 2 [for further detail, see (40, 44)]. Alongside this, each paper was inductively mapped into a technology cluster, depending on the type of technology primarily being used. This was achieved by initially labelling the core technology as described by the authors. These labels were then iteratively developed into overarching categories based on their similarity to other examples.

Within the NICE DHT framework, the categories Drive Clinical Management and Diagnose a Specific Condition overlap conceptually. DHTs classified as Drive Clinical Management may inform clinical assessment, whilst Diagnose a Specific Condition describes DHTs that result in an immediate or near-term action to diagnose, screen, or detect a condition. To support the distinction between the two categories, we classified each DHT on the basis that it would be implemented clinically as described by the authors. DHTs that provided a stand-alone diagnostic or screening outcome, with no integration into traditional assessment pathways, were considered Diagnose a Specific Condition. DHTs that supplemented traditional clinical assessment to produce a probability of diagnosis, or propose a clinical support tool that agglomerates existing clinical data, were classified as Drive Clinical Management.

To support the assessment of study design and quality of evidence, RoB analysis was undertaken using the NIH Study Quality Assessment Tools (48). A RoB assessment is generally not performed in a scoping review unless specifically required to support the study aims (42). However, one objective of this review is to examine the quality of evidence for how DHTs would bring about measurable health benefits. As such, a RoB analysis provided the framework for an objective appraisal of the evidence. The NIH tools provide a suite of tailored checklists for different study designs including controlled interventions (CIV), systematic reviews and meta-analyses, observational cohort/cross-sectional (OCS), case-control (CCO), pre-post with no control group (PPNC), and case series (CAS) studies. Checklists range from 9 items for CAS studies to 14 items for CIV studies, with possible responses to each criterion including Yes, No, Cannot Determine (CD), Not Reported (NR), or Not Applicable (NA). The NIH tools were chosen due to their ability to accommodate multiple study designs within a single framework, which aligns with the heterogeneous nature of study designs collated in a scoping review. The NIH tools can additionally be used to provide an overall rating of papers as either Good, Fair, or Bad. However, given that they are not validated tools, and not considered as robust as study-design specific tools (49–51) we do not categorize studies according to quality. We instead report on the descriptive trends observed across the criterion for the collection of papers. The use of more extensive tools was considered out of scope for the purposes of this study, given that RoB is not typically performed in scoping reviews.

An overview of how each component of the review methodology supports our objectives is presented in Table 3. To support with the synthesis of results, we present our findings following the NICE DHT Framework for classification. As such, the results section is structured such that the answers to our research question, namely the papers identified, common outcome measures, types of technology deployed, study design, and sources of bias, are presented for each NICE DHT category. This enables us to identify trends and insights that were specific to each aspect of the clinical pathway. Full tables containing the information extracted for each paper are available in the Supplementary Material.

Figure 1 shows that after executing the search strategy in each database, 7,982 hits were returned. Following de-duplication, screening for eligibility, and searching through citations of full-text inclusions, 133 papers were included in the review.

Figure 2 shows the cumulative number of publications for each DHT classification. For class C technologies, 63 (43.37%) were classified as Treat Specific Condition, 36 (27.07%) were Drive Clinical Management, 19 (14.29%) were Diagnose a specific condition, and 9 (6.77%) were Inform Clinical Management. For Class B technologies, 5 (3.76%) were Promoting Good Health, and 1 (0.75%) was Communicating about Health and Care.

Alongside charting DHTs into NICE categories, we also categorized papers based on the core technology being deployed, shown in Figure 3. The most used technology was a form of web or app-based cognitive therapy or psychoeducation, used in 29 papers, which were classified predominantly as Treat Specific Condition in 26 cases. This was followed by 22 electroencephalography (EEG) based technologies, used in 7 cases to Treat Specific Condition, 8 cases to Drive Clinical Management, or in 7 cases to Diagnose a specific condition. A series of 16 papers examined the use of a computerized continuous performance test (CPT), either in isolation (n = 5) or while simultaneously (n = 11) collecting other forms of data such as head movement, neuroimaging data, or eye tracking data.

As shown in Table 4, study design varied depending on the NICE DHT classification. For DHTs categorized into Treat Specific Condition, CIVs were dominant, used in 38 studies. In contrast, Drive Clinical Management and Diagnose a Specific Condition were almost all OCS studies, with 34 out of 36 and 18 out of 19, respectively. Promoting Good Health and Inform Clinical Management were each a mix of study designs. The only paper to be categorized as Communicate About Health and Care was a CAS study. Information regarding the average study size, participant age, and % Males for each NICE DHT category is presented in Supplementary Table 1.

Across this category of papers, several subcategories of intervention shared a common primary technology. 27 out of 63 evaluated the deployment of a form of digital cognitive therapy or psychoeducation (52–78), 13 focused on the evaluation of cognitive training programs (79–91), 9 deployed a form of neurofeedback (NF) (92–100), 9 papers used transcranial direct current stimulation (tDCS) or transcranial alternating current stimulation (tACS) (101–109) and 3 used repetitive transcranial magnetic stimulation (rTMS) (110–112). The remaining 2 papers examined the demographic contributors to successful treatment outcomes for remote telepsychiatry (113) and a haptic feedback wearable to improve symptoms of anxiety and focus (114). Key study-level information for each subcategory is available to view in Supplementary Tables 2–6.

Cognitive-Behavioral DHTs (Supplementary Table 2) typically deployed modules to participants grounded in psychotherapeutic approaches. Most drew on Cognitive Behavioral Therapy (CBT) (55, 58–60, 63, 74, 75), with fewer examples referencing Acceptance and Commitment Therapy (65, 66), Dialectical Behavior Therapy (DBT) (71), Self-determination theory (76), Stress management (77), or more general psychoeducational principles (52, 67, 68). DHTs were most frequently deployed as part of a guided program (52, 54, 57, 58, 64–67, 71–74, 76, 77), with fewer examples of DHTs used as a self-guided tool (53, 55, 56, 59, 60, 63, 69, 70, 75), or as companion apps that accompanied traditional face-to-face therapies (61, 62, 68). 1 study additionally developed and deployed a chatbot to accompany self-guided app content (78). Interventions were most commonly 6 weeks in length (Range 3 weeks—16 weeks). Content was mostly developed to address general ADHD-related challenges in 16 cases (53, 55, 56, 58–60, 63, 64, 67–69, 71–74, 78). Fewer interventions focused on specific aspects such as emotion regulation (61, 62), executive functioning (54, 75), and comorbidities including cannabis use disorder (52) and anxiety and depression (70). Effectiveness was largely assessed using self-report measures of ADHD symptoms including the Adult ADHD Self-Report Scale (ASRS) (55, 57, 58, 61, 75), the ADHD Rating Scale (ADHD-RS) (74, 76) and DSM-IV Current Symptoms Scale Self-Report (CSS) (63). Only one paper reported observer-rated symptom severity through the Integrated Diagnosis of ADHD—Revised (IDA-R) framework (68). A majority of papers (53, 54, 56, 59–62, 64, 65, 67, 69–71, 73, 76) reported primary outcomes related to user experience, including feasibility, acceptability, adherence, and user satisfaction. Together, these results indicate a tendency to deliver DHTs as part of a guided program, yet the high proportion of outcomes based around feasibility and acceptability suggests many remain in an early stage of development.

Another 13 papers evaluated cognitive training programs (Supplementary Table 3). These DHTs present the user with increasingly challenging tasks designed to target and strengthen neuropsychological processes implicated in ADHD. 11 programs ran on computer or mobile interfaces (80–83, 85–91), whilst 2 were based on VR paradigms (79, 84). Training most often targeted working memory (79, 81–83, 85–87, 89, 90), with fewer examples targeting attention (80, 84, 88) or executive functioning more generally (86, 87, 91). Outcomes were split into “near-transfer” measures which relate directly to the neuropsychological process being targeted, and “far-transfer” measures which examine whether improvements transfer to a generalized setting. Near-transfer measures were more frequently deployed, in 10 studies, and consisted of tests of cognitive ability such as CPTs (80, 84, 86–88, 91), the Wechsler Adult Intelligence Scale (WAIS) IV Digit Span (82, 83, 89) or Wechsler Memory Scale (WMS) IV Spatial Span (82, 83, 85). Far-transfer measures appeared in fewer cases (n = 8 studies) and relied largely on self-report measures such as the ASRS (82, 83, 85–89), and both the Cognitive Failures Questionnaire (CFQ) and Barkley Deficits in Executive Functioning (BDEF) (82, 83, 85).

A further 9 papers evaluated the deployment of neurofeedback (NF) protocols (Supplementary Table 4). NF operationalizes a neuroimaging modality to produce visual or acoustic stimuli that are fed back to the participant to serve as a basis for training to alter brainwave activity. Various parameters have been identified as targets for ADHD, and in this collection of papers, the theta-beta ratio was most commonly used in 4 studies (92, 97–99), with other examples including sensorimotor rhythm (92, 93, 98), alpha power (99), dorsal anterior cingulate cortex (dACC) activation levels (100), prefrontal HbO2 Concentration (94), and slow cortical potentials (95). Total amount of time spent actively performing NF varied from 8 h to 40 h, with an average across the papers of 18.08 h. Outcome measures again typically consisted of CPTs, used in 6 papers (92, 96–100), self-reports of ADHD symptoms in 6 papers (92, 95, 97–100), and in 3 papers (95, 97, 99), self-reports of symptoms of depression.

A total of 12 studies examined forms of non-invasive brain stimulation (Supplementary Table 5) including transcranial direct current stimulation (tDCS, n = 8), transcranial alternate current stimulation, (tACS, n = 1), or repetitive transcranial magnetic stimulation (rTMS, n = 3). These modalities either apply weak electrical currents to alter excitability (tDCS), alternating electrical currents to stimulate rhythmic electrophysiological activity (tACS), or magnetic pulses (rTMS) to modulate cortical excitability in specific regions. The dorsolateral prefrontal cortex (dlPFC) was the most common region targeted in 7 papers (101, 102, 104, 106, 107, 109, 111), with other areas including regions involved in P300 waveform generation (103), posterior brain regions (105), bilateral prefrontal regions (110, 112), and frontoparietal regions (108). Total amount of time receiving stimulation averaged 5.88 h (Range 0.33–14 h). Consistent with other subcategories of DHTs within Treat a Specific Condition, the most common outcome measures were a combination of ADHD symptom severity self-reports including the ASRS (102, 106, 107) and the Conners Adult ADHD Rating Scale (CAARS) (110, 112) alongside a form of CPT in 2 papers (101, 105).

Altogether, these studies showed a consistent focus on targeting ADHD-related symptoms, whilst relying on reporting efficacy through self-reported symptom measures. Alongside this, feasibility and acceptability were frequently reported, particularly for cognitive-behavioral DHTs. Cognitive training, neurofeedback, and non-invasive brain stimulation DHTs further emphasized near-transfer cognitive measures, and overall, comparatively fewer studies incorporated observer-rated outcomes or evidence of broader functional transfer.

Within this category (Supplementary Table 7), 34 (out of 36) deployed a DHT to collect data used to support a diagnosis of ADHD through machine-learning or logistic regression-based classifier approaches. Approaches were split between using technologies to create unimodal (n = 15 papers) or multimodal datasets (n = 19 papers). Unimodal datasets relied on single sources of data such as EEG in (115–121), digitized standard clinical assessment data (122–124), MRI (125–127), VR (128, 129), CPTs (130, 131), eye tracking (132) and fNIRS (133). Conversely, multimodal datasets typically combined physiological and behavioral measures. The most commonly deployed was the QbTest + (134–139), which combines CPT metrics with head movement measurements, whilst other examples include actigraphy and heart rate data with CPT metrics (140), eye-tracking with CPT metrics (141, 142), VR-based CPT with multiple additional forms of data (head movement, eye tracking, EEG, subjective experience, and fNIRS) (143–145), multiple formats of standard clinical assessment data (self-report data with interview transcripts) (146, 147), and EEG with measures from the Wender-Utah Rating Scale (WURS) (148). Outside of diagnostic assessment, 1 paper examined the use of the QbTest + to evaluate medication response (149), and 1 used MRI in a classification study to discriminate between ADHD subtypes (150).

Classifier performance was evaluated in 30 studies, with the majority reporting accuracy (percentage correctly classified over the total dataset) (115–122, 125–127, 133, 137, 138, 140, 145–148), sensitivity (proportion of true positives over true positives and false negatives) and specificity (proportion of true negatives over true negatives and false positives) (115, 117–120, 123, 125–127, 134–139, 142, 145, 147). Area under curve (AUC) values were reported in 16 papers (115, 120–122, 124–126, 130–132, 138–142, 146), with raw receiver operating characteristics (ROC) presented in 9 papers (115, 120, 122, 124–126, 132, 138, 139). 28 studies derived classifier performance by comparing metrics from an ADHD group against a control group, with Healthy Controls used in 27 studies. 2 studies additionally included participants with Schizophrenia (115) and a range of psychiatric disorders (124), whilst another sought to classify participants into those diagnosed with ADHD or with Autism Spectrum Disorder (138).

Together, these studies generally followed a similar approach wherein DHTs are deployed to augment established diagnostic procedures. This was achieved by leveraging behavioral indices, neurophysiological measures, or both, to support classifier-based discrimination between ADHD and comparison groups.

A further 19 papers also proposed a DHT specifically for diagnostic or screening purposes (Supplementary Table 8). In this group, 17 proposed classifications based on unimodal data, most commonly through EEG in 7 (151–157), MRI in 3 (158–160), CPTs in 3 (161–163), fNIRS in 2 (164, 165), and eye tracking (166) or through psychometric data (167) in 1 each. Fewer papers combined data in multimodal approaches, with 1 combining fNIRS with eye-tracking and CPT data (168), and 1 combining MRI with neuropsychological self-report measures (169).

Accuracy was again the most reported outcome measure in 12 cases (151, 152, 154, 155, 157–160, 164, 166, 167, 169), with 8 further specifying the sensitivity and specificity (151, 153, 154, 157, 159, 160, 162, 168), 7 including an AUC metric (151, 153, 156, 162, 163, 165, 168), and 4 showing the raw ROC (151, 163, 165, 168). All papers included HCs as a comparator group, with 1 study additionally instructing a group to “feign” ADHD to investigate a screening technology that would be sensitive to malingering (162).

Overall, these papers all use similar classification approaches as to those within Drive Clinical Management, and are evaluated in a similar manner. However, these examples differ in that the DHT is not situated as an aid to diagnosis, or adjunct to traditional assessment pathways. Instead, they propose that the DHTs would provide an automated output to facilitate immediate or near-term diagnosis.

A total of 9 papers were classified as Inform Clinical Management (Supplementary Table 9). DHTs in this selection are defined by their purpose of gathering information that is transferred to a third-party to inform clinical decision-making. 5 papers (170–174) proposed the use of tracking tools for longitudinal monitoring of ADHD symptoms, 2 used either SMS (175) or an app (176) to track and improve medication adherence, 1 proposed a chatbot interface for self-screening (177), and 1 implemented a remote smoking monitoring system (178).

Two of the symptom-tracking technologies were early stage DHTs evaluated qualitatively with Backer et al. (170), and Patrickson et al. (171), both recruiting clinicians and end-users to iteratively develop and test app content. Conversely, Surman et al. (172), Sankesara et al. (173), and Ware et al. (174), all pursued a similar approach, wherein they deploy their technologies to collect real-world data for longitudinal symptom monitoring. Outside of symptom monitoring, both Biederman et al. (175), and Carvalho et al. (176), evaluated the effectiveness of their DHTs by examining medication possession rates, whilst the smoking monitoring (178) and chatbot tools (177) were deployed to small pilot cohorts with primary outcomes relating to feasibility.

A further 5 papers were categorized as Promoting Good Health (Supplementary Table 10). Of these, 3 evaluated a form of mobile app that provided resources for concerted information provision (179–181). These examples differed from the psychoeducational content of DHTs classified in Treat Specific Condition given that they focused on improving patients' knowledge of ADHD through non-personalized advice, rather than directly addressing ADHD symptoms. Seery et al. (179), evaluate co-produced psychoeducational content presented through graphical and text interfaces, whilst Luiu et al. (180), investigated user preferences for a prospective app focused on providing psychoeducation for specific dimensions of ADHD centered around the Precaution Adoption Process Model (PAPM) of behavior change. Jang et al. (181), evaluated a form of interactive content delivery through a chatbot, Todaki, that was designed to support the user in learning self-help skills to manage their condition. Outside of psychoeducational DHTs, Lindstedt et al. (182), report on the implementation of a suite of cognitive assistive technologies that were deployed to support daily activities, and Store et al. (183), evaluate the effectiveness of a soft-robotic sleep companion for improving sleep quality.

All studies incorporate investigation into usability and perceived satisfaction or helpfulness, while Lindstedt et al. (182), further examined the impact of assistive technologies on Quality of Life and Store et al. (183), report on changes in sleep measures derived from the Insomnia Severity Index and actigraphy data.

One paper was classified as an evaluation of a DHT for Communicate about Health and Care (Supplementary Table 11). In this paper Adamou et al. (184), surveyed 117 end-users on their experiences of remote diagnostic assessment that was introduced into the specialist adult ADHD and Autism Service as part of the COVID-19 pandemic. They adapted the Telehealth Usability Questionnaire to investigate whether users found remote telecommunication during assessment useful and satisfactory. They identify possible differences in preference across gender, with females more willing to continue conducting remote assessment. They conclude that for ADHD, remote assessments may be feasible given the high levels of reported satisfaction. However, they note in their study, future work should focus on understanding user satisfaction once the patient has progressed further down the clinical pathway. This limitation was due to their study being cross-sectional in design and only surveying participants prior to them receiving their diagnostic outcomes.

For analyzing trends in RoB, we present findings from all study designs for Treat Specific Condition, whilst focusing on OCS studies for Drive Clinical Management and Diagnose a Specific Condition. We place less emphasis on describing trends in bias for Promoting Good Health, Inform Clinical Management, and Communicate about Health and Care, given the lower number of examples within each study design. All RoB data is available to review in Supplementary Tables 12–28.

Across the 63 papers classified as Treat a Specific Condition, our RoB analysis identified several consistent sources of bias. As shown in Figure 4, a lack of participant and intervention provider blinding was most prevalent, either not being implemented or reported in 25 (65.79%) CIV studies. Outcome assessor blinding was similarly uncommon (given the reliance on self-reported outcome measures) with a further 24 (63.15%) CIV, 5 OCS (71.42%) and 13 PPNC (76.47%) papers not including or reporting on using blinded outcome assessors. With regards to sample size, 24 CIV (63.15%) papers did not include power analyses to demonstrate at least 80% power, whilst 12 PPNC (70.58%) papers did not include sufficient justification or analysis of their sample size. For study adherence, 10 CIV (26.32%) papers reported a low level of adherence to the intervention, whilst a further 13 (34.21%) CIV papers did not report on investigations into adherence and 10 PPNC (58.82%) studies had a loss to follow up greater than 20%. Taken together, the high levels of non-blinding may undermine the internal validity of causal claims, whilst the lack of power analyses and investigation into adherence introduces uncertainty whether the observed effects represent true therapeutic benefit that translates into real-world scenarios.

In contrast to Treat a Specific Condition, DHTs which were classified as Drive Clinical Management were predominantly observational in design, with 34 OCS studies and 1 PPNC and 1 CAS study. Within the OCS group, shown in Figure 5, sample size was again a consistent source of bias, with only 3 papers (8.82%) including a justification of sample size or description of power (130, 132, 136). In addition, only 6 studies (17.65%) included adjustment for key confounding variables (124, 128, 130, 132, 134, 135), and 4 (11.76%) included repeated measures of the exposure (i.e., use of the DHT) (120, 126, 136, 139). These DHTs leverage an association between participant features and their ADHD diagnosis, yet these sources of bias may undermine the validity of any association by either not investigating the statistical power of these associations, other explanatory confounding variables, or any temporal instability in the features used for classification. Conversely, the collection's strengths were in specifying the research objectives, study population, and population sources.

Papers within Diagnose a Specific Condition were similarly observational by design, with 18 papers classified as OCS studies and 1 as a CIV study. As with Drive Clinical Management, the most common sources of bias across the OCS papers, shown in Figure 6, included a lack of sample size justification in all 18 papers, a lack of adjustment for confounding in 14 (77.78%) papers, and not measuring the exposure more than once in 15 (83.33%) papers. In comparison to Drive Clinical Management, a greater proportion of papers were less clear on the population sources and inclusion criteria. Various examples did not describe how an ADHD diagnosis was established in participants in sufficient detail [e.g., (151, 158–160, 162, 166)] or being defined through cut-offs on screening measures (167) in comparison to a full, valid, diagnosis.

The remaining papers classified as Inform Clinical Management and Promoting Good Health were each a mix of study designs, whilst the 1 paper classified as Communicate about Health and Care was a CAS study. Common sources of bias were similar to the preceding categories, with a lack of assessor blinding and power analysis most prevalent. Given the low number of papers in each category, caution should be applied in interpreting trends in bias for these categories. However, key areas of bias appear consistent across NICE categories wherein the technology is premised on deployment to and use by an ADHD patient as an end-user.

In this review, we identified 133 papers, and using the NICE classification framework, charted DHTs to identify a concentration of technologies that predominantly aim to treat a specific condition or support/automate diagnosis for adults with ADHD. Our findings present a high proportion of papers focusing on usability, feasibility, and reporting on early-stage trial results. Through our quality assessment, we identified gaps in blinding, adherence, and deployment of DHTs longitudinally. Here, we highlight key priorities for the next wave of research, which should further consider how to generate suitable evidence that supports DHT implementation as valid clinical tools.

Throughout this review, we have highlighted a range of gaps within the evidence base and presented suggestions to improve the development and evaluation of DHTs within each of the NICE classifications. In this section, we highlight two key research priorities to strengthen the development and evidence base for DHTs across the NICE classifications.

There are several limitations in our review to note. Firstly, as a Scoping Review, we do not draw conclusions on the overall efficacy of any of the technologies presented here. As such, the review does not make recommendations or provide implications for direct clinical practice or policy. Rather, we identify and map the current research to provide an up-to-date view of the environment surrounding DHTs for adults with ADHD. Through our analysis, we identify gaps and concerns, which future research should address to provide a robust evidence base that enables implementation of these technologies into a clinical setting.

Secondly, the objective for our research question, and resulting search strategy, was to provide a breadth of DHTs to assess, rather than an in-depth focus on any single category. Previous reviews exist which provide a concerted focus on DHT categories, e.g., (37, 39, 201). However, this review is the first to look more broadly across all stages of the care pathway, enabling us to identify areas where DHTs are more frequently being developed. We justify our focus on adulthood ADHD given that it is currently managed distinctly from pediatric services, and the technologies are more suitably evaluated against the challenges that are faced in adulthood.

Thirdly, we have used three key tools to set the remit of the review. We define a “digital health technology” as a digital tool which provides a health benefit (40), yet there may be examples in the literature which we did not include if their health benefits, direct or otherwise, were not stated or sufficiently described. Next, we used the NICE DHT Framework for classification, which is based on the intended use of a technology in a clinical setting. Despite this, many papers describe results from an applied research setting and do not contextualize how their DHT is situated within a clinical setting. For the purposes of this review, we derived a primary category based on how the technologies were described by the authors, but these categories should not be considered definitive and may differ if implemented clinically. Lastly, we used the NIH Quality Assessment tools to undertake our RoB analysis, which lack validation. This choice is justified considering our objective was not to evaluate efficacy but instead identify trends across a heterogeneous set of study designs. The use of the NIH tools, which have tailored options for each type of study design under a single framework supported this. Future works which systematically examine efficacy may benefit from confirming these trends in particular subcategories using validated tools (49–51).

DHTs offer the potential to provide scalable and efficient solutions that improve access to suitable healthcare for adults with ADHD. This review identified numerous examples of existing research into technologies that could be situated throughout the care pathway to support diagnosis, treatment, monitoring, and self-management of ADHD in adulthood. Our review has highlighted that most DHTs are being developed to support clinical care, being categorized within treatment, diagnosis, and clinical management. However, the outcome measures and sources of bias we identify suggest that these DHTs could be improved by developing and evaluating them with a greater consideration of how they will be adopted into clinical settings and how they will be used by people with ADHD in real-world settings.

For DHTs to realize their potential, we highlight the need for future research that draws more broadly on complementary and multidisciplinary research methods that effectively capture the considerations of all individuals involved in the development and delivery of the technologies. This needs to be combined with robust investigations into participant experiences, and direct participant involvement in the design process. We highlight various frameworks which promote the view that interventions should extend their investigations beyond efficacy and consider the interactions between the technology and the people and systems that currently operate the care pathway.