Gait and balance metrics comparison among different fall risk groups and principal component analysis for fall prediction in older people

Abstract

Background

Falls are a leading cause of morbidity and mortality among older adults, often linked to gait and balance impairments.

Objective

To compare gait and balance metrics across fall risk levels in community-dwelling older adults and identify principal components predictive of fall risk.

Design

Retrospective cohort study.

Setting

General community.

Subjects

Three hundred older adults were stratified into low, moderate and high fall risk groups using the STEADI toolkit.

Methods

Gait and balance metrics were compared across groups. Principal component analysis (PCA) reduced dimensionality, and binary logistic regression assessed the predictive value of components.

Results

High-risk individuals showed slower cadence, shorter step length, wider step width, greater gait variability and increased centre of pressure (CoP) and centre of mass (CoM) sway. PCA identified four gait and seven balance components, explaining 71.62% and 75.88% of variance, respectively. Logistic regression revealed Gait_principal component (PC)2 (instability) (OR = 2.545, P < .001), Gait_PC3 (rhythm control) (OR = 1.659, P = .006), Balance_PC1 (CoP sway during single-leg stance) (OR = 1.628, P = .007), Balance_PC2 (CoM sway velocity variability) (OR = 1.450, P = .032) and Balance_PC4 (CoP sway during double-leg stance, eyes closed) (OR = 1.616, P = .004) as significant predictors. The model achieved 77.2% accuracy, with a sensitivity of 73.1% and a specificity of 79.4%.

Conclusions

Gait instability, rhythm control and increased postural sway are key predictors of fall risk. Integrating gait and balance metrics enhances fall risk stratification, supporting clinical decision-making.

Graphical Abstract

Open in new tabDownload slide

Key Points

• Principal component analysis (PCA) distils gait and balance parameters into key components for fall risk prediction.

• Components like gait instability, rhythm control, centre of pressure and centre of mass sway show distinct effects on fall risk.

• PCA integrating gait and balance factors improves fall risk classification accuracy.

Introduction

Falls among older adults represent a significant public health concern, contributing to morbidity, reduced quality of life and premature mortality [1, 2]. As the global population ages, fall incidence is projected to rise, putting greater pressure on healthcare systems [3]. Approximately one-third of individuals over 65 experience at least one fall annually, with risks increasing due to age or functional impairments [4]. Early identification and preventive measures are crucial for mitigating these outcomes.

Fall risk is often treated as a binary variable, classifying individuals as ‘fallers’ or ‘non-fallers’ [1, 5]. While this dichotomous classification fails to capture the nuanced, multicategory nature of fall risk [6], The STEADI toolkit provides a structured, multicategory fall risk assessment, categorising individuals into low-, moderate- or high-risk groups based on fall history and balance measures [7]. Research shows that those at moderate or high risk are more likely to experience falls compared to low-risk individuals [8, 9]. Importantly, while the STEADI toolkit does not directly measure fall incidence, its comprehensive inclusion of fall history and functional tests allows it to effectively predict fall risk, serving as a reliable surrogate marker for identifying individuals at greater risk for falls.

Falls result from complex interactions among biological, behavioural, environmental and socioeconomic factors, with gait and balance control as key biological determinants [4]. Gait temporal–spatial parameters (e.g. step length, stride velocity, variability and symmetry), provide insights into walking stability and coordination [10, 11]. Similarly, balance metrics, including postural sway displacement/area/velocity, offer objective measures of an individual’s ability to maintain equilibrium [12].

Traditionally, gait and balance control assessments relied on single parameters [1, 5] such as walking speed below 0.8 m/s, which can be considered an independent risk factor for falls and fall-related injuries [1, 10]. Although convenient, these assessments overlook the multifaceted nature of fall risk [6] and fail to capture broader aspects of mobility and postural control [5, 13]. Therefore, a comprehensive analysis integrating gait and balance metrics is needed to better understanding their combined impact on fall risk.

Additionally, traditional fall risk prediction utilises logistic regression paradigm, selecting specific variables as predictors. However, gait and balance parameters encompass various dimensions, including variability, complexity, intensity, symmetry and magnitude [14]. These datasets are often large, nonlinear and highly correlated, posing challenges in selecting feature variables that reflected fall risk during subsequent regression prediction [1]. In this context, principal component analysis (PCA) offers a solution by reducing dimensionality while retaining key variance, facilitating the identification of meaningful predictors. PCA can uncover latent structures within gait and balance data, enhancing model performance in stratifying fall risk [15, 16].

This study aimed to compare gait and balance parameters among older adults classified into low, moderate and high fall risk groups. Additionally, it sought to identify principal components (PCs) associated with fall risk, providing insights into their predictive roles. We hypothesised that [1] gait and balance parameters would differ significantly among fall risk groups [2]. PCs derived from these metrics would reliably predict fall risk, improving stratification accuracy in community-dwelling older adults.

Methods

Participants

Power analysis (G*Power, effect size = 0.25, α = 0.05, power = 0.80) determined a required sample of 85 per group, 300 community-dwelling older adults were recruited finally. Participants were 65 or older, able to walk independently and had normal cognitive function (MMSE ≥24) [17]. Exclusion criteria included significant postural control impairments such as stroke, severe psychiatric disorders (schizophrenia, bipolar disorder, etc.), peripheral neuropathy or serious cardiopulmonary (congestive heart failure, myocardial infarction, etc.)/musculoskeletal conditions (lower limb amputations, etc.) The participants with reduced visual acuity or significant eye pathologies, as well as those diagnosed with conditions known to increase fall risk (such as Parkinson’s disease), were also excluded. The study was approved by the Human Research Ethics Committee of Shanghai University of Sport (No. 102772023RT132).

Fall risk stratification

According to the modified STEADI toolkit [8], older adults are classified into low, medium and high fall risk categories during the preliminary screening phase. Initially, participants are asked if they have: (i) fallen in the past year; (ii) fear of falling; or (iii) feel unsteady while standing or walking. Those who answer ‘no’ to all are considered low risk. A ‘yes’ response triggers further assessment, including the four-stage balance test. Those completing all positions for over 10 seconds proceed to the next stage. This screening differs from our main study, which analyzes COP and COM trajectory changes on a force platform. Participants unable to complete these tests are further evaluated: those reporting a single, noninjurious fall in the past year are classified as medium risk, while those with multiple falls or a hip fracture are deemed high risk.

Data collection and analysis

This study was conducted at a single centre (Shanghai University of Sport), with all data collected from Oct 2023 to Jun 2024. Demographic information was gathered for each participant, including age, sex, height and mass. Gait was assessed using the Zeno Walkway (Protokinetics, USA), a 628 cm × 60.96 cm pressure-sensitive mat with a 120 Hz sampling frequency. Participants walked back and forth at a self-selected comfortable speed, with each trial collecting at least 40 valid steps, repeated three times with 2-minute rests. Spatiotemporal gait parameters were analysed using PKMAS® software.

Balance was assessed with progressively challenging standing tasks under eyes-open (EO) and eyes-closed (EC): double-leg stance (DS), tandem stance (TS), left-leg stance (LS) and right-leg stance (RS), each held for 20 seconds. If a participant could not complete the task, the duration was recorded; those unable to sustain the position for at least 2 seconds were classified as unable to complete the task and were excluded from the final data analysis. Each task was repeated three times with a 30-second rest between trials, with additional rest allowed if needed. CoP and CoM trajectories were captured using the SMART EquiTest Balance Manager System (Natus Medical, USA) at 100 Hz. See Appendices 1 and 2 for details of gait spatiotemporal parameters and balance parameters.

For the Timed Up and Go test (TUGT), participants stood from a stool, walked 3 m forward, turned around, walked back and sat down, with the time recorded in seconds. Three trials were conducted with 30-second rests.

Statistical analysis

Descriptive statistics summarised parameter characteristics, calculating means and standard deviations for normally distributed data and medians with interquartile ranges for non-normally distributed data. One-way analysis of variance (ANOVA) compared gait and balance parameters across fall risk groups, with post hoc Tukey corrections. For non-normally distributed data, the Kruskal–Wallis H test was used, followed by Mann–Whitney U tests for pairwise comparisons. PCA retained components with eigenvalues >1, explaining sufficient variance. Logistic regression identified significant predictors related to fall risk categories. Model fit was assessed using the Hosmer–Lemeshow test, and performance was evaluated with AUC and ROC analyses, determining sensitivity and specificity. Significance level was set at P < .05.

Results

The demographic characteristics of the participants are summarised in Table 1, the bold values indicate significant differences (P < 0.05) among different fall risk groups. No significant differences in height were observed across groups (P = .916). Significant differences were found in age, body mass, body mass index (BMI) and TUGT durations. The high fall risk group was older than the low-risk group (P = .002), had a higher body mass than the medium-risk group (P = .032) and showed higher BMI and TUGT durations than both medium (P < .001) and low-risk groups (P < .001). All participants completed the double-leg stance. For tandem and single-leg stances, we recorded the duration for those unable to maintain balance for 20 seconds. The number of noncompleters, along with the mean and standard deviation of completed durations, are shown in Appendix 3. Given the importance of fear of falling (FoF) in gait assessment, Appendix 4 compares FoF measurements across fall risk groups.

Comparison of gait and balance parameters among different fall risk groups

Significant differences were observed in gait temporal–spatial and variability parameters among different fall risk groups (Appendix 5). Temporal–spatial parameters, including cadence, step/stride length, stride width, step/stride time, stride velocity, stance time, stance/swing percentage and single support percentage, differed significantly between low-, medium- and high-risk groups (all P < .05). Pairwise comparisons showed significant differences between low- and high-risk and medium- and high-risk groups for these parameters, while left/right limb ratios showed no significant differences.

For gait variability (Appendix 6), step length coefficient of variation (CV), stride length CV, step time CV, stride velocity CV, swing time CV and single support time CV were significantly different across fall risk groups (all P < .05). Key differences were noted between low- and high-risk groups for step length CV, step time CV, stride velocity CV and single support time CV and between medium- and high-risk groups for all variability parameters except swing time CV. No significant differences were found in the left/right ratios of these parameters.

Balance metrics also differed significantly among groups (Appendices 7 and 8). The high-risk group showed longer CoP lengths, larger CoP areas and increased anterior–posterior (AP) and mediolateral (ML) displacements during all tasks. Additionally, higher average velocities in AP and ML directions indicated greater sway and instability in the high-risk group.

Principal component analysis of gait and balance parameters

Through ANOVA, 18 gait spatiotemporal parameters and 43 balance parameters with significant differences were found. PCA was conducted on 13 gait spatiotemporal parameters and 26 balance parameters after removing variables with a variance inflation factor (VIF) greater than 10, ensuring the avoidance of multicollinearity. These remaining variables were standardised using z-scores. The Kaiser–Meyer–Olkin (KMO) measure verified the sampling adequacy for both gait (KMO = 0.698) and balance (KMO = 0.811) parameters, while Bartlett’s test of sphericity confirmed the factorability of the datasets (P < .001 for both gait and balance), indicating that PCA was suitable for dimensionality reduction in this study.

Gait parameters

Based on Table 2, Figure 1 and Appendices 9 and 10, the PCA of gait parameters identified four PCs, cumulatively explaining 71.62% of the total variance, each representing key aspects of gait performance.

• Gait_PC1 (36.83%): Represents gait variability, with high loadings on stride length CV (0.828), stride velocity CV (0.783), step time CV (0.753) and step length CV (0.742). Elevated variability in these metrics indicates movement instability, increasing fall risk.

• Gait_PC2 (13.32%): Reflects gait unsteadiness, emphasising single support percent (−0.823), stride width (0.711) and step length (−0.709). Instability during the single-leg support phase compromises gait balance.

• Gait_PC3 (11.78%): Captures gait rhythm control, dominated by cadence (−0.973) and step time (0.947). Poor rhythm disrupts coordinated walking, increasing the likelihood of falls.

• Gait_PC4 (9.69%): Highlights gait phase instability, with key loadings on single support percentage CV (0.930) and stance percentage (0.617), indicating difficulties in transitioning between gait phases, which can lead to balance loss.

Balance parameters

Based on Table 2, Figure 2 and Appendices 11 and 12, PCA of the 26 balance parameters revealed seven PCs, accounting for 75.88% of the variance.

• Balance_PC1 (33.92% variance explained): primarily consists of parameters related to the left-leg stance with eyes open (LS-EO) task, displaying significant loadings including cop_length (loading = 0.883), cop_v_mean_ML (loading = 0.859), cop_area (loading = 0.831) and cop_displacement_ML (loading = 0.667). Higher values indicate poor single-leg stance stability.

• Balance_PC2 (10.85% variance explained): captures variability in CoM sway velocity during double-leg stance (ds_eo_CoM_v_SV, loading = 0.928), CoP area under EO (ds_eo_CoP_area, loading = 0.845) and EC conditions (ds_ec_CoP_area, loading = 0.616). These metrics collectively reflect postural control and the reliance on sensory inputs during balance tasks.

• Balance_PC3 (8.41% variance explained): reflects balance performance during the double-leg stance with eyes open (DS-EO) task, emphasising CoP movement across dimensions, including dseo_cop_length (loading = 0.785), dseo_cop_v_mean_ML (loading = 0.778) and dseo_cop_displacement_ML/AP (loading = 0.775/0.685). Higher values indicate greater instability, with larger sway size and faster ML velocity, suggesting poorer postural control and increased fall risk.

• Balance_PC4 (7.34% variance explained): is primarily derived from metrics related to balance performance during the double-leg stance with eyes closed (DS-EC) task, specifically focusing on mediolateral movement: ds_ec_displacement_ML (loading = 0.886), ds_ec_cop_v_mean_ML/AP (loading = 0.856/0.850). Balance_PC4 encapsulates critical aspects of balance control when visual input is removed, emphasising how individuals manage sway in the ML direction while double-leg stance.

• Balance PC5 (6.62% variance explained): emphasises cop displacement during the tandem stance with eyes open (TS-EO) task, featuring loadings such as tseo_displacement _ML/AP (loading = 0.903/0.884) and tseo_v_mean _ML (loading = 0.657). This PC demonstrates the dual balance requirements in maintaining an upright posture during more challenging standing postures.

• Balance PC6 (4.62% variance explained): focuses on cop movement amplitude during DS-EO, with principal loadings including maximum amplitude (loading = 0.749) and mean amplitude (loading = 0.712). Higher PC6 values typically signify an increase in cop movement amplitude, which may correlate with lower balance control capabilities and an increased fall risk.

• Balance PC7 (4.12% variance explained): primarily reflects displacement in the AP direction during the tandem stance with eyes closed (TS-EC) task, with a significant loading (0.698). It emphasises the importance of controlling forward and backward displacement in more challenging balance tasks.

Logistic regression results

ANOVA showed no significant differences in walking and balance indicators between low and moderate fall risk groups. Binary logistic regression, adjusted for age, BMI and gender, used 11 PCs (4 gait, 7 balance) and TUGT durations as predictors, with fall risk categorised as low/moderate (0) and high [1]. Results showed that Gait_PC2 (β = 0.934, OR = 2.545, P < .001), Gait_PC3 (β = 0.506, OR = 1.659, P = .006), Balance_PC1 (β = 0.488, OR = 1.628, P = .007), Balance_PC2 (β = 0.371, OR = 1.450, P = .032) and Balance_PC4 (β = 0.480, OR = 1.616, P = .004) increased fall risk (Table 3).

The ROC curve analysis yielded an AUC of 0.826, suggesting a strong discriminatory ability between low/medium and high fall risk groups among older people. With an optimal cutoff of 0.35, the model achieved a sensitivity of 73.1%, specificity of 79.4%, accuracy of 77.2%, precision of 64.0% and F1 score of 68.1%.

Discussion

This study examined gait and balance differences among older adults in low, medium and high fall risk groups. PCA reduced gait metrics to four PCs and balance metrics to seven PCs, explaining 71.62% and 75.88% of the variance, respectively. Logistic regression found that Gait_PC1 (gait efficiency) reduced fall risk, while Gait_PC3 (gait variability), Balance_PC1 (postural stability) and Balance_PC3 (balance variability) increased fall risk. The model demonstrated strong predictive performance (AUC = 0.826, sensitivity = 73.1%, specificity =79.4%).

Comparison of gait and balance parameters among different fall risk groups

Our findings align with previous studies showing that high-risk older adults use more conservative gait strategies and show greater instability during gait [11, 18–20] and balance tasks [21]. They exhibited slower cadence, shorter step length, wider stride width, greater variability in these parameters and larger sway displacement and area during balance tests.

Unlike studies focusing solely on spatiotemporal gait parameters [12], this research incorporated temporal–spatial parameters, variability and symmetry to provide a more comprehensive perspective on fall risk. Beyond absolute gait parameters, gait variability (e.g. step time CV, stride velocity CV), reflecting inconsistent and unstable walking rhythms, emerged as a critical factor in distinguishing fall risk levels, reinforcing the importance of fluctuations over time in fall risk assessment [22, 23].

Regarding balance metrics, the high-risk group exhibited larger CoP sway length, sway area and velocities and greater AP and ML displacements, corroborating prior studies [21, 24, 25] identifying postural sway as a key fall predictor. These features likely result from neuromuscular control changes, reduced muscle strength and diminished sensory input [26, 27].

Interpretation of principal component analysis of gait and balance parameters

The application of PCA effectively reduced data complexity while retaining key metrics for predicting fall risk, with four gait and seven balance components explaining over 75% of the variance in each domain.

Four gait components—gait variability, gait instability, rhythm control and phase instability—explained 71.62% of the variance. Gait_PC1, representing variability, accounted for the largest proportion, indicating that higher gait variability increases susceptibility to trips or missteps, elevating fall risk [20]. Gait_PC2, associated with instability during the single-leg support phase, highlights reduced stability as a key risk factor [28]. Gait_PC3, tied to rhythm control, underscores the importance of coordinated gait patterns; poor rhythm increases fall risk during sudden changes, such as turns or crowded navigation [11]. Gait_PC4 reflects instability in phase transitions between stance and swing phases, emphasising the role of consistent timing and strength in maintaining balance and reducing falls [29].

Balance components reflected performance across tasks of varying difficulty and visual conditions, capturing key aspects of postural control. PC1, explaining the largest variance, was dominated by CoP sway patterns during single-leg standing. PC2 focused on variability in CoM sway velocity during double-leg standing, highlighting regulatory strategies for balance. PC3 and PC4 were linked to balance under EO and EC conditions, while PC5 and PC7 captured performance during advanced tasks. PC6 emphasised maximum CoP sway amplitude, reflecting participants’ stability limits.

These findings underscore the multidimensional nature of balance control, such as single- and double-leg standing, variability in CoM sway and responses to visual input changes. Align with previous studies emphasising the role of sensory integration in balance maintenance [30] and the impact of visual conditions on postural strategies [31], our results further highlight the importance of assessing balance tasks under different levels of difficulty to better classify fall risk. The variability in CoM sway emerged as a critical indicator, consistent with prior research [12, 21, 32] identifying rapid and variable CoM movements as predictive of fall risk. This comprehensive approach provides actionable insights for practitioners, offering a framework for assessing and improving balance strategies in older adults.

Logistic regression analysis of gait and balance components

Logistic regression analysis revealed that gait and balance PCs are crucial for predicting fall risk in older adults. Gait_PC2 (single support percent, stride width, step length) and Gait_PC3 (cadence, step time), related to gait instability and rhythm control, significantly increased fall risk. Balance_PC1 and PC4 (postural sway and stability during single and double-leg stances) and Balance_PC2 (CoM sway velocity variability) were also significant predictors. ROC analysis showed an AUC of 0.826, indicating strong discriminatory power between low/medium and high fall risk groups. These findings highlight the importance of integrating multidimensional gait, balance and functional metrics in fall risk assessments.

Consistent with Jayakody et al. [11], who reported that fallers exhibit faster declines in pace and greater gait rhythm variability than nonfallers, our results confirm that gait variability and rhythm control are key predictors of fall risk. Similarly, Jia et al. [33] identified stride length, standing time differences and stride time as protective factors for fall risk. However, unlike studies focused solely on gait parameters, our study incorporated both gait and balance metrics, providing a more comprehensive perspective for fall risk screening in older adults.

Clinical relevance and applicability

Traditional gait and balance assessments, such as clinical observation and the Tinetti scale, are commonly used for fall risk evaluation but can be subjective and may overlook subtle gait abnormalities. Our study introduces a quantitative approach that allows for the detection of minor gait issues, improving the accuracy of fall risk identification and facilitating personalised intervention strategies. Moreover, this quantitative method demonstrated good feasibility and tolerability, with participants reporting no significant discomfort or fatigue, suggesting that it can be effectively implemented in routine clinical practice without burdening patients or healthcare providers. In summary, quantitative gait and balance assessments offer significant value in clinical settings by providing more precise fall risk evaluations and supporting individualised intervention plans, ultimately enhancing patient quality of life.

Strength and limitations

This study analyses gait and balance data from 300 older adults, revealing performance differences across fall risk levels and providing precise insights for personalised interventions. Using PCA, we reduced complex multidimensional data into key components, aiding in the identification of primary features associated with falls and enhancing the clinical relevance of our findings. However, some limitations should be noted. First, the cross-sectional design prevents causal inferences between gait, balance parameters and fall risk. Longitudinal studies are needed to confirm predictive relationships and validate these parameters for risk stratification. Second, we used binary logistic regression to classify participants into low/moderate- and high-risk groups due to the absence of significant differences between low and moderate risk levels. As fall risk is inherently a continuum, future studies with larger samples and multinomial logistic regression could offer a more detailed understanding of variations across risk categories, improving the precision of fall risk assessment.

Conclusions

This study reveals significant differences in gait and balance parameters across fall risk levels in older adults. High-risk individuals exhibited more conservative gait strategies, increased variability and greater postural sway, particularly under challenging conditions. PCA highlighted key fall risk factors, such as gait variability, rhythm control impairments and CoP/COM sway. These findings emphasise the importance of an integrated assessment approach that considers both dynamic gait and balance control, which can enhance fall risk stratification and guide targeted prevention strategies.

Declaration of Conflicts of Interest

None declared.

Declaration of Sources of Funding

This work was supported by Guangdong OPPO Mobile Telecommunications Corp Ltd. under the research project titled ‘Development of a Balance Testing System for Older Adults Based on Inertial Measurement Units and Establishment of a Fall Risk Prediction Model [grant number D5-8002-23-0020]’.

Research Data Transparency and Availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

References

Author notes