CoChat
Sign Up
Cited by 0Medication Adherence and ComplianceMobile Health and mHealth ApplicationsArtificial Intelligence in Healthcare and Education

A scoping review of artificial intelligence-enabled wearables for medication adherence

Open access
Chen, Ying, Zhao, He, Xu, Run, Li, Yiyang, Cai, Shuang, Hu, Yingying·2026·npj Digital Medicine
DOI 10.1038/s41746-026-03246-5
Open this paper in CoChat — free accountOpen PDF

Abstract

Medication non-adherence remains a major global public health challenge with substantial clinical and economic burden. Artificial intelligence (AI)-enabled wearables may offer new opportunities to support medication adherence management. This scoping review assesses the current evidence on AI-enabled wearables for medication adherence, focusing on their applications, performance, and clinical translation. We screened three databases for literature published up to September 18, 2025. A total of 6102 records were identified, and 17 studies met the inclusion criteria. Current applications were concentrated on medication-taking behavior and ingestion event recognition, with fewer studies addressing longitudinal medication adherence assessment or non-adherence risk prediction. No included study evaluated AI-enabled wearables for personalized intervention. Most studies focused on technology development and internal validation. Although reported performance was generally favorable in controlled settings, evidence remains limited for external validation, real-world effectiveness, clinical outcomes, and integration into routine care. These findings suggest that AI-enabled wearables are technically feasible for medication adherence monitoring, but their clinical value remains insufficiently established. Future research should move beyond recognition toward prospective risk prediction, personalized intervention, and clinically integrated decision support.

AI snapshot

Written from the paper’s full text.

TL;DR

This PRISMA-ScR scoping review found that AI-enabled wearables for medication adherence are mostly used for medication-taking behavior and ingestion event recognition, with favorable internal validation performance but little external validation or clinical translation. Across 17 included studies, no study evaluated personalized intervention, and evidence for real-world effectiveness, clinical outcomes, and routine care integration remains limited.

Key findings

  • The search identified 6102 records; after deduplication, 4513 records were screened, and 17 studies met inclusion criteria. Studies spanned 2016–2025, with 14 published since 2021.
  • Applications concentrated on medication-taking behavior and ingestion event recognition; only two studies addressed adherence assessment/pattern characterization, only one addressed non-adherence risk prediction, and no included study evaluated AI-enabled wearables for personalized intervention.
  • Among 13 studies with internal validation, reported performance was generally favorable: SFOA-Bi-LSTM achieved mean accuracy 98.90%, an LSTM-based artificial neural network achieved 93.6% accuracy for medication intake event detection, and CNN-LSTM achieved a median F1-score of 90.5% for multi-gesture classification. However, external validation on an independent dataset was not performed in any of the 17 studies.
  • Only two studies reported clinical outcomes: a pilot randomized controlled trial showed smartwatch intervention improved medication adherence (63.3% vs 43.2%) and symptom control in allergic rhinitis, and another study found smartwatch-measured adherence correlated with higher serum vitamin D levels (r = 0.45).

Methods

  • Scoping review conducted in accordance with PRISMA-ScR; searched PubMed, Web of Science, and IEEE Xplore up to 18 September 2025 for English peer-reviewed original research on AI-enabled wearables for medication adherence. Reviews, study protocols, and wearables lacking AI capabilities were excluded.
  • Two independent reviewers screened titles/abstracts and full texts, resolving discrepancies by consensus or a third reviewer. Data were extracted using a standardized Excel spreadsheet covering study characteristics, model details, application scenarios, outcomes, key findings, translation, ethics, and economics.
  • A formal quality assessment was not performed because the review aimed to outline the research landscape and because methodological and outcome heterogeneity precluded a uniform quality assessment tool.

Limitations

  • Stated in the paper: included studies had small, homogeneous sample sizes; restricted ecological validity from controlled environments or simulated gestures; insufficient external validation and long-term assessment; and concerns about technology reliability. None conducted formal cost-benefit analyses, and broader ethical, privacy, and equity issues were insufficiently addressed.
  • Stated in the paper: participant characteristics were limited, including younger, healthier, right-handed participants and limited racial/ethnic diversity; groups with greater adherence challenges, such as older adults with multimorbidity, people with cognitive impairment, and lower socioeconomic groups, remain underrepresented.
  • Evident from the text: as a scoping review without formal quality assessment, the synthesis cannot weigh study quality; the included evidence is heterogeneous, mostly technology development/validation, and lacks external validation, real-world effectiveness, clinical outcome assessment, and integration into routine care.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41746-026-03246-5

Cites

67 of 67 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 22 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    Most included studies focused on technology development and internal validation.

    “Most studies focused on technology development and internal validation.”
    • Also stated in: § Study characteristics · plain
      “Methodologically, most studies (n = 14) focused on technology development and validation”
    Quote in the paperSubstantial claim
  • Findingreasoning

    There is a lack of reports concerning extensive integration with hospital information systems.

    “there is a lack of reports concerning extensive integration with hospital information systems.”
    Quote in the paperNot a substantial claim
  • Findingown data

    No included study evaluated AI-enabled wearables for personalized intervention.

    “No included study evaluated AI-enabled wearables for personalized intervention.”
    Quote in the paperSubstantial claim
  • Findingown data

    Current applications of AI-enabled wearables for medication adherence were concentrated on medication-taking behavior and ingestion event recognition.

    “Current applications were concentrated on medication-taking behavior and ingestion event recognition, with fewer studies addressing longitudinal medication adherence assessment or non-adherence risk prediction.”
    • Also stated in: § Application and validation of AI-enabled wearables for medication adherence · boosted
      “Current research on AI-enabled wearables for medication adherence demonstrated a pronounced focus on medication-taking behavior and ingestion event recognition”
    Quote in the paperSubstantial claim
  • Findingcitation

    Only a limited number of studies describe specific system prototypes.

    “Only a limited number of studies describe specific system prototypes”
    Quote in the paperSubstantial claim
  • May combine two claims. 27
  • Classified as probably not a substantial claim. 10

  • May combine two claims.

    Findingreasoning

    Most studies have examined clinical integration, but few have validated it through practical implementation.

    “While most studies have examined clinical integration, few have validated it through practical implementation.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingcitation

    A pilot randomized controlled trial showed that smartwatch interventions significantly enhanced medication adherence rates and symptom control in patients with allergic rhinitis.

    “A pilot randomized controlled trial showed that smartwatch interventions significantly enhanced medication adherence rates and symptom control in patients with allergic rhinitis”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingreasoning

    Evidence remains limited for external validation, real-world effectiveness, clinical outcomes, and integration into routine care.

    “Although reported performance was generally favorable in controlled settings, evidence remains limited for external validation, real-world effectiveness, clinical outcomes, and integration into routine care.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingcitation

    Fourteen included studies received ethics committee approval and obtained informed consent from participants.

    “Fourteen research received ethics committee approval and obtained informed consent from participants, meeting basic ethical requirements”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingnone given

    Current evidence remains concentrated on medication-taking behavior and ingestion event recognition, whereas longitudinal adherence characterization, non-adherence risk prediction, personalized intervention, and clinical decision support remain less developed.

    “Within the proposed workflow, current evidence remains concentrated on medication-taking behavior and ingestion event recognition, whereas longitudinal adherence characterization, non-adherence risk prediction, personalized intervention, and clinical decision support remain less developed.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    No included study directly applied LLMs to wearable-based medication adherence management, indicating an important gap for future research.

    “No included study directly applied LLMs to wearable-based medication adherence management, indicating an important gap for future research.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingnone given

    The reviewed studies relied predominantly on AI-based detection of medication-taking gestures, whereas only a few incorporated direct measures such as digital medication systems or biomarker assays.

    “The reviewed studies relied predominantly on AI-based detection of medication-taking gestures, whereas only a few incorporated direct measures such as digital medication systems or biomarker assays.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingreasoning

    Despite progress in the development of AI-enabled wearables, clinical translation remains limited, with little evidence of integration into healthcare systems or long-term clinical use.

    “Despite progress in the development of AI-enabled wearables, clinical translation remains limited, with little evidence of integration into healthcare systems or long-term clinical use.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingnone given

    Most studies reported algorithmic performance or gesture recognition accuracy, but few assessed whether these technologies improved clinically meaningful outcomes, such as symptom control, biomarker improvement, or reduced rehospitalization.

    “Most studies reported algorithmic performance or gesture recognition accuracy, but few assessed whether these technologies improved clinically meaningful outcomes, such as symptom control, biomarker improvement, or reduced rehospitalization.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingreasoning

    Most models were evaluated on internal datasets and performed well primarily under controlled conditions.

    “Most models were evaluated on internal datasets and performed well primarily under controlled conditions.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingnone given

    Several studies discussed remote monitoring, alerts, or decision-support concepts, while few demonstrated operational integration into clinical workflows.

    “Several studies discussed remote monitoring, alerts, or decision-support concepts, while few demonstrated operational integration into clinical workflows.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingnone given

    Study samples were also narrow and often composed of younger, healthier, and more technologically proficient participants.

    “Study samples were also narrow and often composed of younger, healthier, and more technologically proficient participants.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingreasoning

    The current evidence base remains primarily focused on adherence recognition and retrospective assessment, with limited exploration of non-adherence risk prediction.

    “The current evidence base remains primarily focused on adherence recognition and retrospective assessment, with limited exploration of non-adherence risk prediction.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Most studies reported ethics approval and informed consent, but broader issues related to privacy, surveillance, data security, and equitable access were often discussed only briefly.

    “Most studies reported ethics approval and informed consent, but broader issues related to privacy, surveillance, data security, and equitable access were often discussed only briefly.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Camera-based approaches can identify medication objects or tablet types with greater specificity, but also raise privacy and ethical concerns.

    “Camera-based approaches can identify medication objects or tablet types with greater specificity, but also raise privacy and ethical concerns, particularly in home-based or other sensitive clinical settings 44,45 .”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [44]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [45]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationreasoning

    This approach cannot reliably distinguish true medication ingestion from similar confounding behaviors, nor can it confirm whether the ingested item is the prescribed medication at the prescribed dose.

    “This approach cannot reliably distinguish true medication ingestion from similar confounding behaviors, nor can it confirm whether the ingested item is the prescribed medication at the prescribed dose.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Sensor-based inference cannot reliably distinguish true medication ingestion from similar confounding behaviors or confirm the prescribed medication and dose.

  • May combine two claims.

    Premisecitation

    Inertial measurement units are widely integrated into consumer smartwatches and wristbands, making them well suited for large-scale deployment.

    “Inertial measurement units are widely integrated into consumer smartwatches and wristbands, making them well suited for large-scale deployment 43 .”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [43]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationreasoning

    These findings suggest that medication adherence is still frequently inferred from the recognition of specific actions, although adherence also depends on timing, dose accuracy, persistence, intentionality, and clinical context.

    “These findings suggest that medication adherence is still frequently inferred from the recognition of specific actions, although adherence also depends on timing, dose accuracy, persistence, intentionality, and clinical context.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    Patients who are sleeping, exercising, driving, working, or experiencing symptoms may require different forms of adherence support.

    “Patients who are sleeping, exercising, driving, working, or experiencing symptoms may require different forms of adherence support.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Medication non-adherence may result from forgetfulness, misunderstanding of treatment regimens, treatment burden, adverse effects, low perceived necessity, emotional distress, cognitive impairment, or limited social support.

    “Medication non-adherence may result from forgetfulness, misunderstanding of treatment regimens, treatment burden, adverse effects, low perceived necessity, emotional distress, cognitive impairment, or limited social support 51 .”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [51]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation

    Real-world medication-taking behavior is more variable and may be affected by disease-related motor symptoms, such as tremor in Parkinson's disease.

    “whereas real-world medication-taking behavior is more variable and may be affected by disease-related motor symptoms, such as tremor in Parkinson's disease 58 .”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [58]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationcitation

    Although preliminary, these findings suggest that changes in daily activity patterns may precede adherence deterioration and could serve as early indicators of emerging non-adherence risk.

    “Although preliminary, these findings suggest that changes in daily activity patterns may precede adherence deterioration and could serve as early indicators of emerging non-adherence risk.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    The clinical value of AI-enabled wearables will depend on whether their outputs can support decisions without increasing workload or alert fatigue.

    “The clinical value of AI-enabled wearables will depend on whether their outputs can support decisions without increasing workload or alert fatigue.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    Wearables enable longitudinal adherence assessment by capturing dosing patterns and behavioral context, shifting the focus from isolated event detection to adherence as a dynamic behavioral phenotype.

    “Wearables enable longitudinal adherence assessment by capturing dosing patterns and behavioral context, shifting the focus from isolated event detection to adherence as a dynamic behavioral phenotype.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Medication adherence is influenced by multiple psychological, social, cognitive, and environmental factors.

    “Medication adherence is influenced by multiple psychological, social, cognitive, and environmental factors 49,50 .”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [49]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [50]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationreasoning

    Personalized intervention remains one of the least developed but most clinically relevant applications.

    “Personalized intervention remains one of the least developed but most clinically relevant applications.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    This domain gap may substantially reduce model performance in clinical populations, and the reliability of these systems in routine practice remains uncertain.

    “This domain gap may substantially reduce model performance in clinical populations, and the reliability of these systems in routine practice remains uncertain.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Real-world medication-taking behavior is more variable and less scripted than laboratory settings.

  • May combine two claims.

    Interpretationreasoning

    Current AI-enabled wearables are largely limited to event-triggered alerts, which are unlikely to address the heterogeneous causes of non-adherence.

    “Current AI-enabled wearables are largely limited to event-triggered alerts, which are unlikely to address the heterogeneous causes of non-adherence.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    Two independent reviewers screened titles and abstracts based on predefined criteria.

    “Two independent reviewers screened titles and abstracts based on the predefined criteria.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    Full-text assessments were conducted independently by two reviewers.

    “Full-text assessments were also conducted independently by two reviewers.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    For AI-enabled wearables to become clinically useful, adherence data must be integrated into clinical practice rather than remain confined to standalone monitoring systems.

    “For AIenabled wearables to become clinically useful, adherence data must be integrated into clinical practice rather than remain confined to standalone monitoring systems.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    Any discrepancies at either screening stage were resolved through consensus or by consulting a third reviewer.

    “Any discrepancies at either stage were resolved through consensus or by consulting a third reviewer.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing). May combine two claims.

    Premisecitation

    Human-in-the-loop approaches may help clarify responsibility for alerts and reduce inappropriate clinical action, provided they are designed to avoid additional workload and alert fatigue.

    “Human-in-the-loop approaches may help clarify responsibility for alerts and reduce inappropriate clinical action, provided they are designed to avoid additional workload and alert fatigue 67 .”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [67]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductcitation

    This scoping review was conducted in accordance with the PRISMA-ScR guidelines.

    “This scoping review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) guidelines 69,70 .”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [69]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [70]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation

    Health limitations, unfamiliarity with technology, and low digital literacy may further constrain uptake.

    “Health limitations, unfamiliarity with technology, and low digital literacy may further constrain uptake 60,61 .”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [60]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [61]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

35 of 35

  • The paper relies on this without support.

    Real-world medication-taking behavior is more variable and less scripted than laboratory settings.

    “In real-world settings, medication-taking behavior is particularly less scripted and more variable than in laboratory environments.”

  • The paper relies on this without support.

    Recognition accuracy is challenged by motion artifacts, poor contact, and environmental interference.

    “The accuracy of recognition is challenged by issues such as motion artifacts, poor contact, and environmental interference.”

  • The paper relies on this without support.

    Sensor-based inference cannot reliably distinguish true medication ingestion from similar confounding behaviors or confirm the prescribed medication and dose.

    “This approach cannot reliably distinguish true medication ingestion from similar confounding behaviors, nor can it confirm whether the ingested item is the prescribed medication at the prescribed dose.”

  • The paper relies on this without support.

    Different wearable sensing modalities inherently present trade-offs in accuracy, scalability, privacy, and acceptability.

    “Different wearable sensing modalities present different trade-offs in accuracy, scalability, privacy, and acceptability.”

  • The paper relies on this without support.

    Inertial measurement units are low-cost, noninvasive, and relatively privacy-preserving, but require user-specific calibration.

    “They are low-cost, noninvasive, and relatively privacy-preserving, but require user-specific calibration.”

  • The paper relies on this without support.

    The included studies are of sufficient methodological quality to support conclusions about the potential of AI-enabled wearables.

    “A formal quality assessment of the studies was not performed, since the main aim of this review is to outline the research landscape, highlight trends, and pinpoint knowledge gaps.”
6 of 6

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Publication-year count is off by one

    Minor

    Table 1 lists 13 studies published from 2021 through 2025, not 14.

1 of 1

Cites

Flagged first, then by citations

Research works printed but not linked: 3

  1. Gutierrez-Abejon, E., Pedrosa-Naudin, M. A., Fernandez-Lazaro, D. & Alvarez, F. J. Medication economic burden of antidepressant non- adherence in Spain. Front Pharm. 14, 1266034 (2023).
  2. Odhiambo, C. O. et al. Detecting medication-taking gestures using machine learning and accelerometer data collected via smartwatch technology: instrument validation study. JMIR Hum. Factors 10, e42714 (2023). https://doi.org/10.1038/s41746-026-03246-5
  3. Veroniki, A. A. et al. Update to the PRISMA guidelines for network meta-analyses and scoping reviews and development of guidelines for rapid reviews: a scoping review protocol. JBI Evid. Synth. 23, 517-526 (2025). https://doi.org/10.1038/s41746-026-03246-5

Cited by

Still being checked.

Authors
Chen, Y., Zhao, H., Xu, R., Li, Y., Cai, S., & Hu, Y.
Affiliations
  • First Hospital of China Medical University (CN)
  • Affiliated Zhongshan Hospital of Dalian University (CN)
  • Harbin Medical University (CN)
Corresponding author
  • Cai, Shuang
  • Hu, Yingying
ORCID
Venue
npj Digital Medicine
Publisher
Springer Science and Business Media LLC
Published
2026-09-14
ISSN
2398-6352 (ISSN-L)
DOI
10.1038/s41746-026-03246-5
OpenAlex
W7213032452
Semantic Scholar
c4c6dd0497986b02f467995137d044e186e89440
Cited by
0
References
67
FWCI
0.00
Type
journal article
Language
en
Open access
Licence
Locations
npj Digital Medicine (publishedVersion, cc-by)
Topics
  • Health Sciences › Medicine › Family Practice › Medication Adherence and Compliance
  • Health Sciences › Health Professions › General Health Professions › Mobile Health and mHealth Applications
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
Keywords
artificial intelligence, wearable devices, medication adherence, medication non-adherence, medication-taking behavior, personalized interventions, clinical decision support, external validation, real-world effectiveness

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-09.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-09.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review