We chose a scoping review because it provides the flexibility to synthesize heterogeneous evidence across different wearable device types, clinical contexts, and data analytical methods in this rapidly evolving field. We conducted the review following the Joanna Briggs Institute (JBI) manual for scoping reviews [22] and reported it in alignment with the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) guidelines [23]. The protocol was registered on the Open Science Framework (OSF) [24] and published [25] ahead of this review. The PRISMA-ScR checklist is provided in Multimedia Appendix 1.

Our review addressed the following questions:

We adopted the Population, Concept, Context (PCC) framework to guide the development of search strategies and eligibility criteria.

The MEDLINE, Embase, Web of Science, CINAHL, and IEEE Xplore databases were searched from January 2013 to July 2026, as bibliometric analysis indicates a rapid increase in research on wearable technologies for health monitoring since 2013 [29,30]. The search was initially conducted in February 2025 and updated in July 2026 (both by reviewer YL). Search terms were organized into categories, including “wearable*,” “body temperature,” “fever,” “infecti*,” “patient*,” and “hospital.” Specific brands of wearables were identified from previous reviews and added to search strings [7,9]. Gray literature, including conference proceedings, preprints, and dissertations, was also searched. The detailed search strategies for databases and websites for gray literature are provided in Multimedia Appendix 2. The full search strategy was reviewed by a professional medical librarian using the PRESS (Peer Review of Electronic Search Strategies) checklist [31].

We uploaded the search results to Rayyan [32] to remove duplicates and conduct online screening. During the pilot screening, we compiled a preliminary list of potential temperature monitoring devices, discussed which qualified as wearables, and reached consensus on the eligible boundaries before formal screening. Then, a 2-step process, including title and abstract screening and full-text screening, was used. One reviewer (YL) screened all literature, with 20% randomly selected and independently screened by a second reviewer (QZ) to assess screening consistency. The interrater agreement using Cohen κ coefficients was 0.778 and 0.746 for each screening step, indicating substantial agreement. Disagreements were resolved through discussion with a third reviewer (DD). For multiple records derived from the same study, such as conference abstracts and research protocols, the formally published article was retained. The reference lists and citation lists of eligible articles and relevant literature reviews were manually searched to identify additional publications.

Studies were included if they used wearables to monitor patients’ body temperature in acute care hospitals for the purpose of fever or infection management, including but not limited to detection, diagnosis, prediction, treatment, or prognostic assessment. Primary research studies published in English or Chinese were eligible regardless of study design or publication type. Studies were excluded if they focused on monitoring local tissue temperature (eg, wound or foot), reported the technical design or development of wearables, assessed device accuracy exclusively without clinical interpretation of temperature data, or focused on the prevalence of epidemic infectious diseases rather than clinical care.

Data were extracted from the main text and supplementary materials using a predefined data extraction form (Multimedia Appendix 3), including bibliographic and contextual attributes, wearable device characteristics, and strategies for data transmission, storage, and presentation. We also extracted temperature data analysis strategies, including data preprocessing, derived temperature features, and analytic methods. We classified temperature data analyses according to the descriptive, diagnostic, predictive, and prescriptive analytics framework [33,34], and further assessed the maturity stage of wearables’ integration in clinical practice with reference to the World Health Organization’s (WHO) manual, “Monitoring and Evaluating Digital Health Interventions” [35]. We also identified the role of nurses by reviewing the main text, author lists, and affiliations. We did not assess the methodological quality of the included publications in an attempt to map all available evidence, and quality appraisal is not mandatory in a scoping review according to JBI methodology [36]. Data extraction was performed by one reviewer (YL). A second reviewer (QZ) independently extracted data from a random 20% of the included publications to evaluate the reliability of the data extraction process. Discrepancies were discussed to reach a resolution, and a third researcher (DD) was consulted for unresolved discrepancies. A descriptive synthesis of the findings was presented in tables and figures.

A total of 2309 records were identified through database searches. After duplicate removal (n=415), 1948 records (1894 from database searches and 54 from manual identification) were screened by title and abstract. Subsequently, 111 records (84 from database search screening and 27 from other methods) underwent full-text screening. We included 29 publications (from 26 studies) in this review. Figure 1 presents the study selection process. Three publications [6,37,38] reported data derived from the same patient cohort in one study, while 2 publications [39,40] analyzed data from another study.

The 29 included publications (from 26 studies) comprised 25 journal articles (including 1 research letter with original data) and 4 conference abstracts, published in English (28/29, 97%) and Chinese (1/29, 3%) between 2018 and 2026. The included studies were conducted in Asia (9/26, 35%), Europe (8/26, 31%), North America (6/26, 23%), and Oceania (3/26, 12%). The majority were prospective observational studies (20/26, 77%), while 3 were retrospective observational studies, 1 was a randomized controlled trial, and 2 observational studies did not specify whether they were prospective or retrospective. Most studies were single-center (24/26, 92%), with only 2 studies being multicenter. Study populations were adult patients (17/26, 65%), pediatric patients (5/26, 19%), women in labor or pregnancy (3/26, 12%), and older adult patients (1/26, 4%). Clinical settings comprised oncology wards (9/26, 35%), surgical wards (6/26, 23%), medical wards (6/26, 23%), prenatal or delivery rooms (3/26, 12%), and ICUs (2/26, 8%). The primary events of interest were febrile events (15/26, 58%; including fever, febrile neutropenia, cytokine release syndrome [CRS], and intrapartum fever) and infections (11/26, 42%; including sepsis, pneumonia, intraamniotic infection, postoperative infectious complications, and COVID-19).

A total of 18 wearable devices were identified (Table 1), including 4 chest patches, 6 axillary patches, 2 smartwatches, 2 wristbands, 3 upper arm or leg bands, and 1 neck patch. Most devices (11/18, 61%) monitored multiple physiological parameters, while 7 devices were dedicated temperature sensors. Seven devices measured axillary temperature, while 11 devices measured skin temperature at various sites, including the chest (n=3), wrist (n=4), upper arm or leg (n=3), and neck (n=1). Five devices provided estimated core temperature (VitalPatch, CORE, Everion, Moni-Patch, and iThermonitor), with measurements derived from axillary or skin temperature readings. Axillary patches were the most frequently used device, with the highest number of studies reported in oncology wards (Figure 2).

Among the 18 identified wearables, 78% (14/18) were medical-grade and 11% (2/18) were consumer-grade, while the grade was not specified for 2 devices. Ten wearables were approved by the US Food and Drug Administration (FDA), with the majority (8/10) approved as class II devices for use by health care professionals on general patients in health care settings, while 2 were approved for home use or ovulation tracking (Multimedia Appendix 3). However, reporting on the regulatory status and clinical validation of wearables showed gaps. Among the included publications, only 41% (12/29) reported the device grade, and just 35% (10/29) confirmed FDA approval or Conformité Européenne (CE) marking. Only 35% (10/29) of publications validated the wearables by comparing the temperature data with standard clinical measurements as part of their results, and 17% (5/29) cited previously published validation findings. One article noted that the wearable device used in their study had not yet been clinically validated [41], and 2 articles mentioned manufacturer preclinical validation [42,43]. The remaining publications (11/29, 38%) did not mention clinical validation.

There was variation in monitoring frequency and duration across studies. Body temperature was monitored at varying intervals, including every 1, 2, 4, 10, or 30 seconds; every 1, 2, 5, or 15 minutes; or at sampling frequencies of 4 Hz or 25 Hz. The median duration of temperature monitoring per patient ranged from 3 hours to 20 days across studies. Battery life across the included wearables ranged from 1 to 30 days. Holt et al [44] reported that, despite a theoretical maximum battery life of 3 days, the device recorded data only up to 44 hours in practice. To meet the requirements of clinical monitoring durations, different strategies were used depending on the device, such as daily charging [41,45,46], battery replacement [47], or regular device change every 24 hours [39,42,48], 3 days [49,50], or 6 days [51].

Table 3 maps the original WHO stage of maturity for digital health interventions and our modified maturity framework tailored to wearables, outlining 6 phases in a continuum. As specified in the eligibility criteria, studies focused on prototype design, preclinical testing, or only on validating the accuracy of wearables were excluded. Among the 29 included publications, most were classified as being at the clinical validation stage (stage 3a: 9/29, 31%) or clinical research stage (stage 3b: 17/29, 59%). The primary objectives of these studies were to evaluate the performance of wearables for monitoring body temperature in clinical settings or to collect research data, corresponding to the pilot stage in the WHO framework. In 3 studies, wearables were applied in routine clinical practice (stage 4: 3/29, 10%) as a supplement to standard temperature monitoring by nurses, aligning with the demonstration stage. None of the studies in this review had progressed to multicenter implementation or full integration.

The findings show that nurses were engaged in both research and clinical practice related to wearable devices. As researchers (identified in 7 studies), nurses acted as principal investigators who led and conducted the research [47,48,50,55,57,61] or served as clinical research nurses responsible for patient recruitment, device maintenance, and data collection [54,61]. In clinical practice, 12 studies described nurses as direct users of wearable devices for temperature monitoring, evaluating temperature data generated by the devices, and responding to device alerts.

This scoping review assessed 29 publications from 26 studies using wearables to monitor body temperature in acute care hospitals for fever or infection management. We identified a variety of wearables used across diverse patient populations and clinical settings since 2018, indicating growing interest in this emerging technology despite its recent introduction to clinical practice. By mapping the current evidence on wearable temperature data pipelines from sampling through transmission and presentation to clinical use, we also identified overall limited clinical integration and critical gaps at each stage. These findings may serve as a practical reference for the design and implementation of wearable-based monitoring systems within clinical workflows and highlight barriers to be addressed to move toward more sustainable clinical adoption.

This review identified a preference for medical-grade devices in clinical settings, while consumer-grade devices predominated in home-based remote monitoring in oncology [20] or cardiovascular care [64]. This preference reflects the need for accurate temperature data in clinical practice, which influences the selection of wearables and how the data can be used. Wearables generally provide temperature readings by monitoring thermal activity at the skin surface, which is susceptible to ambient conditions and skin perfusion [65]. In order to meet clinical and regulatory standards, medical-grade wearables adopted several strategies to deliver accurate temperature data, including monitoring at more reliable anatomical sites such as the axilla, which closely approximates core temperature when properly acquired [66], and developing algorithms to estimate core temperature from raw sensor signals. Accordingly, the average readings of estimated core temperatures and axillary temperatures were around 36.3 °C to 37.2 °C [40,62], which were mainly considered in studies that presented wearable data in clinical practice.

Unsurprisingly, skin temperatures directly provided by wearables at wearing sites, such as the wrist or chest, remain more variable in terms of validity and clinical utility. The mean skin temperatures were reported around 33.9 °C to 34.8 °C in reviewed studies [43,44], showing significant deviation from commonly used clinical values, particularly when determining fever. Therefore, several studies withheld skin temperature from clinical staff or discarded data from analyses [67-69], highlighting a cautious stance toward its clinical relevance. Some studies limited data collection periods, monitoring wrist skin temperature only during sleep to reduce daily activity influences [43]. However, studies have embraced more innovative strategies to actively use skin temperature, such as identifying elevated values as a predictive signal for fever [55], detecting abnormal circadian rhythms in patients with critical illness [49], and directly using its cutoff values to predict infection [44]. These methods view skin temperature as an independent output pathway of the thermoregulatory system, offering a potentially clinically useful route for its application. Notably, 1 study reported that both axillary and skin temperature emerged as reliable biomarkers, yet prioritized wearable-derived skin temperature over axillary measurements for its higher sampling frequency, continuous availability, and ease of future deployment [59]. Therefore, for clinical interpretation of wearable temperature data, health care providers should not be limited by absolute value comparability to traditional measurements but integrate trend-based and pattern-recognition approaches to expand clinical utility of wearable data across diverse monitoring contexts.

Although medical-grade wearables exhibit accuracy, prolonged usage in real-world clinical environments still presents significant data quality challenges arising from patient nonadherence, poor skin contact, device malfunction, sensor drift, and signal artifacts. Efforts have been made by manufacturers to assign quality scores to each temperature reading via wearables’ firmware [41,45,46,51]. However, the reliability of quality scores has not been adequately validated, and their calculations often lack transparency due to proprietary algorithms and patent protection. As indicated by Koenig et al [41], the wearables’ quality assessment was generally plausible, but it occasionally misclassified vital signs by assigning good quality to incorrectly measured data and poor quality to plausible extreme values. Therefore, studies in this review differed in their use of quality scores. We highly recommend that future studies validate quality scores before deciding whether to incorporate them.

Identifying anomalies should be guided by both data characteristics and clinical domain knowledge, as anomalies may still fall within the physiological boundaries yet represent clinically significant events [15]. The most commonly used method in the reviewed studies relied on thresholds to identify invalid temperature data, which has significant limitations. Removing axillary temperature readings below 35.5 °C may overlook inadvertent hypothermia in patients with critical illness [70] or patients who had undergone surgery [66]. Moreover, recent studies suggest that human basal body temperature has declined over the past century [71]. A systematic review calculated the normal range for axillary temperature to be 35.01 °C to 36.93 °C [72], indicating that a fixed 35.5 °C may not be able to account for individual variations in baseline temperatures. Additionally, the high variability in skin temperature makes it more challenging to determine reasonable thresholds for outlier removal. Instead, specialized artifact detection algorithms have been developed in a broader context. For example, Bardia et al [73] developed algorithms to identify adjacent temperature readings with slopes exceeding 0.08 or an absolute change greater than 0.25 °C, which effectively filtered intraoperative temperature artifacts. Future efforts should develop and incorporate such standardized data cleaning frameworks based on specific clinical contexts and professional knowledge, enabling personalized anomaly detection while preserving clinically relevant events.

After identifying anomalies in temperature data, handling incomplete datasets remains challenging. Most reviewed studies simply excluded cases with high missingness or deleted anomalous window segments. While retaining only valid data is feasible for basic statistical comparisons, removing data disrupts time-series signal continuity and limits specific feature extractions that require continuous data as inputs, such as entropy metrics for nonlinear dynamics [13]. On the other hand, the decision to impute missing data also requires careful evaluation. A study comparing imputation techniques for wireless continuous vital signs found that linear interpolation reduced feature bias (eg, 2-hour window mean and slope) relative to no imputation, whereas other techniques (eg, last observation carried forward and spline interpolation) increased bias, with imputation errors growing significantly with gap length [74]. Some advanced strategies have emerged as alternatives to imputation. For instance, dimensionality and size reduction based on intrinsic and target similarity can be used to eliminate features and samples that are highly sensitive to missing values [75], a principle that may extend to physiological time-series modeling. In addition, algorithms with built-in tolerance to missing data may improve model robustness and generalizability. Overall, we recommend selecting the appropriate preprocessing approach based on the specific downstream task, following careful assessment of potential bias.

Our review found that descriptive analytics remain the predominant approach for analyzing wearable-generated temperature data, with visual inspection of the trends and spikes in temperature curves being a common approach. This approach is intuitive, convenient, and easy to interpret in clinical practice, without requiring additional computational resources. However, it may not allow for detecting a specific pattern or predictive signals preceding infections [41]. In contrast, time series analysis and machine learning models for diagnostic or predictive applications were used in only one-third of the publications reviewed, indicating underutilization of these advanced analytical capabilities.

To leverage the hidden temporal patterns in high-frequency temperature data, machine learning and time series models represent an important methodological evolution. These methods extract temperature features that indicate circadian rhythm variations and pattern similarities, capturing subtle yet crucial indicators of infection that are not discernible through direct visual inspection. Beyond the temperature features identified in this review, studies have examined temperature dynamics in a broader context. For example, Papaioannou et al [76] assessed temperature complexity using wavelet transformation and multiscale entropy to classify patients with systemic inflammatory response syndrome, sepsis, and septic shock. Bhavani et al [77] identified 4 temperature trajectories from hourly recorded EHR data, revealing sepsis subphenotypes with distinct biomarker profiles. These advances highlight the need for further integration of advanced data analytics for temperature temporal pattern mining and knowledge discovery. These algorithms offer promising solutions to tackle data overload challenges by filtering meaningful and actionable information from massive wearable data; however, they still require further integration into computerized decision support systems to enable intelligent analysis and transform data into simpler and more usable forms.

We found that several wearables have been used for temperature monitoring as a supplement to standard nursing measurements, but their integration into routine clinical practice remains limited. The wearable systems were deployed independently, with only 1 study enabling alerts through EHRs [17]. This may stem from health systems’ requirement to establish a dedicated secure network for wearables, separated from the main network to prioritize data privacy [78]. However, this approach creates data silos. Although temperature is a key indicator for infection detection, final clinical diagnosis or treatment decisions require synthesis of wearable data with laboratory findings, clinical assessments, and patient history. Advancing system interoperability is essential to support comprehensive clinical decision-making rather than switching between systems to access fragmented data. Potential solutions include establishing standardized APIs through third-party integration platforms and developing plug-and-play interoperability standards [78]. Additionally, to streamline nursing workflow, more efforts are still needed to achieve automatic uploading of wearable data to nursing records, rather than requiring manual transcription of data from one screen to another. However, as previously discussed, such system integration must incorporate automatic artifact filtering algorithms to prevent recording erroneous data without professional oversight.

To better support clinical practice using wearables, we identified several gaps requiring further attention regarding data access interfaces, data latency, and data security. First, some studies deployed central monitoring systems at the nursing station, which enable simultaneous multipatient monitoring but prevent immediate access to real-time readings when assessing patients at bedside. A hybrid approach combining central and mobile interfaces would optimize both oversight and point-of-care accessibility. Second, some studies used batch uploads at fixed intervals, which conserve energy and extend battery life but may delay clinical reactions to patient deterioration. Transmission frequency must align with clinical context to ensure real-time data streaming, with transparent reporting enabling proper data latency assessment. Third, data security is a growing concern, yet only 10 studies reported specific protection strategies, with 50% (5/10) conducted in Europe under strict EU privacy regulations [20]. There remains a need for stronger adherence to safety and privacy regulations globally.

Overall, the clinical integration of wearables for fever and infection management remains predominantly at an early stage of maturity, aligning with existing evidence from digital health interventions for postoperative monitoring [79] and oncology care [20]. Although the scope of regulatory approval limited some devices’ clinical application, the reasons more frequently reported for not providing data to support clinical decision-making were insufficient data accuracy and limited real-time availability [53]. We also noted that certain devices (eg, SensiumVitals, Everion, and Gen2, listed in Multimedia Appendix 3) are not intended for use in critical care patients, potentially associated with reduced measurement reliability under unstable hemodynamic conditions. This explains their limited use in ICUs (Figure 2) and underscores the need for caution when using wearables for patients who are clinically deteriorating in general wards and avoiding sole reliance on wearable data for clinical decisions. Beyond technical limitations, there are systemic barriers to overcome before wearables can become widely accepted with full clinical integration, such as data ownership and privacy, data standards and interoperability, and reimbursement policies [80]. Additionally, the geographic concentration of reviewed studies in well-resourced areas highlights potential equity challenges, as disparities in device access, digital literacy, and infrastructure may limit scaled implementation and exacerbate health inequalities. Nevertheless, the emergence of wearables in clinical settings shows a developmental trajectory outlined in the maturity framework, which delineates the accumulation of evidence throughout the device lifecycle from initial prototype to full clinical integration.

With the rapid iteration of wearable technology, novel devices are increasingly emerging. While existing evidence mainly focused on lightweight rigid wearables, we identified some textile-based flexible wearables for remote monitoring of infectious diseases in the literature search. However, these textile-based devices are still in the proof-of-concept [81], technical testing [82], and animal experimental stages [83], possibly reflecting the current technological bottlenecks in achieving clinical accuracy and limited maturity in regulatory approval and clinical validation [84]. This development reflects an evolution in wearable technologies toward softer, lighter, and safer sensing materials. In future studies, our maturity framework could serve as a tool to support health care providers in evaluating novel wearables’ readiness for clinical deployment and selecting suitable devices. Noncontact vital sign monitoring technologies, such as camera-based or robotic sensing systems, represent another emerging direction in continuous physiological surveillance in hospital environments [85] and may offer supplementary options for body temperature monitoring in special populations, such as neonates with fragile skin intolerant to adhesive sensors [86], although these technologies fall outside the scope of this review.

As wearables become increasingly used in clinical practice, nurses need to be better prepared for this transformation. Nurses are not only frontline users but also key researchers in this field, with nearly one-third of the studies reviewed being nurse-led. In future clinical deployment, large-scale data streams from wearables will require a robust infrastructure and storage capacity to prevent data loss or overwriting. These data could be incorporated into the clinical research data warehouse as a valuable source. Moreover, to harness the value of big data and AI, it is important for nurses to develop digital literacy and data science competencies to integrate nursing expertise into data interpretation. This will empower nurses to play an active role in wearable-driven health care transformation, rather than being passive end users of monitoring technology.

The observed lack of transparency highlights the need for reporting guidelines specific to clinical research using wearables. Our findings indicate low reporting rates on fundamental aspects, such as device grade (12/29, 41%) and regulatory status (10/29, 34%). While wearable data quality is crucial for appropriate clinical interpretation [87], many publications did not report data accuracy (11/29, 38%) and missing data (8/26, 31%). To improve rigor, the device grade, regulatory status, and clinical validation should be standard reporting elements. For feasibility and pilot studies, data transmission, latency, and loss rates could be taken into assessment and reporting. In predictive modeling research, reporting missing sensor data and any imputation methods is crucial. Data privacy and clinical accessibility could also be incorporated into reporting guidelines.

This review has several limitations. First, we focused specifically on the use of wearables for monitoring body temperature in fever or infection management. The assessment of maturity is limited to this specific area, and maturity stages may differ in other fields. Second, we did not assess the quality or risk of bias of the included studies, which means our findings are based on all available evidence, regardless of its methodological rigor. Third, we only included studies published in English or Chinese, which may have excluded evidence published in other languages. Additionally, only a random 20% of the records were screened and extracted independently by 2 reviewers. While this approach is methodologically acceptable [88], it remains a potential source of bias. Furthermore, we identified the roles of nurses from the main texts, author, and affiliation information. However, confirming the presence of nurses among the authors was not always possible due to the lack of explicitly stated nursing qualifications.

This scoping review summarized ways of using temperature data from wearables across diverse clinical contexts, showing promise in optimizing fever or infection management in acute care hospitals, particularly when integrated with advanced data analytic approaches. However, the overall integration of wearables into clinical practice remains at an early stage of maturity. Key issues include data quality challenges, clinical interpretation of skin surface temperature, and interoperability between wearable systems and EHRs. Future efforts should prioritize standardized data cleaning frameworks and EHR interoperability, and facilitate clinical interpretation of wearable data to promote its active use in decision-making. These efforts will help to unlock the full potential of wearables in driving health care transformation.