The role of AI-integrated wearables in predictive healthcare: a scoping review

Introduction

Background

The development of artificial intelligence (AI) coupled with wearable technology has emerged as a transformative force in healthcare, shifting the paradigm from reactive to predictive and preventive care. With the use of predictive analytics, AI can help staff identify at-risk patients, predict chronic disease course, and improve prediction of outcomes in a variety of clinical settings (1,2). These applications are highly valuable in the real world, as machine learning diagnostics accuracies have surpassed 99% in a variety of chronic kidney disease settings (1) and predictive performance using electronic health records (EHRs) is strong (3).

Unlike earlier AI applications that relied primarily on EHRs or imaging datasets, AI-integrated wearables provide continuous, real-time physiological data directly from individuals, enhancing the granularity and timeliness of health insights.

Simultaneously, wearables have transitioned from basic fitness activity monitors to sensor-rich, AI-enhanced systems capable of continuous monitoring of parameters such as heart rate, respiration, sleep quality, oxygen saturation, and electrocardiogram (ECG) signals (4,5). The use of such wearables, with integrated algorithms, can provide clinicians with health-related information, detect anomalies, and predict diseases such as cardiovascular disease, depression, and sleep apnea (6-8). Although concerns exist regarding data privacy and variability in wearables (9), AI-enabled wearables have the potential to transform predictive healthcare by enabling remote, patient-driven, and proactive care delivery across various medical areas. Nevertheless, AI-enabled wearables stand at the intersection of personalized medicine and digital health, offering a pathway toward scalable, continuous, and equitable predictive healthcare, particularly as 5G connectivity, Internet of Things (IoT) infrastructure, and explainable AI models mature.

For the purposes of this review, “wearable” is defined as body-worn or skin-contact sensor-enabled devices that continuously or intermittently acquire physiological, behavioral, or environmental data for health-related use (10). This includes consumer-grade (e.g., fitness bands, smartwatches, activity bands) and clinical-grade (e.g., continuous glucose monitors, ambulatory ECG/Holter monitors, and pulse oximeters) instruments that have AI or machine learning algorithms to analyze acquired data. Experimental biosensors and research-grade prototypes that exhibited predictive or diagnostic capabilities in health care settings were included. Stationary sensors and non-worn IoT devices were excluded.

Rationale and knowledge gap

AI-enabled wearable technologies are gaining popularity for their potential to revolutionize healthcare through early detection, real-time monitoring, and personalized interventions based on individual needs (11). However, despite rapid technological growth, the existing evidence base remains fragmented and uneven in methodological quality. Prior research has explored AI-integrated wearables in specific domains such as cardiovascular health, oncology, mental health, and sleep medicine (6,7,12), yet these studies often differ in design, outcomes, and reporting frameworks, making cross-domain comparison difficult.

Moreover, much of the available literature consists of conceptual or narrative reviews, with limited empirical validation of device performance or algorithmic transparency (13). The lack of standardized evaluation protocols, population diversity, and reproducible performance metrics further limits the generalizability of findings. There currently lacks a comprehensive scoping review examining the role of these devices in predictive healthcare and consolidating the dispersed evidence into a unified understanding of their diagnostic, monitoring, and personalized care potential.

Additionally, several open questions remain that were not addressed, such as data quality, algorithm bias, interoperability, privacy regulation, and scalability in low-resource settings, that hinder large-scale clinical integration.

This review aims to address these gaps by systematically mapping existing research, identifying dominant themes, methodological trends, and barriers, and building foundational knowledge on how AI-enabled wearables contribute to predictive healthcare delivery.

Objectives

The goal of this study was to systematically map and evaluate how AI, predictive analytics, and wearable devices are used to support early disease detection, more accurate diagnosis, ongoing monitoring, and personalized treatment in different areas of healthcare.

Specifically, this review aims to:

• Identify and analyze existing evidence on how AI and predictive analytics can be applied for early diagnosis and decision-making in healthcare delivery;
• Assess the benefits, limitations, and implementation challenges of AI-enabled wearables for remote monitoring and chronic-disease management;
• Synthesize findings on the integration of machine learning and deep learning algorithms with wearable-derived data to improve predictive accuracy and real-time health insights;
• Map the application of AI-integrated wearables across key medical fields—including cardiology, oncology, endocrinology, neurology, mental health, and sleep medicine—to highlight thematic trends, methodological gaps, and future research priorities.

I present this article in accordance with the PRISMA-ScR reporting checklist (available at https://jmai.amegroups.com/article/view/10.21037/jmai-2025-170/rc).

Methods

Conceptual framework development

A conceptual framework was developed through a literature analysis and thematic mapping of key elements of AI-integrated wearables in predictive healthcare, including real-time monitoring, interpretation, and clinical applications. The relationships between technological capabilities, clinical applications, and system barriers were investigated for their impact on healthcare delivery. The framework builds on the outputs of a scoping review and bibliometric analysis published in this Special Issue and focuses on the early diagnosis and personalized care enabled by wearables and the associated challenges of data privacy, interoperability, and clinical engagement (4,6,7,10).

Evaluation of AI-integrated wearables in predictive healthcare: information sources

Evaluation methodology was based on PRISMA 2020 standards (12) to perform a scoping review in the most prominent academic databases, such as Google Scholar, PubMed, JSTOR, and ProQuest. Boolean operators and relevant keywords like “AI-integrated wearables”, “predictive healthcare”, “remote monitoring”, “personalized medicine”, and “early diagnosis” were applied to extract studies published between 1 January 2016 and 19 July 2025 (the year 2016 was chosen as the start year as this is when AI-enabled wearable devices and predictive analytics started to appear in clinical applications) updated in 21 November 2025. The searches were conducted between 16 July 2025 and 19 July 2025. The review was sensitive to original research, clinical trials, and recent evaluations of technologies playing the role of wearable AI devices to enhance the accuracy of diagnosis, chronic disease management, and health outcomes across various healthcare environments.

Search strategy development

The search strategy was developed through iterative discussion with domain experts and refined during pilot testing for completeness and reproducibility (13). The detailed search formula could be found in Table S1.

Equivalent keywords and subject headings were modified for each database. Database-specific rules for Boolean operators, truncations, and quotation marks were applied. The search strategy was peer-reviewed for completeness using the Peer Review of Electronic Search Strategies (PRESS) checklist.

The reference lists of included articles and related reviews were screened for eligible studies.

Rationale for search timeframe [2016–2025]

By 2016, deep learning architectures such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), and long short-term memory (LSTM) models had become central to processing complex physiological data from wearable sensors (14). Advances in miniaturized, non-invasive sensors enabled continuous monitoring of vital signs (15). Researchers began integrating these architectures for real-time health tracking (16), predictive analytics, and adaptive health systems that improved early disease detection, anomaly recognition, and personalized interventions for chronic disease management (14,16). Earlier literature lacked such AI-driven predictive capabilities.

Eligibility criteria [Population-Concept-Context (PCC) framework]

The PCC framework (17) was used to define inclusion and exclusion criteria for this scoping review. Therefore, studies in which human participants were involved in healthcare or community settings, whether healthy individuals or patients with acute or chronic conditions, provided that wearable devices were used for predictive, diagnostic, or monitoring purposes. The studied concept was AI-enabled wearable technology, which we defined as body-worn or skin-contact sensors that acquire continuous or intermittent physiological, behavioral, or environmental data streams and apply AI, machine learning, or deep learning with the purpose of enabling predictive healthcare applications. Eligible studies had to report outcomes such as prediction, early detection, monitoring of disease progression, or personalized intervention, which are derived from a data stream acquired from wearable devices. Furthermore, studies that reported performance outputs such as accuracy, sensitivity, specificity, or area under the curve (AUC), as well as studies that addressed issues such as implementation feasibility, ethical, or regulatory implications, were included. Conversely, studies that dealt exclusively with imaging-based AI, EHRs, or laboratory data analytics, with no wearable component whatsoever, were excluded. This is because these modalities fall under a different category of methodological, ethical, and regulatory challenges, which are beyond the scope of this review.

For eligibility criteria, the inclusion and exclusion criteria were defined according to the PCC framework recommended for scoping reviews:

• Inclusion criteria:
  • Peer-reviewed studies published in English between January 2016 and July 2025;
  • Studies involving human participants or human-related datasets;
  • Research addressing AI-integrated wearable devices designed for predictive, diagnostic, or monitoring purposes;
  • Studies reporting at least one relevant outcome: diagnostic accuracy (sensitivity, specificity, AUC), clinical utility, disease monitoring, or integration challenges;
  • Both empirical studies (quantitative, qualitative, mixed-methods) and review articles were included to capture the state of evidence and conceptual evolution in this emerging field.
• Exclusion criteria:
  • Non-peer-reviewed sources (conference abstracts, preprints, theses, editorials, commentaries);
  • Studies focusing solely on non-AI wearables (fitness tracking without predictive analytics);
  • Articles using only imaging or EHR-based AI without a wearable data component;
  • Animal studies, simulation-only work, or studies lacking sufficient methodological detail to assess AI integration;
  • Non-English publications or duplicates.

Article selection process

A total of 5,031 documents were initially identified from different academic databases. Two reviewers independently screened titles/abstracts and full texts against prespecified criteria. Conflicts were resolved by discussion. Inter-rater agreement was tracked. During the first screening process, titles and keywords were reviewed for relevance to the topic. After removing 3,531 duplicates and 1,200 irrelevant records, 300 articles remained for review. The abstracts of these studies were reviewed independently by both reviewers to determine whether they met the study inclusion criteria. Articles not meeting the set standards were excluded during the final screening process. Discrepancies were resolved through discussion between the two reviewers. No automation tools were used during the screening or selection process. Ultimately, 51 articles were chosen and included in the qualitative synthesis (Figure 1), after stepwise selection from the remaining articles.

Figure 1 PRISMA-ScR flow diagram illustrating the identification, screening, eligibility, and inclusion of studies. A total of 51 studies met the predefined inclusion criteria. No automation tools were used for screening or selection. The flow follows the PRISMA 2020 and PRISMA-ScR reporting standards.

Risk of bias evaluation

Two reviewers independently assessed the methodological quality of included studies using a structured critical appraisal form adapted from the Joanna Briggs Institute (JBI) checklist for mixed study types (18). Discrepancies were resolved through discussion, and a third senior reviewer was consulted in case of disagreement. No automation or AI-based tools were used in this assessment; all evaluations were performed manually in Microsoft Excel. Given that this work is a scoping review, the aim of quality assessment was to describe the methodological rigor and reporting completeness of included studies rather than exclude studies based on quality scores.

Protocol deviation and clarification

Although the initial protocol prospectively stated that only randomized study types would be included, this criterion was later refined during the screening stage to better align with the scoping review objective. Given the emerging and largely conceptual nature of AI-enabled wearable research, the literature predominantly comprised review, modeling, and observational designs rather than randomized controlled trials (RCTs). Consequently, non-randomized and review studies meeting the PCC framework were included to comprehensively map available evidence on predictive healthcare applications.

Data extraction and data analysis

To assess the role of AI-enabled wearables in predictive healthcare, we extracted data from the primary sources using a rigorous data extraction process. This included author details, research design, application domain, type of technology, study outcomes, and limitations using an Excel matrix. We applied a qualitative synthesis approach (19). For synthesis, studies were grouped, and thematic analysis was used to identify themes around clinical utility, accuracy of the diagnostics, and integration challenges, and content analysis was used to identify enablers and barriers, such as data privacy, algorithm opacity, and clinical uptake (20).

Two reviewers independently performed data extraction to ensure accuracy and consistency. Extracted information was cross-checked, and any discrepancies were resolved through discussion. No AI-assisted or automation tools were employed.

Outcomes sought

We sought outcomes related to (I) diagnostic accuracy (e.g., sensitivity, specificity, AUC); (II) clinical utility and patient outcomes; (III) effectiveness of continuous remote monitoring; (IV) chronic disease management outcomes; (V) integration challenges (e.g., workflow, ethical, regulatory challenges); and (VI) enabling factors (e.g., data quality, transparency of algorithms, interoperability).

Collection of results

We extracted all results relevant to each outcome domain (irrespective of measure, time point, or analysis). When multiple results were presented for the same outcome domain, we extracted primary outcomes as defined by the original study authors. Results of individual studies were tabulated using structured summary tables. In addition to outcomes, we also extracted intervention characteristics (such as type and function of AI-enabled wearable, intended clinical purpose), study characteristics (such as methodology used, year of publication), limitations, and future research. If information about a study was missing or unclear, we used available published data and did not impute values. Contacted authors when information was missing and considered critical. “Critical” information was operationally defined as any data element necessary to (I) determine study eligibility; (II) interpret diagnostic or predictive performance; or (III) assess methodological transparency. Specifically, this included missing details on sample size, data source, AI model type, validation method, or primary quantitative outcomes (e.g., AUC, accuracy, sensitivity, specificity). If we did not get a response, we coded data as “unclear” and contacted them to exclude from quantitative summaries as needed. For diagnostic outcomes, effect measures such as AUC, accuracy, sensitivity, and specificity were extracted or derived. For qualitative outcomes (e.g., integration barriers), thematic counts and narrative synthesis were used.

Because of the high variability in study design, population, outcome, and outcome measurement, and variation in reporting, formal quantification of heterogeneity was not undertaken. Sources of heterogeneity were instead explored narratively by discussing differences in study characteristics (e.g., study design, clinical context, type of AI-enabled wearable, geographical region) and synthesizing results within each synthesis domain (e.g., diagnostic accuracy, clinical utility, integration barriers) separately, when appropriate. Furthermore, we did not perform a formal certainty-of-evidence assessment. No formal sensitivity analyses were conducted because no quantitative meta-analysis was performed.

Data synthesis approach

Because the studies included spanned the methodological spectrum from experimental AI model development (e.g., ResNet101 architectures) to narrative and scoping reviews, we employed a narrative synthesis. Summary quantitative measures (AUC, accuracy, sensitivity, and specificity) were reported descriptively and interpreted within the methodological context of the study’s own design and validation framework. Given the heterogeneity of study design, population, devices, and outcome measures, no meta-analyses or statistical pooling were performed. Therefore, reported quantitative values (e.g., AUC >0.92 or accuracy >95%) refer to individual high-performing studies rather than aggregated estimates. Results were also organized thematically by clinical domain (e.g., cardiology, oncology, mental health) and by design (empirical vs. review).

Presentation of results

Summary results of individual studies were presented in tabulated summary tables. PRISMA 2020 flow diagram (Figure 1) was used to depict the study selection process. Citation frequency and temporal distribution were depicted graphically (Figures 2,3). Themes and subthemes for thematic synthesis were summarized narratively. Where relevant, domains were summarized in thematic maps to illustrate relationships between domains (e.g., diagnostic accuracy, barriers to and enablers of integration).

Figure 2 Annual distribution of selected publications on AI-integrated wearables in predictive healthcare from 2016 to 2025. The data shows a steady increase in publications, with a peak observed in 2024. AI, artificial intelligence.

Figure 3 Citation frequency of included studies on AI-integrated wearables in predictive healthcare. The figure highlights variations in citation counts, with certain publications receiving notably high attention in scholarly discourse. AI, artificial intelligence.

Registration

The registration number with PROSPERO is CRD420251117931.

Results

Study selection, eligibility screening, and rationale for exclusion

The initial phase of our literature search, conducted across a range of comprehensive databases including Google Scholar, PubMed, JSTOR, and ProQuest, appeared to yield an initial pool of approximately 5,031 records. Following the removal of a substantial number of duplicates [3,531] and an additional 1,200 records ostensibly outside the scope based on title and keyword review, the remaining 300 records were advanced for eligibility screening. From this particular interpretive perspective, a further 249 records were subsequently excluded based on title and abstract review. A final set of 51 full-text articles was eligible according to the study’s inclusion criteria. No reports were sought or excluded at the full-text retrieval stage. A detailed depiction of these stages, including the number of records removed, screened, and included, is presented in Figure 1.

Screening outcomes were systematically recorded and verified to ensure consistency between reviewers. Title and abstract screening excluded studies that were irrelevant to AI, predictive analytics, or wearable integration, while full-text review identified 51 studies meeting all criteria. Of these, 6 were empirical or experimental studies reporting quantitative metrics, and 45 were review or meta-analytic studies that conceptually examined AI-enabled wearable applications. These proportions are reflected in Tables 1,2.

Across the eligibility screening process, we identified a number of studies that met many of the inclusion criteria (e.g., reported on an AI-enabled wearable device, were peer-reviewed, etc.) but that were ultimately eliminated from the synthesis due to failing one or more methodological criteria. For example, a number of studies examined wearable technology in general settings (e.g., workplace, education) but did not report on predictive or diagnostic use (e.g., wearing a fitness tracker while recreationally playing sports, not for medical purposes). Other studies lacked sufficient methodological reporting around the AI elements or clinical outcomes, making it difficult to determine applicability for the synthesis. Finally, a few review papers were reporting on theoretical frameworks or speculative future use, but without reported empirical findings or review methodologies. We excluded these studies to maintain methodological rigor and thematic coherence for the synthesis.

Composition of the evidence base

Of the 51 sources in this review, 45 (88.2%) were review-type papers (including systematic, scoping, and narrative reviews), and six were empirical papers reporting primary data collection or experimental validation. In total, these figures show that there remains a predominance of secondary syntheses over original data collection or experimental use in current literature on AI-enabled wearables, and that current conceptual overviews and methodological approaches are not being empirically validated or generalized. This suggests that the use of AI in predictive healthcare with wearables is still in its early stages and is rapidly evolving.

Study characteristics

Characteristics of the 51 included studies are summarized in Table 1. Empirical studies (n=6) reported sample sizes ranging from 7 to 4,036 participants and applied diverse AI approaches, including decision trees, CNNs, and RNNs, often paired with wearable biosensors for vital-sign monitoring. Review studies (n=45) synthesized applications across predictive healthcare, chronic disease monitoring, and AI algorithm development.

To enhance clarity, the original composite was reorganized into two concise summary tables. Table 1 presents empirical and experimental studies reporting quantitative performance metrics, while Table 2 summarizes review and meta-analytic studies that discuss conceptual or methodological advances. This restructuring eliminates redundancy and allows clearer comparison of empirical vs. secondary evidence.

Risk of bias in included studies

Most empirical studies were limited by small sample sizes, lack of external validation, and potential reporting bias. Review studies frequently lacked structured quality assessments, creating a risk of selection and reporting bias. Overall, the evidence base demonstrates promising applications but limited methodological rigor, which may reduce generalizability.

Results of individual studies

Table 1 summarizes the quantitative and qualitative features of the 51 included studies in more detail, showing sample sizes, domains, and reported effect estimates. Key outcomes reported by individual studies are summarized in terms of accuracy/AUC and diagnostic performance, allowing comparisons across diverse AI and wearable technology applications for health care. Although a number of studies report impressive predictions, the studies are limited by small sample sizes, use of their own data for predictions, and variation in the reporting of precision metrics. Most entries are narrative reviews themselves, yielding no original quantitative results, underscoring the need for more empirical research with standardized outcome reporting to strengthen the evidence base.

Synthesis characteristics and risk of bias

The studies synthesized within each thematic synthesis displayed a high degree of variation in study design, approach, and sample size. Empirical studies ranged from small cohorts (n=7 participants) to large EHR datasets (millions of records), using a variety of AI methods including decision trees, CNNs, and RNNs coupled with wearable sensors. JBI critical appraisal checklists revealed frequent occurrences of methodological issues such as small or non-representative sample, external validity missing, performance of algorithm reporting incomplete and potential presence of publication bias. Narrative and scoping reviews often showed selection and reporting bias risk present due to lack of formal quality appraisal. Overall, these sources of variability impact the certainty and generalizability of findings synthesized (Table S2).

Synthesis characteristics and evidence integration

Given the substantial heterogeneity across included studies—spanning experimental, observational, and review-based designs—a meta-analysis was not conducted. Instead, findings were synthesized narratively and thematically. Quantitative metrics (e.g., accuracy, AUC, sensitivity, specificity) reported in individual empirical studies were interpreted descriptively within their respective methodological contexts. For example, Beg et al. (33) and Alam et al. (25) described predictive analytics models that improved diagnostic accuracy by up to 40%, while Osa-Sanchez et al. (8) and Abd-Alrazaq et al. (7) reviewed AI-based wearables achieving moderate-to-high accuracy in detecting sleep apnea and mental health conditions. These data points were used illustratively rather than statistically pooled. Overall, AI-enabled wearables showed directionally consistent benefits in predictive performance and early diagnosis across multiple health domains.

Investigation of heterogeneity among included studies

Due to the substantial methodological and clinical heterogeneity among included studies—varying in study designs, populations, AI models, wearable technologies, and outcome measures—formal statistical investigation of heterogeneity was not feasible. Instead, heterogeneity was explored qualitatively by examining study characteristics and contextual factors. Key sources of heterogeneity included differences in sample size (ranging from small cohorts to millions of EHRs), diversity in AI algorithms (e.g., decision trees, neural networks, deep learning), variation in wearable device types and clinical applications, and inconsistency in reported outcomes and performance metrics. These factors contributed to variability in predictive accuracy and generalizability, underscoring the need for standardized protocols and reporting in future research.

Assessment of reporting bias and certainty of evidence

Risk of bias due to missing results from reporting biases was qualitatively assessed as statistical formal assessment is not possible due to the study designs’ variability and the predominance of narrative studies. Potential biases were: inconsistent outcome reporting, selective display of favorable results, and underreporting of negative results. Most studies did not have pre-registered protocols, thus increasing the risk of selective reporting. Reviews addressing publication bias were few, and hence the validity of their conclusions is not confident.

Formal assessment of evidence certainty was not done due to reasons of synthesis heterogeneity and qualitative analysis. Many promising studies demonstrated the potential of AI-enabled wearable technologies in healthcare. However, small sample sizes, lack of external validation, and varied reporting decrease the confidence in the robustness/generalizability of results. Standardized quality assessments are lacking, and hence the use of frameworks such as GRADE is limited. Future reviews should focus on transparent reporting and application of formal certainty assessments.

Additionally, the exclusion of non-peer-reviewed literature, gray reports, and non-English studies may have introduced publication and language bias. This approach, while ensuring methodological rigor, potentially underrepresents emerging research or early-stage innovations not yet published in academic journals. The decision to exclude studies lacking methodological transparency could also bias the synthesis toward well-documented but potentially less innovative investigations. Future reviews should therefore consider incorporating diverse sources and languages to reduce exclusion-related bias.

Temporal distribution of publications on AI-integrated wearables in predictive healthcare

The trend in publication frequency (shown in Figure 2) indicates the academic attention towards AI-integrated wearables in healthcare. The number of relevant studies started modestly in 2016 and 2019. The number of relevant studies started to increase from 2021 and spiked in 2024 with 21 publications. Although the number dropped slightly in 2025, but the number of studies is high as compared to previous years. The trend indicates the importance of AI-enabled wearables in healthcare innovation due to technological advancements, clinical uptake, and hype around remote monitoring during and post-coronavirus disease 2019 (COVID-19) pandemic. Citation trends and the relative scholarly influence of included studies are presented in Table S3, which highlights large variations in citation counts and identifies the most influential publications in this field.

Key findings

Predictive analytics and AI in healthcare

The existence of recent publications testifies to the great utility or use of AI and predictive tools in early disease detection, accurate diagnosis, and personalized care. Boukenze et al. (1) achieved 99% accuracy in chronic kidney disease prediction with C4.5 decision tree application. Larger applications have been reported by Nwaimo et al. (2) and Shajari et al. (11), such as improving patients’ outcomes and hospital resource utilization with real-life predictive models. Alowais et al. (47) and Khalifa and Albadawy (46) reported the use of AI in personalized care, virtual health, and medication optimization. In contrast, Hirani et al. (48) and Aliferis and Simon (43) call for attention to the ethical oversight, transparency, and trust in the successful deployment of AI in clinical practice.

Wearable technology and remote monitoring

Wearables are revolutionizing healthcare through continuous and non-invasive monitoring of vital signs, enabling decentralized and patient-driven care. Initial wearables provided early warning signs for COVID-19 and chronic disease monitoring (4,5). Recently, wearables are aiding in diagnosis and treatment by guiding actions (28,34). Accuracy of data from wearables varies for sleep and oxygen information (9). Wearables combined with AI provide real-time information for chronic disease monitoring (10). When combined with IoT and 5G, wearables can provide remote, responsive, and scalable information for care (37,38). Although standardization is an issue, wearables are becoming more common in providing preventive, personalized, and data science-enabled healthcare.

Machine learning and deep learning in imaging and EHRs analysis

Within this broader analytical framework, what appears to emerge is that diagnostic workup has evolved, with machine learning and deep learning tools appearing to aid in the substantially quicker and ostensibly more accurate interpretation of medical images and data from EHRs. It emerges from these findings that ResNet101 suggests an accuracy of over 96% in classifying chest X-rays (21), and Grout et al. (3) noted that deep learning models used to predict chronic diseases support AUCs greater than 0.92. This analysis reveals that these tools provide evidence that supports a reduction in errors typically associated with humans and, given the multifaceted nature of this evidence, become a more personalized care provider by considering social determinants and explainable AI. Rahman et al. (45) and Badawy et al. (26) both seem to highlight the usefulness of AI in specialties such as cardiology and hepatology. In addition, tools such as ChatGPT-based chatbots assist in reviewing and documenting patients’ information, which increases efficiency in communication and documentation (42), which in turn increases efficiency in telemedicine. Overall, these tools enable more accurate diagnosis and allowing us to intervene earlier and with data-driven justification in the care of our patients.

Case studies: AI applications in specialized medical domains

AI-enabled technologies are transforming healthcare in several specialist areas via precise measurement, monitoring, and intervention. In cardiology, AI-enabled wearables can detect arrhythmias and heart dysfunction with high sensitivity (6,50) and can be used in conjunction with traditional diagnostics to facilitate remote cardiac rehabilitation (6). In oncology, AI-enabled wearables can track vitals and chemotherapy tolerance as well as monitor biomarkers to guide treatment and enable early detection of adverse reactions (12). In endocrinology, AI improves the diagnosis and monitoring of conditions such as diabetes and thyroid disorders. Methods that use images and sensor information to guide glucose management and classify nodules with high accuracy reduce the need for biopsies (52,53). In neurology, AI enables the diagnosis of neurodegenerative diseases using multimodal data, digital twin, and neuroimaging tools to enable earlier detection and personalized care (58,59). Finally, AI is changing mental health and sleep care: wearables based on heart rate variability (HRV) and activity can be used to detect depression and anxiety (32). AI virtual therapists can provide scalable, personalized care (48), and CNN-enabled wearables based on multi-sensor information can be used to diagnose sleep apnea (8). These technologies enable low-cost, remote care, but raise ethical and policy questions around safety and fairness.

Challenges

However, the application of predictive analytics is subject to several challenges. Beg et al. (33) identified the primary obstacles to the widespread adoption of these methods, including data quality issues, ethical concerns, and integration with existing healthcare systems. The challenges highlight the importance of data governance and ethical guidelines in facilitating the proper implementation of AI. Nevertheless, the advantages of AI are considerable. Laboone and Marques (27) demonstrate the numerous advantages offered by the AI applications in healthcare domains, such as improved diagnostic results, more efficient treatment planning, remote patient monitoring, and more rapid drug discovery.

Discussion

Transformative applications of AI-enabled wearable technologies in healthcare

The observed surge in research from 2021 onward suggests an acceleration of digital health innovation, seemingly linked to the COVID-19 pandemic and a corresponding increase in demand for remote healthcare solutions (37,38). High diagnostic accuracy (1,3) is a hallmark of mature AI algorithms (often decision trees or deep learning) applied on structured clinical data. The success of wearables on multiple cardiology tasks (50), oncology (12), and mental health (32) can be attributed to advances in sensor accuracy and integration of multiple data types. ResNet101 (37) requires an abundance of images in the environment to dominate (which they do), while ethical and implementation concerns (10,38) (standardization, proprietary models, lack of inclusivity) remain the most significant blocking factors to widespread clinical integration.

Synthesis of 51 included studies revealed transformative applications of AI-enabled wearables in predictive healthcare, chronic disease monitoring, and early diagnosis. Empirical studies [e.g., (1,21,22)] reported high diagnostic accuracy of predictive models and wearables, such as 99% chronic kidney disease prediction and 96% COVID-19 X-ray classification accuracy. Review studies [e.g., (2,27)] have confirmed that predictive modeling and wearable integration can improve early disease detection, resource allocation, and patient-centered care.

Moreover, wearable combination with AI can also expand monitoring beyond clinical care. For example, Huhn et al. (28) and Channa et al. (4) highlighted the potential of low-cost and consumer-grade wearables for large population studies (e.g., population-wide analyses in the context of COVID-19) and remote COVID-19 surveillance. Similarly, mental health applications (7,30) and cardiovascular care (49,50) have demonstrated the potential of wearable combination with AI for continuous risk assessment and early warning in large population surveillance for mental and cardiovascular health.

Integration with broader patient-generated clinical data (PCD)

Although this review focuses on AI-enabled wearables, consider these devices as the tip of the PCD ecosystem that also includes genomics, lifestyle logs, environmental exposures, family history, food, movement, and social determinants of health (60). Wearables serve as an easily consumable slice of this data ecosystem by providing a stream of high-frequency, continuous, real-time physiological and behavioral signals (27). When combined with other streams of PCD, such as symptomatic self-reports and genomics, PCD derived from wearables instrumented with physiological sensors may be integrated into longitudinal health records (LHRs) that can be used to inform personalized risk stratification in a dynamic manner (61). In this regard, wearables are both a starting point (Trojan horse) and stimulus for data-centric, holistic healthcare that bridges biological and behavioral to social patient attributes.

Evidence components and research maturity

There are 45 out of 51 papers in this study that were review type, highlighting a characteristic of the current research landscape: research in AI-blended wearables for predictive healthcare is predominantly conceptual (11). Rather than reflecting a lack of originality, this trend demonstrates that research in this area still remains focused on framework development, ethics, and feasibility, with few empirical validations or clinical trials. This publication practice is expected in emerging digital-health domains where standardized evaluation methods, algorithmic transparency, and data governance are still maturing. This scoping review offers a timely synthesis of conceptual and methodological work to guide future experimental and clinical work. Future studies should focus on prospective, multicenter, and validation design studies to build strong evidence and move this field from theoretical synthesis to clinical practice.

Limitations of the evidence

Although this review identified several studies that combined wearable data with EHR or imaging information, such multimodal integrations were not analyzed in depth, as this review focused specifically on AI-enabled wearables. Imaging- and EHR-based AI applications involve distinct technical, ethical, and interoperability challenges that warrant separate systematic investigations.

Despite promising results, the evidence is not without limitations:

• Small and non-representative sample populations and datasets (simulated or proprietary) limit generalization of results (1,21);
• Lack of external and longitudinal validation limits knowledge on real-world performance;
• Variable reporting and lack of pre-registered protocols increase the risk of selective reporting bias, as assessed in our risk of bias evaluation;
• Challenges related to privacy, interoperability, and algorithm explainability raise concerns about clinical adoption and regulatory approval of AI algorithms (27,30).

Methodological patterns and heterogeneity: limitations of the review processes

The evidence base is highly heterogeneous in terms of study design, population, wearables, and analytical design. Across empirical studies, sample sizes varied from small single-cohort studies of n=7 to large EHR-based studies containing millions of individuals (3,24). Across analytical approaches, variety of AI methods were used across studies, ranging from decision trees to CNNs, RNNs, and ensemble models, reflecting fast-moving computational capabilities.

Such heterogeneity presented challenges for meta-analysis and limited our ability to quantify pooled effects. Building on previous work (11,33), variability in performance reporting (accuracy, AUC, precision/recall) as well as lack of standardized evaluation frameworks limited statistical synthesis and external validity.

Implications for practice, policy, and future research

AI-powered wearables move health care closer to proactive, preventive, and personalized care. In chronic disease scenarios, by constantly monitoring patients, wearables facilitate the early detection of arrhythmias, cardiotoxicity, and further progression of the disease, which is beneficial for timely intervention and better patient outcomes (6,31). Public health care benefits may be realized from large population health monitoring through wearables use, which provides population-wide health information, facilitates stratification of risks, and informs resource planning (4,28).

Furthermore, given the potential clinical impact of AI-enabled wearables, health policy should standardize evaluation, data privacy, and interoperability. Because sensitive health data should be protected (10,47), targeted policies and incentives are needed to mitigate potential disparities in access and address AI-enabled wearable adoption gaps among marginalized populations (4,27).

Future research should focus on key aspects to enable reproducibility and support clinical translation:

• Conduct large-scale, longitudinal trials with more demonstrably diverse populations (ostensibly to support generalizability).
• Move towards greater standardization of outcome reporting and validation protocols (building on the foundations provided by PRISMA, CONSORT-AI, and GRADE). From this particular interpretive perspective, improvements in the explainability and ethical governance of AI models will be critical to support patient trust, fairness, and regulatory compliance.
• Finally, more concerted multimodal integration (including the integration of wearable, imaging, genomic, and behavioral data) will be a promising future direction to enable more holistic patient characterization (6,45).

Conclusions

This review showcases the recent and rapid development of AI-enabled wearables in predictive health care, synthesized across bibliometric trends, methodological variation, and clinical use. The emerging body of work demonstrates how AI is enabling early diagnosis, personal intervention, and remote monitoring across a range of diseases. Despite challenges such as data standardization, ethical considerations, and underrepresentation of low-resource settings, the potential benefits are great. As AI technology matures, the development and adoption of its use with wearables and health systems will be enabled to support more proactive, equitable, and patient-centered care. To realize this potential, ongoing research, inclusive validation, and robust regulatory approaches will be important.

Acknowledgments

The author acknowledges the assistance of independent reviewers who verified screening decisions for accuracy. Their contribution was limited to quality control and did not warrant authorship. The author’s ideas and findings in this article do not reflect the views of any affiliation. The author disclaims any guarantees and endorsements, and the publishers, editors, and reviewers disclaim responsibility for any products, tools, and services mentioned by the author.

Footnote

Reporting Checklist: The author has completed the PRISMA-ScR reporting checklist. Available at https://jmai.amegroups.com/article/view/10.21037/jmai-2025-170/rc

Peer Review File: Available at https://jmai.amegroups.com/article/view/10.21037/jmai-2025-170/prf

Funding: None.

Conflicts of Interest: The author has completed the ICMJE uniform disclosure form (available at https://jmai.amegroups.com/article/view/10.21037/jmai-2025-170/coif). The author has no conflicts of interest to declare.

Ethical Statement: The author is accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Boukenze B, Mousannif H, Haqiq A. Predictive analytics in healthcare system using data mining techniques. Comput Sci Inf Technol 2016;1:1-9.
2. Nwaimo CS, Adegbola AE, Adegbola MD. Transforming healthcare with data analytics: Predictive models for patient outcomes. GSC Biol Pharm Sci 2024;27:25-35.
3. Grout R, Gupta R, Bryant R, et al. Predicting disease onset from electronic health records for population health management: a scalable and explainable Deep Learning approach. Front Artif Intell 2023;6:1287541. [Crossref] [PubMed]
4. Channa A, Popescu N, Skibinska J, et al. The Rise of Wearable Devices during the COVID-19 Pandemic: A Systematic Review. Sensors (Basel) 2021;21:5787. [Crossref] [PubMed]
5. Obianyo C, Ezeamii VC, Idoko B, et al. The future of wearable health technology: from monitoring to preventive healthcare. World J Biol Pharm Heal Sci 2024;20:36-55.
6. Sozib HM. Wearable AI for Cardiovascular Health Monitoring: Enabling Early Detection and Prevention. J Comput Sci Technol Stud 2025;7:294-304.
7. Abd-Alrazaq A, AlSaad R, Aziz S, et al. Wearable Artificial Intelligence for Anxiety and Depression: Scoping Review. J Med Internet Res 2023;25:e42672. [Crossref] [PubMed]
8. Osa-Sanchez A, Ramos-Martinez-de-Soria J, Mendez-Zorrilla A, et al. Wearable Sensors and Artificial Intelligence for Sleep Apnea Detection: A Systematic Review. J Med Syst 2025;49:66. [Crossref] [PubMed]
9. Shei RJ, Holder IG, Oumsang AS, et al. Wearable activity trackers-advanced technology or advanced marketing? Eur J Appl Physiol 2022;122:1975-90. [Crossref] [PubMed]
10. Vo DK, Trinh KTL. Advances in Wearable Biosensors for Healthcare: Current Trends, Applications, and Future Perspectives. Biosensors (Basel) 2024;14:560. [Crossref] [PubMed]
11. Shajari S, Kuruvinashetti K, Komeili A, et al. The Emergence of AI-Based Wearable Sensors for Digital Health Technology: A Review. Sensors (Basel) 2023;23:9498. [Crossref] [PubMed]
12. Birla M. Integrating Artificial Intelligence-Driven Wearable Technology in Oncology Decision-Making: A Narrative Review. Oncology 2025;103:69-82. [Crossref] [PubMed]
13. Tricco AC, Lillie E, Zarin W, et al. PRISMA Extension for Scoping Reviews (PRISMA-ScR): Checklist and Explanation. Ann Intern Med 2018;169:467-73. [Crossref] [PubMed]
14. Vallabhuni S, Debasis K. Hybrid deep learning for IoT-based health monitoring with physiological event extraction. Digit Health 2025;11:20552076251337848. [Crossref] [PubMed]
15. Anjum M, Ahmad W, Shahab S, et al. Enhancing wearable sensor data analysis for patient health monitoring using allied data disparity technique and multi instance ensemble perceptron learning. Sci Rep 2025;15:29555. [Crossref] [PubMed]
16. Jain A, Bhullar S. AI-driven wearable health devices with health-aware control and secure Prognostics: Experimental and Simulation-Based Validation. Array 2025;28:100532.
17. Parker M, Fang X, Self-Brown SR, et al. Establishing how social capital is studied in relation to cardiovascular disease and identifying gaps for future research-A scoping review protocol. PLoS One 2021;16:e0249751. [Crossref] [PubMed]
18. Tang X, Zeng Z, Huang H, et al. Quality Appraisal Tools for Quantitative, Qualitative, and Mixed-Methods Studies: A Review and a Brief New Checklist. ECNU Rev Educ 2025. doi: 10.1177/20965311251371227.
19. McMahon K, Clark IN, Stensaeth K, et al. Exploring shared musical experiences in dementia care: A worked example of a qualitative systematic review and thematic Synthesis. Int J Qual Methods 2022;21:16094069221127509.
20. Mikkonen K, Kääriäinen M. Content analysis in systematic reviews. In: Kyngäs H, Mikkonen K, Kääriäinen M, editors. The application of content analysis in nursing science research. Cham: Springer; 2020:105-15.
21. Yenikaya MA, Kerse G, Oktaysoy O. Artificial intelligence in the healthcare sector: comparison of deep learning networks using chest X-ray images. Front Public Health 2024;12:1386110. [Crossref] [PubMed]
22. Gabriel P, Rehani P, Troy T, et al. Continuous patient monitoring with AI: real-time analysis of video in hospital care settings. Front Imaging 2025;4:1547166.
23. Mondal M. Knowledge and associated factors for use of contraceptives among married women of West Bengal, India. Int J Res Granthaalayah 2024;12:1-8.
24. Kaewkannate K, Kim S. A comparison of wearable fitness devices. BMC Public Health 2016;16:433. [Crossref] [PubMed]
25. Alam MA, Nabil AR, Uddin MM, et al. The role of predictive analytics in early disease detection: a data-driven approach to preventive healthcare. Innovatech Engineering Journal 2024;1:105-23.
26. Badawy M, Ramadan N, Hefny HA. Healthcare predictive analytics using machine learning and deep learning techniques: a survey. Journal of Electrical Systems and Information Technology 2023;10:40.
27. LaBoone PA, Marques O. Overview of the future impact of wearables and artificial intelligence in healthcare workflows and technology. Int J Inf Manag Data Insights 2024;4:100294.
28. Huhn S, Axt M, Gunga HC, et al. The Impact of Wearable Technologies in Health Research: Scoping Review. JMIR Mhealth Uhealth 2022;10:e34384. [Crossref] [PubMed]
29. Aziz S, Ali AAM, Aslam H, et al. Wearable Artificial Intelligence for Sleep Disorders: Scoping Review. J Med Internet Res 2025;27:e65272. [Crossref] [PubMed]
30. Kargarandehkordi A, Li S, Lin K, et al. Fusing Wearable Biosensors with Artificial Intelligence for Mental Health Monitoring: A Systematic Review. Biosensors (Basel) 2025;15:202. [Crossref] [PubMed]
31. Nechita LC, Tutunaru D, Nechita A, et al. AI and Smart Devices in Cardio-Oncology: Advancements in Cardiotoxicity Prediction and Cardiovascular Monitoring. Diagnostics (Basel) 2025;15:787. [Crossref] [PubMed]
32. Wang W, Chen J, Hu Y, et al. Integration of artificial intelligence and wearable internet of things for mental health detection. Int J Cogn Comput Eng 2024;5:307-15.
33. Beg SW, Ahmad S, Anwar S, et al. Predictive analysis in medical healthcare: a meta-analysis. Int J Res Granthaalayah 2024;12:9-16.
34. Shang X. The current status and future trends of wearable devices. Appl Comput Eng 2025;120:98-104.
35. Shin G, Jarrahi MH, Fei Y, et al. Wearable activity trackers, accuracy, adoption, acceptance and health impact: A systematic literature review. J Biomed Inform 2019;93:103153. [Crossref] [PubMed]
36. Iqbal SMA, Mahgoub I, Du E, et al. Advances in healthcare wearable devices. NPJ Flex Electron 2021;5:9.
37. Verma D, Singh KRB, Yadav AK, et al. Internet of things (IoT) in nano-integrated wearable biosensor devices for healthcare applications. Biosens Bioelectron X 2022;11:100153.
38. Escobar-Linero E, Muñoz-Saavedra L, Luna-Perejón F, et al. Wearable Health Devices for Diagnosis Support: Evolution and Future Tendencies. Sensors (Basel) 2023;23:1678. [Crossref] [PubMed]
39. Jegan R, Nimi WS. On the development of low power wearable devices for assessment of physiological vital parameters: a systematic review. Z Gesundh Wiss 2023; Epub ahead of print. [Crossref]
40. Guk K, Han G, Lim J, et al. Evolution of Wearable Devices with Real-Time Disease Monitoring for Personalized Healthcare. Nanomaterials (Basel) 2019;9:813. [Crossref] [PubMed]
41. Taherdoost H. Wearable Healthcare and Continuous Vital Sign Monitoring with IoT Integration. Comput Mater Contin 2024;81:79-104.
42. Chakraborty C, Bhattacharya M, Pal S, et al. From machine learning to deep learning: Advances of the recent data-driven paradigm shift in medicine and healthcare. Curr Res Biotechnol 2024;7:100164.
43. Aliferis C, Simon G. Artificial Intelligence (AI) and Machine Learning (ML) for Healthcare and Health Sciences: The Need for Best Practices Enabling Trust in AI and ML. In: Simon GJ, Aliferis C, editors. Artificial Intelligence and Machine Learning in Health Care and Medical Sciences: Best Practices and Pitfalls. Cham: Springer; 2024.
44. Rashid AB, Asma J, Barua K, et al. An Enhanced Deep Learning Framework for Pneumonia Detection in Chest X-rays. SN Comput Sci 2025;6:472.
45. Rahman A, Debnath T, Kundu D, et al. Machine learning and deep learning-based approach in smart healthcare: Recent advances, applications, challenges and opportunities. AIMS Public Health 2024;11:58-109. [Crossref] [PubMed]
46. Khalifa M, Albadawy M. Artificial intelligence for clinical prediction: exploring key domains and essential functions. Comput Methods Programs Biomed Update 2024;5:100148.
47. Alowais SA, Alghamdi SS, Alsuhebany N, et al. Revolutionizing healthcare: the role of artificial intelligence in clinical practice. BMC Med Educ 2023;23:689. [Crossref] [PubMed]
48. Hirani R, Noruzi K, Khuram H, et al. Artificial Intelligence and Healthcare: A Journey through History, Present Innovations, and Future Possibilities. Life (Basel) 2024;14:557. [Crossref] [PubMed]
49. Hiremath S. AI-driven cardiac monitoring: the future of wearable medical devices. J Med Dent Sci Res 2021;8:58-62.
50. Jena N, Singh P, Chandramohan D, et al. Wearable Technology in Cardiology: Advancements, Applications, and Future Prospects. Rev Cardiovasc Med 2025;26:39025. [Crossref] [PubMed]
51. Huang G, Chen X, Liao C. AI-Driven Wearable Bioelectronics in Digital Healthcare. Biosensors (Basel) 2025;15:410. [Crossref] [PubMed]
52. Lu Q, Wu Y, Chang J, et al. Application progress of artificial intelligence in managing thyroid disease. Front Endocrinol (Lausanne) 2025;16:1578455. [Crossref] [PubMed]
53. Oikonomakos IT, Anjana RM, Mohan V, et al. Recent advances in artificial intelligence-assisted endocrinology and diabetes. Explor Endocrine Metab Dis 2024;1:16-26.
54. Belkhouribchia J. Artificial intelligence is going to transform the field of endocrinology: an overview. Front Endocrinol (Lausanne) 2025;16:1513929. [Crossref] [PubMed]
55. Olawade DB, Wada OZ, Odetayo A, et al. Enhancing mental health with Artificial Intelligence: Current trends and future prospects. J Med Surg Public Health 2024;3:100099.
56. Nahavandi D, Alizadehsani R, Khosravi A, et al. Application of artificial intelligence in wearable devices: Opportunities and challenges. Comput Methods Programs Biomed 2022;213:106541. [Crossref] [PubMed]
57. Rizzo M. AI in Neurology: Everything, Everywhere, All at Once Part 3: Surveillance, Synthesis, Simulation, and Systems. Ann Neurol 2025;98:651-67. [Crossref] [PubMed]
58. Gutman B, Shmilovitch AH, Aran D, et al. Twenty-Five Years of AI in Neurology: The Journey of Predictive Medicine and Biological Breakthroughs. JMIR Neurotechnol 2024;3:e59556. [Crossref] [PubMed]
59. Boubga T, Bentaher A, Taous A, et al. Artificial Intelligence in Neurology: Current Applications and Future Prospects. Int J Innov Sci Res Technol 2024;9:104-10.
60. Baumann S, Stone RT, Abdelall E. Smart Healthcare at Home: A Review of AI-Enabled Wearables and Diagnostics Through the Lens of the Pi-CON Methodology. Sensors (Basel) 2025;25:6067. [Crossref] [PubMed]
61. Valero-Ramon Z, Ibanez-Sanchez G, Martinez-Millana A, et al. Personalised Risk Modelling for Older Adult Cancer Survivors: Combining Wearable Data and Self-Reported Measures to Address Time-Varying Risks. Sensors (Basel) 2025;25:2097. [Crossref] [PubMed]