Psychometric Characteristics of Smartphone-Based Gait Analysis in Chronic Health Conditions: A Systematic Review

1. Introduction

Chronic health conditions significantly contribute to reduced mobility and diminished quality of life worldwide, placing a considerable burden on healthcare systems (1, 2). These conditions, prevalent across all age groups, often manifest in gait disturbances, which not only impair daily functioning but also serve as key indicators of overall health and disease progression (3). Gait disorders are linked to adverse health outcomes, including falls, fractures, loss of independence, and mortality. (4–7). For example, studies have shown that slower gait speed is a predictor of frailty and higher mortality risk, while abnormalities in cadence and step length can indicate underlying neuromuscular or cardiovascular issues (6, 7). In this context, a systematic review and meta-analysis revealed that individuals with persistent low back pain exhibit a slower walking speed and a shorter stride length compared to healthy controls without back pain (8). Therefore, precise gait analysis is essential not only for understanding the severity and progression of chronic diseases but also for enabling early intervention to prevent falls and related complications.

Traditionally, gait assessment has relied on sophisticated and expensive technologies, including motion capture systems and force plates, typically available only in specialized clinical or research settings (9). Additionally, these systems require trained personnel and extensive infrastructure, limiting their accessibility and feasibility for widespread clinical use. Moreover, these methods are time-consuming and often impractical for monitoring patients in real-life or home-based environments (10, 11). As such, the need for alternative solutions has become increasingly evident, particularly during global health crises such as the COVID-19 pandemic, which underscored the importance of remote monitoring technologies.

Recent advances in mobile technology have highlighted smartphones as promising tools for gait analysis (12–14). With embedded sensors such as accelerometers and gyroscopes, smartphones offer a cost-effective, portable, and accessible solution for capturing motion data (15, 16). These devices enable both real-time and long-term monitoring of gait parameters in natural environments, providing insights that are often more representative of a patient’s functional abilities (17). Studies in populations such as individuals with Parkinson’s disease, Multiple Sclerosis and/or healthy older adults have demonstrated the utility of smartphone-based gait analysis (18–24).

Earlier systematic reviews have focused on specific populations, such as individuals with Parkinson’s disease or Multiple Sclerosis (25, 12). The results showed that smartphone applications have great potential to validly assess gait and balance impairments in individuals with Multiple Sclerosis and Parkinson's disease, but that further studies are needed to comprehensively evaluate their reliability and sensitivity (25, 12). However, to the best of the authors’ knowledge, no comprehensive review has yet synthesized the psychometric properties of smartphone-based gait analysis across a broad range of health conditions and age groups. Addressing this gap isessential, as it may offer valuable insights into the potential generalisability of smartphone-based gait analysis across a wide spectrum of health conditions and age groups. Therefore, this systematic review aims to critically evaluate the validity, reliability, and sensitivity of smartphone-based gait analysis in young and older adults with different chronic conditions. By synthesizing the existing evidence, this review seeks to assess the clinical potential of smartphone-based inertial measurement units (IMU) for gait monitoring, identify limitations in current research, and propose directions for future studies.

2. Methods

The present systematic review was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (26). Additionally, this study has been registered on the Open Science Framework (OSF) at https://doi.org/10.17605/OSF.IO/QFU8D. Search strategy

A comprehensive literature search was conducted for studies published between 2014 and 2024, utilizing the electronic databases: Web of Science, PubMed, Google Scholar and SportDiscus. Only peer-reviewed studies written in English were included. The timeframe chosen reflects an important consideration: prior to 2014, the integration of smartphones and related technologies into daily life was less widespread, limiting their use in general and specifically in gait analysis. Keywords were collected through expert opinion, literature review, and controlled vocabulary (e.g., Medical Subject Headings [MeSH]). The search was carried out using the Boolean operators “AND” and “OR”. In the following our search strategy applied with PubMed: ("Gait" [MeSH] OR "Gait Analysis" OR "Walking" OR "Locomotion") AND ("Mobile Applications" OR "Smartphones" OR "Wearable Devices" OR "Mobile Health") AND ("Pathological Conditions, Anatomical" [MeSH] OR "Neurological Diseases" OR "Musculoskeletal Diseases" OR "Chronic Diseases" OR "Parkinson Disease" OR "Stroke" OR "Multiple Sclerosis" OR "Arthritis" OR "COPD" OR "low back pain"). Additional manual searches of reference lists were performed. Two review authors (TB and CF) conducted the study selection process independently, who screened the titles and abstracts of all identified records. Each author compiled a list of potentially eligible studies based on the inclusion criteria. These lists were then compared and discussed to reach a consensus. In cases of discrepancy, a third author (HC) was consulted to ensure an unbiased and rigorous selection process.

Inclusion and Exclusion Criteria

The inclusion criteria for eligible studies were elaborated based on the PICOS (Population, Intervention, Comparison, Outcome, Study Design) approach. The following criteria were defined: (1) Population: young and old adult participants (age ≥ 18 years) with chronic health conditions (e.g., Parkinson’s disease, Multiple Sclerosis, musculoskeletal, cardiovascular and metabolic diseases) (2) Intervention: smartphone-based gait analysis using embedded IMUs, (3) Comparison: gold-standard methods for gait analysis (e.g., motion capture systems, force plates) or other validated assessment tools (e.g., trundle wheel or XSens), (4) Outcome: validity, reliability, sensitivity, and feasibility data (5) Study design: observational studies, cross-sectional studies, cohort studies or validation studies.

Studies were excluded if they focused solely on healthy participants, involved individuals under 18 years of age, were not in English, and lacked assessments of reliability, validity, sensitivity, and feasibility. The focus on adults (age ≥ 18 years) was chosen to avoid variability arising from maturation changes in younger populations, which are better suited to separate investigations.

Data Extraction and Analysis

From the included studies, detailed data were extracted, including participant demographics such as sample size, age, sex, and underlying chronic conditions. Information about the study design and context, including whether the study was cross-sectional, observational, or a prospective cohort study, was also recorded. Data were systematically gathered on the devices and applications used, such as the smartphone models, application names, intended users, test location and sensor placement (e.g., sternum or pocket). Key gait parameters, including velocity, step length, and cadence, were noted alongside the gold-standard systems used for validation, such as motion analysis platforms or pressure-sensitive mats. The primary outcomes focused on test–retest reliability, typically reported as intraclass correlation coefficients (ICCs) or typical error of measurement (TEM), and validity metrics, often expressed through correlation or agreement with gold standards. Furthermore, sensitivity, specificity, feasibility, main findings, intended users, and limitations of the studies were extracted. The results of the included studies were qualitatively synthesized. Descriptive analyses were performed to compare reliability and validity outcomes across studies, while variations in device placement and methodologies were highlighted as potential sources of heterogeneity.

Methodological Quality Appraisal

The methodological quality of the included studies was assessed using a modified version of the "Quality Assessment Tool for Observational Cohort and Cross-Sectional Studies" originally developed by the National Institute of Health (27). This tool initially comprises 14 criteria, each rated as "Yes" or "No". The quality of the studies was classified using clearly defined thresholds: studies with a "Yes" response for at least 8 criteria were categorized as "good quality," those with 5 to 7 positive assessments were considered "fair quality," and studies with fewer than 5 positive ratings were classified as "low quality” (28).

3. Results

Study Selection

The study selection process and its outcomes are displayed in Figure 1. The search initially identified a total of 946 studies. After removing duplicates, 867 unique studies remained for further evaluation. Screening the titles and abstracts led to the exclusion of 746 studies. Consequently, the full texts of the remaining 121 studies were reviewed. Of these, 45 studies were excluded for not assessing the reliability or validity of gait measurements, 15 studies were excluded due to their focus on healthy participants, and 7 studies were excluded because they involved participants younger than 18 years. Finally, our systematic literature search resulted in 54 studies eligible for inclusion.

Study Analysis

The results from a total of 54 studies investigating smartphone-based applications for gait analysis are displayed in Table 1. The number of participants varied significantly across studies, ranging from 10 (29) to 1414 participants (30). Overall, 70% (38 studies) of the included studies focused on participants with specific health conditions such as Parkinson's disease (31) or Multiple Sclerosis (32). The remaining 30% (16 studies) analysed either neurological conditions (33) or patients with other conditions like rheumatoid arthritis (34). The average age of participants ranged from 21 years (35) to 73 years (36). Around 29 studies (54%) focused on middle-aged adults (40 to 65 years). Of the 54 studies, 15 studies (28%) were cross-sectional, 11 studies (20%) observational, five (9%) prospective, four validation (7%) studies while 6% (3 studies) used longitudinal designs.

Additionally, 52% (28 studies) (37–43, 31, 44–57, 35, 36, 58–61) of the included studies conducted measurements in clinical settings, 37% (20 studies) (37, 62, 38, 63, 29, 64, 40, 65–69, 20, 70, 71, 30, 53, 72–74) collected data in home environments, and 7% (4 studies) (38–40, 53) combined both clinical and home settings. Parkinson's disease was the most commonly studied health condition, with 46% (25 studies) (62, 37, 63, 29, 41, 65, 31, 45, 68, 75, 20, 69, 70, 48, 71, 30, 53, 52, 76, 72, 56, 77, 36, 60, 61), Multiple Sclerosis was investigated in 10 studies (19%) (38, 39, 64, 32, 66, 67, 78, 55, 73, 74), stroke was the focus of three studies (6%) (44, 46, 54), rheumatoid arthritis was addressed in two studies (4%) (34, 59), as well as pulmonary diseases (4%) (47, 79) and only one study focused on chronic low back pain patients (2%) (80).

The range of smartphone models used in the studies was diverse. 14 studies (26%) employed iPhone models, including older devices like the iPhone 4s (40) and newer models like the iPhone 13 (58). Eight studies (15%) used specialized apps like the mPower app (37, 71), while 16 studies (30%) relied on standard apps such as Google Fit or proprietary applications, as seen in the work from Polese et al. (54). Of the 54 studies, 32 (62%) placed the smartphone at the hip or in the pocket, such as in the works by Adams et al. (37) and Balto et al. (39). Six studies (11%), including e.g., Regev et al. (55), positioned the devices on the chest, while fewer than 5% of included studies (2 studies) examined alternative positions, such as the thigh (43) or the sacrum (41).

Gait speed was measured in 47 studies (87%), including works by Tang et al. (36) and Adams et al. (37). Step length and stride length were a focus in 32 studies (59%), such as Capecci et al. (41) and Mehrang et al. (71). Gait variability and fluctuation width were analysed in 16 studies (30%), including Kim et al. (20) and Pepa et al. (53). Five studies (9%), such as Kim et al. (20), investigated specific parameters like the freezing index. Validation was addressed in 17% of the included studies (9 studies). To validate smartphone measurements, 19 studies (35%) used biomechanical analysis methods such as video analysis (e.g., (41)). Clinical scales like the Expanded Disability Status Scale were used in 20% of the studies (11 studies), like Creagh et al. (66). Wearable sensors such as actigraphs were employed in 24% of the studies (13 studies), as in the work by Brooks et al. (40).

Validation, Reliability and Feasibility Outcomes

Table 2 summarizes the psychometric properties of smartphone-based gait analysis across various pathological conditions. Among the 54 studies included in this review, 70% (38 studies) investigated the validity of smartphone-based gait analysis. These studies primarily evaluated concurrent validity by comparing smartphone-derived metrics with gold standard systems such as video analysis, standalone accelerometers, or force plates. Correlation coefficients (r) ranged from 0.42 to 0.97, highlighting variability in the strength of agreement across different conditions, populations, and measurement protocols. ICC values for agreement varied from 0.53 to 0.96. Additional validity aspects, including criterion (assessed in 28 studies, 52%) and construct validity (assessed in 6 studies, 11%), were investigated in a total of 34 studies (63%). These studies utilized metrics such as sensitivity (ranging from 43% to 98%) and specificity (ranging from 59% to 97%), primarily derived through ROC analyses to evaluate diagnostic accuracy.

Reliability was explored in 27% of studies (15 studies), focusing on test-retest reliability and, less commonly, inter- and intra-rater reliability. ICC values reported for temporal and spatiotemporal gait parameters ranged from 0.53 to 0.95, and the standard error of measurement (SEM) was documented in two studies (4%). Sensitivity and specificity were addressed in 41% (22 studies) and 35% (19 studies) of the included studies, respectively. The reported sensitivity and specificity for distinguishing pathological gait patterns from healthy controls ranged broadly, with values frequently exceeding 90% for detecting conditions such as Parkinson's disease and Freezing of Gait (FoG). However, 11 studies (20%) reported moderate accuracy, emphasizing the need for standardized methodologies and more robust algorithms. Feasibility was positively rated in 94% of the included studies (51 studies). Practical limitations such as sensor placement inconsistencies and environmental interferences (e.g., ambient noise) were noted in 12 studies (22%).

Methodological Quality of the Included Studies

The methodological quality of the included studies is summarized in Table 3. 87% of the studies (47 studies) met requirements such as clear definitions of the target population and adequate sample sizes. However, notable weaknesses were observed in the blinding of assessors and the use of standardized measurement protocols. Only a small proportion of the studies (approximately 20%) reported fully blinded outcome assessments. Statistical analyses, however, were generally considered appropriate and robust, with a variety of approaches used, including linear models (e.g., regressions or ANOVAs), ICC calculations, and ROC analysis. Consequently, the majority (96%) of the included studies (52 studies) were rated as having a “good overall methodological quality” with two studies only rated as having a “fair overall quality”.

4. Discussion

The objective of this study was to critically appraise the psychometric properties of smartphone-based gait analyses across various chronic diseases. Overall, the findings indicate that smartphone applications represent a valid, cost-effective, and accessible alternative to traditional gait analysis methods, such as motion capture systems or force plates, for assessing gait parameters across different chronic diseases, both in clinical and home environments. Additionally, the reliability of smartphone-based gait analysis can generally be rated as good, emphasizing the potential of smartphone applications to deliver consistent results. However, it is important to note that reliability has been studied less extensively than validity. Sensitivity and specificity were investigated in 37% of the included studies, with the majority demonstrating high accuracy in distinguishing pathological from healthy gait patterns. Feasibility was confirmed in 94% of the studies, underscoring the widespread acceptance and practicality of these technologies across diverse settings.

The use of smartphone applications has gained remarkable attention over the last couple of years. This is because these devices offer a unique combination of portability, cost-efficiency, and accessibility, making them suitable for both clinical assessments and home-based monitoring. Their widespread adoption has been driven by advancements in sensor technology, allowing precise and comprehensive data collection directly from everyday devices. Furthermore, the integration of machine learning algorithms has enhanced their ability to process and analyze gait-related data with high accuracy, further cementing their utility in modern healthcare practices. Among the included studies, 70% evaluated validity by comparing results with clinical or biomechanical reference systems. Brooks et al. (40) examined the concurrent validity and demonstrated a high correlation between app-based and clinical walking distance measurements (r = 0.89) in patients with congestive heart failure and pulmonary hypertension aged between 25 and 76 years. Likewise, Polese et al. (54) reported excellent agreement between Google Fit and actual step counts measured with video analyses (ICC = 0.93) in stroke survivors patients aged 62 years. Furthermore, Tang et al. (36) reported strong correlations between a smartphone and the XSens system in Freezing of Gait (FoG) detection (r = 0.86 to 0.97) in Parkinson patients aged 73 years, reflecting a good concurrent validity. Moreover, Chen et al. (65) reported using a self-designed framework for mobile phones a sensitivity of 97.3% and a specificity of 97.1% in classifying Parkinson’s severity based on gait variability. Such results underscore the excellent discriminative validity and the potential of smartphone-based apps to replace traditional gait analysis systems, particularly in resource-limited settings where access to expensive technologies like force plates or motion capture systems is restricted. Despite these positive findings, some studies highlighted some limitations. For instance, Balto et al. (39) conducted a study including patients with Multiple Sclerosis (46 ± 10 years) and found that certain apps, such as Health, Health Mate and Moves, did not exhibit significant correlations with walking speed. This highlights the importance of carefully selecting appropriate algorithms, applications, and devices. It is indeed evident that not all smartphones and apps are equally suitable (39), which can significantly influence validity and reliability outcomes. These findings highlight the versatility of specific smartphone applications in capturing gait-related parameters.

Reliability was investigated in only 27% of the included studies, with consistently positive results. This indicates that, despite its relevance, reliability was relatively neglected. Bourke et al. (64) reported ICC values ranging from 0.53 to 0.96 for spatiotemporal parameters measured with the FLOODLIGHT software on the smartphone. Creagh et al. (67) demonstrated for patients with Multiple Sclerosis aged 40 years excellent reliability of step counts with an ICC of 0.91. Similar results were reported by Serra-Ano et al. (56), who found ICC values ranging from 0.89 to 0.92 for gait analyses using the FallSkip® app in patients with Parkinson disease aged 69 years. In the same sense, Tang et al. (36) reported high reliability in FoG detection (ICC = 0.768–0.896) for persons aged 73 ± 9 years. Overall, there is evidence that smartphone-based gait analyses provide consistent results, even though data in some areas, such as inter-rater reliability (2 of 54 studies [4%]), remain limited and methodological standardization is still lacking. However, the question remains as to why so few studies have investigated reliability. This could suggest that many apps and devices are designed primarily for single measurements and have not been adequately validated for long-term use or in varying contexts. Future research should focus more on inter- and intra-rater reliability, particularly for applications intended for clinical practice or remote monitoring.

The ability of smartphones to differentiate pathological from healthy populations was examined in 37% of the included studies (20 studies). For example, Arora et al. (29) revealed sensitivity and specificity values of over 90% in distinguishing Parkinsons patients from controls. Similar findings have been reported by Tang et al. (36). Likewise, Mehrang et al. (71) and Pepa et al. (53) demonstrated moderate-to-high accuracy of smartphone algorithms in identifying Parkinson’s disease based on gait parameters, with sensitivity reaching up to 84.9% and specificity up to 95.2%. Additionally, Creagh et al. (67) demonstrated that smartphone-based models could classify gait abnormalities in patients with mild to moderate Multiple Sclerosis with high precision (88%). Such findings are particularly promising for the early detection of gait disorders in the general population. However, results varied depending on the target population and algorithms used, as illustrated by Polese et al. (54), who reported excellent agreement between the actual step count counted with video analysis and the step count from the mobile phone in stroke survivors (ICC = 0.93), in contrast to Balto et al. (39), where certain applications failed to correlate with walking speed in Multiple Sclerosis patients. This emphasizes the importance of tailoring algorithms to specific diseases and populations.

The feasibility of smartphone-based gait analyses was addressed in almost all studies (51 of 54 (94%)), with the majority highlighting their ease of use at home and in clinical settings. Of note, device placement varied considerably across the included studies, from pockets to the hip or chest. While Kim et al. (20) found that the position of the smartphone did not significantly affect FoG detection, Creagh et al. (67) and Brinkløv et al. (81) stressed that inconsistent placements could lead to significant measurement errors. Additionally, Brinkløv et al. (81) showed that placing smartphones in jackets caused higher measurement errors. This highlights the need for standardized methods. Additionally, Abujrida et al. (62) reported that noise disturbances in home environments could affect data quality. Nonetheless, there is evidence indicating high participant acceptance and satisfaction with smartphone use in home settings (70). Such positive feedback is crucial for promoting acceptance of these technologies, especially in remote monitoring. However, only 19% of the included studies examined the feasibility and acceptance of the technologies. This is a critical aspect, as patients with motor impairments such as tremor or rigidity might face challenges in using such tools. Future studies should prioritize addressing the specific needs of the target population to ensure that these technologies are both accessible and practical for all user groups. Additionally, the role of smartphones in long-term monitoring and personalized treatment remains a promising but underexplored topic.

5. Limitations and Future Research Perspectives

Certain limitations of smartphone-based gait analysis are evident and warrant further investigation. While validity was the primary psychometric feature investigated—reported in 70% of the included studies—reliability received comparatively little attention, with only 27% of studies addressing it. Among these, inter-rater reliability was particularly underrepresented, investigated in just 4% of the studies. This imbalance suggests a need for future research to adopt a more comprehensive approach to psychometric evaluation, ensuring that reliability aspects are adequately explored. In addition, a key challenge lies in the lack of standardization across studies, particularly regarding device placement, algorithm design, and measurement protocols. Inconsistent methodologies can significantly influence the validity and reliability of findings, limiting their generalizability. Additionally, user experience and satisfaction with smartphone technologies remain underexplored, especially in populations with significant mobility challenges, such as those with advanced motor impairments. Future research should address these gaps by prioritizing the development of standardized methodologies and refining algorithms tailored to specific diseases and populations. Furthermore, feasibility studies should focus on the needs of diverse user groups to ensure accessibility and practicality in both clinical and home environments. Long-term investigations into the effectiveness of smartphone-based technologies are also needed to evaluate their potential for personalized treatment and proactive management of chronic conditions. These advancements could promote the widespread adoption of smartphone-based gait analysis tools and enhance their impact on healthcare delivery.

6. Conclusions

This review underscores the growing role of smartphone applications in assessing gait parameters across various chronic conditions. Overall, the summarised evidence supports their capability to deliver valid, reliable, and sensitive measurements of gait impairments, as well as their utility in distinguishing pathological gait patterns from those of healthy individuals. These findings emphasize the potential of smartphone-based gait analysis to contribute meaningfully to clinical practice and remote monitoring. Smartphone applications have the potential to transform gait analysis by providing accessible, cost-effective, and scalable solutions for monitoring patients in their natural environments.