The scoping review was conducted following the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) guidelines [17].

To ensure a comprehensive identification of relevant literature, we developed a systematic search strategy in accordance with the PRISMA-S (Preferred Reporting Items for Systematic Reviews and Meta-Analyses and search extension) guidelines [18]. We expanded our search beyond standard biomedical databases to include engineering and technology-focused repositories to capture sensor-based research. The electronic databases searched included PubMed, Embase, Web of Science, and IEEE Xplore.

The search strategy was designed to capture the intersection of 4 key concepts. First, we defined the population, specifically targeting patients with knee osteoarthritis. Second, we identified the target symptoms, which included pain, stiffness, fatigue, and ROM. Third, we established the monitoring context, focusing on daily, continuous, or ambulatory settings. Finally, we specified the technology, encompassing wearable sensors, mobile apps, and physiological signals.

The search string used a combination of medical subject headings and free-text keywords, including specific sensor modalities (eg, “accelerometry,” “photoplethysmography,” and “electromyography”) and digital tools (eg, “smartphone applications”), as detailed in the Introduction section. The search was updated and re-run on February 5, 2026. To ensure the review focused on contemporary sensor technologies relevant to current clinical practice, the search was limited to studies published in the last 5 years (from January 1, 2019, to the search date). This timeframe was selected to align with the rapid proliferation of wearable technology and smartphone integration in health care. The full search terms are reported in Multimedia Appendix 1 and were adapted for the specific requirements of each database.

To fully adhere to the PRISMA-S reporting guidelines, we explicitly state the following methodological details regarding our search process: the search strategy was developed collaboratively by the research team but was not formally peer-reviewed by an independent medical librarian (PRESS [Peer Review of Electronic Search Strategies]). No specific methodological search filters (eg, study design filters) were applied, nor were the search strategies adapted from prior literature reviews. Additionally, our search was restricted to the above databases; we did not search clinical trial registries, browse unindexed online resources, or contact authors and manufacturers for additional data, as our focus was strictly on peer-reviewed published evidence. To address potential gaps in electronic indexing and maximize coverage, we also performed a manual hand-search of the reference lists of included studies and relevant systematic reviews to identify additional eligible articles. Furthermore, regarding other search specifics, all databases were searched independently on their respective platforms; no simultaneous multidatabase searching via a single platform was performed. No additional information sources or search methods beyond those explicitly described were used. Additionally, no formal update search protocols or automated email alerts were implemented throughout the review process.

All identified citations were collated and uploaded into EndNote (Clarivate Analytics), and duplicates were removed. Specifically, the deduplication process involved using EndNote’s automated “Find Duplicates” function, followed by a manual verification by the reviewers to ensure accuracy. To minimize selection bias and ensure rigor, the screening process was conducted in two distinct stages. In the first stage, which involved title and abstract screening, 2 independent reviewers (BC and YW) evaluated the records against the predefined inclusion criteria. In the second stage, representing the full-text review, potentially eligible articles were retrieved and assessed in detail for final inclusion. Any disagreements between the reviewers at either stage were resolved through discussion or, if necessary, consultation with a third reviewer to reach a consensus.

Studies were selected based on the following inclusion and exclusion criteria, explicitly designed to identify technologies suitable for the objective, ecological monitoring of symptom-related variables. For the population, studies needed to focus on adults diagnosed with knee osteoarthritis. Regarding technology and feasibility, the study had to use noninvasive wearable sensors (eg, IMUs and smartwatches) or mobile health technologies (eg, smartphone apps) that are portable and capable of unsupervised data acquisition without requiring constant health care professional involvement, thereby ensuring applicability for free-living monitoring. Conceptually, the study needed to investigate objective biomechanical or physiological digital markers and explicitly assess their association with subjective symptoms. In terms of context, we included studies conducted in free-living environments (home or community) or laboratory settings, provided that the laboratory studies simulated daily living activities (eg, level walking and sit-to-stand tasks) to validate the sensor’s potential for real-world application. Finally, eligible study designs were restricted to original quantitative or qualitative research published in English.

Conversely, studies were excluded based on several criteria. Regarding the population, studies were excluded if participants had undergone total knee arthroplasty, ensuring our focus remained on monitoring natural disease progression. We also excluded studies based on technological constraints; specifically, if the monitoring method relied on stationary clinical infrastructure (eg, optoelectronic motion capture systems and nonportable force plates) that precludes ecological integration into free-living environments, or if it relied solely on manual clinical assessment tools (eg, hand-held dynamometers, visual inspection scales, or manual goniometry performed by a therapist), as these do not represent automated, continuous monitoring technologies. Furthermore, regarding intervention focus, studies were excluded if their primary aim was to evaluate therapeutic interventions, such as rehabilitation programs, pharmacological clinical trials, acupuncture, or radiological diagnostics, unless sensor-derived symptom markers were explicitly investigated as a primary outcome. Finally, publication types such as literature reviews, study protocols, editorials, or conference abstracts without available full texts were excluded.

Data extraction was conducted using a standardized data charting form developed specifically for this review. To ensure consistency and rigor, the form was pilot-tested on a random sample of 5 studies. Data were extracted by one reviewer (BC) and verified by a second reviewer (YW) for accuracy. Any discrepancies were resolved through discussion.

We extracted data at two levels: study characteristics and specific monitoring variables. Specifically, the extracted study characteristics included the author, year, country, study design, setting (laboratory vs free-living), and participant demographics such as sample size, age, sex, and BMI. Regarding the specific monitoring variables, we systematically categorized the data into 3 main areas. First, for monitoring technology, we recorded the sensor type (eg, IMU, electromyography [EMG], and photoplethysmography [PPG]), body placement (eg, lumbar, shank, and wrist), sampling frequency, and duration of monitoring. Second, for symptom assessment, we extracted the method of symptom quantification (eg, VAS, WOMAC, and NRS), the specific symptom domains measured (such as pain, stiffness, and fatigue), and the frequency of data collection (eg, momentary, daily, or weekly). Finally, we documented the key findings, focusing on the reported statistical associations between the objective sensor-derived variables and the subjective symptom scores.

As this is a scoping review aiming to map available literature rather than assess efficacy, a formal risk of bias assessment (eg, Cochrane ROB) was not performed. However, to address the rigor of the included evidence, we classified studies based on their design (eg, proof-of-concept, validation study, and longitudinal monitoring) and setting (controlled laboratory vs unsupervised free-living environment). This classification allows for a nuanced understanding of the current maturity of sensor-based symptom monitoring.

In accordance with JBI (Joanna Briggs Institute) guidelines for scoping reviews, the extracted data were analyzed descriptively. We used simple frequency counts to summarize study characteristics, sensor modalities, and targeted symptom domains. Furthermore, a basic qualitative content analysis was applied to map and categorize the findings into distinct physiological and biomechanical domains, identifying overarching trends and evidence gaps. We synthesized the results narratively, grouping studies based on the primary physiological or biomechanical domain targeted by the sensor. To align with the extracted data, the findings were categorized into five functional domains. First, we identified biomechanical and functional changes, which encompass gait kinematics (eg, speed and stride length), joint angles, and functional movement patterns measured by motion sensors. Second, the domain of neuromuscular control and postural stability includes muscle activity patterns (eg, surface electromyography [sEMG]), postural sway metrics, and balance parameters that serve as proxies for pain-induced inhibition or instability. Third, we grouped physiological markers, capturing autonomic responses such as heart rate variability (HRV), skin temperature, and electrodermal activity, used to infer pain-related stress. Fourth, physical activity patterns were categorized to include gross activity levels, step counts, and energy expenditure reflecting functional capacity. Finally, the fifth domain addresses sleep and circadian interactions, detailing actigraphy-derived sleep quality metrics and their temporal relationship with symptom fluctuations.

Furthermore, to visualize the extent and nature of the available literature, we constructed an Evidence Gap Map. This synthesis matrix cross-referenced the identified sensor modalities against the five clinical symptom domains, enabling the systematic identification of areas with technological saturation versus those with critical evidence gaps.

The search strategy and selection process were conducted in accordance with the PRISMA-S [18]. Figure 1 illustrates the PRISMA flow diagram. The comprehensive electronic search across multiple databases yielded a total of 3085 records. Following the removal of duplicates, 2452 unique records remained for title and abstract screening. Of these, 2396 records were excluded as they did not meet the primary inclusion criteria based on the title and abstract review.

Consequently, 56 full-text articles were retrieved and assessed for eligibility. At this stage, 40 studies were excluded. To strictly follow the feasibility requirements for free living monitoring, studies relying solely on nonportable, stationary laboratory equipment (eg, force plates and isokinetic dynamometers) or those lacking direct correlation between sensor data and clinical symptoms were excluded. Finally, a total of 16 studies [19-34] met all inclusion criteria and were included in this scoping review.

This scoping review synthesized the current evidence on sensor technologies for monitoring knee osteoarthritis symptoms in free-living settings. Relative to our initial objective of evaluating the feasibility and validity of these technologies, our primary finding—visualized in the evidence gap map (Figure 3)—is a distinct technological imbalance. Research is heavily saturated with biomechanical studies using IMUs to quantify gait and movement deficits [20,21,23,26,27,30,32]. In contrast, there is a notable lack of investigations using physiological sensors (eg, ECG and bioimpedance) to monitor systemic symptoms such as fatigue, stress, and sleep quality [28,29,31,32]. While biomechanical markers have demonstrated robust discrimination between patients with osteoarthritis and healthy controls [19], the link between physiological sensor data and patient-reported symptom fluctuations remains an emerging field requiring further validation.

While the primary focus of this review is sensor-based monitoring, discussing PROMs is methodologically and clinically unavoidable. At present, no noninvasive sensor can directly quantify a subjective experience (such as pain or fatigue) without patient input. Therefore, PROMs remain the indispensable ground truth against which all objective sensor metrics must be validated [3,4]. However, our review reveals two critical limitations of relying solely on subjective reports. First, subjective perception may diverge from physiological reality. A prime example from our review is the observation in sleep monitoring: while subjective sleep quality strongly predicted morning pain, objective actigraphy metrics (eg, sleep duration) did not [32]. This discrepancy is clinically revealing: it suggests that for some patients, the perception of poor sleep (distress) drives pain more than the actual physiological lack of sleep. Without objective sensor data to contrast with the subjective report, a clinician might incorrectly prescribe sedatives for insomnia, whereas the sensor data would reveal that the patient is sleeping but unrefreshed.

Furthermore, historically, pain was the primary outcome for knee osteoarthritis symptoms management [1], which can also be seen from Table 1, with the vast majority of included studies focusing on pain as the major symptom outcome. However, our review highlights that fatigue, stiffness, and sleep disturbances are equally debilitating contributors to disability [4]. Subjective recall of these fluctuating symptoms may be biased and varies between patients [8]. Sensors offer a unique advantage here: they can continuously quantify symptom drivers, such as micro-arousals in sleep (Actigraphy) [32] or elevated muscle coactivation (sEMG) [24,25,34], before they manifest as conscious pain. This evidence from our review demonstrates that objective sensor metrics and subjective reports provide different and additive information. Therefore, rather than viewing sensors as a complete replacement for PROMs, we advocate for a necessary hybrid approach. In this paradigm, continuous passive monitoring via sensors can successfully bridge the temporal gaps between clinical visits, significantly reducing the patient’s active reporting burden (eg, the need for highly frequent questionnaire inputs), while periodic PROMs continue to capture the essential subjective ground truth. Indeed, several studies included in this review have already laid the groundwork for this by successfully pairing sensor data with standard subjective questionnaires (eg, WOMAC, VAS, and KOOS) [19,25,30]. Building upon this existing foundation, future research should standardize and advance toward a dynamic hybrid monitoring paradigm, where subjective and objective measures are collected simultaneously and continuously, creating a personalized symptom fingerprint that neither method could achieve in isolation.

We acknowledge that high-precision laboratory instruments, such as force platforms and isokinetic dynamometers, remain the gold standard for biomechanical assessment [11]. However, their stationary nature limits them to capturing only a snapshot of the patient’s capacity in a controlled, artificial environment. This highlights the unique advantage of sensor-based monitoring: it prioritizes ecological validity to capture continuous, real-world data (eg, how a patient walks when tired, or how sleep quality affects morning stiffness) [10]. For the specific purpose of symptom monitoring, wearable sensors offer a vital complementary approach rather than a standalone solution. It is crucial to recognize that knee osteoarthritis symptoms are inherently multifactorial and driven by complex biopsychosocial mechanisms [35].

However, it is crucial to emphasize that the relationship between gross physical activity and multidimensional osteoarthritis symptoms is not strictly linear or inherently coherent. For instance, recent evidence demonstrates that in clinical knee osteoarthritis, the total volume of physical activity is often not associated with knee pain, functional decline, or health-related quality of life [36]. This lack of a direct correlation highlights a critical conceptual boundary: simple, isolated objective metrics, such as daily step counts, cannot serve as standalone proxies for the complex, biopsychosocial spectrum of osteoarthritis symptoms [26]. Instead, the true value of wearable technology lies in capturing context-dependent and qualitative markers [23,31], prioritizing ecological validity over the snapshot precision of stationary laboratory instruments like force platforms.

Consequently, wearable devices are not expected to perfectly correlate with or entirely replace subjective patient-reported outcomes. Instead, their expected validity lies in capturing the objective behavior and physiological correlates of these symptoms. By providing this continuous, objective dimension, wearables are expected to yield partial but clinically meaningful correlations with subjective experiences [26]. Future research should thus focus on using lab standards to validate these portable metrics, ensuring that this hybrid approach—integrating objective sensor data with subjective patient reports—maintains comprehensive clinical reliability [37].

Beyond biomechanical assessments, physiological sensors offer a vital systemic view of the disease burden. However, the specific pathophysiological pathways linking systemic physiological signals, such as HRV derived from ECG, to localized joint disease remain to be fully characterized [28]. Growing evidence suggests that knee osteoarthritis is associated with broader systemic responses, particularly regarding autonomic nervous system dysregulation and central sensitization [28,29]. Thus, physiological sensors may offer valuable insights into the systemic burden of pain, rather than the structural pathology of the joint itself. Therefore, wearable ECG and bioimpedance sensors are not intended to monitor the joint structure itself, but rather the physiological cost of pain—the systemic stress and inflammatory burden that patients experience [28,29]. By capturing these objective physiological markers, we can potentially identify patients who may benefit more from stress-management interventions than mechanical offloading [38].

However, it is important to note that monitoring itself does not constitute an intervention. The transition from passive data collection to active symptom management requires a combination of different technologies and strategies. Specifically, future research should explore integrating these continuous monitoring paradigms with just-in-time adaptive interventions in mobile health [39]. In this model, real-time objective sensor data and subjective reports act as tailored triggers to deliver personalized, timely behavioral support and stress-management interventions, thereby creating a true active feedback loop.

This scoping review provides a comprehensive synthesis of the current state of sensor-based monitoring technologies, categorized by their relevance to specific clinical symptoms. A key strength is the multidimensional approach, moving beyond pain to include fatigue and physiological stress. However, limitations exist. First, the heterogeneity of sensor modalities and study designs precluded a meta-analysis of sensor-symptom correlations. Second, most included studies were observational, limiting our ability to infer causality between sensor metrics and symptom changes. Third, we excluded non-English studies, which may have omitted relevant technological advancements from other regions. Finally, many free-living studies relied on short monitoring durations (eg,<7 d), which may not fully capture the long-term fluctuating nature of knee osteoarthritis symptoms.

In conclusion, this scoping review synthesizes the current landscape of sensor-based monitoring for knee osteoarthritis by mapping technologies across a multidimensional symptom spectrum, including fatigue, sleep quality, and physiological stress. By constructing an evidence gap map, we identified a distinct saturation in biomechanical gait analysis alongside a critical gap in physiological monitoring. Crucially, the observed discrepancies between subjective and objective data underscore why neither assessment method is sufficient in isolation. The primary real-world implication of these findings is the necessity of a hybrid monitoring paradigm. For sensor technology to successfully translate into daily patient care, it must integrate objective ecological data with patient-reported outcomes to enable digital phenotyping—distinguishing patients driven by mechanical instability from those driven by systemic stress. Furthermore, while the current literature primarily consists of observational studies that successfully fulfill their aims of passive data collection, a vital future direction is the development of adaptive, event-based sampling, where sensor-detected anomalies trigger real-time subjective assessments. Ultimately, this hybrid approach will facilitate a shift from merely tracking disease progression toward supporting just-in-time adaptive interventions. By combining continuous monitoring with distinct interventional strategies, future systems can provide active, closed-loop feedback, thereby empowering patients to self-manage the complex and fluctuating nature of their condition.